Pogorelov Et Al 2015 Cut-Cell Method Based Large-Eddy Simulation of Tip-Leakage Flow PDF

Cut-cell method based large-eddy simulation of tip-leakage flow
Alexej Pogorelov, Matthias Meinke, and Wolfgang Schröder
Citation: Physics of Fluids 27, 075106 (2015); doi: 10.1063/1.4926515

View online: http://dx.doi.org/10.1063/1.4926515
View Table of Contents: http://scitation.aip.org/content/aip/journal/pof2/27/7?ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in

An unstructured mesh arbitrary Lagrangian-Eulerian unsteady incompressible flow solver and its
application to insect flight aerodynamics
Phys. Fluids 28, 061901 (2016); 10.1063/1.4949547
Large-eddy simulation of 3-D corner separation in a linear compressor cascade

Phys. Fluids 27, 085105 (2015); 10.1063/1.4928246
Subgrid-scale eddy viscosity model for helical turbulence

Phys. Fluids 25, 095101 (2013); 10.1063/1.4819765
Large-eddy simulations of a turbulent Coanda jet on a circulation control airfoil

Phys. Fluids 22, 125105 (2010); 10.1063/1.3526757
High-order large-eddy simulation of flow over the “Ahmed body” car model
Phys. Fluids 20, 095101 (2008); 10.1063/1.2952595
Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions. Downloaded to IP:
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
PHYSICS OF FLUIDS 27, 075106 (2015)
Cut-cell method based large-eddy simulation

of tip-leakage flow
Alexej Pogorelov,a),b) Matthias Meinke,b),c) and Wolfgang Schröderb),d)
Institute of Aerodynamics and Chair of Fluid Mechanics, RWTH Aachen University,
Wüllnerstraße 5a, 52062 Aachen, Germany
(Received 20 March 2015; accepted 24 June 2015; published online 9 July 2015)
The turbulent low Mach number flow through an axial fan at a Reynolds number of
9.36 × 105 based on the outer casing diameter is investigated by large-eddy simu-
lation. A finite-volume flow solver in an unstructured hierarchical Cartesian setup
for the compressible Navier-Stokes equations is used. To account for sharp edges, a
fully conservative cut-cell approach is applied. A newly developed rotational periodic
boundary condition for Cartesian meshes is introduced such that the simulations are
performed just for a 72◦ segment, i.e., the flow field over one out of five axial blades
is resolved. The focus of this numerical analysis is on the development of the vortical
flow structures in the tip-gap region. A detailed grid convergence study is performed
on four computational grids with 50 × 106, 250 × 106, 1 × 109, and 1.6 × 109 cells.
Results of the instantaneous and the mean fan flow field are thoroughly analyzed
based on the solution with 1 × 109 cells. High levels of turbulent kinetic energy and
pressure fluctuations are generated by a tip-gap vortex upstream of the blade, the
separating vortices inside the tip gap, and a counter-rotating vortex on the outer cas-
ing wall. An intermittent interaction of the turbulent wake, generated by the tip-gap
vortex, with the downstream blade, leads to a cyclic transition with high pressure
fluctuations on the suction side of the blade and a decay of the tip-gap vortex. The
disturbance of the tip-gap vortex results in an unsteady behavior of the turbulent
wake causing the intermittent interaction. For this interaction and the cyclic transition,
two dominant frequencies are identified which perfectly match with the characteristic
frequencies in the experimental sound power level and therefore explain their physical
origin. C 2015 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4926515]
I. INTRODUCTION
Due to extremely complex geometries and highly unsteady turbulent structures, the flow field in
turbomachines has always been a big challenge in research. Flow phenomena near the tip-gap region
of axially rotating blades are a significant source of inefficiency and therefore one of the major prob-
lems in turbomachinery research. Aerodynamic losses, rotating instabilities, and blockage induced by
the tip-leakage flow strongly affect the performance of turbomachines.1–3 Another important aspect
is the contribution of the tip-gap flow to the broadband noise emission.4,5
The tip-leakage flow has been extensively studied in the past to achieve a better understanding
of the various phenomena. In general, the studies are performed either experimentally,6–15 encoun-
tering huge difficulties inside the tip gap due to access and safety reasons, or numerically using
methods based on the Reynolds-averaged Navier-Stokes (RANS) equations.1,2,16,17 Since the funda-
mental RANS formulation in conjunction with standard closure models suppresses the resolution of
the relevant temporal and spatial length scales of the turbulence spectrum, which is determined by
a) Electronic mail: a.pogorelov@aia.rwth-aachen.de

b) http://www.aia.rwth-aachen.de.
c) Electronic mail: m.meinke@aia.rwth-aachen.de
d) Electronic mail: office@aia.rwth-aachen.de
1070-6631/2015/27(7)/075106/21/$30.00 27, 075106-1 © 2015 AIP Publishing LLC
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
075106-2 Pogorelov, Meinke, and Schröder Phys. Fluids 27, 075106 (2015)
the highly complex and unsteady nature of the flow field, e.g., the interactions of the vortical flow
structures caused by the tip gap with the boundary layer of the outer casing wall or the neighboring
blades, RANS simulation results are not reliable to capture this class of highly unsteady flows.18–20
Whereas the computational resources required for a direct numerical simulation (DNS) are beyond
today’s computer capabilities, the large-eddy simulation (LES) concept can be used to accurately
predict the turbulent flow field around the blades if 80% of the energy spectrum is resolved.21
Nevertheless, only a few authors have used LES to investigate the complex flow phenomena
inside the tip-gap region. You et al.22–25 investigated the tip-leakage flow of a linear cascade with
a moving end wall. The underlying mechanisms for low-pressure fluctuations downstream of the
rotor near the end wall were analyzed. They observed a good agreement of their numerical results
with experimental data in terms of velocity, Reynolds stresses, and energy spectra. Furthermore, they
showed the mechanisms for the generation of vorticity and turbulent kinetic energy to be independent
of the tip-gap size and identified the velocity gradients to be the major causes for viscous losses in
the cascade end wall region. They also described a pitchwise low frequency wandering motion of
the tip-gap vortex. Boudet et al.26–28 performed computations for a single blade focusing on flow
spectra and noise generation. Using a zonal RANS/LES approach, the tonal and broadband con-
tent of the flow could be captured. A Fourier decomposition over the whole domain highlighted the
existence of a wandering motion of the tip-gap vortex, which they assumed to be a major contribu-
tion to noise radiation. It is fair to state that most of the previous numerical studies were performed
using structured mesh approaches for relatively simple geometries, i.e., single blades or linear cas-
cades. Although much progress has been made by previous studies, the detailed dynamics of the
tip-leakage flow such as the growth of the turbulence intensity in the tip-gap area and its decay
in the wake, the prospective interaction of the tip-gap vortices with the downstream blades, or the
impact of the vortical structures in the tip-gap region on the noise radiation is still not fully under-
stood.
Compared to previous investigations, the current LES computations based on Cartesian grids
are applied to more complex geometries and the resolution is extremely high such that the temporal
generation and the spatial development of the intricate flow structures are captured. For the finest
simulation, a grid with approximately 1.6 × 109 cells is used. The number of required time steps is
approximately 1.2 × 106 and about 128 TB of disk space is needed to store data for two full rotations
of the blades. The complex blade geometry together with the resolution of the tip-gap region makes
the generation of a high quality structured grid very tedious. Therefore, unstructured grids are used
since they can be generated fully automatically and very efficiently on parallel computers. Cartesian
cell based unstructured grids are advantageous compared to standard unstructured meshes which are
based on tetrahedral cells, since they possess due to their rectangular shape lower numerical diffu-
sion, i.e., they ensure that the computation is second order, which is of interest especially for high
Reynolds number turbulent flows. For this reason, Cartesian based unstructured meshes are used in
this study.
In the following, a parallelized numerical methodology based on hierarchical Cartesian grids
with the capability to automatically generate a computational grid and to accurately predict the un-
steady flow field in complex turbomachinery applications using LES is described. A detailed grid
convergence study is performed to investigate the sensitivity of the grid resolution on the numerical
results. The objective of this study is twofold. On the one hand, it is to be shown that the unstruc-
tured Cartesian approach is extremely efficient in analyzing such an intricate flow field problem like
the ducted rotating axial fan and on the other hand, the understanding of the temporal generation
and the spatial development of the complex turbulent flow structures in the tip-gap region is to be
improved.
The article presents the following structure: first, the numerical methods of the grid generator and
the flow solver are introduced; second, the numerical setup and boundary conditions are presented.
In Sec. IV, a thorough grid convergence study is performed, before a detailed analysis of the instan-
taneous and time-averaged flow field quantities in the tip-gap region is given. Finally, the essential
findings are summarized and some conclusions are drawn.
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
II. COMPUTATIONAL METHODOLOGY

