F. Kafyeke
Manager, Advanced Aerodynamics
Bombardier Aerospace
Fassi.Kafyeke@notes.canadair.ca
Tel: (514) 85552189
Introduction
The design of a modern airplane brings together many
disciplines: structures, aerodynamics, controls,
systems, propulsion with complex interdependencies
and many variables. This paper illustrates the use of
CFD in the solution of a number of aircraft integration
problems. These include the design of transonic wings,
the installation of the power plant, the prediction of
wing deformation and the prediction of maximum lift
characteristics of complete aircraft configurations.
Wing Design
In 1978, the Canadair Challenger became the first civil
aircraft to fly with a supercritical wing designed with
CFD methods; these methods included Jameson's
FLO22 full potential code for transonic wings, the
BGK 2D transonic airfoil code, and Canadair's
MDRAG panel program for 2D highlift airfoils. Since
then, CFD has been used as the principal tool for
aerodynamic design and development of several new
Bombardier jet aircraft:
The 50passenger CRJ200 Regional Jet (1992);
The longrange, highspeed, Global Express Business
Jet (1999);
The 70passenger regional jet CRJ700 (2001);
The 86passenger regional jet CRJ900 (2002);
The Challenger 300 SuperMidsize Business Jet
(2003),
The approach to aircraft design was traditionally based
on wind tunnel testing with flighttesting being used
for final validation. CFD emerged in the late 1960's. Its
role in aircraft design increased steadily as speed and
memory of computers increased. Today CFD is a
principal aerodynamic technology along with wind
tunnel testing and flighttesting. Stateoftheart
capabilitiy in each of these technologies is needed to
achieve superior performance with reduced risk and
low cost.
Bombardiers capability in CFD has grown since it first
became involved in it in 1968. Several codes were
developed inhouse and others acquired from outside to
provide a comprehensive toolbox for the analysis and
design of complete aircraft and its components. These
methods, validated by wind tunnel and flight test data
from the successive aircraft programs, are used in the
design of new aircraft.
Advanced aerodynamic wing design methods
To design transonic wings, several methods are used.
The most common is the wing shape optimization
program ALLOP developed inhouse. Using a
gradientbased optimizer, it is used to match a user
supplied target pressure distribution. The control points
of a NURBS representation of the geometry are used as
design variables in order to optimize the pressure
distribution locally or globally. The ALLOP optimizer
can call a variety of 2D and 3D high speed or low
speed aerodynamic analysis codes. These codes return
results that are used to calculate the current value of an
objective function to minimize. Geometric constraints
are imposed using penalty functions.
Several developments were made to the representation
of wing shape in order to include typical manufacturing
constraints on the aerodynamic lines. The wing
geometry is defined by a set of defining sections. Each
of these sections is represented by a 2D NURBS and
the wing surface is obtained with linear lofting between
the defining sections. A methodology was put in place
in order to easily impose manufacturing constraints on
a 3D wing geometry. The constraints, defined in
planes, are built with conics and/or lines. These conics
are themselves defined by parameters (slopes and
eccentricity) that can act as design variables during the
optimization.
A second method, INDES [1], an inverse design code
originally developed by Tohoku University, in Japan,
was linked to two transonic analysis codes. First,
INDES was linked to the MGAERO 3D Euler code for
complete aircraft of Analytical Methods Inc. The
MGAERO version used included a boundarylayer
coupling introduced at Bombardier [2]. INDES was
also linked to Bombardiers KTRAN transonic small
disturbance code [3] and FANSC Euler/NavierStokes
code for complete aircraft [4]. INDES is also used to
match a given target pressure distribution. Figure 1
shows how a typical target pressure distribution on a
3D wing section is achieved by application of INDES
and KTRAN. The advantage of using INDES resides in
its rapid convergence.
Figure 1: Optimization with INDES and KTRAN using
target pressure distributions. A typical pressure distribution
obtained on the wing is shown.
A third method, Epogy/AeroPointer [5], licensed
from Engineous Corporation, is an optimization
environment capable of performing multidisciplinary
optimizations using a global parameter as an objective
function, such as the total aircraft drag or weight.
AeroPointer was linked to Bombardier's KTRAN and
FANSC transonic analysis codes. The different
capabilities of these methods are complementary, and
each can be used effectively in the overall wing design
process.
The main differences between ALLOP and INDES, the
two methods for optimizing pressure distributions, is
their relative speed of execution and flexibility. Since
it is an inverse method, INDES is significantly faster.
