See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/227018239

Flow Simulations of an Axial Transonic Compressor Stage

Chapter · January 2009

DOI: 10.1007/978-3-540-69182-2_21

CITATIONS READS

0 72

2 authors, including:

Jens Iseler
Dassault Systèmes
7 PUBLICATIONS   20 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Jens Iseler on 08 May 2015.

The user has requested enhancement of the downloaded file.

Flow Simulations of an Axial Transonic
Compressor Stage

Jens Iseler and Reinhard Niehuis

Abstract This paper refers to numerical investigations of a transonic compressor

stage for homogeneous and disturbed inflow conditions. The simulations were con-
ducted with the Navier-Stokes Solver TRACE and compared with the experimental
results, obtained from measurements accomplished at the DLR in Cologne. In the
first part of the planned investigations, the flow behavior of the transonic compres-
sor stage was predicted for undisturbed inlet conditions at design point using steady
state simulations. This will be followed by simulations at operation points near insta-
bility onset. The second part includes simulations with disturbed inflow conditions.
Here, transient simulations are scheduled for predicting correctly the complex flow
behavior. Right now, only one blade passage is simulated due to the assumption
of periodicity. In case of disturbed inflow conditions, this assumption is no longer
valid. Therefore, the whole blade row has to be simulated which leads to a high
computational effort.

1 Introduction

In order to decrease costs during the design and developing process of modern air-
crafts, high efficient and accurate design tools are necessary. These tools will allow
the prediction of the flow behavior at the aircraft body and its airfoils as well as
inside of the jet engines for a whole flight mission. For simulating the flow physics
correctly, the interaction of the inner jet engine flow and the outer flow around
the aircraft has also to be taken into account. Accurate predictions are especially
demanded during states where high aerodynamic loads are present. One of these
critical phases is the take-off procedure, where highly turbulent air with strongly
varying total pressure may enter into the jet engine. This inlet distortions increase
the risk of compression system instabilities like stall and surge. In case of a surge
scenario, parts of the compressor like rotor or stator blade could be destroyed. Thus
the prevention of those phenomena during all flight stages represents a major job
in the design and development of jet engines. Inlet distortions are composed of to-
tal pressure-, angle- and total temperature distortions. All three disturbances have a
reduction of the operation range in common. In order to predict accurately the cre-
ation and the migration of the inlet disturbances and its impact on the compressor,

J. Iseler · R. Niehuis
Institut für Strahlantriebe, Werner-Heisenberg-Weg 39, 85577 München, Germany
e-mail: [email protected]; [email protected]

261
262 J. Iseler, R. Niehuis

the outer aerodynamics in front of the engine and flow into the jet engine have to
be simulated simultaneously. Right now, most of the numerical methods are spe-
cialized either to inner or to outer aerodynamics. Thus one possibility to solve this
problem is to couple numerically two codes- one created for inner, the other for
outer flow dynamics. This strategy is pursued by members of a DFG project, where
a coupling of the DLR codes TRACE and TAU is planned. The creation and de-
velopment of the inlet distortions will be simulated with TAU, while TRACE pre-
dicts the impact on the compressor stage. The first steps of the project, which have
been realized currently, contain validation tests with the TRACE code concerning
the reaction of the compressor to inlet distortions. This validation tests refer to a
data set of a transonic compressor stage, which was experimentally investigated for
homogeneous and disturbed inflow conditions at the Institute of Propulsion Tech-
nology in Cologne. The inflow disturbances are realized numerically by defining
a total pressure field at the entry plane of the computational grid. In this paper,
numerical results concerning the investigation of the transonic compressor stage
will be discussed and compared with experimental data. All numerical results are
gained from three dimensional computations. In case of distorted inflow conditions,
additionally a complete circumferential resolution of the unsteady flow is neces-
sary.