A. Grid generation
The generation of the unstructured computational grid for the finite-volume flow solver is per-
formed by a parallel grid generator developed by Lintermann et al.29 The method automatically gener-
ates high resolution hierarchical Cartesian grids with multiple levels of refinement on an arbitrary
number of central processing units (CPUs) even for very complex geometries, which is one of the main
advantages compared to structured boundary fitted meshes. All grid cells are stored in a hierarchical
octree data-structure. For wall-bounded flows of high Reynolds number, Cartesian grids typically
contain more cells than structured grids, due to the isotropic refinement in all three spatial directions,
which, however, has the advantage of a higher accuracy. Furthermore, the grid generation and flow
simulation can be performed on different CPUs and the number of CPUs for the grid generator and
the flow solver can be arbitrary.
The grid generator requires a description of the computational domain in Standard Tessellation
Language (STL) format, which is a triangular representation of 3D surface and can be easily extracted
from a computer-aided design (CAD) model. Let Γ be the computational domain occupied by the
fluid phase and ∂Γ the surface described by the STL. The grid generation procedure starts with a
single cube with a side length l 0, which encloses the overall geometry, and is performed in three main
steps. First, the cube is uniformly refined by subdividing each cube into eight equally sized sub-cubes
until a specified initial refinement level is reached. A cell length at the k-th refinement level can be
computed by l k = l 0/2k . This first step is performed by each CPU. In the second step, the initial grid is
decomposed on the number of available CPUs using an efficient algorithm for domain decomposition
based on Hilbert’s curve. Each CPU uniformly refines its part of the domain up to a required base
refinement level. In each refinement step, cells which are located outside of the domain of integration
Γ are identified by a ray tracing technique and removed. In the last step, local refinement zones along
wall boundaries can be added to the isotropically refined base grid, by further refining the boundary
cells intersecting ∂Γ, followed by a smoothing procedure across the interface of different domains
to adjust the thickness of the refined layer, which requires the generation of halo and window cells.
Additionally, local internal refinement zones can be specified.
The efficiency of the grid generator was demonstrated by Lintermann et al.29 on different high
performance computers. Computational grids for a cubic test case with up to 640 × 109 cells could
be generated on the CRAY System HERMIT at the High Performance Computing Center Stuttgart
(HLRS) using 112768 CPUs in less than 300 s.
B. Mathematical model
We consider the three-dimensional Navier-Stokes equations of an unsteady viscous compress-
ible fluid flow of an ideal gas, which can be written for an infinitesimal control volume dV in their
non-dimensional form as
  
∂Q
dV + H · ndA = bdV, (1)
V ∂t ∂V V
with the vector of the conservative variables Q = [ρ, ρu, ρE]T and the density ρ, the velocity vector u,
and the total specific energy E = e + u2/2 containing the specific energy e. Furthermore, the quantity
inv
n is the outward unit normal vector on dA. The flux tensor H can be decomposed into an inviscid H
vis
and a viscous part H defined as
ρu + 0 +
inv vis * 1 *.
H +H = ... ρuu + p /// + .. τ /// . (2)
Re0
,u(ρE + p)- ,τu + q-
The Navier-Stokes equations are solved in a rotating frame of reference such that a body force b is
added to the right-hand side to take into account Coriolis and centrifugal forces
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
* 0 +
b = ... 2ρΩ × u − ρΩ × Ω × r /// , (3)
,(−2ρΩ × u − ρΩ × Ω × r) · u-
where Ω is the angular velocity vector.
All variables are non-dimensionalized by fluid properties at the state of rest denoted by the
subscript “0.” The Reynolds number is obtained by Re0 = ρ0a0l r e f /η 0 using the speed of sound
a0 = γp0/ρ0 with the ratio of specific heats γ = 1.4 and the characteristic length l r e f . The stress

tensor τ for a Newtonian fluid with zero bulk viscosity can be expressed as
2 ( )
τ=η (∇ · u) I − η ∇u + (∇u)T . (4)
3
According to Fourier’s law, the conductive heat flux reads
λ
q=− ∇T, (5)
Pr (γ − 1)
where ∇T is the gradient of the static temperature, Pr = η 0c p /λ 0 = 0.72 is the Prandtl number, and c p
is the specific heat capacity at constant pressure. For a constant Prandtl number, the thermal conduc-
tivity is λ(T) = η(T) and the dynamic viscosity η is calculated using Sutherland’s Law30
1+S
η(T) = T 3/2
, (6)
T+S
with S = 111K/293K for air at moderate temperatures. The system of equations is closed by the
equation of state for an ideal gas, i.e.,
p
e= . (7)
ρ (γ − 1)
C. Numerical method
Navier-Stokes equations (1) are solved on a fixed non-isotropic hierarchical Cartesian grid using
a cell-centered finite-volume discretization, where the computational grid acts as an implicit filter by
decomposing the turbulent flow field into a resolved part containing the dominant large structures
and a non-resolved sub-grid scale part. The sub-grid scale part which contains the cumulative effect
M
of the smallest scales must be modeled and is denoted by the flux tensor H . The LES formula-
tion in this paper is based on the monotone integrated LES (MILES) approach.31 In this method,
the subgrid-scale terms are not explicitly modeled. Instead, suitable discretization schemes act as an
implicit subgrid-scale model by numerically dissipating the energy at the smallest scales. The MILES
approach has been successfully applied to various complex turbulent flow problems and is able to
produce highly accurate results.32–35 At wall boundaries, cut cells are introduced. That is, cells which
intersect the surface ∂Γ are cut by computing their cutting points with the STL and discarding the
part which is located outside of Γ, such that the boundary ∂Γ is approximated by piecewise linear
segments. Based on the cutting points, the new volume V , surface areas A, and boundary-surface
normals n of the cut cell can be computed. Furthermore, the cell-center is shifted to the new center
of gravity xc = V1 V xd Ṽ . For further details, the reader is referred to Ref. 36.
The surface fluxes are computed using an upwind-biased scheme. The primitive variables at
the cell surfaces are obtained by a second-order accurate monotonic upstream-centered scheme for
conservation
 laws (MUSCLs) scheme, where the left and right values at each surface centroid xs =
1
A A xd Ã, marked by the subscripts L and R, are extrapolated from the cell-centers of the two facing
cells as
PsL/R = PcL/R + ∇P |cL/R ·(xs − xcL/R ). (8)
Cell-centered gradients ∇P | c of each primitive variable ξ ∈ P in cell c are computed by a weighted
least-squares reconstruction37
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50

