CENTRIFUGAL COMPRESSOR DESIGN AND TESTING IN THE FINNISH HIGH

SPEED TECHNOLOGY

J. Larjola, J. Backman, H. Esa, H. Pitkänen, P. Sallinen and J. Honkatukia

Lappeenranta University of Technology, Department of Energy Technology
P.O. Box 20 , 53851 Lappeenranta, Finland
E-mail: [email protected]

ABSTRACT
This paper reports on the work carried out in the NOMENCLATURE
Laboratory of Fluid Dynamics, at Lappeenranta University of e specific internal energy
Technology (LUT), which has been designing and testing u,v,w Cartesian velocity components
compressors, turbines and pumps since 1982. The Laboratory x,y,z Cartesian coordinates
has been responsible of large number of centrifugal compressor p pressure
design and testing in Finland. These compressors are usually
r
r radius
directly connected to a high speed electric motor. Speed control t time
is used instead of IGV and diffuser vane control and this sets y+ non-dimensional distance
some additional requirements e.g. for compressor performance E total internal energy
map shape. One dimensional computation, giving the basic F,G,H inviscid flux vectors
geometry and performance map of compressor, is made with a Fv,Gv,Hv viscous flux vectors
program developed by LUT. Then a three-dimensional program, Pinput input power
developed also by LUT is used to define the whole compressor Q source term
wheel geometry. The wheel geometry data is used in the flow R residual
and the structure analyses. The compressor flow is calculated S cell face area
with a three-dimensional CFD-program (Finflo), which has U solution vector
been designed by Helsinki University of Technology and V volume
r
modified for centrifugal compressor flow together with LUT. V absolute velocity
The final compressor geometry is measured in the test facility at R density
Lappeenranta University and the test results are used to develop γ ratio of specific heats
the design loop. Up to this date, around 15 high speed π pressure ratio
compressors have been aerodynamically designed and tested in Ω angular velocity
this design loop. Typical pressure ratio of the compressors
ranges from 1.6 to 2.5.