INDES will converge or achieve its best result in some
20 calls to the analysis code. In comparison, ALLOP
requires hundreds of function calls, with the length of
the optimization depending on the number of design
variables. ALLOP optimizes the location of the control
points of a NURBS representation of the geometry, so
the more points are used or the greater the number of
airfoil sections, the longer the optimization will last.
INDES may or may not achieve a given target pressure
distribution. If INDES does not converge on the
specified target, the best that can be done is to modify
the target pressure distribution itself. ALLOP is more
flexible because it can be restarted with a different set
of design variables, and will usually continue to
converge towards the target. Typically, for a complex
design, ALLOP must be restarted a number of times,
and the complete process may last in the order of two
or three days. Unlike INDES, ALLOP will always
produce smooth airfoil sections since it uses NURBS to
describe the geometry. Another advantage of ALLOP
is that it allows the user to work on a portion of a wing,
or on a part of an airfoil section. For instance, the user
may optimize only the upper surface of the wing, or
only the leading edge, etc. For these reasons, INDES
will typically be used to initiate an optimization
process, because it does a good part of the work in a
short period of time. ALLOP is then used to refine the
design.
The major disadvantage of using either ALLOP or
INDES is the requirement to define a target pressure
distribution. This is not only a time consuming process
but it also assumes that the designer has enough
experience to know what an optimal pressure
distribution is for a specific wing. In contrast, no target
pressure distribution is required when using the
AeroPointer software. The latter is capable of
performing a multidisciplinary optimization by
minimizing a global parameter, such as a combination
of the total drag and the weight of an aircraft. This
capability is very useful since it makes no assumptions
about the pressure distribution, and it effectively
automates the design process.
Automation in the design process is important not only
from the point of view of efficiency, but also because it
makes multidisciplinary optimization possible, since
the latter can only be done through the minimization of
global parameters. AeroPointer achieves this capability
through the use of a hybrid optimizer that combines the
capabilities of genetic, gradient, and simplex methods.
AeroPointer also allows the user to define any
geometric parameter as a design variable or a
constraint, and can perform weighted multipoint
optimizations. Naturally, the quality of the final design
will depend on the fidelity of the analysis code and on
the topology of the design space. Typically, a careful
selection of design variables and constraints is required
to ensure a successful optimization, and the
methodology developed for one application may not
necessarily be optimal for another. In some cases,
AeroPointer will produce a good design but one that
can clearly be improved in some areas. In such a case,
an AeroPointer optimization would be followed by the
further optimization of the pressure distributions using
either INDES or ALLOP.
The current best approach therefore is to initiate the
design process with AeroPointer to define the general
characteristics of the optimal design in an MDO sense
(Figure 2). This process would typically be initiated
with a low fidelity analysis code and completed with a
higher fidelity code whenever practical. Once the
general characteristics of the configuration have been
x/c
C
p
,
C
p
t
a
r
g
e
t
Cp
starting point
Cp
target
Cp
final solution
defined, any further improvements required in the
pressure distributions could then be achieved using
INDES if possible or ALLOP when detailed
refinements are required (Figure 3).
Figure 2: (a) Initial business jet configuration; KTRAN
solution; M=0.8 CL=0.5. (b) Configuration optimized with
AeroPointer/KTRAN; M=0.8 CL=0.5
Figure 3: Pressure distribution achieved with an MDO sense
optimization compared to a final design including a locally
refined pressure distribution.
CFD flow analysis and drag prediction
The accurate prediction of drag and its different
components is essential in wing design. Bombardier
CFD development efforts have concentrated on the Full
Aircraft NavierStokes Code FANSC [4]. The program
uses multiblock structured grids, with an unstructured
block topology, i.e. it allows any number of blocks to
merge at the same location. It uses a cellcentered finite
volume approach with a choice of spacediscretization
schemes and an explicit RungeKutta timemarching
method. The code can be run in Euler mode, in Euler
mode with boundarylayer coupling and in Navier
Stokes mode. The NavierStokes code uses the Spalart
Allmaras and k turbulence models [4].
To be useful in a realistic design environment, the
FANSC code was made robust for solutions on
complex aircraft geometry. Boundary conditions
include noslip and slip walls, transpiration wall for
boundarylayer coupling, symmetry and degenerate
lines and points, Riemann and engine inlet/outlet
boundary. The code allows also the specification of
multiple boundary conditions on each block face. Its
run time efficiency was considerably improved by
adding coarse grain parallelization on blocks (3.6 out
of 4 CPUs) and vectorization (94% efficient). Several
preprocessing features were implemented to detect
errors in the topology or grid file. Specific solver/grid
quality criteria were constructed to ensure successful
flow analyses.