∇ξ | c = Cl (ξl − ξ c ) (9)
l ∈N c
using all direct neighbors of the cell denoted by Nc . The reconstruction constants can be obtained by
solving a linear equation system
ACl = Wl 2dxl , (10)

with the distance vector dxl = xl − xc and the tensor A = l ∈Nc Wl 2dxl dxl , where Wl 2 is the squared

reconstruction weight the neighboring cell l. Further details on the solution of the linear system and
the computation of the weights can be found in Ref. 38.
inv
For the computation of the inviscid flux vector denoted as Finvi =H · ei , a modified version
39
of advection upstream splitting method (AUSM) of Liou and Steffen in a low dissipation version
suitable for LES proposed by Meinke et al.32 is used,
1 LR L
i =
Finv [M (f + f R ) + |MiL R |(f L − f R )] + pi , (11)
2 i
where i = 1, 2, 3, MiL R = 12 [(ui /a) L + (ui /a) R ] and
f = [ρa, ρau, ρa(E + p/ρ)]T . (12)

The term for the pressure is obtained by
) * 0 +
1 R 1
 ( ) (
pi = p + χ(ui /a) + p
L L
− χ(ui /a) R .. e // .
. i/ (13)
2 2
o
, -
The parameter χ determines the numerical dissipation of the scheme and is set to 0.5 at surfaces be-
tween cells of different refinement levels and 1/96 everywhere else. The viscous terms are discretized
by central differences.36 The overall spatial accuracy of this approach is second order as shown in
Ref. 38.
For the explicit temporal integration of Eq. (1), a second-order accurate 5-stage Runge-Kutta
scheme40 optimized for stability
Q(0) = Q(n),
Q(k) = Q(0) − α k ∆t R ∆Q(k−1), k = 1, 2, 3.., 5, (14)
Q (n+1)
=Q (5)
with α = ( 41 , 61 , 38 , 12 , 1)T is used. The superscript n denotes the time level t = n∆t and R ∆ is the approx-
imation of the cell surface integral. The time step ∆t is computed via the Courant-Friedrichs-Lewy
(CFL) stability constraint
h
∆t = CFL min , (15)
i=1,2,3 |ui | + a
where h is the length of the smallest non-cut cell of the computational domain.
The aforementioned cut-cell approach can produce arbitrary small cells, which can lead to numer-
ical instabilities using the time step from Eq. (15). To ensure stability without reducing the time step
to extremely small values, a flux redistribution method is applied. This flux redistribution approach
was developed by Schneiders et al.38 where the details of the formulation are given.
III. PROBLEM DEFINITION AND COMPUTATIONAL SETUP

A. Geometry, flow parameters, and boundary conditions
The investigated flow configuration is a low Mach number axial standardized fan test rig, which
has been used for acoustic measurements by, e.g., Sturm and Carolus.41 Fig. 1 schematically shows
the axial fan, which has five twisted blades, only one of which is resolved in the following study
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 1. Lateral and front views of the axial fan.
to reduce the computational costs of the detailed LES computations. The diameter Do of the rotat-
ing outer casing wall, note that a rotating frame of reference is used, is 300 mm and the inner hub
diameter is Di = 0.45Do . A fixed operating point is investigated, defined by a flow rate coefficient of
Φ = π 24DV̇3 n = 0.165, where n is the rotational speed and a tip-gap size of s/Do = 0.01. The result-
o
ing Reynolds number based on the rotational velocity and the diameter of the outer casing wall is
ρπ D 2 n
Re = η o = 9.36 × 105 and the Mach number is M = π Dao n = 0.136.
A ghost-cell procedure is used to impose the boundary conditions at cut cells as discussed in
Ref. 38. At the fan inflow, a fixed axial velocity corresponding to the flow rate coefficient of Φ = 0.165
superposed with a tangential velocity π Dv to n = D2ro due to the rotation of the outer casing wall is given.
Additionally, the density is defined and a zero-pressure gradient normal to the inflow plane is applied.
At the outflow, the static pressure is fixed and for the other primitive variables the gradients normal
to the outflow plane are set to zero. To eliminate numerical wave reflections caused by the inflow and
outflow boundary conditions, sponge layers are used.42 Along the walls, an adiabatic no-slip boundary
condition is imposed.
Simple rotationally periodic boundary conditions as known from structured periodic grids cannot
be applied to Cartesian grids, since a Cartesian grid itself is not rotationally periodic for arbitrary
section angles. Therefore, a newly developed periodic boundary condition for the circumferential
direction formulated and implemented for arbitrary refined Cartesian meshes is introduced. The basic
idea of the boundary condition is schematically illustrated in Fig. 2 for a two-dimensional domain,
which is subdivided into two subdomains Γ1 and Γ2 for a parallel computation on two CPUs. The
angle between the periodic boundaries of the domain, which are marked by red solid lines, is denoted
as θ per . An overlapping region with cells located outside of the red lines is used to impose the periodic
FIG. 2. Illustration of the periodic boundary condition on an unstructured Cartesian mesh.
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
boundary condition. These cells are called periodic cells and are not used in the temporal integration
of the Navier-Stokes equations. To obtain a periodic flow field, all primitive variables in the periodic
cells have to be determined and updated after each Runge-Kutta step. At least two cell layers are
required to maintain the second-order accuracy for all spatial gradients.
Several steps are carried out to obtain the primitive variables from the internal domain between
the periodic boundaries. We consider one periodic cell in the domain Γ1, which is denoted pc in Fig. 2.
In the first step, the cell center coordinates of the periodic cell are rotated by the angle of periodicity
θ per . The rotated coordinates, denoted by P, the periodic cell ID, and the domain ID of Γ1, are then
exchanged with all other subdomains. In the simple problem illustrated in Fig. 2, the information is just
sent from Γ1 to Γ2. Using a kd-tree search algorithm,43 which allows to find cells within logarithmic
time, the domain and the corresponding internal cell denoted cc, which surrounds the coordinate P,
are identified. Information, including the corresponding cell ID, the periodic cell ID, and both domain
IDs belonging to these cells, is subsequently stored in a connectivity matrix generated by domain Γ2
such that the searching procedure has to be performed only once.
The following step is performed after each Runge-Kutta update. Since the rotated cell center
coordinates of the periodic cell P and the cell center coordinates of the corresponding internal cell
do not match, each primitive variable ξ has to be reconstructed from the adjacent cells located in the
integration domain using a non-linear least-squares interpolation method. For this purpose, a linear
equation system has to be solved,
ξ k = α0 + α1∆x k + α2∆ y k + α3∆x k ∆ y k + α4(∆x k )2 + α5(∆ y k )2, k = 1, 2, . . . , N, (16)
where N denotes the number of reconstruction cells and ∆x , ∆ y define the distance between the
k k
k-th reconstruction cell and the coordinate P. Two layers of cell neighbors are used for the reconstruc-
tion. Having solved the linear system, the value of the primitive variable at coordinate P is given by
ξ P = α0. This procedure can be directly extended to three space dimensions. After the reconstruction,
all primitive variables at coordinate P are sent from Γ2 to Γ1 and prescribed to the cell center of the
periodic cell. Performing these steps for all periodic cells ensures a rotationally periodic flow field.
This formulation of a periodic boundary condition in an unstructured Cartesian mesh method has
been successfully used in Ref. 20.
B. Computational setup
Computations were performed on four different refined grids with approximately 50 × 106 (grid
G1), 250 × 106 (grid G2), 1 × 109 (grid G3), and 1.6 × 109 cells (grid G4). Fig. 3 shows an example
FIG. 3. Geometry and computational mesh of the fan configuration, resolving a 72◦ section: (a) overall Cartesian grid; (b)
axial cut through the computational grid.
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 4. Resolution inside the tip-gap region for the four meshes: (a) grid G1; (b) grid G2; (c) grid G3; and (d) grid G4.
of the overall computational domain resolving a 72◦ section of the fan, i.e., only the flow field over
one single blade is considered, the mesh used in this study, and an axial cut through the grid. Local
boundary refinement towards the walls is used to resolve the boundary layers. The grids consist of four
refinement levels, a minimum base refinement level in the internal region, and a maximum refinement
level along the walls with a smooth transition between them, as shown in Fig. 3(b). The first three
grids G1, G2, and G3 were generated by successively increasing the minimum and the maximum
refinement levels. For the finest grid G4, additionally, the grid inside the tip-gap region was isotrop-
ically refined using the wall resolution in the entire tip-gap region and by the keeping the rest of the
grid identical to the G3 formulation. The grid resolutions inside the tip-gap region for all meshes are
shown in Fig. 4.
Four full rotations of the rotor were simulated for each computational grid. After two full rota-
tions, the flow field was fully developed for all grids, and over the next two full rotations, data have
been collected for the statistical analysis of the flow field. For each grid, 2000 samples were recorded.
Information about the grids and the resources required for the four computations are given in Table I.
For the finest grid, the cell length in wall units is ∆x +i < 3 at the wall and increases up to ∆x +i < 24 in
the coarsest region away from the wall. The maximum stable time step size based on the rotational
velocity and the diameter of the outer casing wall was ∆tπn = 1.936 × 10−5 for the finest grid. This
time step size was chosen for all computations to minimize the influence of the temporal discretization
error for the comparison of the results obtained on the four meshes. Since the number of CPUs for
each simulation was chosen such that the number of cells per CPU is similar for all cases, the total
CPU times of approximately 250 h for four full rotations of the rotor were almost identical.
TABLE I. Parameters defining the computational grids and required resources.
Number of cells Refinement levels Smallest cell (wall) Number of CPUs Disk space
grid G1 ≈0.5 × 108 9-12 4.0 × 10−4 D o 1 608 ≈4 TB