1
 compressors have been aerodynamically designed and tested in
INTRODUCTION this design loop.
In the centrifugal compressor the flow enters axially, but is
then turned radially out from the impeller. The static pressure 1. ONE-DIMENSIONAL FLOW DESIGN
rises due to the decrease of the relative velocity, but also due to In the early 80’s the first compressor design was based on
the centrifugal forces. After the impeller, the flow enters a the theory of one-dimensional flow in centrifugal compressors.
radially annular vaned or vaneless diffuser. In vaneless diffuser, The design team gathered data from literature and visited to
the flow is slowed by the free vortex. The compressor needs a other universities and was convinced that the design of a
gathering body, volute, to assemble the flow to the next stage or centrifugal compressor starts with the definition of the initial
to the outlet piping. parameters, which are used to calculate the one-dimensional
The first part of the paper deals how the design of a flow in the corresponding program. Gradually, we were able to
centrifugal compressor starts with the definition of the initial have data also from our own measurements and the team
parameters, which are used to calculate the one-dimensional developed the program design principles further.
flow in the corresponding program. This program, developed by
LUT, uses empirical data from literature and test results from
previous design to give preliminary results about the
compressor performance.
Compressors, which design is presented here, are usually
directly connected to a high speed electric motor. Speed control
is used instead of IGV and diffuser vane control and this sets
some additional requirements e.g. for compressor performance
map shape. Benefits of this High Speed Technology are
presented in details in Larjola (1986 and 1988), Larjola and
Backman (1995) and Larjola et al. (1995).
It is required, that at the design pressure ratio the
compressor has an operating range 1:2 in mass flow, when
optimal speed control is used. It is also required, that the high Fig. 1 Result window of the one-dimensional flow
efficiency range of compressor is as wide as possible. The design program
initial parameters of compressor impeller geometry and flow are The one-dimensional flow design program, developed by
used in the three-dimensional program, which is used to define LUT, uses empirical data from literature and test results from
the whole compressor wheel geometry. This program, previous design to give preliminary results about the
developed also by LUT, calculates the blades and wheel compressor performance as well as some impeller and diffuser
passages in details, which can be examined in more details to geometrical parameters. The calculation routine is sequential,
find any non-conformity and to optimize the blade thickness iterative and quick with modern personal computers.
distribution. The wheel geometry data is used in the flow and The input data includes compressor pressure ratio, inlet
the stress and vibration analyses. temperature, inlet pressure, fluid (humid air is the most
The compressor flow is calculated with three-dimensional frequently used fluid), rotational speed and mass flow. The
CFD-program (Finflo), which has been designed by Helsinki program includes variables that can later be altered: specific
University of Technology and modified for centrifugal speed, different geometrical ratios, number of blades in the
compressor flow together with LUT. The program also impeller and diffuser, loss constants, inlet flow angles, volute
calculates the flow in the stator of the compressor. Normally the pressure recovery factor, etc.
computation is made time-averaged, but also some time- Although the program has been programmed in the DOS
dependent calculations have been made. Results of CFD- environment, it has all the results onscreen (Fig. 1) and the main
calculation and one-dimensional calculation are compared, and calculation results are:
the design procedure repeated, if necessary, see Pitkänen et al.  Pressure, temperature, enthalpy, specific volume in the inlet
(1999a, 1999b and 1999c). (1) and outlet (2) of the impeller wheel, outlet of the
The high complexity of the flow, especially in the rotating diffuser (3) and volute (4). Pressure, temperature and
impeller, makes the CFD modeling very difficult. There has enthalpy are given with the static and stagnate conditions (1
been done much research in analysing the flow and the different and 01)
viscous phenomena, but there are still no programmes that can  Main geometry parameters for the impeller, diffuser and
calculate the time dependent flow explicitly. volute. Also, compressor specific speed and power
The final compressor geometry is tested in the test facilities consumption
at Lappeenranta University and the test results are used to  Isentropic efficiency and pressure ratio in each point,
develop the design loop. Up to this date, around 15 high speed pressure recovery in the diffuser and volute

2
 Velocity triangles, deviation angle and relative Mach When the blade designer accepts the blade geometry the
number in the impeller wheel inlet and outlet numerical geometry data can then be generated for compressor
flow CFD analysis, finite element method (FEM) based stress
The program calculates the off-design performance and vibration analysis (FEM) and computer aided
automatically and produces the compressor performance map manufacturing (CAM). If any non-conformity is found in blade
for further studies on the performance. geometry or the results from either CFD or FEM analysis do not
meet the requirements, the design procedure must be repeated
2. THREE-DIMENSIONAL GEOMETRY DESIGN partially or totally, even starting from one-dimensional flow
The high-speed research team has been involved in the design.
impeller research since 1982, first with radial turbines and then
with centrifugal compressors. The full geometry design is vital
for CFD analysis and manufacturing and predicting the
operational strength of the impeller.
The impeller blade geometry for centrifugal compressors is
generated using a software developed by LUT. Before the year
1998 the three-dimensional geometry modeling program
developed by LUT has been used in minicomputers. The need
for using microcomputers for impeller blade design arose after
the mid-1990’s and thus a new software for microcomputers
was developed at LUT.

Fig. 3. View of one flow passage (full blade suction

side to splitter pressure side) of an impeller.