The large CPU time of NavierStokes computations
still precludes their inclusion in routine design and
optimization loops, unless and adjoint formulation of
the derivatives is used. Euler/boundarylayer
computations are used instead. A boundarylayer code
was developed and coupled with FANSC first through
the use of a direct Viscous/Inviscid Interaction (VII)
scheme [6]. The coupling uses a transpiration velocity
approach, with no need to regenerate a new mesh at
every VII cycle. Since a direct VII procedure fails
when separated flow is encountered, as often found
during design iterations, an inverse boundarylayer
code was also coupled with FANSC using a quasi
simultaneous VII scheme. The viscous flow is solved
with the CIBL3D inverse code, developed by Cebeci et
al. at California State University, Long Beach [7].
With this code, separated boundary layers can be
computed with accuracy comparing favorably with
more timeconsuming NavierStokes computations for
many cases of interest. This is illustrated in Figure 4,
showing a pressure distribution computed at midspan
of a Challenger wingbody configuration.
To be used effectively in aerodynamic design loops,
CFD codes must produce accurate, reliable and
repeatable drag estimates. Drag modules were
constructed as a postprocessing step to the
Euler/Boundarylayer solutions. They include a semi
empirical module for fuselage and nacelle drag, a
Multhopp algorithm for induced drag, Locks method
for computation of wave drag based on shock strength
and a SquireYoung module for the computation of
wing and tailplane viscous drag. Farfield methods for
the induced drag and the computation of wave drag
from the integration of entropy variation across shock
waves were also investigated.
(a)
(b)
x/c
C
p
AeroPointer
Refined Design
x/c
C
p
0.25 0.5 0.75 1
1.5
1
0.5
0
0.5
1
Experiments
FANSC Euler/BLDirect
FANSC Euler/BLSemiInverse
FANSC NavierStokes
Figure 4: Euler/Boundary layer and NavierStokes
computations of flow on the Challenger wing/body
configuration Mach 0.82, Alpha = 1.5 degrees, Rec = 6
Million, station at 40.5% of the wing semispan.
More recently, investigations were made in the
prediction of drag from direct integration of pressures
and skin friction obtained with a highaccuracy Navier
Stokes solution. The difficulties of predicting drag
through NavierStokes computations were illustrated at
two AIAA Drag Prediction Workshops held in June
2001 in Anaheim, California [8] and in June 2003 in
Orlando, Florida [9]. In the first workshop, FANSC
was used to predict the drag polar of a DLRF4 wing
body for which experimental results had been collected
in NLR, ONERA and DRA wind tunnels. Drag was
obtained from integration of pressure and skin friction
coefficients, as specified for the workshop. Initial
predictions made by FANSC with the grid supplied by
the workshop organizers showed drag levels much
higher than experimental values (DPW grid results, in
Figure 5). A new mesh of the DLRF4 generated using
Bombardiers MBGRID program [10] was prepared.
The mesh had good orthogonality on the solid surfaces,
106 chord wall spacing, 3.8 Million mesh points, an
open wing tip and a blunt trailing edge. Calculations
with the same program on this mesh showed excellent
correlation with the experimental values (BBD grid
results, in Figure 5). The main difference between the
two grids was in the orthogonality near solid surfaces.
FANSC showed excellent convergence characteristics
(density and turbulent viscosity) on the Bombardier
generated grid and on the workshop supplied grid,
despite its excessive skewness. Integrated lift and
pitching moment predictions on the workshop grid
were accurate. Mesh skewness introduced
discretization errors on the skinfriction evaluations
whereas pressure drag was correctly predicted on both
grids. This illustrates the great care that must be
exercised if drag from NavierStokes computations is
used as the function to be minimized in a wing design
process. One must ensure that mesh modifications do
not introduce variations not due to the wing geometry
changes.
C
D
C
L
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ExpNLR
ExpONERA
ExpDRA
FANSCNSDPW grid
FANSCNSBBD grid
FANSCEuler/BLBBD grid
M=0.75, Re
c
=3*10
6
Figure 5: FANSC prediction of drag polar for the DLRF4
configuration, Mach 0.75, Reynolds number 3 Million.
