You are on page 1of 14

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

net/publication/306077584

Influence of the Coriolis Force on the Flow in a Low Pressure Turbine Cascade
T106

Conference Paper · June 2016


DOI: 10.1115/GT2016-57399

CITATIONS READS
6 835

3 authors, including:

Denis Koschichow Jochen Fröhlich


CGNEV Dresden GmbH Technische Universität Dresden
10 PUBLICATIONS 92 CITATIONS 369 PUBLICATIONS 6,783 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Denis Koschichow on 12 August 2016.

The user has requested enhancement of the downloaded file.


Proceedings of ASME Turbo Expo 2016: Turbomachinery Technical Conference and Exposition
GT2016
June 13- 17, 2016, Seoul, South Korea

GT2016-57399

INFLUENCE OF THE CORIOLIS FORCE ON THE FLOW IN A LOW PRESSURE


TURBINE CASCADE T106

Oliver Baum, Denis Koschichow∗ and Jochen Fröhlich


Institute of Fluid Mechanics
Technische Universität Dresden
George-Bähr-Straße 3c
01062 Dresden, Germany
Email: denis.koschichow@tu-dresden.de

ABSTRACT NOMENCLATURE
The paper presents numerical investigations of a model 1, 2 Indices identifying inlet, outlet
setup conceived to investigate the influence of the Coriolis force α Angle of incidence
on the secondary flow in a low pressure turbine cascade. It ad- δ Boundary layer thickness
dresses the question in which sense and by how much results in ζ2 Stagnation pressure loss coefficient, (P1 − P2 )/(0.5ρu22 )
a linear cascade may differ from the situation in a rotating cas- ν Kinematic viscosity
cade. For this purpose highly resolved Direct Numerical Simu- ρ Density
lations of the flow within a T106A passage close to the end- τw Wall shear stress
wall were conducted for two cases with and without rotation and hΦi Time averaged quantity
hence with and without Coriolis force at a Reynolds number of Ω Angular velocity
20, 000. Comparing non-rotating and rotating turbine passage, cax Axial chord length
several effects are detected: First, the Coriolis force causes tran- Cp Static pressure coefficient, (p − p2 )/(P1 − p2 )
sition of the horseshoe vortices, so that the region between the D Dean number
blades becomes much more turbulent and an explanation for the f Body force vector
destabilization is provided. Second, the strong radial flow caused LE, TE Leading edge, trailing edge
by the Coriolis force suppresses the laminar separation of the p Static pressure
boundary flow at the suction side. Third, in the case with rota- P Stagnation pressure
tion, the large-scale secondary motion creates higher stagnation t Time
pressure loss than in the reference case and is responsible for a Ta Taylor Number
complete redistribution of the flow field in the passage. An ad- u Velocity vector
ditional test case with opposite rotation was computed for com- ux , uy , uz Velocity components in x-, y-, z-direction
pleteness. u∞ Inflow velocity outside boundary layer
x, y, z Axial, pitchwise, spanwise/radial direction

∗ Address all correspondence to this author.

1 Copyright © 2016 by ASME


Introduction The present paper considers a model configuration of a lin-
Besides investigations of rotating systems, numerical and ear low pressure T106A turbine cascade. To allow exact com-
experimental investigations of linear cascades represent an im- parison of the flow with and without rotation the geometry is left
portant topic in turbomachinery research. The model of the lin- unchanged when introducing rotation. This constitutes an easy
ear cascade is commonly used because it offers many advantages and illustrative way of separating the influence of the Coriolis
over a rotating machine. The major ones are an easy accessibility force from all other issues and is one of the originalities of the
during experimental measurements and an operational simplicity paper.
for numerical issues. According to Todd et al. [1], the first studies
dealing with a linear cascade of blades date back to the late twen-
ties [2, 3]. The linear cascade was designed for investigations of Equations and numerical method
two-dimensional flow as it predominates in the midspan region A flow of an incompressible fluid trough a passage of a lin-
of a real turbine or compressor stage. In the last four decades, ear cascade can be computed using the Navier-Stokes equations.
the experimental and numerical studies on the linear model have Simulating the flow in the reference system of a rotating blade
made great advances so that most of the effects in the midspan requires accounting for the apparent forces caused by the rota-
region were understood and could be directly transferred to real tion of the reference frame, i.e. the Coriolis force and the cen-
machines. As the secondary flow generates between 30% and trifugal force. For incompressible flow the centrifugal force does
50% of the total aerodynamic losses [4,5] there is still a potential not affect the velocity field as it can be expressed as the gra-
for optimization. Due to good experiences with the linear cas- dient of a potential and can therefore be assigned to a reduced
cade it seems plausible to use this model also for investigating pressure [14], here still denoted p. The resulting equations for
the secondary flow. In the past, numerous experimental and nu- constant density and constant viscosity read
merical studies considered the three-dimensional secondary flow
in both the linear turbine cascade and the linear compressor cas- ∂u 1 1
cade [6–9]. Some interesting effects of the rotor-stator interac- + (uu · ∇) u = − ∇p + ν∇2 u + f cor , (1)
∂t ρ ρ
tion could be shown. However, a link to the real machine or a ∇·u = 0, (2)
transfer of the data to a rotating system are missing. The reason
is that the complex, highly three-dimensional secondary flow in
a linear cascade differs from that in an axial stator [10]. Signifi- where u is the velocity in the rotating frame of reference. The
cant differences of the stagnation pressure loss and the exit flow Coriolis force is given by [15]
deviation could be found between the planar cascade and annular
cascade [11]. Furthermore, additional effects of the centrifugal f cor = −2ρ (Ω
Ω × u) . (3)
force and Coriolis force should be taken into account if a swirl
flow is considered in its relative frame of reference. A rotation of the system around the x-axis, i.e. Ω = (Ω; 0; 0)T
At first sight, the principal influence of the centrifugal force with constant angular speed Ω leads to the simplification
in a rotating row is simple and well-known. The centrifugal
force causes a radial pressure gradient. The velocity inside the T
boundary layer is lower than outside. A secondary flow and a ra- f cor = 2ρ Ω 0; uz ; −uy . (4)
dial migration of the boundary layer is generated due to a radial
imbalance between the centrifugal force and pressure gradient in The in-house code LESOCC 2 (Large Eddy Simulation on
the boundary layer [12,13]. While in a stator, the boundary layer Curvilinear Coordinates version 2) [16] was used to solve (1),(2),
moves to the hub region, it moves towards the blade tip in a ro- and (4) employing a cell-centered finite volume method of sec-
tor [13]. Since, the Coriolis force does not only appear in the ond order with central schemes for the convective and the dif-
profile boundary layer but also in the endwall region, where the fusive fluxes. Block-structured grids with hexagonal cells were
boundary layers undergo changing and the secondary flow arises used. Time advancement in the code was performed by a sec-
between the pressure and suction side of two adjacent profiles, ond order three-stage Runge-Kutta scheme with a pressure cor-
the situation is very complex. In fact, any deviation of the flow rection in the last sub-step. The solver was validated in a large
from a uniform rotating motion in the entire flow field leads to number of earlier studies [16–18]. In [6, 7, 19] the method was
a contribution of the Coriolis force. In particular, the impact of successfully employed for DNS of the flow in a linear turbine
the Coriolis force on the direction of the flow and the level of cascade near the endwall, even with incoming turbulent wakes,
turbulence is practically relevant, as these issues are relevant for and showed good agreement with the corresponding experimen-
pressure losses and heat transfer. Here, the question of whether tal data. For the present study, the code was supplemented with a
and how information from a linear cascade can be transferred to Coriolis force term. This was thoroughly validated as described
the rotating configuration is of major concern. in the following section.