3. THREE-DIMENSIONAL FLOW DESIGN

Modern computers and softwares have made the numerical
testing of the compressor geometry design possible. Generally,
it is much more economical to find out the locations in the
geometry needing some modifications compared to the real
laboratory experiments. The computational fluid dynamics
(CFD) has been a part of the radial compressor design process
Fig. 2. Axial view of an impeller blade. at Lappeenranta University of Technology since 1995.
The result of the 3D design program, i.e., the boundary
Generating the impeller blade geometry for centrifugal curves of the full and splitter blades, can directly be read into
compressors proceeds as follows. The main geometry data of the pre-processor, which converts the data into the Cartesian
the impeller, such as the main dimensions and the blade angles coordinate system. Both the full and splitter blades consist of
obtained from one-dimensional flow design (see chapter 1), is four boundary curves: the hub and shroud curves on pressure
used as input data for three-dimensional modeling. Blade design and suction surfaces, respectively.
is started with defining the camber lines for the hub and shroud
of an impeller blade. The three-dimensional modeling software
provides a graphical blade design environment for modifying
camber line geometry.
The next stage is to define the thickness distributions for
the hub and shroud of the blade and they can be modified
graphically, as well. Based on the camber lines and the
thickness distributions the suction and pressure side co-
ordinates for the hub and shroud are finally calculated. The
blade can be displayed using various projections (Fig. 2 and 3).

3
 solution of the Reynolds averaged Navier-Stokes equations with
Cartesian velocity components
∂ U ∂ ( F − Fv ) ∂ ( G − Gv ) ∂ ( H − H v )
+ + + =Q (1)
∂t ∂x ∂y ∂z

where the solution vector U = [ ρ , ρ u, ρ v , ρ w, E ] , and F, G,

T

H are the flux vectors in each coordinate direction, respectively.

r r r r
Here ρ is the density, V = ui + vj + wk is the absolute velocity,
and E is the total internal energy. The rotational speed of the
r r
domain is Ω × r = [0, − Ωz , Ωy] and the convective, i.e. the
T

r r r
relative speeds are V − Ω × r . In the case of the steady or quasi-
steady computation a source term Q = [0, 0, ρ Ωw, − ρ Ωv , 0] is
Fig. 4. One impeller passage. Splitter blade in the T

middle.
introduced, but in the time-accurate calculation Q = 0. The
quasi-steady calculations are performed in the rotating
The grid generation is made by an automatic generation
coordinate system with Cartesian velocity components and
tool of a commercial turbomachinery oriented software, IGG
modified convective speeds. In the case of the time-accurate
(1997). The main input parameters for the grid generation
solution the convective speeds are identical to those in a
besides of the blade geometry data are the number of cells at
rotating coordinate system. Hence, the only difference between
each block, and the type of the grid clustering near the solid
these two approaches concerns the source term Q.
walls. The tip-clearance size and the leading edge rounding are
The pressure is calculated from an equation of state. For a
specified also at the input data file. The grid covers also part of
the inlet duct and the vaneless diffuser. Typically only single perfect gas p = (γ − 1) ρ e , where γ is the ratio of specific heats
impeller passage is modeled, Fig. 4. and e is the specific internal energy. The molecular viscosity is
The effect of the grid quality has to be known in the CFD calculated from Sutherland's formula. The Reynolds stresses in
simulations. In order to capture the real physical phenomena, the viscous fluxes Fv, Gv and Hv are modeled using Boussinesq's
e.g., boundary layer thickness and secondary vortices in the approximation. The algebraic turbulence model developed by
blade passage, the number of cells should be large enough. But, Baldwin and Lomax (1978) is applied in order to calculate
when the number of cells grows, the calculation time increases turbulent viscosity. The implementation of the model is given in
rapidly. The grid size is, therefore, a compromise between time Reunanen (2000).
and resolution of the result. Most of the radial compressor The quasi-steady approach and the time-accurate solution
impeller simulations have been made by using a grid, where the utilize the same basic steady-state algorithm. In the present
size of the cross-sectional grid of the blade passage between solution, a finite-volume technique with a structured grid is
two adjacent full blades is 64x32x64 in azimuth, spanwise and applied. The flow equations have a discrete form
streamwise directions, respectively. Besides the grid size, height
∑ − S ( F$ − F$v ) + Vi Qi = Ri
dU i
of the first cell on the solid surface is important. The size has to Vi = (2)
be set in order to get the y+ value near unity. If first cell is dt faces
bigger, the solution could underestimate the losses. where the sum is taken over the faces of the computational
After the grid has been generated, the boundary conditions cell.
for the flow solver have to be defined. The values for rotational In the evaluation of the inviscid fluxes in a moving grid,
speed, mass flow rate, total inlet temperature, and static Roe's method, Roe (1981) is applied. A MUSCL-type approach
pressure at vaneless diffuser outlet are taken from 1D program. has been adopted for the evaluation of the primary flow
Thus, inlet pressure, as well as the outlet temperature, are the variables (ρ, u, v, w, p) on the cell surfaces. The viscous fluxes
results of the CFD analysis. All the solid walls at the impeller are evaluated using a thin-layer approximation. The thin-layer
are defined as rotating wall excluding the shroud surface. model is activated in all coordinate directions.
The flow solver FINFLO, developed in Helsinki University Equation (2) is integrated in time implicitly by applying the
of Technology, has been used in the compressor simulations, DDADI-factorization, Lombard et al. (1983). The resulting
Siikonen and Pan (1992). It has been applied for many kind of implicit stage consists of a backward and forward sweep in
both internal and external flow simulations. every coordinate direction. The boundary conditions are treated
explicitly, and a spatially varying time step is utilized.
In the steady-state solution as well as in the time-accurate Furthermore, a multigrid acceleration of convergence is
simulation the impeller is assumed to rotate around the x-axis utilized.
with an angular velocity Ω. The simulation is based on the