grid G2 ≈2.5 × 108 10-13 2.0 × 10−4 D o 7 992 ≈20 TB
grid G3 ≈1.0 × 109 11-14 1.0 × 10−4 D o 31 992 ≈80 TB
grid G4 ≈1.6 × 109 11-14 1.0 × 10−4 D o 51 192 ≈128 TB
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 5. Strong scaling experiment for grid G3 with 1.0 × 109 cells. Simulations were performed for 1000 time steps using
five numbers of cores, i.e., 5472, 10 968, 21 936, 32 880, and 91 872; ideal speedup (red line) and achieved speedup (blue line
with square dots).
Strong scaling experiments were conducted on the CRAY System HORNET at the High Perfor-
mance Computing Center Stuttgart (HLRS). For the computational 1 × 109 cell grid G3, five core
numbers were used, i.e., 5472, 10 968, 21 936, 32 880, and 91 872. The overall speedup, including
the wall-clock time for 1000 time steps, as a function of the number of cores shown in Fig. 5 proves
the good scalability of the LES code.
IV. RESULTS
We turn now to discuss the large-eddy simulation solutions of the flow field through an axial fan.
First, the overall flow field is introduced and compared to experimental data. Then, a grid convergence
study is conducted, followed by a detailed analysis of the time-averaged and instantaneous flow field
quantities in the tip-gap region.
A. Overall flow field analysis

LES results of four computations based on different grid resolutions are presented for the axial
fan configuration with a tip-gap size of s/Do = 0.01. Fig. 6(a) illustrates the operating line of the
∆p 4V̇
FIG. 6. Pressure coefficient Ψ = 2 versus flow rate coefficient Φ = for s/D o = 0.01. ∆p = p st at, ou t −
( π2 ρ D 2a n 2) π 2 D 3o n
p 0, i n is the difference between the static pressure at the outlet p st at, ou t and the stagnation pressure at the inlet p 0, i n of
the fan; (a) overall operating map; (b) detailed view close to the operating point; experimental data from Ref. 44; experiment
(black dot), grid G1 (gray diamond), grid G2 (green square), grid G3 (red down-pointing triangle), and grid G4 (blue
up-pointing triangle).
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
fan, which is defined by the dimensionless pressure gain Ψ = 2

∆p
with ∆p = pst at,out − p0,in
( π2 ρ D 2a n 2)
versus the flow rate coefficient Φ = π 24DV̇3 n . Additionally, the LES results of the four computations
o
are shown for a fixed operating point at a flow rate coefficient Φ = 0.165. A good agreement between
the experimental and numerical results can be observed for all grids. Even a more detailed view in
Fig. 6(b) evidences only small deviations from the experiment. By successively increasing the grid
resolution, the LES results asymptotically converge towards the experimental results.
Fig. 7 shows the instantaneous and time-averaged vortical structures of the flow field in an axial
fan visualized by the Q-criterion45 and the relative Mach number for the 1 × 109 cell grid G3. Several
physical phenomena are resolved by the LES. The tip-gap vortex generated near blade tip is evident.
The tip-leakage flow caused by the pressure difference between the pressure and suction side feeds
this vortex which initiates a highly unsteady and turbulent wake region. The turbulent wake inter-
acts with the neighboring blade leading to a transition region on the suction side of the blade. The
counter-rotating vortex is generated by a separation on the rotating outer casing wall. The turbulent
FIG. 7. Contours of the Q-criterion of the instantaneous (a) and time-averaged (b) flow field colored by the relative Mach
number showing the vortical structures of the flow field in an axial fan (grid G3).
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 8. Contours of the vorticity magnitude of the instantaneous flow field in a cylindrical slice at 90% span (grid G3). Note
that a perspective view is illustrated.
wake at the trailing edge of the blade passes the neighboring blades without any interaction, whereas
the passage vortex causes a turbulent transition of the boundary layer on the hub.
To visualize the turbulent structures in the wake region more clearly, Fig. 8 illustrates the contours
of the vorticity magnitude of the instantaneous flow field in a cylindrical slice at 90% span for grid
G3. Strong vorticity is generated by the tip-gap vortex and its turbulent wake on the suction side,
which interacts with the neighboring blade.
B. Influence of the grid resolution on the time-averaged flow field