2 Copyright © 2016 by ASME


Validation 1 1 ~
x
A suitable configuration for validating the additional Cori-
olis body force term in (1) is the Dean flow in a rotating pipe, us
0.5
shown in Fig. 1. It was selected since it covers the same effects 0.5 11.832
11.7
as addressed later on with both the curvature of the streamlines

LESOCC 2
10.5
and the Coriolis force generating the secondary flow in the pipe, 0 9
7.5
so that this problem is physically close to the main configuration 6

x
~
investigated below. 0 4.5
3
-0.5 1.5
0
~
-0.5 y
us u ̃y
-1 -0.5 0 0.5 1
u x̃
s ~

Daskopoulos and Lenhoff [21]


y
-1 -0.5 0 0.5 1
z r

x
y
Ω
x
Ω
a R x̃
z y ̃y

FIGURE 1. ROTATING PIPE FLOW CONFIGURATION FOR


VALIDATION. FIGURE 2. 2D RESULTS FOR ROTATING PIPE FLOW: AXIAL
VELOCITY us AND STREAMLINES OF THE SECONDARY FLOW
WITH D = 100 AND Ta = −1000 AND COMPARISON WITH REF-
ERENCE DATA FROM [21].
The setup consists of a circular pipe with the radius of cur-
vature R much larger than the cross section radius a, such that
κ = a/R  1. Density and viscosity are supposed constant. The LESOCC2
Daskopoulos & Lenhoff
flow is driven by imposing a Poiseuille flow at the inlet and com- 12

puted until full development in s (Fig. 1). The curvature of the Ta = +500
streamlines in the pipe generates a positive radial pressure gradi-
ent ∂ p/∂ r leading, when fully developed, to a two-dimensional
8
secondary flow in the cross section of the pipe (Fig. 2). Without Ta = +1000
rotation of the pipe the flow
√ is characterized by the Dean number
uy~

D = −(∂ p/∂ s) a3 /(ρν 2 ) 2κ [20], and a value of D = 100 was


chosen here. Ta = 0
4
This configuration is then extended by rotating the pipe
around the x-axis with constant angular speed Ω, so that the
Coriolis force has a major impact on the secondary flow pat-
tern. This was characterized
√ in [21] using the Taylor number 0
Ta = (2aΩR/ν) 2κ, which can be seen as a non-dimensional -1 -0.5 0 ~ 0.5 1
y
measure of rotational speed, with Ta = 0 identifying the steady
curved pipe described first. Several cases with D = 100, R/a = FIGURE 3. QUANTITATIVE RESULTS FOR ROTATING PIPE
100, and different values of Ta (−1000, 0, 500, 1000) were com- FLOW WITH D = 100 AND Ta ≥ 0 AND COMPARISON WITH REF-
puted. The section covered in s was 0◦ . . . 50◦ . The grid contained ERENCE DATA FROM [21].
5, 445 points in the cross section and 1, 750 points in s direction.
Perfect agreement with the reference data of [21] was achieved
as shown in Fig. 2 and Fig. 3.