Figure 5 shows that an excellent drag polar was also
obtained on the DLRF4 configuration with FANSC
running in Euler/boundarylayer mode with the post
processing drag formulas. The grid required for this
calculation was much simpler, with 1.3 million grid
points. A solution for one angle of incidence is
obtained in 0.45 hours on 8 CPUs of a Cray SV1
computer instead of the 6 hours required by the Navier
Stokes calculation. The Euler solution required 600
Mbytes memory instead of the 2.2 Gbytes required by
the NavierStokes analysis. There are therefore
advantages in using Euler/boundarylayer methods in
large parts of the wing design process.
In the second workshop, in 2003, results on the DLR
F6 wingbody and wingbody engine configurations
were obtained on several meshes. Comparison of the
flow solutions computed on all the above meshes
showed that good results were obtained with the
MBGRID mesh. The residuals for calculations on a
wingbody converge to machine accuracy with a linear
convergence rate (Figure 6). Accurate flow solutions
were obtained on the DLRF6 wingbodyunderwing
engine, as shown in Figure 7. However, the predicted
drag, shown in Figure 8 for various grids, still requires
improvements, particularly for the wing/body/engine
case.
Iterat ions
D
e
n
s
it
y
R
e
s
i
d
u
a
l
0 2500 5000 7500 10000 12500 15000
16
14
12
10
8
6
4
2
0
RMS
MAX
DLRF6 WingBody
Mach 0.75 =0 Re=2.4x10
6
Figure 6: Convergence of the residuals on DLRF6 wing
body configuration (Mach 0.75, Re=3.0 Million).
Figure 7: FANSC results on the DLRF6 wingbodyengine
configuration (Mach 0.75, Re=6.0 Million).
Figure 8: FANSC results on the DLRF6 wingbodyengine
configuration (Mach 0.75, Re=6.0 Million).
Additional efforts in drag prediction are focused on
deriving a rigorous mathematical model for computing
the induced, viscous and wave drag from Navier
Stokes solutions. The method uses cellbycell entropy
integration to compute the drag generated within each
computational cell. Using conservative laws, the
irreversible viscous and wave drag flow phenomena
can be computed separately. The reversible drag
phenomenon, induced drag, is computed using cross
flow analyses. One major difficulty is the
identification of the spurious drag produced by the
artificial dissipation present in the flow solver to
stabilize the iterative flow procedure.
Despite considerable progress made to date, the use of
NavierStokes methods in aircraft design integration is
still a challenge. The best approach seems to be the use
of a full suite of low and high fidelity codes, starting
for instance with low fidelity codes and finishing with
the more sophisticated methods. The implementation of
3D NavierStokes codes in design requires
computation of the sensitivities through the solution of
adjoint equations. This approach works very well for
the design of twodimensional single and multi
element airfoils and for simple wing/body
configurations. However, for complete aircraft
configurations, it still requires and a significant
upgrade of the available computing hardware to be
fully effective.
Aerodynamic analysis of transonic flexible wings
One important aspect of wing design is fluidstructure
interaction. The objectives are the prediction and
minimization of wing weight, the prediction of the
wing structural deformation under loads (Figure 9) and
the influence of this deformation on the aerodynamic
load distribution. The prediction of the bending and
twisting of wings was achieved by coupling the
transonic CFD code KTRAN [3] with a thinwalled
structural analysis program TWSAP [11]. The linear
structural capabilities of the NASTRAN structural
analysis software were utilized to predict the bending
and twisting of simplified finite element models (stick
models) of actual wings. Deformations predicted using
these stick models were in very good agreement with
the results of full Finite Element Models (FEM) [11].
Results obtained for the static equilibrium
(convergence) state of the Challenger and Global
Express wings in 1gflight were found to be in very
good agreement with experimental data. In a following
step, the methodology was extended to predict the
aeroelastic deformation of wings at the conceptual
design stage. The program generates conceptual
layouts of wing structural components and creates a
beam finite element model of the wing structure
(Figure 10). To establish the accuracy of the stick
CD
C
L
0.015 0.02 0.025 0.03 0.035 0.04 0.045
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
FANSCMBGRID 7.8 10
6
pts
FANSCMBGRID 3.9 10
6
pts
FANSCMBGRID 1.0 10
6
pts
Experiments
FANSCMBGRID 6.0 10
6
pts
WB
WBE
model designed by TWSAP, the predicted wing
bending and twisting were compared with results
obtained with the full finite element model of the real
wing structure, without winglets. It can be seen in
Figure 11 that the accuracy obtained is sufficient for
preliminary design purposes.
Figure 9: Static aeroelastic deformation of a transonic wing
computed by the KTRAN/TWSAP/NASTRAN package at
Mach 0.80 and CL=0.5
Figure 10: Challenger wing structure and examples of
conceptual structures generated by the TWSAP program.