4
 The time-accurate simulation is based on a dual time-
stepping, i.e. it shares the same basic features described above,
but now the pseudo time-integration is performed inside a
physical time step and a true time derivative is added on the
right-hand side of Eq. (2). A three-level fully implicit scheme is
applied for the time-integration, Hoffren et al. (1995). More
details of the solution methods are given in Siikonen (1995) and
Reunanen (2000).
Two multigrid levels are typically used in the simulation.
The first few hundred iterations are calculated using a coarse
grid, which contains only half of the grid points at each grid-
index direction. The result of the coarse grid is used as the
starting point for the dense-grid simulation.
It is practical to start the CFD analysis from the near-
optimum operating point locating at the middle of the
performance map. If the simulation converges normally, the Fig. 5. Comparison between CFD predicted and
result is saved, and the next operating point can be simulated measured data of the centrifugal compressor stage.
for instance using the same speed and outlet pressure, but little
bigger or smaller mass flow rate until the operating limits are
reached. 4. TEST FACILITIES
The calculation time is strongly related to the grid, and The high-speed technology compressors have been
computer speed. Usually the simulation of one operating point commercially manufactured since 1996 in Finland. The LUT
can be done overnight with a modern workstation without using research group has designed hundreds of different compressor
parallelization. wheels with the one-dimensional flow design program for
The CFD result is mainly used for the preliminary different production plans. From previous basic design, 35 have
verification of the compressor impeller performance. It gives been designed with full geometric details and 20 have been
estimations for the pressure ratio and efficiency levels at gone through Finflo calculations.
selected operation points. It also predicts the operation range The commercial compressors have been tested both for
between stall and choking limits. According to the CFD results, general and individual performance. The test facility for general
it is possible to make modifications to the geometry. performance is situated at LUT and it is built to give accurate
It should be noted that the full stage performance information from compressor inlet and outlet conditions.
estimation with CFD would need also the volute analysis. The compressor test facility can be used both for
However, it is not always modeled due to the complicated and overpressure and vacuum compressors. In a standard measuring
time-consuming grid generation process. If the volute is procedure the compressor thermal efficiency, compressor total
analyzed separately, the combination of the impeller and volute efficiency, mass flow rate, pressure differential produced by
results is possible but then some simplifications have to be compressor and the electric input power are measured. Main
accomplished Pitkänen et al (1999c). The full stage simulation parts of the measuring arrangements are described in figures 6
is often too complex and slow task for design purposes, but it and 7.
will come more common in future. Currently the volute is taken The fluid flow is measured with standard orifice and
into account roughly by using an estimated pressure recovery nozzle. Mass flow measurement is measured by means
coefficient. Result of that method is shown in Fig. 5. explained standards ISO 5167, VDI 2041 chapter 5.2. VDI
In practice, experimental studies can not be fully replaced 2041 standard tells the special measurement of the gas flow. It
by the numerical simulations because there are always some gives technique how to measure the mass flow of the gas when
options, which may have an effect to the result, e.g. grid density, the orifice or nozzle is at the beginning of a pipeline.
grid quality, and models of turbulence. At present, performance The compressor performance is measured by means
of the manufactured devices, which are going to be installed explained standards ISO 5389, VDI 2045 and ASME Power
into the plant, is always verified by the measurements. CFD Test Code. The measured values are corrected to the same inlet,
results have to be compared with that data. temperature and air humidity.