3 Copyright © 2016 by ASME


Turbine configuration simulated
As motivated above, the goal of the present work is to high-
light the differences to be expected when inferring from the situa- u∞ α
tion in a linear cascade, featuring secondary flow, to the situation c
in the corresponding rotating machine. To meet this objective
the well investigated linear low pressure turbine (LPT) cascade
T106A [6,8] is assumed to be a rotor (Fig. 4). A simulation of the
linear case is conducted and compared to a simulation in a frame
of reference rotating with constant angular velocity Ω around an
endwall
axis parallel to the x-axis. The rotational frequency was cho-
δ z
sen to be N = 3, 000 rpm which can be seen as a lower limit for
y x
stationary gas turbines. The frequency is made dimensionless ac- c ax
cording to Ω = −2πNcax /u∞ , yielding a value of Ω = −0.337.
The minus sign results from the definition of the coordinate sys-
tem and the technically correct sense of rotation of a turbine.
This case (labeled Rotating) is compared to the reference case
without rotation, Ω = 0, (labeled Non-Rotating). Another case FIGURE 4. GEOMETRY OF T106A TURBINE PASSAGE.
of the same rotation rate but in opposite direction (labeled In-
verse) was computed as well for illustration. Figure 4 shows the
investigated passage with the coordinate system and information wall-normed direction z. A laminar boundary layer is prescribed
on the geometry and boundary conditions. The x-axis is parallel at the inlet according to
to the machine axis, which is also the axis of rotation. The y-
axis points in circumferential direction. The blades are assumed 
z  z 3  z 4
being mounted on the hub. |uu|1 (z) 2 − 2 + for z ≤ δ
= δ δ δ (6)
All given lengths or positions are normalized by the ref- u∞ 1 for z > δ ,
erence length Lref := cax ≈ 85.88mm, which is defined by the
physical axial chord length cax . The height of the computational
domain is cax , with a symmetry condition imposed at this posi- which is a good approximation of the Blasius boundary layer
tion to limit the computational cost. A no-slip condition is ap- with boundary layer thickness δ [22]. Here, u∞ represents
plied at the endwall located at z = 0. Furthermore, a periodic the freestream velocity. Velocity values are normalized by this
boundary condition in the pitchwise direction y and a convective freestream velocity i.e. uref := u∞ = 80.0m/s. At this point,
boundary condition at the outlet at x = 2 cax are applied. the value δ is chosen to ensure that significant secondary flow
The geometry described is the one of the linear cascade fea- is generated, in particular the horseshoe vortex. Moreover, the
turing parallel blades. In a rotating machine the blades would be boundary layer thickness is chosen representative for existing en-
positioned at an angle pointing in radial direction. In the present gines, here δ = 14.6mm = 0.17 cax . By doing so, the displace-
study the geometry is not altered between the Non-Rotating and ment thickness and the momentum thickness are δ1 = 0.051 cax
the Rotating case. This corresponds to the limit of very large and δ2 = 0.02 cax , respectively. The Reynolds number based on
hub radius. Note that the Coriolis force term does not contain the axial chord length and the inlet freestream velocity equals
the radius as an input. Hence, any difference in the flow is only 20, 000. This is lower than in actual machines but allows per-
caused by the Coriolis force without interference with geomet- forming DNS, which is necessary to resolve the transitional flow
rical changes. Maintaining the geometry, is seen as a particular in this setting at reasonable cost. Figure 5 depicts the numerical
advantage here as it allows clean conclusions and easy reproduc- grid. The total number of the finite volume cells equals approxi-
tion by others. mately 59 million. The wall resolution in wall units at the suc-
In the same spirit, the inlet velocity is imposed independent tion side is n+ + +
1 < 0.30, ∆t < 0.6, and at the endwall z1 < 0.58
+
of the rotation rate at x = −0.7 cax at the inlet and z1,max = 4.00 the maximal value in the passage.
Here, the index 1 identifies the distance of the cell center of the
wall-adjacent cell from the wall, while ∆t + refers to the cell size
u|1 = (cos α |uu|1 , sin α |uu|1 , 0)T , (5)
in tangential direction, and n to the normal direction along the
blade. The mesh generator developed by Wu [23] was used to
where α denotes the inflow angle. For the T106A turbine profile generate this grid.
with a stagger angle of 59.28◦ , zero incident flow is obtained with For the case Non-Rotating, a start-up period of eleven pas-
α = 37.7◦ . The absolute inflow velocity is made a function of the sage flow-through times (cax /u∞ ) was computed. The statistics

4 Copyright © 2016 by ASME


1 pletely. Near the endwall, the plot of the TKE reveals that the
turbulence is only generated by the mutual interaction of parts
of the horseshoe vortex, the secondary cross flow and the inflow
0.5
boundary layer. Upstream and downstream of the near-wall re-
gion, influenced by the horseshoe vortex, no TKE generation can
y/cax

be detected. The upstream flow and the boundary layer are lami-
nar so that before transition only little fluctuations around vortex
0 A and B are observed. Only close to B slight perturbations can
be seen near the leading edge. The same holds for the corner
region at the suction side related to the other leg of the horse-
-0.5
shoe vortex. Further downstream, around x ≈ 0.5 cax . . . 0.6 cax ,
massive turbulent fluctuations are observed in the region of vor-
-0.5 0 0.5 x/cax 1 1.5 2 tices A and B which turn around each other and eventually merge
(Fig. 6(e)). The resulting turbulent vortex as a whole is further
FIGURE 5. NUMERICAL GRID FOR TURBINE PASSAGE, WITH lifted away from the wall and impacts on the suction side over a
ONLY ONE OUT OF EIGHT GRID LINES PLOTTED. broad range. Still further away from the endwall the flow along
the suction side of the blade remains laminar almost until the rear
end of the blade. Here, the adverse pressure gradient causes an
were collected over a period of twelve flow-through times. The open separation. The separated flow is unstable and undergoes
time step was determined dynamically to satisfy a CFL number transition. Due to the relatively low Reynolds number the flow
smaller than 0.6 yielding ∆t ≈ 3.0 · 10−4 cax /u∞ . This simula- does not attach again.
tion consumed 77 hours wall clock time on a high-performance Figures 6(a) and 6(d) depict the case Rotating with activated
system with 336 cores (Intel E5-2690, Sandy Bridge). The re- Coriolis force. Here, the secondary flow pattern differs consid-
sults of the simulation Non-Rotating were used as initial con- erably from the case Non-Rotating. The two stable parts of the
dition for both cases Rotating and Inverse. In these cases, the horseshoe vortex seen in Fig. 6(b) are transformed into a single
simulation was conducted for seven flow-through times as well large turbulent structure (Fig. 6(a)). The position of the pressure
before time averaging was, again, performed over twelve passage side leg of the horseshoe vortex A in the passage is similar to the
flow-through times. corresponding vortex A in the reference case, and the vortex is is
larger. A comparison between Fig. 6(d) and 6(e) shows that the
vortex system impinging on the suction side is lifted up towards
Vortex systems the midspan region compared to the case Non-Rotating. As an
Figure 6 provides an overview over all three cases computed additional result of this lift up, the near-wall fluid is less turbu-
showing instantaneous vortex structures by means of the λ2 - lent from the center of the passage even until the wake region of
criterion [24] and near-wall Turbulent Kinetic Energy (TKE). blade.
The case Non-Rotating, shown in Fig. 6(b) and 6(e) is taken as Finally, Fig. 6(c) and 6(f) show the same data for the case
a reference. Two dominant vortices, labeled A and B, rotate Inverse. A rotation of the turbine in opposite direction is tech-
counter-clockwise with respect to their flow direction. They both nically not realistic but addressed here for illustration. Figure
represent the pressure side leg of the horseshoe vortex generated 6(c) and 6(f) for Ω > 0 show exactly the opposite tendency com-
by the incoming boundary layer. Due to the vicinity of the end- pared to the same rotation rate in the other direction (Fig 6(a)
wall, vortex A and vortex B induce tertiary vortices upstream of and 6(d)): Turbulence is largely reduced apart from the endwall.
their position. The values of λ2 for the iso-surface was chosen Furthermore, the vortex system is pushed towards the endwall, so
so as to provide a clear view, which does not, on the other hand that the near-wall TKE magnitude increases, which is detectable
detect weaker vortices. A closer analysis (not shown here) iden- again even in the wake region, seen in the picture.
tifies the horseshoe vortex as a system of even more vortices, also
observed by Wang et al. [25]. Both vortices, A and B, are trans-
ported towards the suction side by the secondary cross flow at the Mechanism of vortex destabilization
endwall which is generated by the pressure gradient in pitchwise The marked effect which the Coriolis force has on the tran-
direction. Vortex A is lifted along the profile and away from the sition of the horseshoe vortices can be explained as follows using
wall in addition. Vortex B remains closer to the endwall and does the sketches in Fig. 7. Assume the axis of the vortex being in x-
not interact with vortex A until it reaches the suction side of the direction. Then, it is parallel to the axis of rotation of the system.
adjacent blade. Hence, although the horseshoe vortex occurs as a According to (3) the resulting Coriolis force is perpendicular to
single effect, the behavior of its individual parts can differ com- the axis of rotation and the velocity of the horseshoe vortex. In

