Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

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

net/publication/224986654

Improved CFD predictions for high lift flows in the European project
EUROLIFT II

Conference Paper · June 2007


DOI: 10.2514/6.2007-4303 · Source: DLR

CITATIONS READS

25 290

6 authors, including:

Peter Eliasson Pietro Catalano


Swedish Defence Research Agency CIRA Centro Italiano Ricerche Aerospaziali
113 PUBLICATIONS 1,307 CITATIONS 86 PUBLICATIONS 1,177 CITATIONS

SEE PROFILE SEE PROFILE

Jorge Ponsin
Instituto Nacional de Técnica Aeroespacial
27 PUBLICATIONS 181 CITATIONS

SEE PROFILE

All content following this page was uploaded by Peter Eliasson on 19 October 2014.

The user has requested enhancement of the downloaded file.


25th AIAA Applied Aerodynamics Conference AIAA 2007-4303
25 - 28 June 2007, Miami, FL

Improved CFD Predictions for High Lift Flows in the


European Project EUROLIFT II

Peter Eliasson1
FOI, Swedish Defence Research Agency, SE164 90, Stockholm, Sweden

Pietro Catalano2
CIRA, via Maiorise, 81043 Capua, Italy

Marie-Claire Le Pape3
ONERA, BP 72, 92322 Châtillon CEDEX, France

Judith Ortmann4
DLR, German Aerospace Research Center, Braunschweig 38108, Germany

Emilio Pelizzari5
Alenia Aeronautica S.p.A. , Corso Marche 41, 10146 Torino, Italy

and

Jorge Ponsin6
INTA, Torrejón de Ardoz, Madrid, 28850, Spain

In the framework of the European High Lift Programme EUROLIFT II the numerical
and experimental validation of high lift configurations initiated in EUROLIFT are being
pursued. This present contribution describes a summary of the CFD improvements carried
out in this project. The improvements cover turbulence modeling and grid generation. More
sophisticated turbulence models beyond the eddy viscosity concept are evaluated as well as
unsteady effects and efficiency enhancement through wall function approaches. Different
techniques to reduce the number of nodes in unstructured grids have been implemented in
partners grid generators, mainly through the stretching of elements and introduction of
hexahedral elements. The overlapping grid technique for high lift flows are demonstrated as
well. Efficiency enhancement through wall functions show that the cost can be reduced to
about 50% or more. The computed results agree well with results from near wall resolved
grids in the linear range but the maximum lift is over-predicted. Explicit algebraic Reynolds
stress models as well as full differential Reynolds stress models improve the predictions in
the maximum lift region compared to standard eddy viscosity models. Introducing
stretching of unstructured grids along leading and trailing edges as well as introducing
hexahedral elements can substantially reduce the number of grid nodes while maintaining
the solution accuracy. Overlapping grid techniques for high lift flows show good results but
some grid dependencies are observed.

1
Research leader, Division of Defence & Security, Systems and technology.
2
Research engineer, Computational Fluid Dynamics Laboratory.
3
Research engineer, CFD Department.
4
Research Engineer, Institute of Aerodynamics and Flow Technology, Lilienthalplatz 7.
5
Research engineer, TAS Department.
6
Research engineer, Fluid Dynamics Branch.

1
American Institute of Aeronautics and Astronautics

Copyright © 2007 by P. Eliasson. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
I. Introduction

C ALCULATING viscous fluid flows over high lift configurations is still a challenge in CFD. The difficulties in
simulating these flows come from the complexity of both the geometry and flow physics. In particular, the
multiple elements with small gaps give rise to multiple wakes, flow separation, laminar/turbulent transition,
shock/boundary layer interaction etc., where many of these phenomena interact with each other. Since the fluid
dynamics is dominated by viscous phenomena, only high-fidelity simulations based on the Navier-Stokes equations
can provide the required accuracy to obtain realistic CFD solutions.
The numerical simulation of the flow field around high lift configurations based on the Reynolds Averaged
Navier-Stokes (RANS) equations has made significant progress during the last decade1. Until the beginning of the
EUROLIFT project in early 2000, most of the European high-lift activities had been devoted to two-dimensional
computations2. The need for an extension to three dimensions as well as a state-of-the-art experimental database
stimulated the launch of the EUROLIFT programme that was funded by the EC as part of the 5th European
framework program. A close coupling and harmonization between experimental and numerical activities was
attempted in the project. CFD was brought into a more daily use in EUROLIFT and introduced in three dimensions.
Mainly through hybrid Navier-Stokes technology, it has become possible to compute viscous flows about take-off
and landing configurations within a reasonable time frame and with sufficient reliability. The work carried out in the
EUROLIFT project has resulted in several publications, an overview is given in references3-6.
EUROLIFT II7 is a European High Lift Programme in the 6th European framework following the work initialized
in EUROLIFT. The numerical and experimental investigations in EUROLIFT left many important questions
unanswered being pursued in EUROLIFT II. The work is divided into three major work packages. In the first work
package the experimental results from EUROLIFT are further evaluated by numerical calculations covering
Reynolds number effects, transition influence and prediction and wind tunnel installation effects as well as model
deformation. In the second work package the geometrical complexity of the high lift configurations is extended by
attaching a pylon/nacelle with a strake. Both experimental and numerical investigations of these configurations have
been conducted. The work package also covers a new flap design by optimization based on numerical investigations
with experimental verification. In addition, a slat less configuration was studied with passive and active flow control
on the main wing leading edge. The flow control was experimentally carried out based on initial numerical
investigations. In the third work package partners tools were improved. Transition prediction methods incorporated
into partners CFD software as well as improvements for conducting the wind tunnel experiments are examples of
such improvements. The present contribution is also part of this work package and concerns improvements of
partners CFD prediction tools with respect to turbulence modeling and grid generation.
EUROLIFT has demonstrated that the main differences between different numerical results come from the
modeling of turbulence3, 4. As a next step in EUROLIFT II, more sophisticated turbulence models beyond the eddy
viscosity concept have been introduced and evaluated. These cover both Explicit Algebraic Reynolds Stress Models
(EARSM) and full Differential Reynolds Stress Models (DRSM). In addition to steady RANS methods unsteady
effects in RANS and hybrid RANS/LES methods have been introduced and evaluated. Efficiency improvements
though different wall function approaches have been considered by several partners by reducing the near wall
normal grid resolution.
It has also been established in EUROLIFT that high quality results on unstructured grids can be obtained but at
the expense of larger grids. Several partners have worked with approaches to reduce the required number of grid
nodes by increasing the grid anisotropy by stretching elements along leading and trailing edges as well as
introducing hexahedral elements. Another approach being investigated is the overlapping grid technique where a
number of structured blocks are introduced where many of these block partly or completely overlap.
In the following sections the consortium of the European partners will be briefly described. After this follow
results from the turbulence modeling and grid generation parts with concluding remarks at the end.