Next, the influence of the grid resolution on the overall time-averaged flow field is investigated.
Fig. 9 depicts axial distributions of the time-averaged Mach number M and turbulent kinetic energy
k upstream and downstream of the blade at three radial locations, i.e., 20%, 50%, and 80% span,
and the azimuthal location θ = −45◦ for all four grids. The turbulent kinetic energy per unit mass is
defined by
1
(< u ′x 2 > + < ur′ 2 > + < uθ′ 2 >),
k= (17)
2
where <> denotes a time-averaged quantity. The surface of the twisted blade is located between
x/Do = 0.61 and x/Do = 0.63.
Upstream of the blade, a smooth increase of the Mach number towards the suction side of the
blade is observed at all radial locations. The Mach number peak increases with radius due to the
rotation of the blade. Downstream of the blade, the Mach number is decreased and different wakes
can be noticed. At 20% span, the wake produced by the passage vortex generates a minimal Mach
number and an increase of turbulent kinetic energy at x/Do = 0.82. At 50% span, the blade wake
generated at the leading edge is captured at x/Do = 0.89. It is weaker compared to the passage vortex
wake and shows a lower peak of turbulent kinetic energy. Closer to the casing wall at 80% span,
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 9. Axial distributions of the time-averaged Mach number and turbulent kinetic energy at three radial locations and
θ = −45◦ for different grid resolutions: (a) 20% span; (b) 50% span; (c) 80% span; grid G1 (red line), grid G2 (green dotted
line), grid G3 (blue dashed-dotted line), and grid G4 (black dashed double-dotted line).
the turbulent wake generated by the tip-gap vortex is observed, which causes much larger values of
turbulent kinetic energy. It is evident from Fig. 9 that the coarsest grid resolution G1 does not suffice
to capture the wakes in detail. Neither the Mach number nor the turbulent kinetic energy distributions
are accurately predicted. However, the results based on grids G2, G3, and G4 agree very well.
Additionally, radial distributions of the time-averaged Mach number and the turbulent kinetic
energy are illustrated at two axial locations and θ = −45◦ in Fig. 10. Fig. 10(a) depicts the distribu-
tions upstream of the blade close to the blade surface at x/Do = 0.60 and Fig. 10(b) the distributions
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 10. Radial distributions of the time-averaged Mach number and turbulent kinetic energy upstream (a) and downstream
(b) of the blade at θ = −45◦ for different grid resolutions; grid G1 (red line), grid G2 (green dotted line), grid G3 (blue
dashed-dotted line), and grid G4 (black dashed double-dotted line).
downstream of the blade through the turbulent wake generated by the tip-gap vortex at x/Do = 0.683.
In Fig. 10(a), a pronounced peak in turbulent kinetic energy caused by the tip-gap vortex is observed
close to the casing wall at r/Do = 0.47. Fig. 10(b) shows two peaks in the turbulent kinetic energy
distribution, one at the hub due to the turbulent boundary layer and one close to the casing wall due
to the wake generated by the tip-gap vortex. Similar to the axial distributions, the coarse grid G1 is
not sufficient to capture the flow structures that are relevant for the turbulent kinetic energy level in
the tip-gap region, which is why the turbulent kinetic energy near the casing wall is underpredicted.
The peak at the hub, however, is well captured.
Globally speaking, an asymptotic behavior of the results computed by various mesh resolutions
is shown and only small differences are observed between the findings on grid G3 and G4. Note that
the analysis of distributions at other circumferential angles θ confirms the above behavior. Hence, the
comparison corroborates a sufficient grid resolution for grid G3 below the blade tip.
C. Influence of the grid resolution on the flow field in the tip-gap region
As mentioned before, the flow field in the tip-gap region is of high importance, since it strongly
affects the efficiency and noise emission. Therefore, the impact of the grid resolution on the local
effects inside the tip-gap region is analyzed in detail. The rotation of the blades yields a high local
Reynolds numbers on the surface of the blade close to the tip-gap region. However, due to the small
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 11. Radial distributions of the time-averaged Mach number and turbulent kinetic energy at three circumferential
locations in the tip-gap region: (a) θ = −55◦; (b) θ = −45◦; (c) θ = −35◦; grid G1 (red line), grid G2 (green dotted line),
grid G3 (blue dashed-dotted line), and grid G4 (black dashed double-dotted line).
size of the tip gap, the Reynolds number based on the maximum mean velocity inside the gap vg, ma x
ρv s
at Θ = −45◦ and the tip-gap size s is Re s = g , ηma x = 1.16 × 105. To quantify the influence of the
grid resolution, Fig. 11 shows the radial distributions of the time-averaged Mach number and turbu-
lent kinetic energy at the same circumferential locations inside the tip gap. The azimuthal location
θ = −55◦ is near the leading edge of the blade and θ = −35◦ close to the trailing edge. Due to the
twisted shape of the blade, the axial locations change in the circumferential direction. The turbulent
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
kinetic energy peak generated inside the tip gap is caused by the separated tip-leakage flow. It is
much larger compared to peaks from the passage vortex, the blade wake, or the wake from the tip-gap
vortex. At all circumferential locations, this peak is located close to the blade-tip surface. As expected
from the previous results, grid G1 is too coarse to capture the complex phenomena inside the tip
gap. This is valid for the Mach number and the turbulent kinetic energy distributions. The turbulent
kinetic energy is strongly underpredicted on G1, especially close to the blade-tip surface. Comparing
the results from grids G2, G3, and G4, only slight differences in the radial Mach number distribution
are present. The turbulent kinetic energy is more sensitive to the grid resolution. Close to the leading
edge at θ = −55◦, the influence of the grid resolution is relatively small. Further downstream, i.e., at
θ = −45◦ and θ = −35◦, the differences become larger. The relative change of the peak value of the
turbulent kinetic energy, however, decreases at all three locations showing the converging behavior
of the solutions.
D. Flow field near the tip-gap region