Figure 11: Comparison of wing bending predicted by the
conceptual stick model and the full FEM of the Challenger
wing without winglets.
Powerplant Integration
The integration of aftfuselagemounted engine
nacelles requires a major effort to reshape the fuselage
in the nacelle/ pylon area and optimize the pylon
design. This process can be illustrated by the
development of the Global Express. The Bombardier
Global Express is a business jet specifically designed
to provide comfortable ultralong range at a higher
speed than any other aircraft in its class. With a
Maximum TakeOff Weight of 93,500 lbs, the Global
Express can fly eight passengers and four
crewmembers over a distance of six thousand nautical
miles, at Mach 0.85. The aircraft was also designed to
provide excellent runway performance. The main aero
dynamic features of the aircraft are: a high aspect ratio,
35degree swept wing with winglets, using third
generation supercritical airfoils, a Ttail configuration
and two fuel efficient turbofan engines mounted on the
aft fuselage to keep the wing free from adverse
nacelle/engine interference.
Figure 12: Cartesian and cylindrical grids used by
Bombardiers KTRAN transonic small disturbance code for
full aircraft configurations.
The objective in this case was the elimination of
undesirable shock waves that appeared on the lower
surface of the pylon and the nacelle at cruise conditions
above Mach 0.8. The shaping of the fuselage was first
carried out with the aid of the fast transonic small
disturbance code KTRAN [3]. This code uses an
internally generated Cartesian and cylindrical mesh
system that includes an overall coarse grid and fine
grids for the wing, the body/nacelle and for the
winglets, as shown in Figure 12. The pylons are not
included in the calculations. Figure 13 shows isobars
obtained with KTRAN on fuselage and nacelle at Mach
0.85 at three stages of the development of the Global
Express.
(a) Rigid wing geometry
(b) Wing geometry under loads
X
YZ
464
488
512
537
561
582
4047
4066
4085
4108
4131
4154
4178
4199
4221
4243
4265
4287
4390
4415
4439
26
X
Y Z
X
YZ
Challenger
Wing Structure
Stick
Model
Wing Structure Conceptual Models
0
2
4
6
8
10
12
0 50 100 150 200 250 300 350 400
Wing Station
D
e
f
l
e
c
t
i
o
n
(
I
n
.
)
Full FEM
TwsapDesign
Figure 13: Isobars obtained with KTRAN on fuselage and
nacelle at three stages of the development of the Global
Express configuration; Mach 0.85 cruise conditions.
The fuselage shape obtained from these calculations
was used as input to the FANSC Euler/NavierStokes
code to check the flow situation with the addition of
the pylon. The pylon and the nacelle position, in terms
of incidence and toeout angle were optimized with the
aid of FANSC. Figure 14 shows the Euler bodyfitted
structured mesh generated for the Global Express
complete configuration. Figure 15 shows the isobars
obtained on fuselage, nacelle and pylon with the
FANSC Euler/Boundary Layer code, at Mach 0.85
cruise conditions, at three stages of the aircraft design.
Figure 14: Multiblock structured grid generated for the
Global Express complete configuration using Bombardiers
MBGRID program.
Figure 16 shows comparisons of FANSC calculations
with wind tunnel test results obtained at the ARA
transonic wind tunnel (Bedford, U.K.).It can be seen
that FANSC predicts well the shock on the lower side
of the original pylon, and the shockless pressure
distribution of the optimized pylon at Mach 0.85 cruise
conditions.
Figure 15: FANSC study of Global Express aft fuselage
pressure distributions at three stages of the design. Mach 0.85
cruise conditions.
Figure 16: Channel pressure distributions at M=0.85;
Comparison of FANSC computations with test data.
HighLift Systems Design
The development process of an aircraft highlift
configuration includes CFD design and analysis, wind
tunnel tests and flight tests, as illustrated by the
development of the flaps and slats of Bombardiers
CRJ700 70passenger Regional Jet (Figure 17)[12].
Figure 17: Bombardier CRJ700 with flaps and slats
deployed in landing configuration.