II. Consortium
The consortium involved in CFD improvements consists of 8 partners. The partners activities are listed in the
table below. Six partners represent European research institutes: CIRA, DLR, FOI, INTA, NLR, ONERA. Alenia
Aeronautica represents the aircraft industry and IBK is an SME (Small and Medium Enterprise).
The first half of the project was devoted to implementation followed by a preliminary and final validation in the
second half. Many of the new features have already been developed in other European projects applied to other
types of applications, one example of this is the project FLOMANIA8 in which several partners participated
implementing EARSM and DRSM turbulence models. To assess the improvements any new implementation should

2
American Institute of Aeronautics and Astronautics
be evaluated and compared to results from EUROLIFT or to results generated in the beginning of the project based
on existing capabilities.
All partners use in-house CFD methods based on finite-volume techniques, usually with explicit time marching
and convergence acceleration based on FAS multigrid. The flow solvers for unstructured grids9, 10 are node and edge
based for hybrid grids with different elements and make use of dual grids agglomerated to coarser control volumes
for the multigrid algorithm.

Table 1. Partners activities and grid type used.


Partner Activity Grid type
Turbulence modeling
CIRA Wall functions Structured
DLR DRSM extension Structured
FOI Wall functions; EARSM/DRSM extension Unstructured
IBK Hybrid RANS/LES Unstructured
INTA Wall functions Structured
ONERA EARSM/DRSM extension; URANS Structured
Grid generation
Alenia Anisotropic grid generation Unstructured
DLR Hexahedral elements Unstructured
FOI Anisotropic grid generation Unstructured
NLR Anisotropic grid generation Unstructured
ONERA Overlapping grid generation Structured

III. Turbulence modeling


The focus of the development and validation of turbulence models is on higher order models beyond the eddy
viscosity concept. Most partners have worked with steady RANS applications, one partner used unsteady RANS
(URANS) in the investigations of a 2D high lift configuration. Although one partner has chosen to apply a
RANS/LES approach (Detached Eddy Simulation, DES) to the same 2D configuration extended to 3D in the
calculations, these results are not reported here since the computed results did not improve the results compared to
steady RANS results. There are, however, unresolved questions concerning time step, convergence criteria,
evolution in time, numerical dissipation etc. to be resolved in future studies.

A. Wall functions
Three partners (CIRA, FOI, INTA) are involved with wall functions in their respective flow solver. CIRA and
INTA use their respective codes for structured grids while FOI uses a code for unstructured grids. In common for
the three approaches is that they all, in theory, can be applied to any near wall y+-values and that they approach the
standard no-slip boundary conditions as the grid is refined normal to the wall.
CIRA has implemented a so-called scalable wall functions approach11. The velocity scale used for the evaluation
of the friction velocity is based on the solution of the turbulent kinetic energy in the logarithmic region of the
boundary layer. This velocity is always positive and allows addressing the issue of the classical wall function
formulations at separation and reattachment points of the flow. The viscous distance (y+) of the first cell close to a
solid boundary is assumed to be not lower than the intersection between the log and the viscous layers. The viscous
layer is taken into account by computing the friction velocity through a blending between the linear ad log law of the
velocity. This method has been implemented into the CIRA flow solver ZEN in conjunction with k-ε, k-ω, and
Spalart-Allmaras turbulence models. At the wall, a Neumann boundary condition is used for k while ω and ε are
imposed in the first cell by a blending between the solutions of the viscous and log regions of the boundary layer.
The working variable of the Spalart-Allmaras turbulence model is calculated in the first cell by considering a linear
variation with the distance from a solid boundary. All the models have been applied to simulate the transonic flow
around 2D and 3D configurations12. In this paper results for high-lift flows obtained using the k-ω TNT turbulence
model13 are presented.
FOI uses a universal wall function approach in which a finite velocity is prescribed on the wall. The method is
based on a formulation by Rung14,15 and extended in EUROLIFT II. The approach has been implemented for any

3
American Institute of Aeronautics and Astronautics
turbulence model based on k and ω. For the results presented in this paper it has been applied to the
Wallin&Johansson EARSM16 with the Hellsten k - ω EARSM model17, in the following denoted Hellsten EARSM.
The universal wall function approach is based on a universal wall law that is an extension of the log law down to the
wall. A finite velocity is prescribed on the wall. The production in the first interior node is modified according to the
wall law and the value of ω is prescribed in the first interior node by adding the viscous and log solutions.
INTA has implemented the so-called generalized wall function approach. This wall function formulation is
similar to that used by FOI but it accounts specially for the effect of adverse and favorable pressure gradients
including those associated with zero skin friction. Therefore, it can be applied to complex flows with acceleration,
deceleration and recirculation. These flow features are common in high lift flows, where strong acceleration and
deceleration may be present at the leading edges of the slat and main components. Flow separation can occur for
flow conditions close to maximum lift on the element upper surfaces and recirculation is present in the coves.
The formulation of the Generalized wall function18,19 is based on the definition of a special velocity scale
consisting of the sum of the friction velocity and a characteristic velocity built via dimensional analysis from the
pressure gradient and kinematic viscosity at the wall. Contrary to the usual wall function formulation, which uses
only the friction velocity as velocity scale, the new velocity scale is never zero (even with flow separation). Based
on an order of magnitude analysis of the RANS equations in the near-wall region, Shih et al.19 have derived a set of
matched asymptotic expansions which are valid for the whole turbulent inner layer.
This generalized wall function has been implemented into the INTA RANS solver for the Spalart-Allmaras and
SALSA turbulence models20. Details concerning the numerical implementation can be found in reference21. The
performance of this wall function formulation has first been
assessed with a high lift 2D test case consisting of the A310
section at landing configuration. The aim of the computations
is to compare the performance of the generalized wall function
against a standard wall function approach with the original
Spalart-Allmaras22 turbulence model. The wall function
computations were carried out on a mesh where 20 ≤ y+ ≤ 40 at
the first node off the wall, while for the computation without
wall function 1 ≤ y+ ≤ 3. A comparison of the three approaches,
concerning the lift polar computation, is depicted in Fig. 1. The
generalized wall function, in this case, shows an improvement
over the standard wall function approach which does not take
into account the effect of the pressure gradient on the friction
velocity determination. Furthermore, the results obtained with
this new formulation are comparable, at least in the linear Figure 1. Computed Lift coefficient for the
range, to those obtained using the original low-Re model. The A310 landing configuration. Comparison of
over-prediction of lift at lower angles is a well-know computed low Re. results with results using
phenomenon due to the under-estimation of the flap standard and generalized wall functions (INTA).
separation4.
All partners used the same 3D test case for validation, a simplified take-off configuration with full span slat and
flap. The Mach number is M∞=0.174 and the selected Reynolds number is Re=15×106, corresponding to the test
conditions in the ETW wind tunnel. Two grids generated for a lower Reynolds number, Re=1.34×106, were used in
the evaluations of the partner’s wall functions. Corresponding experimental and numerical investigations have been
carried out for this Reynolds number as well although not reported here.
The two evaluated grids, one structured multiblock grid and one unstructured grid, are of similar size and consist
of about 3 million nodes. The surface grids of the two grids can be seen in Fig. 2. The average y+-values are about
y+~10 with the highest values at the leading edges which is also displayed in Fig. 2. Two partners are using the
structured grid (CIRA, INTA) and one partner the unstructured grid (FOI). CIRA uses the TNT k-ω model by Kok13,
INTA uses the SALSA turbulence model20 and FOI uses the Hellsten EARSM16,17.
The integrated lift and drag forces are displayed in Fig. 3 where the results from the three partners are cross
plotted. Results from FOI on a near wall resolved grid y+~1 on a refined unstructured grid are displayed as well.
This grid has similar number of nodes on the surface and is refined mainly in the wall normal direction and has
about 8 million nodes in total.