5 Copyright © 2016 by ASME


Z

X
uz/uref TKE/uref
2
Y
1.4 0.07
1.0 0.06
0.6 0.05
0.2 0.04
A -0.2 B 0.03
0.02
B

__
-0.6

__
__

0.01

__
-1.0

__
__

0.00

_
_
__
__

___ ___
A A
(a) Rotating: Ω < 0 (frontal view) (b) Non-Rotating: Ω = 0 (frontal view) (c) Inverse: Ω > 0 (frontal view)

X
Y

2(
w⃗
2 ( ×ω
w⃗ × ⃗ )
B ω
⃗)

A B A
___

__
_
_
(d) Rotating: Ω < 0 (lateral view) (e) Non-Rotating: Ω = 0 (lateral view) (f) Inverse: Ω > 0 (lateral view)

FIGURE 6. TWO VIEWS OF THE COHERENT VORTEX STRUCTURES REPRESENTED BY THE λ2 -CRITERION (TOP ROW: FRONTAL
VIEW; BOTTOM ROW: LATERAL VIEW). THE ISOSURFACE λ2 = −500 IS COLORED BY THE SPANWISE VELOCITY uz . A PLANE NEXT
TO THE ENDWALL SHOWS NEAR-WALL TKE BY A CONTOUR PLOT AND FLUID MOTION VIA SURFACE STREAMLINES. THE PRES-
SURE SIDE SURFACE OF THE PROFILE AND BLOCKS DOWNSTREAM OF THE PASSAGE ARE REMOVED FOR BETTER OVERVIEW.
VORTICES A AND B ARE DISCUSSED IN THE TEXT.

the case Rotating the Coriolis force always points outwards in


Vortex
Vortex
centrifugal direction. It thus induces an expansion of the vortex,
reduces the pressure on the axis of the vortex and destabilizes it
such that transition is supported (Fig. 7(a)). As a result the vortex Vortex
Vortex f fcorcor f corf cor
becomes turbulent much earlier. In contrast, Fig. 7(b) shows the
case with opposite rotation of the vortex and the frame of ref-
erence. Here, the Coriolis force acts in centripetal sense. This
causes the stabilization and the compression of the vortex. In-
deed, the eddies shown in Fig. 6(c) and 6(f) for the case Inverse Ω<0 Ω>0
Ω>0
Ω<0
are very stable. They only destabilize after interaction with the (a) (b)
boundary layer of the suction side of the adjacent profile. Figure
6 shows that the horseshoe vortices are not perfectly aligned with FIGURE 7. SKETCH OF THE IMPACT OF THE CORIOLIS
the x-axis, but they have a substantial component in this direction FORCE ON A VORTEX IN DIRECTION OF THE AXIS OF ROTA-
so that the argument holds. TION.

6 Copyright © 2016 by ASME


1
<uz>/uref
0.8 0.8
0.7
0.6
0.5
0.4 0.6

z/cax
0.3
0.2
0.1
0
0.4
-0.1
-0.2
-0.3 0.2
-0.4
-0.5

0
1 0.8 0.6 0.4 0.2
y/cax
(a) Rotating: Ω < 0 (b) Non-Rotating: Ω = 0 (c) Inverse: Ω > 0

FIGURE 8. STREAMLINES OF THE MEAN SECONDARY FLOW, AS DEFINED IN THE TEXT, IN THE PLANE x = 0.5 cax FOR THE THREE
CASES INVESTIGATED. THE MAGNITUDE OF THIS VELOCITY CAN BE ASSESSED BY THE VALUES OF THE VERTICAL COMPONENT
< uz > SHOWN IN FORM OF A CONTOUR PLOT IN ALL GRAPHS.