2 Simulation Methods for Transonic Turbomachinery Flow

The flow field encountered in a turbomachinery is one of the most complicated in the
field of fluid dynamic practice. The flow is viscous, compressible, unsteady and due
to the engine construction fully three dimensional. This flow features lead to phe-
nomena which influence the flow through the blade rows of a transonic compressor.
Figure 1 shows some of the most important flow phenomena appearing at a transonic
compressor stage: the wall boundary layers at casing and hub, the secondary flow,
boundary layers at the blade surface, mixed subsonic and supersonic flow fields and
shocks. Additionally the shape of the blade profile and blade geometry alternating
in radial direction influences the flow distribution. The appearance of these flow
phenomena is always combined with the generation of fluid losses. Generally, there
exist two different loss types: cascade losses and wall losses. Wall losses appear near
casing and hub walls and are composed of friction losses at the surfaces, separation
losses at blade-wall corners and losses due to secondary flow and tip leakage flow.
Cascade losses represent all losses emerging at a single blade element outside of the
near wall region. Those losses include profile losses, shock losses induced by shock
systems in the blade passage and shock-boundary layer interactions. Nowadays, per-
forming two dimensional simulations of airfoils using RANS codes represent a stan-
dard procedure in the design process. Those simulations allow a good prediction of
the cascade losses in an affordable time frame. However, the influence of near wall
flow phenomena and its interaction with the cascade flow cannot be derived from
these calculations. Since the total pressure losses caused by tip leakage flow can
be—depending from the operation point—of the same magnitude as the cascade
Flow Simulations of an Axial Transonic Compressor Stage 263

Fig. 1 Secondary flow in a compressor cascade [1]

losses, a detailed knowledge of these wall effects is necessary. This leads to the
request of unsteady fully three dimensional flow simulations. Despite the high stan-
dard of modern computing systems, performing of three dimensional simulations
is still combined with a huge computational effort. Therefore, simplification tech-
niques like the assumption of periodicity (periodic boundaries) are frequently used.
Nevertheless, the usage of periodic boundaries for real turbomachinery applications
is normally coupled with a scaling of the geometry since a one-to-one matched in-
terface between rotor and stator row is demanded. Unfortunately, the scaling itself
involves new problems: Considering an axial machine, the scaled geometry may not
correctly represent the flow field in the original machine since the scaled blade chord
leads to a varied Reynolds number condition. Secondly, the axial gap between the
rotor and the stator is changed due to the scaling. This may lead to a different blade-
row-interaction and finally to different blade loadings. Another technique that can
be used for unsteady flow simulations is the phase-lag method [2]. Here, a scaling
is not necessary since the interface between the rotating and stationary components
need not to be one-to-one matched. Thus the instantaneous flow field in the entire
stage can be reconstructed by the passage flow at different time steps. Nowadays,
the phase-lag method is available in several numerical codes in order to permit ac-
curate unsteady simulation of real turbomachinery applications. For instance, Wang
[3] used successfully the phase-lag method implemented in the RANS-code TRACE
to simulate the first three blade rows of the IDAC 3.5-stage axial compressor [4].

3 Description of the Test-Case

The original stage [5], as shown in Fig. 2, was designed for a spool speed of
20260 rpm with a total pressure ratio of 1.51 at an equivalent mass flow of 17.3 kg/s
264 J. Iseler, R. Niehuis

Fig. 2 Cross section view of

the DLR compressor stage

Fig. 3 Test facility at the DLR Cologne

under standard reference conditions with 288 K and 101325 hPa. The rotor diame-
ter is 398 mm with a hub-to-tip ratio of 0.5 and the maximum blade tip speed is up
to 421 m/s. Overall 28 blades (plane b to d in Fig. 2) with nearly DCA-profiles at
the hub as well as MCA profiles at the tip and a maximum chord length of 60 mm
were used. The stator blade (plane e to g) row consists of 60 blades with NACA-65-
profiles with a maximum blade thickness of 8 percent and a chord length of about
30 mm. In Fig. 3, a schematic view of the compressor test facility with its drive
system and the test rig is presented. Due to a 15:1 gear ratio of the gearbox, speeds
up to 22500 rpm can be realized at the rotor. Thus, stage pressure ratios up to 2.0
are achievable. Pressure and temperature measurements as well as measurements of
flow direction were conducted. For undisturbed inflow conditions, radial surveys of
total pressure were performed upstream of the rotor, between rotor and stator and be-
hind the stator. Surveys of the total temperature were carried out downstream of the
stator row. The flow direction was recorded only between rotor and stator (deduced
from five-hole-probe measurements). Besides the investigation with undisturbed in-
flow condition, measurements at the compressor stage were also accomplished for
steady inlet distortions [6]. Total pressure distortions as well as distortions of the
Flow Simulations of an Axial Transonic Compressor Stage 265

Fig. 4 Overall performance

of the transonic compressor
stage

incoming flow direction were generated upstream of the compressor stage, in order
to investigate their influence on the overall compressor performance and on the flow
behavior. Total pressure distortions were generated by a wire grid, which represents
a simple and non-expensive method. The intensity of pressure distortions could be
varied by changing the grid permeability (58% and 51%). The circumferential extent
of these distortions was chosen as 60 and 120 degree. The generation of distortions
concerning flow direction was performed by installing inlet guide vanes (same axial
position as the grid) with different chord lengths. In Fig. 4, the measured character-
istics of the transonic compressor stage with and without inlet distortions are shown
at 70%, 85% and 100% design speed. The characteristics for disturbed inlet con-
ditions show a significant decrease of the total pressure for high rotational speeds.
Additionally, a movement of the surge line up to higher mass flow rates is obvious,
which finally leads to a decrease of the stable operating range. In contrast to the
70% and the 85% speed line, the 100% speed line displays a significant dependency
of the grid permeability and the circumferential extent of the distortion.