4
American Institute of Aeronautics and Astronautics
Figure 2. KH3Y 3-element take-off configuration. Left: Surface grids of the structured and unstructured grids for
which wall functions are applied .Right: surface y+-values on structured grid. M∞=0.174 and Re=15×106 (CIRA).

Figure 3. Integrated forces for the KH3Y 3-element take-off configuration. Cross plot of three calculations
using wall functions (CIRA, INTA on structured grid, FOI on unstructured grid). Near wall resolved results are
displayed as well where y+~1 (FOI on another unstructured grid). M∞=0.174 and Re=15×106. Left: Lift versus angle
of attack. Right: Lift versus drag.

The numerical results are rather close to each other and reveal that the lift is fairly well predicted with wall
functions in the linear range of the lift curve. The drag, however, is consistently over-predicted by all numerical
results with the largest over-prediction obtained by the results from INTA. From earlier investigations of this
configuration23 it has been established that the major part of this over-prediction comes from the installation of the
half model in the wind tunnel for which cross-flow velocity components are generated in the plane of symmetry.
These velocity components cause higher inboard velocities and hence lower pressure at the slat and main wing
leading edges which in turn results in an over-prediction of drag with up to 10%. The influence from the installation
on the lift is small in the linear range, the main influence is in the maximum lift region for which a delay with about
one degree is encountered. This is further discussed below where different turbulence models are compared.

5
American Institute of Aeronautics and Astronautics
Although the lift is fairly well predicted with wall functions in
the linear range, a too high maximum lift compared to
experimental data is obtained. Only the results using a near-wall
resolved grid results in a maximum lift that agrees well. A closer
inspection of the two solutions from FOI with and without wall
functions reveals that in general the pressures as well as the skin
friction distributions are very similar. The major differences
between the solutions come from the skin friction and this
difference is amplified as the angle of attack. The difference
becomes most obvious at maximum lift for which there is a clear
difference in the skin friction distribution close to the tip, as
displayed in Fig. 4. In the solution with the near wall resolved grid
there is clear reduction in skin friction behind the suction on the
slat element which, at higher angle of attack, contributes to the Figure 4. Skin friction distribution at
flow separation on the slat causing the lift to break down. Since maximum lift for KH3Y. Results at 89%
this phenomenon is missing in the results using wall functions, the span with and without wall functions (FOI).
lift continues to grow and the maximum lift is over-predicted. M =0.174 and Re=15×106.

Hence, calculations using wall functions, as implemented, are not
well suited for reliable predictions of maximum lift.
One of the motivations for using wall functions is the reduced computational effort required due to the less
refined grids required in the wall normal direction. This could also lead to less iterations since the grid will not be as
stretched and hence the grid maximum aspect rations are reduced. For the high lift calculations carried out here,
about the same number of iterations were required as with a near wall resolved grid. Hence, the reduction in
computational effort comes mainly from the fact that smaller grids with fewer nodes are used, in the calculations
carried out here a reduction of about 50% is obtained using wall functions.

B. Turbulence modeling beyond the eddy viscosity concept


In EUROLIFT mostly eddy viscosity models were employed for the prediction of high lift flows in three space
dimensions. Some more sophisticated models were introduced, mainly in two dimensions. It was established that the
main differences between different numerical results come from the modeling of turbulence3,4. Hence, more
sophisticated models have been introduced in EUROLIFT II and validated in three dimensions. Both Explicit
Algebraic Reynolds Stress Models (EARSM) and full Differential Reynolds Stress Models (DRSM) have been
investigated.
FOI has investigated the performance of EARSM16 together with a length-scale determining equation based on ω
derived by Hellsten17. This model has been evaluated on the same 3D test case used above for the evaluation of the
wall functions, the KH3Y 3-element take-off configuration. Comparisons with the most commonly used models, the
Spalart-Allmaras model22 and the Menter SST model24, are displayed in Fig. 5 for this configuration. The grid used
in this investigation is an unstructured hybrid grid with about 8 million nodes and is the same grid that produced the
near-wall resolved results (y+~1) in the previous section on wall functions. The computations were carried out with
steady state calculations assuming fully turbulent flow.
The error bars denote the amplitude of the oscillations in the steady state calculations where they failed to
converge and give an indication in what range corresponding unsteady results could be within. Small differences are
observed between the turbulence models in the linear range, at the lowest angle lift and drag values almost fall on
top of each other. There is a difference in the maximum lift region though. Both Spalart-Allmaras and Menter SST
models produce a higher lift compared to the results with Hellsten EARSM and one might conclude that these
models give an improved prediction since the results are closer to the experimental lift values. The drag is
consistently over-predicted by all models, in the order of 10% higher than experiments.
The over-prediction of drag, however, is a consequence of the installation of the half model in the wind tunnel.
An investigation of these installation effects of this model has been carried out by conducting CFD calculations
inside the ETW wind tunnel and with comparisons to free flight calculations23. As indicated above, the installation
of the model causes cross flow velocity components in the plane of symmetry leading to a local shift of the angle of
attack at inboard stations with higher leading edge velocities and lower pressure which effect mainly the drag but
also the lift. The effect of the installation can be seen in Fig. 6 where free flight and in-tunnel CFD results are
compared to experimental results. Note that both numerical in-tunnel forces and experimental values are corrected to
correspond to free flight with the same correction and that the experimental and free flight EARSM results are the
same as in Fig. 5.

6
American Institute of Aeronautics and Astronautics
Figure 5. Integrated forces for the KH3Y 3-element take-off configuration. Evaluation of three turbulence
models, Hellsten EARSM, Spalart-Allmaras and Menter SST models (FOI on unstructured grid). M∞=0.174 and
Re=15×106. Left: Lift versus angle of attack .Right: Lift versus drag.
Figure 6 reveals that the effect from the installation in the wind tunnel is much larger than the effect of the
turbulence model. Now the in-tunnel results with EARSM are in a very good agreement with experimental lift
values. The over-prediction of drag is entirely due to the installation and a good agreement is obtained with the in-
tunnel results. The maximum lift is shifted to a lower angle for the in-tunnel results and the actual maximum lift
value is very well reproduced in the in-tunnel results.
Hence the higher lift obtained by the Spalart-Allmaras and Menter SST models in Fig. 5 should not be
considered as an improvement although they are closer to experiments. On the contrary, the Hellsten EARSM model
produces good results in the in-tunnel calculations with very good experimental agreement. One might argue that,
considering the results in Fig. 5, both the Spalart-Allmaras and Menter SST models would probably produce too
high lift values when carried out inside the wind tunnel with less accurate agreement with experimental forces
compared to the results with Hellsten EARSM. For a fair comparison and to avoid speculation, however, these
calculations should be carried out and compared to the in-tunnel results with the Hellsten EARSM model.