Large-Scale secondary flow the separation at x ≈ 0.75 cax . The separation is laminar (hence
The displacement of the vortices discussed in relation to not visible in Fig. 6(e)) with subsequent transition and turbulence
Fig. 6 is due to a large-scale secondary motion induced by the only close to the trailing edge. This is a consequence of the low
Coriolis force illustrated in Fig. 8. This figure shows streamlines Reynolds number here. For the case Rotating (Fig. 9(d)) the re-
of the secondary flow (< uy >sec , < uz >sec ) with gion of distorted fluid motion near the suction side is significantly
increased in spanwise direction. Additionally, the streamlines
here experience an upward deflection, caused by the large-scale
< uy >sec = < uy > − < uy >midspan ,
flow discussed above (Fig. 8). For inverted rotation (case In-
< uz >sec = < uz > , (7) verse), Fig. 9(f) depicts a downward redirection of the near-wall
flow beyond x ≈ 0.5 cax . The strong spanwise secondary flow
where < uy >midspan is the value of < uy > of the case Non- suppresses the separation of the flow on the suction side in both
Rotating at z ≈ cax and a function of y. Figure 8(b) shows the cases with rotation.
situation for the case Non-Rotating. Apart from the localized Now the situation on the pressure side is discussed. In
vortex, centered around y ≈ 0.36 cax and z ≈ 0.11cax , at this ax- Fig. 9(b), considering the case Non-Rotating, again a small area
ial position no large-scale motion is seen. For the case Rotating, of flow separation is visible in the corner of leading edge and
instead, large-scale motion is seen, albeit at an intensity smaller endwall (red line). In the case Rotating the near-wall flow along
than the localized vortex. This flow lifts up vortex A such that its the pressure side at the leading edge is in upstream direction. In
center now occurs at y ≈ 0.37 cax , z ≈ 0.15 cax . the case Inverse practically no separation occurs at the leading
Figure 8(c) shows that the position of the horseshoe vortex edge. For both rotating cases the radial deflection of streamlines
has changed to y ≈ 0.32 cax , z ≈ 0.07 cax at x = 0.5 cax . Further- is inverted compared to the suction side. Additionally, Fig. 9(c)
more its size is smaller and the intensity lower. Also, the large- shows the influence of the upper symmetry boundary condition
scale secondary flow determined according (7) turns in opposite for span heights above z > 0.9 cax .
direction compared to Fig. 8(a). Figures 9(g) - (i) show the magnitude of the mean wall
shear stress, τw , at the endwall in the passage together with wall
streamlines. Near the suction side the level of τw is small in the
Flow near solid surfaces case Rotating, larger in the case Non-Rotating, and still larger
The magnitude of the wall shear stress and associated in the case Inverse. This is due to the different spanwise lift-up
streamtraces on pressure side, suction side, and endwall are of the horseshoe vortex in the different cases (Fig. 6). Remote
shown in Fig. 9. The suction side is discussed first: For the case from this area the level of τw behaves similarly, which is due
Non-Rotating, Fig. 9(e), the region where the horseshoe vortex to reduced near-wall mean velocity in the case Rotating (and in-
impinges on the suction side is clearly recognizable by the de- creased in the case Inverse), as seen in Fig. 13(a), (b) below. Also
viation of the wall streamlines. The shear stress has its maxi- the streamlines show that the flow near the endwall changes when
mum in this area. Further away from the endwall, the horizon- the Coriolis force is present. In the case Non-Rotating (Fig. 9(h))
tal streamlines confirm a mostly undisturbed flow, terminated by

7 Copyright © 2016 by ASME


LE 7E LE Z
TE LE TE
X
Y

(a) Rotating: Ω < 0 (pressure side) (b) Non-Rotating: Ω = 0 (pressure side) (c) Inverse: Ω > 0 (pressure side)

LE 7E LE Z
TE LE TE
X
Y

(d) Rotating: Ω < 0 (suction side) (e) Non-Rotating: Ω = 0 (suction side) (f) Inverse: Ω > 0 (suction side)
Y

SP2 SP1
SP1 SP1 Z
X
_____

_____
_____
_____

_____

_____
_____
_____

X
_

X
_

X X

(g) Rotating: Ω < 0 (endwall) (h) Non-Rotating: Ω = 0 (endwall) (i) Inverse: Ω > 0 (endwall)


w 0 0.005 0.01 0.015 0.02

FIGURE 9. AVERAGE FLOW: MAGNITUDE OF THE WALL SHEAR STRESS, τw , AND SURFACE STREAMLINES AT THE PROFILE
SURFACE AND AT THE ENDWALL. THE RED CONTOUR LINES IN (a) - (f) INDICATE WHERE THE x-COMPONENT OF THE MEAN
WALL SHEAR STRESS VANISHES. IN THE VIEWS OF THE ENDWALL, THE DOMAIN WAS DUPLICATED.