5
 All measuring data is collected with data logger into a
computer in which the needed parameters are calculated. The
measured signals are analyzed with PC-program that saves the
original data into a file, calculates the compression process and
shows the most significant process parameters. The program
also saves the stationary measured data points into a file that
can be utilized by other program to show the compressor map.
Absolute pressure and pressure difference are measured by
Rosemount and Druck transmitters. Temperatures are measured
by K-type thermocouples. Relative humidity is measured by
Vaisala humidity sensor. Compressor shaft speed is measured
with magnetic pulse generator installed permanently to the
compressor shaft and compressor electrical input power is
measured with LEM Norma D4000 power analyzer.
The uncertainty of the mass flow measurement can be
determined with the help of the standards /ISO 5167/ and /ISO
5168/. Performing partial differentiation uncertainties of the
other terms can be calculated. Current uncertainties of the
Fig. 6. Compressor test facility layout. various terms are:
When the compressor is used with different inlet - uncertainty at the 95.4% confidence level in input power
conditions, the aerodynamical performance differs from that on measurements is about ± 0.3%
the test facility. This will be taken into account with calculation - uncertainty of the electromechanical efficiency is about ±
in which similarity laws are used. Conversion of the test results 0.9% when Pinput is 45 kW and electromechanical efficiency
from test conditions to the specified conditions, assuming is 88%
constant polytropic efficiency. - uncertainty of the flow rate is about ± 0.9%
- uncertainty of the pressure ratio is about ± 0.6%,
- uncertainty of the isentropic efficiency is about ± 1.2%,
- uncertainty of the compressor total efficiency is about ±
0.9%.

5. CONCLUSIONS
This paper reports on the work carried out in the
Laboratory of Fluid Dynamics, at Lappeenranta University of
Technology (LUT), which has been designing and testing
compressors, turbines and pumps since 1982. The Laboratory
has been responsible of large number of centrifugal compressor
design and testing in Finland. These compressors are usually
directly connected to a high speed electric motor. Instead of
IGV and diffuser vane control speed control is used, which sets
some additional requirements e.g. for compressor performance
map shape. It is required, that at the design pressure ratio the
Fig. 7. Schematic of compressor test facility. compressor has an operating range 1:2 in mass flow, when
optimal speed control is used. It is also required, that the high
Previously mentioned assumption is generally only possible efficiency range of compressor is as wide as possible.
when similarity of the flow is assured for the conversion of a Efficiencies obtained are from 80 to 82% in this operating range
test point to specified conditions, i.e. when the following main (vaneless diffuser). Calculated values with the own programs
requirements are fulfilled /ISO 5389 Annex D, Similarity of the used in design are within 2 to 4% from the measured ones in
flow, Similar velocity triangles/, most cases.
a) geometrical similarity The final compressor geometry is tested in the test facilities
b) equal isentropic exponent at Lappeenranta University and the test results are used to
c) equal enthalpy rise coefficients and flow coefficients develop the design loop. Up to this date, around 15 high speed
d) equal Mach numbers compressors have been aerodynamically designed and tested in
e) equal Reynolds numbers this design loop. Typical pressure ratio of the compressors