Figure 6. Wind tunnel installation effects for the KH3Y 3-element take-off configuration. In-tunnel CFD
calculations (corrected to free flight) and free flight calculations compare to corrected experimental data (CFD
calculations from FOI on unstructured grids) .M∞=0.174 and Re=15×106. Left: Lift versus angle of attack .Right:
Lift versus drag.
DLR has implemented a full Differential Reynolds Stress Model in their structured flow solver FLOWer. The
model is the SSG/LRR-ω model developed in the European project FLOMANIA8. The model is a combination of
the Speziale-Sarkar-Gatski (SSG) model being developed for homogeneous flows and the Launder-Reece-Rodi

7
American Institute of Aeronautics and Astronautics
(LRR) model using the Menter blending function between the models and a length scale determining equation based
on ω. Details about the model is given in reference 25.
The validation in 3D is carried out for the same configuration as above, the KH3Y 3-element take-off
configuration, but at a lower Reynolds number, Re=1.4×106. A structured multiblock grid consisting of about 3.8
million nodes was used in the validation, the grid was locally refined with 5 overlapping blocks to avoid some initial
convergence problems. Steady state calculations were carried out assuming fully turbulent flow and by using low
speed preconditioning to speed up the convergence.
Figure 7 displays the integrated forces compared to experimental and numerical results with two other turbulence
models, the Spalart-Allmaras and the standard Wilcox k-ω model. The error bars indicate the amplitude in the
oscillations of the results where the calculations failed to converge. Using DRSM two distinct solutions were found
depending on how the calculations are initiated. The upper branch with higher lift is obtained by initializing the
computations by the solution from the previous lower angle of attack. The flow finally separates and the lift breaks
down at an angle of attack about 2.5 degrees later than in the experiments. The lower branch is obtained by
initialization from free stream or from a higher angle.

Figure 7. Integrated forces for the KH3Y 3-element take-off configuration. Evaluation of three turbulence
models, DRSM SSG/LRR-ω, Wilcox k-ω and Spalart-Allmaras models (DLR on structured grid). M∞=0.174 and
Re=1.4×106. Left: Lift versus angle of attack .Right: Lift versus drag.
The lower branch with earlier lift break down is an
improvement compared to the results obtained with the
Spalart-Allmaras and the Wilcox k-ω models for which the
maximum lift is predicted late, up to 5 degrees too late with
the Spalart-Allmaras model. Similar to the results obtained
earlier in Fig. 5 – 6 for the same configuration at a higher
Reynolds number, the numerical results over-predict the
drag. The experiments were conducted with the same wind
tunnel model but in a different tunnel, the LSWT tunnel in
Bremen. It is likely to believe that there is still an influence
from the installation of the half model in the tunnel since
there is an over-prediction of the drag here as well. The
over-prediction is smaller though compared to the results at
Figure 8. Skin friction lines and distribution for
the higher Reynolds number. In Fig. 8 the skin friction
two solutions at the same angle of attack for the
distributions are displayed of the two separate solutions
KH3Y 3-element take-off conf. (DLR). M∞=0.174
obtained at the lowest angle of attack. The solution with
and Re=1.4×106. Left: higher lift. Right: lower lift.
lower lift show a separation on the slat and rear main wing
at about 70-80% wing chord, the other solution with higher lift has attached flow.
Another comparison between turbulence models has been carried out by ONERA who compare a full Reynolds
stress model developed in FLOMANIA8, an EARSM model coupled with the k-kl two-equation model together with
two eddy viscosity models, the Spalart-Allmaras mode and the k-kl two-equation model without the EARSM
extension.

8
American Institute of Aeronautics and Astronautics
A comparison between these models have been performed in 3D for the low Reynolds number (Re=1.4×106)
KH3Y three-element configuration with slat and flap deployed for landing. Two solutions are available at the lowest
angle of attack only, the Spalart-Allmaras model and the two-equation k-kl eddy viscosity model. The calculations
converged well with DRSM but with EARSM there were convergence problems at higher angles of attack resulting
in large oscillations in the integrated forces. To overcome this problem, unsteady calculations were carried out for
these angles. These unsteady calculations eventually converged into almost steady results at the expense of rather
extreme computing costs. The integrated forces obtained by ONERA for this configuration can be seen in Fig. 9.

Figure 9. Integrated forces for the KH3Y 3-element landing configuration. Evaluation of four turbulence
models, DRSM, EARSM k-kl, Spalart-Allmaras model and eddy viscosity k-kl models (ONERA on structured grid).
M∞=0.174 and Re=1.4×106. Left: Lift versus angle of attack .Right: Lift versus drag.
Results with the Spalart-Allmaras model and the two-
equation k-kl eddy viscosity model show almost identical
values of the forces at the lowest angle of attack. Both
DRSM and EARSM give a fairly good estimate of the
forces in line with results presented by other partners for
similar configurations. The drag is over-predicted by all
numerical results, similar to the results presented above.
The DRSM, however, predicts a lower drag at the lowest
incidence compared to the other models. DRSM predict a
maximum lift that is slightly higher three degrees later
compared to the experimental results. EARSM, on the
other hand, predicts a lower maximum lift only one degree
later than in experiments. In Fig. 10 the skin friction Figure 10. Skin friction distribution for two
contours are shown for the calculations with DRSM and solutions for the KH3Y 3-element landing
EARSM at the incidence where EARSM predicts configuration. (ONERA). M∞=0.174 and
maximum lift are displayed. In both calculations there is a Re=1.4×106. Left: EARSM. Right: DRSM.
tendency to onset of separation at the outer part of the
wing at the rear part of the main wing. The main difference between the solutions occurs at the main wing close to
the fuselage for which the solution with EARSM predicts an area with separated flow which explains the lower lift.

C. Unsteady effects using RANS


ONERA has carried out 2D unsteady RANS (URANS) computations using for the GARTEUR A310 landing
case2,26. The objective was to see if the inclusion of unsteady effects could improve the predictions for this test case.
ONERA has used an in-house CFD code27 for structured multiblock grids for this investigation. For the far field
boundary conditions the computational domain has been extended up to 50 chords using overlapping Cartesian
grids.
The Menter SST k-ω model was used in the calculations. Earlier, calculations with the Wilcox k-ω model had
been carried out but no unsteadiness was observed with this model. Unsteadiness was mainly observed at the lowest

9
American Institute of Aeronautics and Astronautics
computed angle and beyond maximum lift. The unsteady pressure
distribution for the lowest computed angle is displayed in Fig. 11.
The lift oscillates with amplitude of about ΔCL=0.1 at this
incidence and the pressure distributions reveal that the oscillations
occur on the flap where the flow is separated. The agreement with
experimental pressure is very good. Although there is a
fluctuation on the flap, a large separated area is predicted.
Typically, as observed in most other calculations, this separation
is very small or not existing at all4.