Initial
Fuselage/Pylon
Intermediate
Fuselage/Pylon
Final
Fuselage/Pylon
Initial
Fuselage/Pylon
Intermediate
Fuselage/Pylon
Final
Fuselage/Pylon
Initial
Fuselage/Pylon
Intermediate
Fuselage/Pylon
Final
Fuselage/Pylon
(x/c)
N61 nacelle
(x/c)
N61
nacelle
C
p
Test Data Above Pylon
Test Data Below Pylon
MBTEC Above Pylon
MBTEC Below Pylon
0
(x/c)
N61 nacelle
(x/c)
N61
nacelle
C
p
Test Data Above Pylon
Test Data Below Pylon
MBTEC Above Pylon
MBTEC Below Pylon
0
Test Data Above Pylon
Test Data Below Pylon
MBTEC Above Pylon
MBTEC Below Pylon
0
The CRJ700 was designed to carry seventy passengers
and three crewmembers over a range of 1685 Nautical
Miles (1985 NM for the extended range version) at
Mach number 0.78. Maximum operating Mach number
is 0.83 and maximum cruising altitude 41,000 feet. The
aircraft, designed with a Maximum TakeOff Weight
(MTOW) of 73,000 lbs (75,250 lbs for the extended
range version), is powered by two fuselage mounted
General Electric CF348C1 turbofans, each
developing 12,670 lbs thrust. The takeoff distance at
MTOW is 5,130 ft and the landing distance, at the
Maximum Landing Weight of 67,000 lbs is 4850 ft.
Figure 18 shows the planform definition of the CRJ
700 highlift system. The trailing edge flaps are
doubleslotted hinged flaps. The flap deflection angles
are 10 and 20 degrees for takeoff, 30 degrees for
approach and 45 degrees for landing. The leading edge
slats cover the full span except for an inboard segment
of the wing, which was left unprotected to improve the
stall characteristics. The slats are tapered, covering
15% of the wing chord at the break and 17% chord at
the wing tip. The slat deflections are 20 degrees for 0
degrees and 10 degrees flap settings and 25 degrees for
all other flap settings.
Figure 18: CRJ700 flaps and slats planform definition
The development of the highlift systems was
conducted using Computational Fluid Dynamics (CFD)
methods. Initial work focused on the design of multi
element airfoils at various span wise stations on the
wing. Design variables included the shapes of the slat
and flap and the optimum gap, overlap and deflection
angle of each one. The CEBECI or CSU code, a
viscous panel method with strong boundary layer
coupling developed by T. Cebeci and his team at the
University of California in Long Beach [7] was used
for rapid evaluation of alternative designs. The MSES
viscous Euler code of Drela [13] and the NSU2D 2D
unstructured grid NavierStokes code developed by D.
Mavriplis [14] were used for the detailed design and
final verification of flaps and slats. The accuracy of
three codes was verified using wind tunnel test data
obtained on a Bombardier experimental highlift model
[15]. Figure 19 shows that all three codes predict fairly
well the characteristics of the singleelement airfoil.
The figure gives also results obtained with Tornado, a
multiblock structured grid NavierStokes code
developed by University of Toronto [16]
Figure 19: Lift and drag predictions of various CFD codes
on a clean airfoil configuration; comparison with wind tunnel
data on a Bombardier experimental airfoil.
The main aerodynamic issues of highlift system
development are: the prediction of the C
Lmax
and stall
characteristics of complete aircraft configurations with
thin swept wings; the prediction of roughness effect on
the C
Lmax
of these wings and the prediction of liftto
drag ratio of takeoff configurations.
One method frequently used to estimate the 3D
maximum lift of a full aircraft involves a combination
of a panel method coupled with 2D empirical
correlations, as reported in the work of Valarezo and
Chin [17]. The original work was used to predict the
maximum lift of arbitrary 3D configurations without
roughness, based on the Pressure Difference Rule
concept. This is an empirically derived correlation that
relates the maximum pressure differential as predicted
by simple linear panel methods to the experimentally
measured maximum lift condition. In other words, for
a given Reynolds and Mach number combination,
empirical curves have been determined which give the
maximum inviscid adverse pressure gradient a
corresponding 2D boundary layer can sustain before
separation occurs. This empirical limit is then used to
determine the maximum lift of a 3D configuration
based on the sectional pressure distributions.
Figure 20 shows the good correlation obtained between
predictions and wind tunnel test results for a CRJ200
Inboard Flap
Slat #3
Multi Function Spoilers
Outboard Flap
Aileron
Slat #2
Slat #1
Ground Spoilers
cruise configuration. In [18], a simple method is
introduced as an extension of the Pressure Difference
Rule that allows the estimation of the maximum lift of
an aircraft with wing leading edge contamination.