6
ranges from 1.6 to 2.5 and rotational speed from 25,000 rpm to Pitkänen, H. Esa H., Sallinen P., Larjola J., 1999c, ”CFD
40,000 rpm. analysis of a centrifugal compressor impeller and volute”. Int.
Gas Turbine & Aeroengine Congress, Indianapolis, June 7 -
ACKNOWLEDGMENTS June 10. ASME publ. 99-GT-436, 8 p.
Reunanen A., Pitkänen H., Siikonen T., Heiska H., Larjola
The authors wish to acknowledge the Technology Development J., Esa H., Sallinen P., 2000, “Computational and experimental
Centre of Finland and High Speed Tech Oy LTD for the Comparison of different volute geometries in a radial
financial support for this work. The authors also wish to thank compressor,” Accepted to: Int. Gas Turbine & Aeroengine
Professor T. Siikonen of Helsinki University of Technology for Congress, May 8-11, 2000, Munich, Germany, ASME Paper
providing the basic solver for CFD-calculations. 2000-GT-0469
Roe P., 1981, “Approximate Riemann Solvers, Parameter
REFERENCES Vectors and Difference Schemes”, Journal of Computational
Baldwin B., Lomax H., 1978, “Thin Layer Approximation Physics, Vol. 43, pp. 357-372.
and Algebraic Model for Separated Turbulent Flows”, AIAA Siikonen T., 1995, ”An Application of Roe’s Flux-
Paper 78-275. Difference Splitting for the k-ε Turbulence Model”,
IGG, 1997, Interactive Geometry Modeller and Grid International Journal for Numerical Methods in Fluids, Vol. 21,
Generator, User Manual, Version 3.6, NUMECA International pp. 1017-1039
Hoffren J., Siikonen T., Laine S., 1995, ”Conservative Siikonen T., Pan H., 1992, "Application of Roe's Method
Multiblock Navier-Stokes Solver for Arbitarily Deforming for the Simulation of Viscous Flow in Turbomachinery,"
Geometries”, Journal of Aircraft, Vol. 32, No. 6, pp. 1342-1350 Proceedings, First European Computational Fluid Dynamics
Larjola J., Backman J., 1995, ”Oil free bearings in high Conference, Ch. Hirsch et al., ed., Brussels, Belgium, pp. 635-
speed technology". Stockholm Power Tech, June 18-22, 1995, 641
International Symposium on Electric Power Engineering,
invited speakers' sessions, pp. 63-68
Larjola J., Alamäki J., Sallinen P., 1995, ”High speed
pumps in water jet cutting and in water jetting”. - Proceedings
of the 4th Pacific Rim International Conference on Water Jet
Technology, April 20-22, 1995, Shimizu, Japan. Part l: Future
applications of the jetting technology (Invited papers), pp. 47-
56
Larjola J. (editor), 1988, ”Proceedings of Conference on
High Speed Technology”, August 21-24, 1988, Lappeenranta,
Finland. Lappeenranta University of Technology, Department
of Energy Technology, publ. No ENTE D-15, 288 p.
Larjola, J., 1986, ”High-Speed Turbomachinery and Its
Application in the Conversion of Energy”, ASME Winter
Annual meeting, Anaheim, December 7-12, 1986. ASME publ.
86-WA/FE-4
Lombard C., Bardina J., Venkapathy E., Oliger J., 1983,
“Multi-Dimensional Formulation of CSCM – An Upwind Flux
Difference Eigenvector Split Method for the Compressible
Navier-Stokes Equations”, 6th AIAA Computational Fluid
Dynamics Conference, Massachusetts, pp. 649-664.
Pitkänen H., Esa H., Sallinen P., Larjola J., Heiska H.,
Siikonen T., 1999a, ”Time-accurate CFD analysis of a
centrifugal compressor”. Fourth Int. Symposium on
Experimental and Computational Aerodynamics of Internal
Flows, Dresden August 31 to September 2. Proceeding Vol. 2,
pp. 130-139
Pitkänen H., Esa H., Reunanen A.., Sallinen P., Larjola J.,
1999b, ”Experimental and numerical investigation of a
centrifugal compressor volute”. Fourth Int. Symposium on
Experimental and Computational Aerodynamics of Internal
Flows, Dresden August 31 to September 2. Proceeding Vol. 2,
pp. 120-129