IV. Grid Generation

Figure 11. Cp distribution for A310 3-


D. Unstructured grid generation with anisotropic elements element landing configuration. (ONERA).
In EUROLIFT it was established that high quality results on M =0.174 and Re=4.2×106. URANS results

unstructured grids can be obtained at the expense of larger grids. with Menter k-ω model.
Hence, several partners in EUROLIFT II have worked with
different approaches to reduce the required number of grid nodes
by increasing the grid anisotropy, applying a span wise stretching to elements in the whole wing (Alenia) or mainly
along leading and trailing edges and tip (FOI, NLR). Different values of maximum stretching were attained by
partners on high lift configuration geometries, in the order of 6-10 for NLR, 10 for FOI and 20 for Alenia. Below the
results by Alenia are highlighted.
Alenia Aeronautica has developed a fully automatic three-dimensional anisotropic unstructured grid generation
algorithm based on the domain boundary curvature. The approach consists in augmenting the domain to be meshed
with a metric field M(x) (i.e. to define a Riemannian structure on the domain) and in requiring that all the mesh
edges have the same length with respect to M(x) (typically unit edges lengths in M(x) are chosen). The metric field
on the boundary is computed according to the principal curvatures and directions of the boundary surface. An
original anisotropic treatment for “nearly-flat” surfaces has been developed which consists in generating the mesh in
the parametric domain associated to the surface and in mapping the resulting mesh onto the real one. Using the
surface mesh as input data it is then possible to build a background grid to interpolate the value of the boundary
metric field at any other point. The domain can then be discretized by using the 3D anisotropic mesh generator.
More details about the approach are given in reference28.
To evaluate this approach a 3D viscous anisotropic hybrid grid, made of tetrahedral and prismatic elements near
the surfaces, has been generated for the KH3Y three element take-off configuration. The surface grid consists of 35
500 nodes and 88 500 anisotropic triangular elements, with a stretching value up to 20. The 3D volume is filled with
about 0.9 million nodes and 2.9 million elements, divided into 1.7×106 tetrahedral and 1.2×106 prismatic elements.
The viscous part of the grid contains 16 layers of prismatic cells.
A comparison with computations from EUROLIFT on the same configuration but on a different, more isotropic
has been made. The main features of the two grids are compared in Table 2. The isotropic grid has been created by
FOI. It has more than three times as many nodes in the total grid and 5.7 as many nodes on the surface. The surface
grids of the two grids are shown in Fig. 12.

Table 2. Main features of two computational grids for the KH3Y 3 element take-off configuration.
Grid Nodes Boundary nodes Elements Prisms Tetras
Stretched (Alenia) 919 000 35 500 2 966 000 1 183 000 1 783 000
Isotropic (FOI) 3 078 000 201 300 8 382 000 4 682 900 3 593 200

The numerical flow simulations have been performed using Alenia Aeronautica UNS3D finite volume, node
centred solver for the Navier-Stokes equations, with a k-ω EARSM turbulence model16,17. The free stream Mach
number is M∞=0.174 and the Reynolds number is Re=1.4×106.
The integrated forces are displayed in Fig. 13 with experimental comparison. The forces obtained with the new
grid are close to the forces obtained with the original larger grid. The experimental agreement is improved with the
new grid since the lift is slightly higher and the drag is reduced. Unfortunately, the computations were not carried on
beyond maximum lift. The results agree very well with other numerical results for this case. There is a small under-
prediction of lift and an over-prediction of drag as observed for most other computed results.

10
American Institute of Aeronautics and Astronautics
Isotropic

Stretched Stretched Isotropic

Figure 12. KH3Y 3-element take-off configuration. Stretched (Alenia) and isotropic surface grid. (FOI).

Figure 13. Integrated forces for the KH3Y 3-element take-off configuration. Evaluation of computations on
an anisotropic grid (0.92 M nodes) and an isotropic grid (3.1 M nodes). M∞=0.174 and Re=1.4×106. Left: Lift versus
angle of attack. Right: Lift versus drag.
These results show clearly that the accuracy in the
computed results is maintained when stretching the
elements along the leading edges. The calculations also
show that accurate high lift calculations can be obtained
with rather small meshes, in this case on a mesh with less
than 1 million nodes and a 70% reduction of the total grid
nodes compared to the original grid. Results from the two
other partners (FOI, NLR) involved with grid node
reduction by stretching surface elements along the leading
edges confirm that the grid can be stretched with
maintained accuracy, the challenge is to actually achieve
this stretching in the grid generator in an automatic
manner.
In addition NLR has gained experience by finding Figure 14. Illustration of span-wise stretched
more optimal grid settings in a grid approach that surface grid with structured order surface elements
employs span-wise stretching in combination with at the leading edges of the slat and nacelle. (NLR).

11
American Institute of Aeronautics and Astronautics
structured ordered triangles at the leading edges of high-lift wing elements. To make this approach feasible, new
algorithms have been introduced in the NLR in-house Delaunay-based tetrahedral grid generator29. This has been
used to generate three common meshes for the geometrically most complex cases, i.e. a high-lift configuration
including a nacelle, slat tracks, flap track fairings and pressure tube bundles in a relatively short frame of time (two
weeks). A part of a surface grid is displayed in Fig. 14.

E. Unstructured grid generation with hexahedral elements


DLR has worked with the grid generator Centaur to include hexahedral elements at leading edges to increase the
stretching ratios and perhaps also the accuracy. In addition, hexahedral elements are introduced in boxes in the flow
field with the purpose to better resolve vortices to reduce the diffusion of these. These regions are marked in Fig. 15-
16 below which is the KH3Y 3-element landing configuration with the pylon/nacelle added at a slat cut-out. The
Centaur software (http://www.centaursoft.com) is used to generate all grids in this investigation.

Figure 15. Introduction of hexahedral elements for the KH3Y 3-element landing configuration with
pylon/nacelle. Inclusion at leading edges (DLR).

Figure 16. Introduction of hexahedral elements for the KH3Y 3-element landing configuration with
pylon/nacelle. Inclusion at wakes behind the slat horn and strake. Left:, marked in red. Mid: cut at constant span,
grid without hexahedral elements. Right: cut at constant span, grid with hexahedral elements (CAD sources marked).
A large number of grids has been generated and validated on this configuration. An original grid was used as a
starting grid, a hybrid grid with prismatic and tetrahedral elements and 15.6 million nodes. This grid was
successively modified by adding hexahedral elements at the leading edges, adding hexahedral elements in the wakes
and by refining the tetrahedral grid on the suction side outside the prismatic layer. The grids were all generated for a
low Reynolds number, Re=1.3×106, with 30 prismatic (or hexahedral) layers and the grid sizes vary between 14.0
and 17.5 million nodes.
The maximum stretching of the surface quadrilaterals (from which the hexahedral elements are grown) is 6.3 due
to limitations in the grid generator. A small improvement in resolution is obtained by using hexahedral elements
along the leading edges as well as a reduction in number of nodes, about 10%. A hexahedral box around the vortices
increases the resolution and reduces the dissipation of these, illustrated in Fig. 17.

12
American Institute of Aeronautics and Astronautics
Figure 17. Vorticity contours for the KH3Y 3-element landing configuration with pylon/nacelle. Comparison
of the strake vortex at a stream-wise cut above the wing. Left:, original grid. Mid: grid with a hexahedral box. Right:
grid with a refined hexahedral box (DLR).
A summary of the lift and the lift to drag ratio for the results on some of the grids is given in Fig. 18.
Experimental values are included as well. The lift is under-predicted and the experimental maximum lift is never
reached in the computations. The reason behind this is unclear, as presented earlier the maximum lift has been
reached in most calculations but quite often at a higher angle of attack than in experiments.