4 Flow Solver Information

The CFD simulations are conducted with the RANS code TRACE [7] developed by
the DLR’s Institute of Propulsion Technology in Cologne. This code allows the sim-
ulation of a multistage three-dimensional periodically unsteady and transitional tur-
bomachinery flow. TRACE is a time-accurate and parallelized computational fluid
dynamics solver specialized to turbomachinery flows. In the version used, it solves
the three-dimensional Reynolds-averaged viscous Navier-Stokes equations in the
rotating frame of reference for compressible fluids on multi-block structured grids.
266 J. Iseler, R. Niehuis

The discretization of the convective fluxes is based on the TVD-Upwind scheme by

Roe, which is combined with a MUSCL extrapolation scheme in order to obtain sec-
ond order accuracy in space. The viscous derivatives are discretized using a second
order central-differences scheme. Furthermore, non-reflecting boundary conditions
are implemented at the inlet and outlet boundaries. Steady state simulations are re-
alized by using the mixing plane concept proposed by Denton [8], where relative
and absolute system (here rotor and stator row) are coupled by the transfer of cir-
cumferential averaged flow variables. Unsteady calculations are performed with an
implicit dual time stepping method. The time-accurate coupling of the rotor and
stator system is accomplished by the patched-cells-algorithm. In order to simulate
the turbulent effects, the Wilcox k-ω turbulence model [9] as well as an extended
version of the well-known Spalart and Allmaras model [10] is available. In the cur-
rent investigation, the Kato-Launder extension of turbulence production is used to
suppress overproduction of turbulent kinetic energy at the blade leading edges. The
CGNS (CFD General Notation System) file format is employed for the storage of
the grid geometry and the flow field solution. The solver is written in the C pro-
gramming language and embedded in a parallel environment. The code is hardware
independent and may therefore be run on any processor architecture. A very efficient
parallel scalability is reached on machines ranging from small computing systems to
massively parallel super-computers with several hundred processors like the hlrb2
system of the LRZ in Munich.

5 Computational Grid of the Transonic Compressor Stage

A multi-block structured grid, built with the G3DMESH mesh generator, was used.
For stator and rotor blades, an OCH grid topology was chosen. A high grid res-
olution was performed near the blade surfaces in order to obtain Low Reynolds
conditions. At hub and casing surfaces, the grid was designed for the usage of wall
function. This resulted in a grid with 1.2 million grid points, intended for flow sim-
ulations with undisturbed inflow conditions. Figure 5. displays the whole stage as
a three dimensional plot. Around the rotor blade an O-type grid with 310 discrete
nodes was applied, as can be seen in Fig. 6. This high point density enables a good
resolution of the boundary layer, the wake flow of the blade and the shock boundary
layer interaction. The O-grid is surrounded by a C-type block, while the remaining
regions are filled with H-type blocks. The tip region is covered by a special H-grid
block, as shown in Fig. 7. 45 elements are used to resolve the radial extension of
the stage. 15% of these elements are reserved for the resolution of the tip-casing
gap. The inflow conditions are specified by radial distribution of total pressure, total
temperature and flow angles derived from experimental data. At the outlet plane, the
static pressure measured at the casing surface is prescribed for all radial positions.
Flow Simulations of an Axial Transonic Compressor Stage 267