In the following, the flow field near the tip-gap region and the development of the tip-gap vortex
are further investigated based on the LES results on grid G3. Fig. 12 illustrates the development of
the tip-gap vortex in the circumferential direction near the blade tip. The contours of the turbulent
kinetic energy in Fig. 12(a) and pressure fluctuations in Fig. 12(b) are shown in eight planes starting
from θ = −60◦ near the leading edge up to θ = −25◦ near the trailing edge. Additionally, streamlines
of the projected time-averaged velocity field are depicted. A main tip-gap vortex, which starts near
the leading edge of the blade and grows in the circumferential direction, is evident. It increases its
distance to the blade surface while propagating in the circumferential direction. Several vortices inside
the tip gap, which separate from the tip surface near the trailing edge and interact with the tip-gap
vortex, and a counter-rotating vortex caused by a separation at the outer casing wall at θ = −45◦ and
θ = −40◦ upstream of the tip-gap vortex are observed. Furthermore, a wake region generated by the
break-up of the tip-gap vortex of the neighboring blade, which interacts with the leading edge close
to the tip gap, is illustrated at θ = −60◦.
As indicated in Fig. 12(a), high turbulent kinetic energy is generated by the tip gap and the
separated vortices inside the gap. From θ = −55◦ up to θ = −35◦, the tip-gap vortex gains turbulent
kinetic energy from the leakage flow by interacting with the separated vortices shed from the gap.
Downstream of θ = −35◦, the tip-gap vortex oscillates and starts to decay such that the turbulent
kinetic energy decreases and the region of influence still increases. The remaining vortical structures
in the wake originating from the neighboring blade interact with the leading edge and generate strong
fluctuations on the suction side. Another source of turbulent kinetic energy is the separation region at
the outer casing wall leading to a counter-rotating vortex. In the center of the counter-rotating vortex,
FIG. 12. Development of the tip-gap vortex and the flow field in the gap region in eight planes from θ = −55◦ near the leading
edge to θ = −25◦ near the trailing edge. (a) Contours of the turbulent kinetic energy; (b) contours of the pressure fluctuations.
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 13. Contours of the pressure coefficient of the instantaneous flow field at three time steps, i.e., t 0, t 0 + ∆T , and t 0 + 2∆T ,
1
where ∆T πn ≈ 10 , to show the unsteady behavior of the tip-gap vortex and the separated vortices inside the tip gap at
θ = −45◦; black solid lines denote the contours of the Q-criterion.
the turbulent kinetic energy is very low, while close to the separation point a region of high turbu-
lent kinetic energy is observed. The comparison of Figs. 12(a) and 12(b) shows that strong pressure
fluctuations are generated in regions of high turbulent kinetic energy in the tip-gap vortex, e.g., on
the suction side of the blade due to the interaction with the wake. However, the pressure fluctuations
in the tip-gap vortex already start to decay at θ = −40◦, where the turbulent kinetic energy still in-
creases. Furthermore, the locations of the maxima of the turbulent kinetic energy and the pressure
fluctuations do not correlate. At θ = −40◦, two maxima in the distribution of the pressure fluctuations
are observed near the core of the tip-gap vortex, while the turbulent kinetic energy possesses only
one maximum. The magnitude of the turbulent kinetic energy intensity caused by the separation of
the counter-rotating vortex is comparable to that inside the tip-gap vortex, which is not true for the
pressure fluctuations.
To analyze the time dependent behavior of the tip-gap vortex and the separated vortices inside the
2(p−p i n)
tip gap, Fig. 13 depicts the contours of the pressure coefficient Cp = ρ(π Dst at,
n)2
and the Q-criterion
o
FIG. 14. Temporal variation of the flow field visualized by the pressure coefficient and the Q-criterion at three time steps,
1
i.e., t 0, t 0 + ∆T , and t 0 + 2∆T , where ∆T πn ≈ 10 ; (a) contours of the Q-criterion and the pressure coefficient near the tip-gap
region showing the unsteady behavior of the tip-gap vortex; the oscillating trajectory of the vortex is indicated by a dashed
line; (b) contours of the Q-criterion on the suction side of the blade and near the tip-gap region showing the blade-wake
interaction and turbulent transition.
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 15. Definition of the locations of points P1 and P2.
denoted by black solid lines at θ = −45◦ and three time steps, i.e., t 0, t 0 + ∆T, and t 0 + 2∆T, where
1
∆T πn ≈ 10 . Due to the flow from the pressure to the suction side of the blade, a separation region
develops inside the tip gap generating large scale vortical structures denoted by low pressure regions.
The comparison of the flow structures at t 0, t 0 + ∆T, and t 0 + 2∆T evidences the highly unsteady
behavior inside the tip gap. Another low pressure region generated by the main tip-gap vortex is shown
on the suction side of the blade. Size and location of the low pressure region inside the vortex core
depend on time. From t 0 to t 0 + ∆T, the tip-gap vortex moves closer to the blade and at t 0 + 2∆T,
it almost disappears. This temporal decay is caused by the interaction of the turbulent wake of the
neighboring blade with the leading edge which is shown in Fig. 14 at the same three time steps.
Fig. 14(a) illustrates the instantaneous contours of the pressure coefficient and the Q-criterion in the
overall tip-gap region. The instantaneous behavior of the tip-gap vortex and its turbulent wake is
highlighted by the dashed line that emphasizes the oscillating trajectory of the vortex. In addition,
Fig. 14(b) shows the blade-wake interaction by the instantaneous contours of the Q-criterion on the
suction side of the blade. At t 0, just before the interaction, a relaminarizing turbulent region is observed
on the suction side of the blade, which had been generated by a previous interaction. At t 0 + ∆T, the
wake interacts with the blade generating another transition near the leading edge, which propagates
FIG. 16. Velocity and pressure fluctuations at location P1 defined in Fig. 15 showing the blade-wake interaction: (a) fluctu-
ations of the streamwise u ′x /π D o n (blue dashed-dotted line), radial u ′r /π D o n (red line), and azimuthal u θ′ /π D o n (black
dashed line) velocity components; (b) pressure fluctuations 2p ′/ρ(π D o n)2.
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 17. Autocorrelations of the velocity and pressure fluctuations at location P1 defined in Fig. 15: (a) autocorrelation of
the streamwise R u ′x, u ′x (blue dashed-dotted line), radial R u ′r , u ′r (red line), and azimuthal R u ′ , u ′ (black dashed line) velocity
θ θ
fluctuations; (b) autocorrelation of the pressure fluctuations R p ′, p ′.
downstream over the blade surface. Furthermore, the tip-gap vortex is disturbed by the interaction
with the wake such that the trajectory of the tip-gap vortex and its turbulent wake is strongly excited.
This effect causes an intermittent interaction between the wake and the adjacent blades. At t 0 + 2∆T,
the flow field close to the blade tip is highly turbulent resulting in the pronounced decay of the tip-gap
vortex which is illustrated in Fig. 13. It is also shown that the interaction shifts the stagnation point
at the leading edge to the pressure side.
To further investigate the blade-wake interaction, pressure and velocity signals are determined
at two locations of the flow field defined in Fig. 15. Fig. 16 shows the time signals of the components
of the velocity fluctuations ui′(t) and the pressure fluctuations p′(t) over two full rotations recorded
on the suction side of the blade at location P1. The signals of the velocity components clearly show
several cyclic disturbances that are associated with the interaction discussed above. To analyze the
time signals, autocorrelations
< ui′(t)ui′(t + τ) > < p′(t)p′(t + τ) >
Ru ′i,u ′i (τ) = and R p′, p′(τ) = (18)
< ui′2(t) > < p′2(t) >
are computed. Their illustration in Fig. 17 indicates a periodic behavior.
Therefore, energy spectra are determined by a Fourier transformation of the autocorrelations,
which are depicted in Fig. 18 as a function of the Strouhal number based on the rotational velocity
and the diameter of the outer casing wall Sr = f /πn. Two dominant peaks related to the intermittent
interaction and cyclic transition are observed for all spectra at Sr 1 = 2.37 and Sr 2 = 3.47. Fig. 19
shows the experimental specific sound power L w45−s pec , which is the sum of the sound power levels
measured on the pressure side L w4 and the suction side L w5 for two flow rate coefficients Φ = 0.165
and Φ = 0.195. The experiments have been conducted by Zhu and carolus44 at a constant rotational
speed of n = 50 1s . Further details about the experimental data can be found in Ref. 44. As discussed
in Ref. 20, for the increased volume flux Φ = 0.195, the turbulent wake generated by the tip-gap
vortex is located further downstream and passes the adjacent blades without interaction. The specific
sound power for Φ = 0.165 shows two additional peaks, which are not observed for Φ = 0.195. Note
that these peaks occur at the same Strouhal numbers as the peaks in the energy spectra in Fig. 18.
FIG. 18. Energy spectra computed by Fourier transformations of the autocorrelations R u ′ , u ′ and R p ′, p ′ at location P1
i i
defined in Fig. 15, on the suction side of the blade: (a) energy spectra E u ′x, u ′x (blue dashed-dotted line), E u ′r , u ′r (red line),
and E u ′ , u ′ (black dashed line); (b) energy spectrum E p ′, p ′.
θ θ
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
FIG. 19. Specific sound power level for two flow rate coefficients φ = 0.165 (blue dashed-dotted line) and φ = 0.195 (red
line); experimental data from Ref. 44; grey vertical stripes indicate the spectral broadening of the modes at Sr 1 and Sr 2.
Furthermore, compared to the peaks at the blade passing frequency and its harmonics, which are
also depicted in Fig. 19, these peaks are less sharp. The perfect match of these frequencies explains
the physical origin of the peak values in the experimental sound power level. These findings lead
to the conclusion that the blade-wake interaction has a dominant effect on the noise generation for
Φ = 0.165.
In addition, time signals of the components of the velocity fluctuations ui′(t) and the pressure
fluctuations p′(t) were determined at location P2 defined in Fig. 15 inside the tip-gap vortex to analyze
the wandering motion of the tip-gap vortex. Fig. 20 shows the energy spectra at location P2. The same
dominant frequencies at the Strouhal numbers Sr 1 = 2.37 and Sr 2 = 3.47 are observed. However, the
peaks do not exist in all autocorrelations. At Sr 1, only the radial component has a peak, whereas for
Sr 2, the peak occurs in the radial and the axial components. Furthermore, the energy E p′, p′ has a much
lower peak at Sr 1 compared to Sr 2.
Similar observations have been made by Boudet et al.28 They found by spectral analysis one
frequency related with the wandering of the tip-gap vortex, which dominated the sound pressure
spectrum. This frequency was assumed to be excited by turbulence from the casing boundary layer
and/or the wake of the adjacent blade and to be a precursor of a rotating instability. Furthermore, the
hypothesis was stated that by reducing the flow rate coefficient, the trajectory of the tip-gap vortex is
shifted upstream and due to its wandering motion quasi-periodically excites the tip-gap region of the
downstream blade, which is confirmed by the findings in this study. The dominant frequency found
in Ref. 28 was higher than the blade passing frequency. The corresponding Strouhal number based
on the chord length Sr c = 1.0 is in agreement with the current findings. Based on the chord length at
midspan, which is approximately 90 mm in this configuration, the Strouhal numbers of the dominant
modes are Sr c1 = 0.71 and Sr c2 = 1.04.
FIG. 20. Energy spectra computed by Fourier transformations of the autocorrelations R u ′ , u ′ and R p ′, p ′ at location P2
i i
defined in Fig. 15, inside the tip-gap vortex: (a) energy spectra E u ′x, u ′x (blue dashed-dotted line), E u ′r , u ′r (red line), and
E u ′ , u ′ (black dashed line); (b) energy spectrum E p ′, p ′.
θ θ
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
V. CONCLUSIONS
Large-eddy simulations of the flow field in a rotating 5-blade axial fan have been presented for
a fixed operating point with a flow rate coefficient of Φ = 0.165 and tip-gap size of s/Do = 0.01.
Computations have been performed for one single blade using a periodic boundary condition in the
circumferential direction. A grid convergence study on four computational grids with different refine-
ment levels shows the capability of the LES to capture the main features of the flow field. The tem-
poral and spatial developments of a tip-gap vortex upstream of the blade, several separated vortices
inside the tip gap, and a counter-rotating vortex on the outer casing wall, which lead to high levels of
turbulent kinetic energy and pressure fluctuations, are analyzed in detail. Furthermore, an interaction
of the turbulent wake generated by the tip-gap vortex with the downstream blade triggers a cyclic
transition and high pressure fluctuations on the suction side of the blade and a massive decay of the
tip-gap vortex. Two dominant frequencies are identified for the intermittent interaction and the cyclic
transition. They perfectly match with the characteristic frequencies in the experimental sound power
level, which explains the physical origin of these peak values. The current results confirm findings
from the literature that the blade-wake interaction is one of the major sources for the noise generation.
In future studies, LES results will be used to determine the acoustic field based on computational
aeroacoustics (CAA) methods.
ACKNOWLEDGMENTS
This study was funded under Grant No. 17747N (L238) by the German Federal Ministry of Eco-
nomics and Technology via the “Arbeitsgemeinschaft industrieller Forschungsvereinigungen Otto
von Guericke e.V.” (AiF) and the “Forschungsvereinigung Luft- und Trocknungstechnik e.V.” (FLT).
The authors also wish to thank the High Performance Computing Center Stuttgart (HLRS) and the
Jülich Supercomputing Center (JSC) for providing the computing resources.
1 J. Storer and N. Cumpsty, “Tip leakage flow in axial compressors,” J. Turbomach. 113, 252–259 (1991).
2 J. Storer and N. Cumpsty, “An approximate analysis and prediction method for tip clearance loss in axial compressors,” J.
Turbomach. 116, 648–656 (1994).
3 R. Mailach, I. Lehmann, and K. Vogeler, “Rotating instabilities in an axial compressor originating from the fluctuating blade
tip vortex,” J. Turbomach. 123, 453–463 (2001).