Figure 18. Integrated forces for the KH3Y 3-element landing configuration. Evaluation of computations on
different grids. M∞=0.175 and Re=1.3×106. Left: Lift versus angle of attack. Right: Lift to drag ratio versus angle of
attack.
The predicted forces on the different grids are clustered fairly
well together with some exceptions. With the increased resolution
on the leading edges using the hexahedrons with the high
anisotropy of 6.3 the stall could be shifted to a higher angle of
attack compared to the results with the original
prismatic/tetrahedral grid. A lower lift is obtained for the solution
where the tetrahedral grid has been refined on the suction side at
the end of the prismatic/hexahedral layers. Still, this solution has a
constant lift to drag ratio which is an improvement although the lift
moves further away from experimental values. Good agreement is
in general observed between computed and experimental pressure.
The main difference is observed in one section at 28% span in the
vicinity of the pylon/nacelle, this is visualized in Fig. 19 for the Figure 19. Pressure distribution at
initial and the hexahedral grid with refined tetrahedral grid. The lowest computed angle for KH3Y 3-element
plot reveals that there is an under-prediction of the suction for both landing configuration. Results with initial
grids contributing to the low predicted lift. The main difference and refined grid, 28% span (DLR).

13
American Institute of Aeronautics and Astronautics
between the pressure distributions from the two solutions is observed on the flap where the grid with refined
tetrahedral part gives an improved prediction with a higher pressure due to improved wake resolution. Obviously,
more work should be devoted to resolve the underlying problem with the under-prediction of the total lift.

F. Overlapping grid generation


In EUROLIFT II the overlapping grid technique, or Chimera approach, has been applied to a slat-less two-
element high lift configuration by ONERA. Sub Boundary layer Vortex Generators (SBVG’s) are attached close to
the leading edge to delay separation and to increase the maximum lift to regain some of the high lift performance
that is lost when the slat is removed. The Chimera approach is used by generating separate grids for the two
elements and the SBVG’s, being combined in calculations with and without the SBVG’s.
In the Chimera approach, independently generated overlapping grids are combined in a CFD calculation30-33. It
allows, for example, flow computations around bodies in relative motion without re-meshing31, and computations of
configurations where geometrical details are added can be easily computed by successively adding the
independently generated grids. In the present approach, the grids consist of structured multi-block grids where the
blocks may be either patched (not overlapping, node to node correspondence on the block boundaries) or
overlapping. Grids in overlapping areas must have a similar spatial resolution and overlapping areas must be
sufficiently wide to permit exchange between the grids. Information between overlapped blocks is exchanged at
each time step in the flow computation by a conservative variable interpolation of two kinds: a classical overlapping
in the interior of other domains, and overlapping of solid bodies causing holes in the overlapping grid. These holes
are blanked and not computed by the flow solver.
The Chimera method is available in the elsA software34 which solves the compressible, three-dimensional
Reynolds averaged Navier-Stokes equations in a cell centered finite-volume formulation. The spatial discretization
uses multi-block structured meshes. The solution of the RANS system with a one or two-equation turbulence model
is based on a decoupled approach allowing the use of different spatial discretizations: the Jameson scheme for the
RANS system and a first or second order upwind scheme for the turbulence equations. Time integration may be
conducted either by a backward Euler scheme or by an explicit 4-step Runge-Kutta scheme. LU implicit acceleration
techniques are associated with the backward Euler scheme, whereas IRS is associated with the Runge-Kutta
integration. The FAS multigrid method is also implemented to accelerate the convergence.
An algorithm in elsA searches for overlapped cells and for nodes inside solid bodies to be blanked. To accelerate
the searching, bodies are represented by simple geometrical shapes as juxtapositions of parallelepipeds. In addition,
a preconditioning by a Cartesian grid locally refined is used. Different techniques, like extrapolation, implicit
interpolation and scheme degeneration, permit to diminish the constraints to two cell overlapping32. No special
treatment is applied in the boundary layer region where the grid is highly stretched and where strong flow gradients
exist, these regions are detected though. Interpolation between “Euler” and “Navier-Stokes” regions is avoided by
extending the holes and the blanking of blocks overlapping boundary layer regions. elsA returns a list of interpolated
and blanked cells with interpolation coefficients. It also detects non-interpolated cells, called orphan cells, and cells
which are interpolated in already interpolated cells. If such cells exist, the user has to manually change the
interpolation exchange or locally refine the grid.
The SBVG test case to which Chimera has been applied is a slat-less two-element AFV configuration with a
constant chord and a wing sweep angle of 40° 35. This configuration has been examined in a related flow control task
in EUROLIFT II where both passive (through SBVG’s as in this investigation) and active flow control (through
constant blowing) has been applied to the leading edge of the main wing with the purpose to increase maximum lift
and delaying flow separation. A row of vortex generators is placed at the leading edge, in the calculations only one
SBVG is included in the computational domain assuming infinite sweep with periodic boundary conditions span
wise. Hence, the computational domain contains only a small slice of the wing with a span wise extent of 0.012 m
being the distance between two SBVG’s. The chord of the wing is 0.6537 m and the span of the complete wing is
2.0 m. The free stream Mach number is M∞=0.146, the Reynolds number is Re=3.3×106.
The configuration without SBVG is represented with three overlapping structured blocks, Fig. 20, created with a
transfinite method grid generator36 consisting of a background grid and a grid for the wing and flap. To reduce the
influence of the far field boundary, the background grid extends 100 chords. Wing and flap meshes are nearly
independent, the node number and their distribution for each body have been chosen to have a regular overlapping
grid which is illustrated in the right-most grid in Fig. 20 where blanked areas have been left out. Both wing and flap
grids are made of a C-mesh topology around the bodies and an H-mesh topology for the base areas. Outer
boundaries of the wing and flap grids are located about two chords away. The main wing grid contains, in two
dimensions, about 13300 cells and the flap 13000 cells. The size of the first body cell in the normal direction is
equal to 2.5×10-6 normalized by the chord length, which leads to y+ values in the range 0.5 ≤ y+ ≤ 2.5. To create a

14
American Institute of Aeronautics and Astronautics
grid in three dimensions, these meshes are duplicated on 12 planes with a sweep angle of 40°. The whole
computational domain without the SBVG contains about 440 000 cells. Of these, about 27 000 cells (about 6%) are
interpolated where 11 500 belong to the wing mesh, 9 100 to the flap mesh, and 5 200 to the background mesh.

Figure 20. AFV wing-flap configuration overlapping meshes. 400 000 cells in 3 blockss and 13 planes, 27 000
interpolated cells.
In-between two lateral planes, the SBVG is placed on the wing with an angle of 20 degrees to the flow, near the
wing leading edge, see Fig. 21. The height of the SBVG is 0.001 m and equals approximately the height of the
boundary layer. It has a triangular shape with no thickness in the calculations and the length is 0.01 m. To avoid
blanking due to the lateral planes, the SBVG mesh has been generated exactly between these planes. The SBVG
mesh is composed of eight blocks. The near wall normal resolution is similar to the wing mesh. The number of cells
for the SBVG mesh is about 116 000, which leads to about 556 000 cells for the whole configuration, and which
represents about 21% of the total cell number. In total elsA finds 561 interpolated cells due to blanking of the
SBVG, and about 14 900 interpolated cells representing about 13% of the cells in the SBVG grids.