Aircraft certification regulations stipulate that an
aircraft handling characteristics and performance must
be determined for flight in icing conditions and in
roughness conditions. The roughness can be caused by
ice, frost, deicing and antiicing fluids used prior to
takeoff, insect contamination, paint and surface
irregularities and leading edge damage such as that
produced by a hailstorm. Roughness on the wing
leading edge affects the stall characteristics of an
aircraft and its performance.
The method is based on a combination of the Pressure
Difference Rule, using a threedimensional panel
method, with results of a twodimensional interactive
viscousinviscid CFD procedure developed by Cebeci
[7]. The code is able to predict aerodynamic
performance of single and multi element airfoils,
including stall, with and without surface roughness,
with sweep effects (Quasi 3D), for steady flows. The
code uses a Hess and Smith panel method to calculate
the inviscid flow field with a simple KarmanTsien
compressibility correction formula. A twodimensional
compressible boundary layer code operating in an
inverse mode, is coupled to the panel method. Michels
formula is used for transition prediction and an
improved Cebeci Smith model is used for turbulence
modeling with roughness effects. Roughness effects are
introduced using the CebeciChang model [19]. The
equivalent sand grain roughness is the input
characterization parameter for the code.
Figure 20: Predicted maximum lift with and without
roughness; comparison with experimental data
The method is illustrated in Figure 21, using a model of
the M100 ONERA3 wing/body test article. In this
application, the VSAERO panel method of Analytical
Methods Inc. is used. An initial VSAERO analysis is
first conducted to determine the critical spanwise
location where the maximum pressure difference
occurs. Based on the local chord Reynolds number at
that critical section, a 2D analysis is conducted to
determine the incremental effects due to roughness
with on maximum lift. The figure shows the 2D lift
curves calculated with and without contamination. This
increment is applied to the original limit ?Cp curve
and compared with the original spanwise distributions
of ?Cp to determine the new maximum lift point with
contamination. Figure 21 shows the limit ?Cp curves
with and without roughness as well as the spanwise
distributions of ?Cp as calculated using VSAERO for
several angles of attack. Finally, Figure 21 shows the
predicted maximum lift for the configuration with and
without roughness. The methodology was validated
using the results of the wind tunnel tests carried out on
a 1/3 scale model of a regional jet. Tests were
conducted at Mach 0.15 and mean chord Reynolds
number of 2.72 million, for various levels of wing
contamination. Figure 20 shows the comparison of
predicted and experimentally measured maximum lift
coefficients with and without contamination for the
cruise configuration. The relative loss in lift due to
contamination compares well with experiment,
although the absolute levels are slightly overpredicted
in this case.
Figure 21: Application of the extended pressure difference
rule to the prediction of maximum lift with roughness effects.
An application of a NavierStokes method to the
investigation of an aircraft maximum lift is reported in
[20]. The NSU3D unstructured NavierStokes solver is
used for the study [21]. It uses an edgebased, vertex
centred finitevolume scheme for space discretisation
C
L
Pr edicted C
LM AX
( No Roughness)
Predicted C
L MAX
(W ith Roughness)
Symbo ls  Exper iment
C
p
m
i
n

C
p
T
E
0.00 0.25 0.50 0.75 1.00
15
10
5
0
Limit C
p
Curve
1 No Roughness
2 With Roughness
=12 CL =1.19
=8 C
L
=0.89
=10 C
L
=1.04
CRITICAL
2
1
C
L
5 0 5 10 15 20
0.0
0.5
1.0
1.5
Predicted C
LMA X
(No Roughness)
Predicted C
LMA X
(with Roughness)
3
D
P
A
N
E
L
M
E
T
H
O
D
C
L
0 5 10 15 20
0.0
0.5
1.0
1.5
2.0
2.5
Effect
of
Roughness
2D
and a multistage RungeKutta technique for time
integration with point or line preconditioning. An
agglomeration multigrid algorithm is implemented for
convergence acceleration. Two turbulence models are
implemented: the SpalartAllmaras model and the
Wilcox k model. Both can be used with or without
wall functions.
The Challenger wing/body/nacelle configuration was
selected to investigate the ability of NSU3D to predict
flows at high angles of attack up to and beyond stall.
The geometry modelled represents the wind tunnel
model including flap fairings and flowthrough
nacelles. The flow conditions of the wind tunnel data
used for comparison are a Mach number of 0.25 and a
Reynolds number of 2.2106 , based on the wing mean
aerodynamic chord. The stall pattern on this model is
typical of transonic jets with no slats or leading edge
flaps. A leading edge flow separation, due to the
bursting of a laminar short bubble, causes a sudden loss
of lift at stall.