Fig. 5 3D view of the

transonic compressor stage

Fig. 6 OCH-Mesh of the

rotor hub section

Fig. 7 OCH-Mesh of the

rotor tip section
268 J. Iseler, R. Niehuis

6 Numerical Results and Discussion

During the first period of the DFG-Project “Numerische Modellierung der Strö-
mung in einem Triebwerks-Fan bei inhomogenen Zuströmbedingungen”, validation
tests with the RANS code TRACE are intended. Several operating conditions of
the transonic compressor stage are planned to be simulated. Currently, there ex-
ist numerical results for design conditions. Therefore, the discussion will relate to
conditions, where the stage reaches the maximum efficiency. Right now, the numer-
ical results show an overestimating of the isentropic efficiency with ηis = 86.5%
instead of ηis,ex 84.2%, deduced from the experimental data. The predicted total
pressure ratio Π = 1.58 differs only 1% from the value determined by Dunker. Fig-
ure 8 shows the circumferentially averaged radial distribution of the calculated and
measured total pressure Pt and total temperature Tt . Pressure measurements were
performed between rotor and stator stage at x = 0.113 m as well as downstream
of the compressor stage at the axial position x = 0.173 m. Total temperature mea-
surements were only performed behind the stator row (same position as the pressure
measurements). The solid line represents the distribution at the axial gap, the dash
dotted lines represent measurement (red) and simulation (green) downstream of the
stator row. Between rotor and stator, measured and calculated total pressure show
a different behavior for the first 50% radial height. The measured total pressure is
slightly decreasing (0.5% compared to the near-hub value) up to 35% radial height
and increases afterwards, while the simulated pressure indicate a opposite distribu-
tion with a maximal pressure at 22.6% radial height. These too large pressure values
are possibly the result of a positive incidence. Near hub, the not considered leakage
flow coming from the gap between rotor disk and the inlet hub (plane a to b in Fig. 2)
could also be a reason for different pressure values. From mid span to blade tip, both
measured and simulated values rise up to 88% blade height. However, the simulated
pressure values are permanently smaller than the measured ones. Obviously, the
simulated shocks and its interactions with the boundary layer are overestimated in
this part and lead to higher total pressure losses. Due to the dash-dotted lines, mea-
sured and simulated total pressure losses between stator inlet and outlet are very

Fig. 8 Radial distribution of measured and calculated total pressure Pt and temperature Tt
Flow Simulations of an Axial Transonic Compressor Stage 269

Fig. 9 Total pressure

distribution downstream of
the rotor stage at x = 0.163 m

Fig. 10 Measured (left) and simulated Mach number distribution at 18% span

similar. The temperature distributions show generally a good agreement, particu-

larly for the first 50%. The highest differences are observed near the blade tip with
2%. The used wall function at the tip region could be a reason for that inaccuracy
which leads to a minor work production.Figure 9 represents the simulated total pres-
sure distribution downstream of the compressor stage at x = 0.163 m. This picture
clarifies the circumferentially variation of Pt caused by the wakes of the two stator
blades. The low values inside the wakes result from viscous effects at stator suction
side and pressure side and also from secondary flow effects near hub and shroud.
At design conditions, one can see that the loss regions at hub and shroud are nearly
of the same magnitude and extent. Figures 10–13 show the measured and simulated
Mach number distributions for design conditions at different radial positions. The
experimental data was deduced from L2F-measurements. Figure 10 displays the
distribution at 18% span. In both cases, a local supersonic region at the suction side
270 J. Iseler, R. Niehuis

Fig. 11 Measured (left) and simulated Mach number distribution at 45% span

near leading edge can be seen. Due to the positive incidence in the simulated case,
the supersonic region is located closer to the leading edge. Another consequence
of this incidence is the reacceleration of the suction side flow downstream of the
shock region at 15% chord length. The simulated Mach number levels in the rotor
passage indicate only small differences compared to the measured ones. At 45%
span, the simulated relative inlet Mach number is close to 1. Here, a detached shock
upstream of the rotor leading edge can be observed. However, the experimental data
show a detached slight bow shock at the suction side sector and a passage shock at
the pressure side section. The occurrence of this different shock phenomenon is due
to a higher inlet Mach number, which is supercritical at 45% span. None the less,
a good agreement is given for the passage flow downstream of the passage shock.
For 69% and 89% span, measurements and simulations show a flow field combined
with a complex shock system. Due to inlet Mach numbers higher than Ma = 1.1,
a bow shock and expansion waves are spreading from the blade leading edge. The
bow shock is detached from the leading edge. Due to the surface curvature of the
neighboring blade, expansions waves are present upstream of the leading edge. The
suction side part of the bow shock is bowed and weakened by the characteristics
coming from the blade leading edge. The pressure side section of the bow shock-
the passage shock- may be strengthened by those characteristics before hitting the
neighbor suction side [11]. At 69% span, the passage shock represents a normal
shock (measurement and simulation).
Consequently, a deceleration can be observed in the blade passage downstream
of the passage shock resulting from the diffuser shape of the blade channel. Due
to Fig. 12, a general good agreement is obtained for the Mach number distribution
at 69% span. Maximum speed as well as the deceleration rate in the passage is
nearly identical. At 89% radial height, the passage shock represents a bowed shock,
which is confirmed by the simulation. Nevertheless, a different flow behavior exists
Flow Simulations of an Axial Transonic Compressor Stage 271