Figure 21. Grid topology and surface mesh for the SBVG on the two-element AFV configuration. 116000
cells in 8 blocks.
Calculations have been produced with and without SBVG’s. The two-equation k-ω turbulence model from
Wilcox and the one-equation Spalart-Allmaras model have been used. Computations have been done using an Euler
backward algorithm associated with a LU implicit relaxation technique with CFL = 100. An adiabatic no-slip
boundary condition is applied on bodies except on the bases governed by a slip condition.
The integrated lift of all computations is displayed in Fig. 22. Without a SBVG, the maximum lift is obtained at
an incidence of α = 9º using the Spalart-Allmaras turbulence model, beyond that the lift decreases and falls rapidly
at α = 10.5º where flow separation occurs. To estimate the effects of the SBVG, the 8 blocks containing the SBVG
are added to the computational domain. Calculations with these blocks included are first carried out at maximum lift
where the SBVG is supposed to have the largest effect and at stall where flow separation without SBVG occurs. The
calculations are carried out by first ignoring the effect from the SBVG by using through-flow boundary conditions
through the SBVG, similar to what is done at a periodic boundary. In a second step, the effect of the SBVG is added
by using standard no-slip wall boundary conditions. Adding the SBVG blocks in the computations apparently has an
effect of the results. Figure 22 reveals that with these blocks added ignoring the SBVG, a decrease of lift about
ΔCL=0.06 is obtained compared to the results without the SBVG using the Spalart-Allmaras model. This indicates
that there is a mesh dependency effect caused by the SBVG blocks. Including the SBVG in the calculations causes a

15
American Institute of Aeronautics and Astronautics
small increase of the computed lift by about 1% compared to the results with the same grid ignoring the SBVG.
Hence there is a small but positive effect of the SBVG at this angle. Beyond maximum lift, at α = 10.5º, the
computed results with and without the SBVG show separated flow indicating that the SBVG with this shape, this
size and in this position is not efficient to prevent flow separation.

Figure 22. Lift polar and pressure distributions for the two-element AFV configuration. Left: lift vs. angle of
attack for all computations. Mid: Cp distribution at α = 9º with Spalart-Allmaras model. Right: Blow-up.
The pressure distributions in Fig. 22 shows that the highest suction giving the highest lift is the results without
the SBVG and the SBVG blocks. There is a small but notable effect from the SBVG comparing the two results
where the SBVG blocks are included. Apparently, the influence from the mesh or interpolation is greater than the
influence from the SBVG itself. Applying a different turbulence model does not change the conclusions. The results
using the Wilcox k-ω turbulence model, included in the lift polar in Fig. 22, show the same mesh dependency but
with higher lift values and reversed effect of the SBVG block causing higher lift than with no blocks. In addition, a
later stall is obtained with this model. The relative gain from the SBVG is the same, about 1% increase in lift.
Figure 23 gives results obtained for α = 9º with and without SBVG with the same mesh. The SBVG diverts the
streamlines in the very near-wall part of the boundary layer causing pressure variations on both sides of the SBVG.
Very small scales have been chosen for the drawings to see the small SBVG effect.

Figure 23. Cp distribution for α=9° without (left) and with SBVG (mid). Streamlines near the SBVG (right).

V. Conclusions and Perspectives


A summary of achievements on CFD improvement activities for high lift flows in the project EUROLIFT II is
presented, the improvements turbulence modeling and grid generation. Turbulence modeling include the validation
of wall function for high lift calculations, the introduction of advanced turbulence models beyond the eddy viscosity
concept through two-equation Explicit Algebraic Reynolds Stress Models (EARSM) and full Differential Reynolds
Stress Models (DRSM), unsteady RANS (URANS) computations and the introduction of hybrid RANS/LES to high
lift flow computations. In grid generation the objectives were to reduce the number of nodes in unstructured grids by
anisotropic span-wise stretching, introduction of hexahedral elements and application of the overlapping grid
(Chimera) technique to high lift flows.

16
American Institute of Aeronautics and Astronautics
In general the objectives for both turbulence modeling and grid generation have been met although many
questions still remain to be answered.
Wall functions have been applied to high Reynolds number high lift applications on grids generated for lower
Reynolds number. A reduction of the computational effort of about 50% is obtained and the agreement with
experimental data and near wall resolved (y+~1) numerical results are satisfactory in the linear range of the polar. At
the maximum lift region, however, a consistent over-prediction of maximum lift is obtained by all partners for the
considered test case. With a near wall resolved grid the maximum lift is well predicted. Hence, it is concluded that
wall functions, as implemented and in particular where y+~10 , are not suitable for maximum lift predictions.
Both EARSM and DRSM have shown improved results compared to simpler models like the Spalart-Allmaras
model. It has been demonstrated that EARSM provides a good estimate of maximum lift for which other models
over-predict it. In a similar manner, DRSM has been shown to get an earlier lift break down with closer
experimental agreement. However, wind tunnel installation effects dominate over the turbulence model effects. To
have reliable conclusive results against existing experimental data, future investigations should be carried out inside
the wind tunnel as initiated in another task in the project.
Unsteady RANS (URANS) simulations were carried out, the investigation was mainly limited to 2D but also
some URANS calculations in 3D were performed. Good results were obtained in 2D for the A310 landing
configuration with the Menter SST model. In 3D, URANS was used by one partner who obtained large oscillations
in the convergence of the RANS calculations. The URANS calculations, however, converged to almost steady
solutions. Unsteady calculations will become more and more important in the future as more unsteady flows will be
computed and as unsteady phenomena are revolved with increased grid densities.
An attempt to use hybrid RANS/LES (DES) for high lift calculations was made for the A310 configuration. The
DES calculations did not improve the prediction of this test case compared to RANS results and no results are
presented here. However, there is a potential in applying hybrid RANS/LES to high lift predictions in the maximum
lift region. There are several question marks that can be focused upon in future studies like the selection of time
step, convergence criteria, evolution in time, numerical dissipation, selection and generation of grids etc.
The total number of nodes can be reduced by using stretched surface elements span wise. The solution accuracy
is not degraded as the elements are stretched with increased anisotropy. One partner managed to reduce the number
of nodes by almost 70% to a grid with less than one million nodes and the results compare very well to the ones on a
grid with 3 million nodes. The main difficulty is to incorporate this functionality in a grid generator.
Hexahedral elements are included in unstructured grids to resolve leading edges and to resolve wake vortices. A
small reduction in the number of nodes is obtained and the quality of the computed results is maintained or
improved. As for the unstructured grids, the main challenge is to get the generation of hexahedral elements to work
as desired in the grid generator. Apparently, more work needs to be devoted to grid generation.
The overlapping grid technique, or Chimera technique, has been applied in 3D with structured multi block grids.
Good results have been achieved with overlapping grids but the results also show that overlapping grids can
introduce grid dependency. Insertion of extra blocks for a given problem affects the results and hence more work is
required on interpolation, conservation etc. to overcome this.
We are still on the learning curve to generate good grids with a reasonable amount of nodes distributed where
they are needed. Grid generation is an area where much more work is required to generate good grids and to spend
grid nodes economically. The effects from different grids are often larger than the effects of different turbulence
models. With a sufficiently refined grid the difference between different turbulence models is concentrated mainly
in the maximum lift region. Perhaps the main challenge in the near future is to generate these well resolved grids.
Another issue to bring forward is the amount of manual work involved in grid generation. Grid adaptation is to a
large extent made by hand based on experience and on gradients in the solution and geometries. A more automated
procedure with advanced sensors that automatically distribute nodes economically is desired.