4 R. Camussi, J. Grilliat, G. Caputi-Gennaro, and M. C. Jacob, “Experimental study of a tip leakage flow: Wavelet analysis
of pressure fluctuations,” J. Fluid Mech. 660, 87–113 (2010).

5 M. Jacob, J. Grilliat, R. Camussi, and G. Caputi Gennaro, “Aeroacoustic investigation of a single airfoil tip leakage flow,”
Int. J. Aeroacoust. 9, 253–272 (2010).

6 M. Inoue, M. Kuroumaru, and M. Fukuhara, “Behavior of tip leakage flow behind an axial compressor rotor,” J. Eng. Gas
Turbines Power 108, 7–14 (1986).

7 A. Goto, “Three-dimensional flow and mixing in an axial flow compressor with different rotor tip clearances,” J. Turbomach.
114, 675–685 (1992).

8 R. C. Stauter, “Measurements of the three-dimensional tip region flow field in an axial compressor,” J. Turbomach. 115,
468–476 (1993).
9 S. Kang and C. Hirsch, “Experimental study on the three-dimensional flow within a compressor cascade with tip clearance:
part I-velocity and pressure fields,” J. Turbomach. 115, 435–443 (1993).

10 S. Kang and C. Hirsch, “Experimental study on the three-dimensional flow within a compressor cascade with tip clearance:
part II-the tip leakage vortex,” J. Turbomach. 115, 444–452 (1993).

11 S. Kang and C. Hirsch, “Tip leakage flow in linear compressor cascade,” J. Turbomach. 116, 657–664 (1994).
12 B. Lakshminarayana, M. Zaccaria, and B. Marathe, “The structure of tip clearance flow in axial flow compressors,” J.
Turbomach. 117, 760–771 (1995).

13 C. Muthanna and W. J. Devenport, “Wake of a compressor cascade with tip gap, part 1: Mean flow and turbulence structure,”
AIAA J. 42, 2320–2331 (2004).

14 Y. Wang and W. J. Devenport, “Wake of a compressor cascade with tip gap. part 2. effects of endwall motion,” AIAA J. 42,
2332–2340 (2004).
15 C. W. Wenger, W. J. Devenport, K. S. Wittmer, and C. Muthanna, “Wake of a compressor cascade with tip gap. part 3.
two-point statistics,” AIAA J. 42, 2341–2346 (2004).

16 M. R. Khorrami, F. Li, and M. Choudhan, “Novel approach for reducing rotor tip-clearance-induced noise in turbofan
engines,” AIAA J. 40, 1518–1528 (2002).

17 N. Gourdain, F. Wlassow, and X. Ottavy, “Effect of tip clearance dimensions and control of unsteady flows in a multi-stage
high-pressure compressor,” J. Turbomach. 134, 051005 (2012).

18 T. Carolus, M. Schneider, and H. Reese, “Axial flow fan broad-band noise and prediction,” J. Sound Vib. 300, 50–70 (2007).
19 J. Tyacke, P. Tucker, R. Loveday, N. Vadlamani, R. Watson, I. Naqavi, and X. Yang, “Large eddy simulation for turbines:
Methodologies, cost and future outlooks,” J. Turbomach. 136, 061009 (2013).
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50
20 A. Pogorelov, M. Meinke, W. Schröder, and R. Kessler, “Cut-cell method based large-eddy simulation of a tip-leakage vortex
of an axial fan,” AIAA Paper 2015-1979, 2015.
21 S. B. Pope, Turbulent Flows (Cambridge University Press, 2000).
22 D. You, R. Mittal, M. Wang, and P. Moin, “Computational methodology for large-eddy simulation of tip-clearance flows,”
AIAA J. 42, 271–279 (2004).

23 D. You, M. Wang, P. Moin, and R. Mittal, “Effects of tip-gap size on the tip-leakage flow in a turbomachinery cascade,”
Phys. Fluids 18, 105102 (2006).

24 D. You, M. Wang, P. Moin, and R. Mittal, “Large-Eddy simulation analysis of mechanisms for viscous losses in a turbo-
machinery tip-clearance flow,” J. Fluid Mech. 586, 177–204 (2007).