8 Copyright © 2016 by ASME


a saddle point can be located at x ≈ −0.38 cax , y ≈ 0.72 cax the separation at the suction side. However, the strong secondary
(SP1). Regarding the case Rotating, the saddle point SP1 moves flow generates more losses than the open flow separation in the
away from the leading edge (x ≈ −0.4 cax , y ≈ 0.68 cax ) and case without the Coriolis force. Finally, an analysis of the exit
additionally a second saddle point is formed at x ≈ −0.03 cax , flow angle (not presented here) depicts that the entire exit plane
y ≈ 0.79 cax (SP2). In contrast, for the case Inverse, the saddle is affected by the large-scale secondary motion in the case Ro-
point SP1 shifts towards the leading edge, here to x ≈ −0.31 cax , tating and the deviation from the design exit flow angle can be
y ≈ 0.77 cax . observed at each z-location. For the case Inverse, in contrast, the
influence on the exit flow angle is confined to the endwall region
and the upper boundary.
Losses
The stagnation pressure loss coefficient is one of typical per-
formance criteria. For an incompressible fluid, it can be writ- Pitchwise averaged statistics
ten as ζ2 = (P1 − P2 )/(0.5ρu22 ), where P = p + 0.5ρu2 is the Temporal and pitchwise averaged results are shown in
stagnation pressure. Here, ζ2 is defined in the relative frame of Fig. 11 - 13. Here, velocity components, the magnitude of the
reference to be consistent between the three different cases pre- velocity, the Turbulent Kinetic Energy (TKE) and the static pres-
sented. The comparison of the loss coefficient in Fig. 10 reflects sure are presented for the three cases studied. Figures 11, 12 and
the massive differences caused by the Coriolis force. In the case 13, depict the same data at different locations in the flow, x = 0,
Non-Rotating (Fig. 10(b)), the wake of the profile is broad due to x = 0.5 cax and x = cax , respectively.
the separation at the suction side. The interaction of the horse- In Fig. 11, the deviation of the data for the case Rotating
shoe vortex, the secondary cross flow and the boundary layer from those of the reference case Non-Rotating can be observed
result in a core of increased losses in the region z ≈ 0.2 . . . 0.5 cax for all values reported. The Coriolis force occurs upstream the
and y ≈ 0 . . . 0.5 cax . Below the core, a small region in the wake passage due to the velocity component uy (Fig. 11(b)) according
with lower stagnation pressure losses appears because the lami- to (3). It generates a spanwise velocity uz (Fig. 11(c)), which
nar secondary flow predominates between the core of the vortex is up to 10% of the velocity magnitude. Due to the symmetry
and the endwall, i.e. for z ≈ 0.05 . . . 0.2 cax and y ≈ −0.2 cax . . . 0. condition at midspan, the Coriolis force then indirectly causes a
Closer to the endwall, losses increase again due to the corner radial redistribution of the flow. For the case Rotating, the axial
vortex between the endwall and the suction side. ζ2 is relatively velocity undergoes a deceleration in the region of the boundary
small outside the wake, because no further effects appear. layer and an acceleration in the midspan area, as shown in the
Figure 10(a) depicts the effects of the large-scale secondary Fig. 11(a). A further effect of the Coriolis force can be ob-
motion induced by the Coriolis force in the case Rotating. The served in Fig. 11(f). A positive radial pressure gradient occurs
pressure loss core is shifted towards midspan. It is stronger for the case Rotating. It is created by the velocity in pitchwise
and its size is larger than in the Non-Rotating case due to in- direction, uy , according to ∂p/∂t ≈ −2Ωuy with its bulk value
creased secondary flow and earlier transition of the horseshoe −2Ωuy ≈ 0.24. In the region beyond the boundary layer and
vortex. The large-scale motion (Fig. 8(a)) generates the trailing the secondary flow, the radial pressure gradient is in the range
edge vortex behind the profile via the strong opposite flow along of ±[0.25 . . . 0.4], with positive values for the case Rotating and
the rear part of the blade, positive in z on the suction side, neg- negative values for the case Inverse. Recall that the centrifugal
ative on the pressure side. It yields higher losses in the region force was not retained in the simulation as it has a potential and
z ≈ 0.05 . . . 0.3 cax and y ≈ −0.20 . . . 0.05 cax . The level of ζ2 then is compensated by an additional hydrostatic pressure gra-
outside the regions affected by the wake is higher than in the dient without inducing a corresponding flow. This was verified
reference case because of the large-scale secondary motion. numerically by a test with the centrifugal term retrained yielding
exactly the same  flow as without in the validation exercise de-
Again, in the case Inverse (Fig. 10(c)), the Coriolis force has
scribed above. The origin of the horseshoe vortex is a wall nor-
the opposite effect compared to the case Rotating. The region of
mal gradient of the stagnation pressure along the leading edge of
higher ζ2 is due to the horseshoe vortex and remains near the end-
the blade, generated by the boundary layer velocity profile. The
wall. Closer to the wall, the trace of the corner vortex is visible.
additional radial pressure gradient created by the Coriolis force
Since no separation occurs, the wake can be clearly identified in
(Fig. 11(f)) now adds up to the former and can influence the gen-
ζ2 and is thin compared to the other two cases.
eration of the horseshoe vortex at the leading edge. Quantifica-
Span- and pitchwise averaged values of ζ2 are 0.06, 0.13 and
tion of this effect is difficult, though. Hence, the Coriolis force
0.12 for the cases Non-Rotating, Rotating, and Inverse, respec-
has not only a direct impact on the horseshoe vortex as shown
tively. The flow of the case Non-Rotating generates the smallest
in Fig. 6 and described above, but also an indirect effect by the
stagnation pressure losses, despite the fact that the flow sepa-
influence of the radial pressure gradient.
rates at the suction side for this case only. The large-scale mo-
tion caused by the Coriolis force leads to the disappearance of

9 Copyright © 2016 by ASME


2

0.6
0.5
0.4
0.3
0.2
0.1
0

(a) Rotating: Ω < 0 (b) Non-Rotating: Ω = 0 (c) Inverse: Ω > 0

FIGURE 10. STAGNATION PRESSURE LOSS COEFFICIENT ζ2 DOWNSTREAM THE PASSAGE AT x = 1.2 cax .

1 1

0.8 0.8
Non-Rotating
Rotating 0.6

z / cax
0.6
z / cax

Inverse

0.4 0.4

0.2 0.2

0 0 -0.8 -0.6 -0.4 -0.2


0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0 0.2 0.4 0.6 0.8 1 1.2
<ux>/uref <uy>/uref <ux>/uref <uy>/uref

(a) (b) (a) (b)


1 1

0.8 0.8

0.6
z / cax

0.6
z / cax

0.4 0.4

0.2 0.2

0 0 0 0.5 1 1.5 2
-0.1 0 0.1 0.2 0.3 0.4 0 0.5 1 1.5 2 -0.1 0 0.1 0.2 0.3 0.4
<uz>/uref <|u|>/uref <uz>/uref <|u|>/uref
(c) (d) (c) (d)
1 1
Non-Rotating
0.8 0.8
Rotating
Inverse
0.6 0.6
z / cax
z / cax

0.4 0.4

0.2 0.2

0 0 -0.5 0 0.5 1 1.5


0 0.02 0.04 0.06 0.08 -0.5 0 0.5 1 1.5 0 0.02 0.04 0.06 0.08
TKE / u ref2 <p> TKE / u ref2 <p>
(e) (f) (e) (f)

FIGURE 11. RADIAL DISTRIBUTION OF TIME- AND PITCH- FIGURE 12. RADIAL DISTRIBUTION OF TIME- AND PITCH-
WISE AVERAGED QUANTITIES AT x = 0: (a) - (c) VELOCITY WISE AVERAGED QUANTITIES AT x = 0.5 cax : (a) - (c) VELOC-
COMPONENTS, (d) VELOCITY MAGNITUDE, (e) TKE AND (f) ITY COMPONENTS, (d) VELOCITY MAGNITUDE, (e) TKE AND
STATIC PRESSURE. (f) STATIC PRESSURE.