Acknowledgments
This work has been carried out within the EUROLIFT II project (European High Lift Programme II). It has
partly been supported by the European Commission under contract No. AST-2004-502896.

References
1
Rumsey, L. R. and Ying, S. X., “Prediction of high lift: review of present CFD capability, Progress in Aerospace Sciences”,
Vol. 38, 2002, pp. 145 - 180.
2
Lindblad, I.A.A. and De Cock, K.M.J., “CFD prediction of Maximum Lift of a 2D High Lift Configuration”, AIAA 1999-
3180, 1999.
3
Rudnik, R. “CFD Assessment for 3D High Lift Flows in the European project EUROLIFT”, AIAA 2003-3794, 2003.

17
American Institute of Aeronautics and Astronautics
4
Eliasson, P. “CFD Improvements for High Lift Flows in the European project EUROLIFT”, AIAA 2003-3795, 2003.
5
Perraud, J., Moens and F. Séraudie, A. “Transition on a High Lift Swept Wing in the European project EUROLIFT”, AIAA-
2003-3796, 2003.
6
Rudnik, R., Eliasson, P. and Perraud, J. “Evaluation of CFD methods for transport aircraft high lift systems”, The
Aeronautical Journal, 2005, pp. 53-64,
7
EUROLIFT II-European High Lift Programme II, Annex B “Description of Work”, A Specified target Research Project of
the 6th European Framework Program, GRD-2004-502866, 2004.
8
Haase, W., Aupoix, B., Bunge, U. and Schwamborn, D. “FLOMANIA – A European Initiative on Flow Physics Modelling”,
Notes on Numerical Fluid Mechanics and Multidisciplinary Design, Springer, Vol. 94, 2006.
9
Eliasson, P., “EDGE, a Navier-Stokes solver for unstructured grids”, Proc. to Finite Volumes for Complex Applications III,
ISBN 1 9039 9634 1, 2002, pp. 527-534.
10
Schwamborn, D., Gerhold, T., Heinrich, R., “The DLR TAU-Code: Recent Applications in Research and Industry”,
ECCOMAS CFD Conference, 2006.
11
Grotjans H. and Menter F. “Wall Functions for General Application CFD Codes”, Proc.of ECCOMAS 98 Conference,, pp.
1112-1117.
12
Catalano P. “Transonic Flow simulations using Wall Functions”, 2nd European Conferecne for aErospace Sciences, July
2007, Brussels
13
Kok, J.C., “Resolving the Dependence on the Free-Stream Values for the k-ω Turbulence Model”, AIAA Journal, Vol. 38,
No. 7, 2000, pp. 1292-1295.
14
Rung, T. “Formulierung universeller Wandrangbedingungen für Transportgleichungsturbulnzmodell”, TU-Berlin
Institutsbericht Nr. 02/99, 1999.
15
Morgenweck, D. “Investigation and Validation of a Hybrid Wall Boundary Condition in Turbulent Flows”, TU-Berlin
Report, Matrikel Nr. 197433, 2005.
16
Wallin, S., and Johansson, A., V. “An explicit algebraic Reynolds stress model for incompressible and compressible
turbulent flows”, J. Fluid Mech, Vol. 403, pp. 89-132, 2000.
17
Hellsten, A., “New Advanced k-ω Model for High Lift Aerodynamics”, AIAA Journal, Vol. 43, No. 9, pp. 1857-1869,
2005.
18
Shih, T., Povinelli, L., Liu, N. Potapczuk, N., and Lumley, J., “Turbulent Surface Flow and Wall Function”, AIAA 99-
2392, 1999.
19
Shih, T., Povinelli, L. and Liu, N. “Application of Generalized Wall Function for Complex Turbulent Flows ”, Journal of
Turbulence , 2003.
20
Rung, T., Bunge, U., Schatz, M. and Thiele, F. “Re-statement of the Spalart-Allmaras Eddy-Viscosity Model in Strain-
Adaptative Formulation ”, AIAA Journal, Vol 41, Nº 7, 2003.
21
Ponsin, J. “Application of a Generalized Wall Function for High Lift Flows ”, EUROLIFT Technical report, Nº D3.2.1-5,
2007.
22
Spalart, P., R. and Allmaras, S., R. “A One-Equation Turbulence Model for Aerodynamic Flows”, AIAA 92-439, 1992
23
Eliasson, P. “Numerical Validation of a Half Model High Lift Configuration in a Wind Tunnel”, AIAA 2007-262, 2007.
24
Menter, F. “Two-Equation Eddy Viscosity Models for Engineering Applications”, AIAA Journal, Vol. 32, No. 8, 2005, pp.
1598-1605.
25
Eisfelt, B. and Brodersen, O., “Advanced Turbulence Modelling and Stress Analysis for the DLR-F6 Configuration”, AIAA
2005-4747, 2005.
26
Thibert, J. J. “The GARTEUR High Lift Research Programme”, AGARD CP-515, High Lift Aerodynamics, 1993.
27
Couaillier V., “Numerical simulation of separated turbulent flows based on the solution of RANS/Low Reynolds two-
equation model”, AIAA 1999-154, Reno, 1999
28
Selmin, V., Pelizzari, E. and Ghidoni, A. “Fully Anisotropic Unstructured Grid Generation with Application to Aircraft
Design”, in proc. ECCOMAS 2006 Conference, P. Wesseling et al eds. , TU Delft, The Netherlands, 2006.
29
van der Burg, J. W., van der Heul, D., and Schippers, H. “Experiences with unstructured grid coarsening for conformal
antennas”, in Proceedings of the 9th International Conference on Numerical Grid Generation in Computational Field
Simulations, San Jose, USA, June 2005. Also issued as NLR Technical Publication 2005-193.
30
Steger J. and Benek J.A., “On the use of composite grid schemes in computational aerodynamic”, Computer Methods in
Applied Mechanics and Engineering, Vol. 64, 1987.
31
Le Pape M.C., Darracq D. and Guillen Ph., “Design of a Chimera unsteady code, application to store separation”, 7th
International Symposium on CFD, Beijing (China), September 15-19, 1997.
32
Le Jeanfaivre G., Benoit Ch. and Le Pape M.C., “Improvement of the robustness of the Chimera method”, 32nd AIAA Fluid
Dynamics Conference and Exhibit, Saint-Louis (USA), June 23-27, 2002.
33
Fillola G., Le Pape M.-C., Montagnac M., “Numerical Simulations around Wing Control Surfaces”, ICAS Congress 2004.
34
Fillola Cambier L. and Gazaix M., “elsA : an Efficient Object-Oriented Solution to CFD Complexity”, 40th AIAA
Aerospace Science Meeting and Exhibit, Reno (USA), January 14-17, 2002.
35
Thibert, J. J. , Reneaux, J., Moens, F. and Preist, J. “ONERA Activities on High-Lift Devices for transport Aircraft”, CEAS
Forum on High-Lift and Separation Control, ONERA – TP 1995-48, 1995.
36
Jacquotte, O.P., Montigny-Rannou F. and Coussement G., “Generation, optimization and adaptation of multiblock grids for
complex configurations”, Survey on Mathematics for Industry, Vol. 4, pp.267-277, 1995.

18
American Institute of Aeronautics and Astronautics

View publication stats

You might also like