25 D. You, M. Wang, P. Moin, and R. Mittal, “Vortex dynamics and low-pressure fluctuations in the tip-clearance flow,” J.
Fluids Eng. 129, 1002–1014 (2007).

26 J. Boudet, J. Caro, L. Shao, and E. Lévêque, “Numerical studies towards practical large-eddy simulation,” J. Therm. Sci.
16, 328–336 (2007).

27 J. Boudet, J. Caro, and M. Jacob, “Large-eddy simulation of a single airfoil tip-clearance flow,” AIAA Paper 2010-3978,
2010.
28 J. Boudet, A. Cahuzac, P. Kausche, and M. C. Jacob, “Zonal large-eddy simulation of a fan tip-clearance flow, with evidence
of vortex wandering,” J. Turbomach. 137, 061001 (2015).

29 A. Lintermann, S. Schlimpert, J. H. Grimmen, C. Günther, M. Meinke, and W. Schröder, “Massively parallel grid generation
on hpc systems,” Comput. Methods Appl. Mech. Eng. 277, 131–153 (2014).
30 W. Sutherland, “The viscosity of gases and molecular force,” Philos. Mag. 36, 507–531 (1893).
31 J. P. Boris, F. F. Grinstein, E. S. Oran, and R. L. Kolbe, “New insights into large eddy simulation,” Fluid Dyn. Res. 10,
199–228 (1992).
32 M. Meinke, W. Schröder, E. Krause, and T. Rister, “A comparison of second- and sixth-order methods for large-eddy simu-
lation,” Comput. Fluids 31, 695–718 (2002).

33 N. Alkishriwi, M. Meinke, and W. Schröder, “A large-eddy simulation method for low mach number flows using precon-
ditioning and multigrid,” Comput. Fluids 35, 1126–1136 (2006).

34 F. Rütten, W. Schröder, and M. Meinke, “Large-eddy simulation of low frequency oscillations of the Dean vortices in
turbulent pipe,” Phys. Fluids 17, 035107 (2005).

35 P. Renze, W. Schröder, and M. Meinke, “Large-eddy simulation of film cooling flows at density gradients,” Int. J. Heat Fluid
Flow 29, 18–34 (2008).

36 D. Hartmann, M. Meinke, and W. Schröder, “A strictly conservative Cartesian cut-cell method for compressible viscous
flows on adaptive grids,” Comput. Methods. Appl. Mech. Eng. 200, 1038–1052 (2011).
37 D. J. Mavriplis, “Revisiting the least-squares procedure for gradient reconstruction on unstructured meshes,” AIAA Paper
2003-3986, 2003.
38 L. Schneiders, D. Hartmann, M. Meinke, and W. Schröder, “An accurate moving boundary formulation in cut-cell methods,”
J. Comput. Phys. 235, 786–809 (2013).

39 M.-S. Liou and C. J. Steffen-Jr., “A new flux splitting scheme,” J. Comput. Phys. 107, 23–39 (1993).
40 A. Jameson and D. Mavriplis, “Finite volume solution of the two-dimensional euler equations on a regular triangular mesh,”
AIAA J. 24, 616–618 (1986).

41 M. Sturm and T. Carolus, “Tonal fan noise of an isolated axial fan rotor due to inhomogeneous coherent structures at the
intake,” Noise Control Eng. J. 60, 699–706 (2012).

42 J. B. Freund, “Proposed inflow/outflow boundary condition for direct computation of aerodynamic sound,” AIAA J. 35,
740–743 (1997).
43 K. Bentley, “Multidimensional binary search trees used for associative searching,” Commun. ACM 18, 509–517 (1975).
44 T. Zhu and T. H. Carolus, “Experimental and numerical investigation of the tip clearance noise of an axial fan,” in
Proceedings of ASME Turbo Expo 2013 (GT2013-94100), San Antonio, Texas, USA, 3-7 June 2013.
45 J. Jeong and F. Hussain, “On the identification of a vortex,” J. Fluid Mech. 285, 69–94 (1995).
134.130.190.40 On: Wed, 29 Jun 2016 12:56:50

Pogorelov Et Al 2015 Cut-Cell Method Based Large-Eddy Simulation of Tip-Leakage Flow PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Pogorelov Et Al 2015 Cut-Cell Method Based Large-Eddy Simulation of Tip-Leakage Flow PDF

Caricato da

Copyright:

Formati disponibili

Cut-cell method based large-eddy simulation of tip-leakage flow

Alexej Pogorelov, Matthias Meinke, and Wolfgang Schröder

Citation: Physics of Fluids 27, 075106 (2015); doi: 10.1063/1.4926515

Articles you may be interested in

Large-eddy simulation of 3-D corner separation in a linear compressor cascade

Subgrid-scale eddy viscosity model for helical turbulence

Large-eddy simulations of a turbulent Coanda jet on a circulation control airfoil

Cut-cell method based large-eddy simulation

a) Electronic mail: a.pogorelov@aia.rwth-aachen.de

1070-6631/2015/27(7)/075106/21/$30.00 27, 075106-1 © 2015 AIP Publishing LLC

II. COMPUTATIONAL METHODOLOGY

ACl = Wl 2dxl , (10)

f = [ρa, ρau, ρa(E + p/ρ)]T . (12)

III. PROBLEM DEFINITION AND COMPUTATIONAL SETUP

FIG. 1. Lateral and front views of the axial fan.

FIG. 2. Illustration of the periodic boundary condition on an unstructured Cartesian mesh.

TABLE I. Parameters defining the computational grids and required resources.

grid G1 ≈0.5 × 108 9-12 4.0 × 10−4 D o 1 608 ≈4 TB

A. Overall flow field analysis

fan, which is defined by the dimensionless pressure gain Ψ = 2

B. Influence of the grid resolution on the time-averaged flow field

D. Flow field near the tip-gap region

FIG. 15. Definition of the locations of points P1 and P2.

tip vortex,” J. Turbomach. 123, 453–463 (2001).

of pressure fluctuations,” J. Fluid Mech. 660, 87–113 (2010).

Int. J. Aeroacoust. 9, 253–272 (2010).

Turbines Power 108, 7–14 (1986).

114, 675–685 (1992).

part I-velocity and pressure fields,” J. Turbomach. 115, 435–443 (1993).

part II-the tip leakage vortex,” J. Turbomach. 115, 444–452 (1993).

Turbomach. 117, 760–771 (1995).

AIAA J. 42, 2320–2331 (2004).

two-point statistics,” AIAA J. 42, 2341–2346 (2004).

engines,” AIAA J. 40, 1518–1528 (2002).

high-pressure compressor,” J. Turbomach. 134, 051005 (2012).

Methodologies, cost and future outlooks,” J. Turbomach. 136, 061009 (2013).

AIAA J. 42, 271–279 (2004).

Phys. Fluids 18, 105102 (2006).

machinery tip-clearance flow,” J. Fluid Mech. 586, 177–204 (2007).

Fluids Eng. 129, 1002–1014 (2007).

16, 328–336 (2007).

of vortex wandering,” J. Turbomach. 137, 061001 (2015).

lation,” Comput. Fluids 31, 695–718 (2002).

ditioning and multigrid,” Comput. Fluids 35, 1126–1136 (2006).

turbulent pipe,” Phys. Fluids 17, 035107 (2005).

Flow 29, 18–34 (2008).

J. Comput. Phys. 235, 786–809 (2013).

AIAA J. 24, 616–618 (1986).

intake,” Noise Control Eng. J. 60, 699–706 (2012).

Potrebbero piacerti anche