10 Copyright © 2016 by ASME


1
In the cases investigated a laminar boundary layer was im-
posed at the inlet. The Turbulent Kinetic Energy can hence only 0.8
be generated by the breakdown of the large horseshoe vortex
0.6

z / cax
structures, by the dynamic motion of them, or by transition in
boundary layers. The plot in Fig. 11(e) supported by Fig. 6 shows 0.4
that the TKE increases in the region of the horseshoe vortex. In
0.2
the case Rotating, it breaks down at a very early stage. Hence, the
TKE increases very fast in the leading edge region. In the case 0
0 0.2 0.4 0.6 0.8 1 1.2 -1.8 -1.6 -1.4 -1.2 -1
Inverse, the horseshoe vortex remains stable over a long stretch. <ux>/uref <uy>/uref
Compared to the reference case Non-Rotating the maximal value (a) (b)
of the TKE in Fig. 11(e) is increased by a factor of 1.3 for the 1
case Rotating and decreased by a factor of 3 for the case Inverse.
When passing the passage, the flow pattern becomes much 0.8

more complicated, so that the distinction between the cause and 0.6

z / cax
the effect becomes more difficult. Figures for the radial veloc-
0.4 Non-Rotating
ity at the middle of the passage (Fig. 12(c)) and of the trailing Rotating
edge (Fig. 13(c)) show an upward radial flow for the case Rotat- Inverse
0.2
ing and a downward flow for the case Inverse. These radial flows,
0
also seen in Fig. 8, are the reason for the different radial loca- -0.1 0 0.1 0.2 0.3 0.4 0 0.5 1 1.5 2
<uz>/uref <|u|>/uref
tions of secondary flow structures shown in Fig. 6(d-f) discussed
above. The persistent radial flow results in larger deviations of (c) (d)
the axial velocities of cases Rotating and Inverse from the case 1
Non-Rotating as shown in the Fig. 12(a). The position of the lo-
0.8
cal minimum of the static pressure in Fig. 12(f) corresponds to
the location of the horseshoe vortex. This graph, hence, clearly 0.6
z / cax

demonstrates the impact of the Coriolis force on the radial dis- 0.4
placement of the vortex discussed above with Fig. 6. In Fig. 12(e)
a marked increase of the TKE can be observed for the case Non- 0.2
Rotating at z ≈ 0.08 cax . Figure 6(e) illustrates that at this axial 0
0 0.02 0.04 0.06 0.08 -0.5 0 0.5 1 1.5
position the transition of the vortex to turbulence occurs. TKE / u ref2 <p>
At the outlet x = cax (Fig. 13), the radial distribution of the
(e) (f)
axial and pitchwise velocities are affected by the Coriolis force
but in a very complicated way so that a detailed discussion is te- FIGURE 13. RADIAL DISTRIBUTION OF TIME- AND PITCH-
dious. However, the radial distributions of the TKE and the static WISE AVERAGED QUANTITIES AT x = cax : (a) - (c) VELOCITY
pressure confirm the observations made above. In the case Non- COMPONENTS, (d) VELOCITY MAGNITUDE, (e) TKE AND (f)
Rotating, a double peak of the TKE reveals the position of the STATIC PRESSURE.
horseshoe vortex. The fairly constant, elevated level of the TKE
beyond z ≈ 0.8 cax results from the transition of the separated
boundary layer flow at the suction side of the profile discussed Surface pressure on the blade
above. In fact, the boundary layer even separates in this region. Figure 14 shows the different pressure distributions on the
For the case Rotating the double peak in the TKE is observed blade for all three cases at a span height of z = 0.1 cax , 0.4 cax ,
further away from the wall. A small region of the turbulent flow and 0.8 cax . The plotted static pressure coefficient is computed as
is observed in the upper part, certainly influenced by the condi- Cp = (p − p2 )/(P1 − p2 ), where p is the local static pressure on
tion at the upper boundary, though. In this case separation of the the blade, p2 the pitchwise and spanwise averaged static pressure
boundary layer along the suction side wall is suppressed as dis- at the outlet of domain and P1 the average total pressure at the
cussed before with the wall shear stress in Fig. 9. For the case inlet of domain. Note that for Cp the same discussion applies as
Inverse the TKE profile shows a lower position of the vortex. for ζ2 on page 9. All considerations only refer to the relative
Furthermore, the suction side boundary layer remains fully lami- frame of reference, so the dynamic pressure contribution in P1
nar and attached in the wide range of span z > 0.35 cax . Similarly does not account for the rotational velocity, which is present in
as with Fig. 12(f), the local minimum in the mean static pressure both rotating cases.
of Fig. 13(f) relates to the vortex center, again illustrating the dis-
location.

11 Copyright © 2016 by ASME


1 1 1

Non-Rotating
0.5 0.5 Rotating 0.5
Inverse
Cp

Cp

Cp
0 0 0

-0.5 -0.5 -0.5

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x/cax x/cax x/cax

(a) z = 0.1 cax (b) z = 0.4 cax (c) z = 0.8 cax

FIGURE 14. WALL STATIC PRESSURE COEFFICIENT AROUND THE PROFILE AT THREE DIFFERENT z-POSITIONS.