Fig. 12 Measured (left) and simulated Mach number distribution at 69% span

Fig. 13 Measured (left) and simulated Mach number distribution at 89% span

downstream of the passage chock. In opposite to the measurements, the simulated

passage flow field decelerates to subsonic speeds. Afterwards, the flow reaccelerates
up to Ma = 1.07. This reacceleration results from a channel contraction caused by
flow separations near the trailing edge. The supersonic flow is finally terminated
by a normal shock near the rotor trailing edge. The measurements do not confirm
the existence of a subsonic region downstream of the passage shock. None the less,
the terminating normal shock is located nearly at the same position (2% deflection).
272 J. Iseler, R. Niehuis

Regarding the overall development of Mach number distribution at the rotor blade,
a satisfying agreement with the measurements can be found.

7 Summary

Detailed numerical simulations were performed on a transonic compressor stage

and compared with experimental data. The experimental investigations were per-
formed for different operation conditions. Detailed measurements exist for design
conditions and near the instability onset. Additionally, the compressor stage was in-
vestigated for disturbed inflow conditions, in order to observe its influence on the
instability onset. Right now, numerical results exist for the design point with undis-
turbed inflow conditions. The simulations were performed with the RANS code
TRACE, using a multi-block structured grid with 1.2 million nodes. The code was
applied in steady mode with activated turbulence model. The comparison of cir-
cumferentially averaged total pressure and total temperature values indicates that
the differences are of an acceptable magnitude (max. 2%). These differences are
possibly due to the neglected leakage flow between inlet hub and rotor disk and
to the usage of the wall function for the tip region. The simulated Mach number
distributions at several blade heights show a satisfying agreement with the experi-
mental data deduced from L2F-measurements. Remarkable deflections appear only
near hub, where a positive incidence was observed. This incidence leads to a pre-
mature appearance of the supersonic region on the blade suction side. For this year,
further numerical simulations for disturbed and undisturbed inflow conditions are
planned. For disturbed inflow conditions the whole rotor and stator row has to be
simulated in unsteady mode. Therefore, high computational resources, as they are
provided by the Leibniz Rechenzentrum, are necessary to perform the simulations
in an acceptable timeframe.

Acknowledgements The authors wish to acknowledge the support of the Deutsche Forschungs-
gemeinschaft (DFG) for the research program partly reported in this paper. The work was per-
formed within the joint project “Numerische Modellierung der Strömung in einem Triebwerks-Fan
bei inhomogenen Zuströmbedingungen”.

References
1. D.C. Wisler, Advanced Compressor and Fan Systems (General Electric Aircraft Engine Busi-
ness Group, Cincinnati, 1988)
2. J.I. Erdos, E. Alzner, M. McNally, Numerical solution of periodic transonic flow through a fan
stage. AIAA J. 15(11), 1559–1568 (1977)
3. Z. Wang, R. Niehuis, Untersuchung instationärer Strömungsphänomene in mehrstufigen Axi-
alverdichtern. FVV-Bericht Heft 820 (2006)
4. A. Bohne, R. Niehuis, Experimental off-design investigation of unsteady secondary flow phe-
nomena in a three-stage axial compressor. ASME Paper GT 2004-53100 (2004)
 Flow Simulations of an Axial Transonic Compressor Stage 273

5. R. Dunker, Theoretische und experimentelle Untersuchungen über die Verdichtungsstoßver-

luste in transsonischen Axialverdichtern. Forschungsbericht DFVLR Cologne DFVLR-FB 88-
38 (1988)
6. M. Lecht, Beitrag zum Verhalten von Axialverdichterstufen bei stationärer Störung der Zus-
trömung. Forschungsbericht DFVLR Cologne DFVLR-FB 83-39 (1983)
7. H. Yang, D. Nürnberger, H.-P. Kersken, Towards excellence in turbomachinery CFD: A hy-
brid structured-unstructured Reynolds-averaged Navier-Stokes solver. ASME Trans. J. Turbo-
mach. 128, 390–402 (2006)
8. J.D. Denton, The calculation of three dimensional viscous flows through multistage turboma-
chines. ASME paper 90-GT-19 (1990)
9. D.C. Wilcox, Turbulence Modeling for CFD, 2nd edn. (DCW Industries, Anaheim, 1998)
10. P. Spalart, S. Allmaras, A one-equation turbulence model for aerodynamic flows. Technical
Report AIAA-92-0439 (1992)
11. W. Hofmann, Topologie und Entstehung von Blattspitzenwirbeln in transsonischen
Verdichterrotoren und ihr Einfluss auf die Stabilitätsgrenze. Dissertation RWTH Aachen,
2006

View publication stats