In all three cases the Cp distribution on the pressure side is study shows that issues related to secondary flow can perhaps
quantitatively the same and just shifted between the cases. On the not be represented well. Beyond the induced secondary flow, the
contrary, there is a significant change on the suction side: in the transition behavior of coherent vortices is substantially altered
case Rotating the average level of the wall pressure coefficient with rotation.
decreases with z. For inverse rotation the trend is the opposite. The paper provides a simple yet instructive test case high-
This leads to an increased blade load on the upper part of the lighting the phenomenon. This setup together with the numerous
blade in the case Rotating and on the lower part of the blade in statistical data provided constitutes an efficient validation case
the case Inverse. for other methods, such as RANS or hybrid models. In future
work, a geometry closer to the annular cascade will be simulated
with more realistic inlet boundary layers, free stream turbulence
Conclusions and with higher Reynolds numbers.
To assess the impact of the Coriolis force on the secondary
flow in a linear turbine cascade three cases were investigated and
compared. Starting from a reference case without Coriolis force
ACKNOWLEDGMENT
(Ω = 0) two additional cases with rotation (Ω < 0) and counter
rotation (Ω > 0) of the passage were computed and a substantial The present work is funded by the German Research Foun-
change of the flow with rotation was observed. dation (DFG) under grant FR 1593/6-2. The computations were
Altogether, the Coriolis force is of major importance for the performed on a Bull Cluster at the Center for Information Ser-
secondary flow at the endwall and the whole passage. Further- vices and High Performance Computing (ZIH) at TU Dresden.
more, the transition behavior of the horseshoe vortex is changed The authors also thank S. Heitkam for helpful comments on the
substantially. As a result, the internal structure of the horseshoe manuscript.
vortex is changed between the Non-Rotating and the Rotating
case. A consequence of all these effects is also that the horseshoe
vortex is lifted with rotation and that separation and transition on REFERENCES
the suction side are altered. Specifically, the stronger transition [1] Todd, K., Eng., M., and A.M.I.Mech.E., 1947. “Practical
of the horseshoe vortex yields an overall increased level of losses. aspects of cascade wind tunnel research”. Proceedings of
Although the detailed interaction between all involved flow fea- the Institution of Mechanical Engineers, 157, pp. 482–497.
tures is hard to analyze and to quantify, the differences between [2] Stodola, A., 1927. Steam and Gas Turbines. McGraw-Hill.
the cases examined are so significant that the simulations truly [3] Habris, R. G., and Fairthorne, R. A., 1928. “Reports and
demonstrate the impact of the Coriolis force on the secondary memoranda, no. 1206”. Aeronautical Research Committee,
flow pattern and the losses in a turbine passage. This sheds new 1(33), p. 286.
light on the correspondence between linear cascade and rotating [4] Fottner, L., 1989. “Review on turbomachinery blading de-
turbine stage. Hence, while two-dimensional flow in a rotating sign process”. AGARD Lecture Series: Blading Design for
machine can well be modeled with a linear cascade, the present Axial Turbomachines, 167.

12 Copyright © 2016 by ASME


[5] Sharma, O., and Butler, T., 1987. “Predictions of endwall Mathematics and Mechanics, 13, pp. 311–312.
losses and secondary flows in axial flow cascades”. Journal [20] Daskopoulos, P., and Lenhoff, A., 1989. “Flow in curved
of Turbomachinery, 109, pp. 229 – 236. ducts: bifurcation structure for stationary ducts”. Journal
[6] Koschichow, D., Kirik, I., Fröhlich, J., and Niehuis, R., of Fluid Mechanics, 203, pp. 125–148.
2014. “DNS of the flow near the endwall in a linear low [21] Daskopoulos, P., and Lenhoff, A., 1990. “Flow in curved
pressure turbine cascade with periodically passing wakes”. ducts. part 2. rotating ducts”. Journal of Fluid Mechanics,
ASME Paper GT2014-25071. 217, pp. 575–593.
[7] Koschichow, D., Ciorciari, R., Fröhlich, J., and Niehuis, R., [22] Pohlhausen, K., 1921. “Zur näherungsweisen Integration
2015. “Analysis of the influence of periodic passing wakes der Differentialgleichung der laminaren Grenzschicht”.
on the secondary flow near the endwall of a linear LPT cas- ZAMM - Journal of Applied Mathematics and Mechanics
cade using DNS and U-RANS”. ETC2015-151, Proceed- (in German), 1(4), pp. 252–290.
ings of the 11th European Conference on Turbomachinery [23] Wu, X., and Durbin, P., 2001. “Evidence of longitudinal
Fluid Dynamics and Thermodynamics. vortices evolved from distorted wakes in a turbine passage”.
[8] Kirik, I., and Niehuis, R., 2015. “Experimental investiga- Journal of Fluid Mechanics, 446, pp. 199–228.
tions on effects of unsteady wakes on the secondary flows [24] J.Jeong, and Hussain, F., 1995. “On the identification of a
in the linear T106 turbine cascade”. ASME Paper GT2015- vortex”. Journal of Fluid Mechanics, 285, pp. 69–94.
43170. [25] Wang, H. P., Olson, S. J., Goldstein, R. J., and Eckert, E.
[9] Krug, A., Busse, P., and Vogeler, K., 2015. “Experimental R. G., 1997. “Flow visualization in a linear turbine cascade
investigation into the effects of the steady wake-tip clear- of high performance turbine blades”. Journal of Turboma-
ance vortex interaction in a compressor cascade”. Journal chinery, 119(1), January, pp. 1–8.
of Turbomachinery, 137, June.
[10] Cumpsty, N. A., 1989. Compressor Aerodynamics. Long-
man Scientific and Technical.
[11] Moustapha, S. H., Paron, G. J., and Wade, J. H. T.,
1985. “Secondary flows in cascades of higly loaded tur-
bine blade”. Journal of Engineering for Gas Turbines and
Power, 107, October, pp. 1031–1038.
[12] Scholz, N., 1965. Aerodynamik der Schaufelgitter. Braun
Verlag.
[13] Amecke, J., and Kost, F. “Rotating annular cascade”. Ad-
vanced Methods for Cascade Testing, AGARD-AG, 328.
[14] Greenspan, H., 1990. The Theory of Rotating Fluids. At
the University Press.
[15] Vavra, M., 1974. Aero-thermodynamics and flow in turbo-
machines. R.E. Krieger Pub. Co.
[16] Hinterberger, C., Fröhlich, J., and Rodi, W., 2008. “2D
and 3D turbulent fluctuations in open channel flow with
Reτ =590 studied by Large Eddy Simulation”. Flow, Tur-
bulence and Combustion, 80(2), pp. 225–253.
[17] Wang, P., Fröhlich, J., Michelassi, V., and Rodi, W., 2007.
“Large Eddy Simulation of variable density turbulent ax-
isymmetric jets”. In Proceedings of 5th Int. Symp. on Tur-
bulent Shear Flow Phenomena, R. Friedrich, N. Adams,
J. Eaton, J. Humphrey, N. Kasagi, and M. Leschziner, eds.,
pp. 1049–1054.
[18] Wissink, J., and Rodi, W., 2008. “Numerical study of the
near wake of a circular cylinder”. International Journal of
Heat and Fluid Flow, 128(4), Oktober, pp. 668–678.
[19] Koschichow, D., Fröhlich, J., Ciorciari, R., Kirik, I., and
Niehuis, R., 2013. “Numerical and experimental investi-
gation of the turbulent flow through a low-pressure turbine
cascade in the endwall region”. Proceedings in Applied

13 Copyright © 2016 by ASME

View publication stats

You might also like