Figure 22: Poststall isobars and skinfriction lines on a
Business Jet cleanwing model at Mach 0.25 and a=14.21
o
;
NSU3D NavierStokes solution with a k? turbulence model.
The relative performance of the SpalartAllmaras and
k turbulence models in predicting the lift variation
with incidence was evaluated on this mesh. Poststall
isobars and skinfriction lines computed at a=14.21
o
using the k turbulence model are shown on Figure
22. The predicted lift variation with incidence for the
two turbulence models is compared with the
experimental data in Figure 23. These results were
obtained with the assumption of fully turbulent flow.
At incidences up to 10, both turbulence models predict
lift fairly well. At higher incidences, however, the
predicted lift is lower than the experimental data before
stall, with the oneequation SpalartAllmaras model
results being worse than those obtained with the two
equation k model. Both models underpredict the pre
stall lift coefficient, due to an excessive amount of
predicted separated flow on the outboard wing. None
of the numerical results predicts the sudden drop of lift
after stall, but the SpalartAllmaras predictions show a
kink in the lift variation shortly after the experimental
stall incidence.
Figure 23: Comparison of predicted lift curves with
experimental data
Modelling laminar flow at the leading edge of the wing
improves marginally the results but it is fair to
conclude that the present models need improvements
before they can predict correctly the maximum lift
behavior of threedimensional configurations.
Figure 24: VSAERO solution for a CRJ700 early landing
configuration, flap setting: 45 degrees, slat setting: 28
degrees Mach 0.2, Alpha=10 degrees
The analysis and design of the CRJ700 highlift
system was therefore carried out using the extended
()
C
L
0 5 10 15 20
Wind tunnel data
k, fullyturb.
SA, fullyturb.
CL601 WBN
B82 W50 F5 N51 P54
M = 0.25 Re
MAC
= 2.210
6
Mesh 2.5.2
pressure difference rule. Figure 24 shows a VSAERO
model of the landing configuration of the CRJ700.
The predicted C
Lmax
of a clean configuration compared
very well with high Reynolds number wind tunnel data,
as shown in Figure 25.
Figure 25: Comparison of wind tunnel data with theoretical
prediction of CLmax for the CRJ700 clean configuration at
wind tunnel conditions: Mach 0.20, chord Reynolds number
6.5 million.
Figure 26: Comparison of trimmed lift curves from the
IAR 5ft wind tunnel test with flight test data.
The highlift system design was ultimately validated
during an extensive flight test program. The aircraft is
fitted with a stall protection system including a stick
shaker for stall warning and a stick pusher for stall
recovery. The stick pusher activates ahead of natural
stall for the clean configuration and postnatural stall
for all configurations with the slats deployed. Figure 26
shows lift curves from a highReynolds number wind
tunnel test [12], trimmed for the most forward centre of
gravity, compared to equivalent data from performance
flight tests. This figure shows good correlation for all
flap angles.
Conclusions
The aerodynamicist designing and developing the
detailed geometry of a high performance aircraft is
faced with several issues, some of which are: how to
achieve the optimum cruise configuration for the
mission required, how to account for the flexibility of
the wing, how to install large turbofans with minimum
interference, how to define optimum flaps and slats for
field performance? Aerodynamic computations with
CFD methods provide answers to many of these
questions. The judicious use of CFD methods, from
panel methods to NavierStokes solver, allow the
aerodynamicist to arrive at feasible configurations with
minimum cost and risk.
The technology for simulating wing flows has
progressed to a point where routine NavierStokes
analyses of complete aircraft configurations are
possible. However, progress is still required to make
the prediction of drag from these methods more
reliable. The full advantage of NavierStokes methods
will be reached when they will be used routinely for
3D full aircraft design. This requires the calculation of
design sensitivities using the solution of adjoint
equations and even more powerful computers.
Acknowledgement
The author would like to acknowledge the contribution
of the staff of Bombardier Advanced Aerodynamics,
notably Pat Piperni, Eric Laurendeau, Marc Langlois,
Jose Boudreau, David Leblond and Cedric Kho.
(deg)
L
i
f
t
C
o
e
f
f
i
c
i
e
n
t
(
C
L
)
Predicted C
Lmax
(deg)
L
i
f
t
C
o
e
f
f
i
c
i
e
n
t
(
C
L
)
      5 x 5 Wind Tunnel
Flight Test
s
l
a
t
2
5
o
f
l
a
p
4
5
o
s
l
a
t
2
0
o
f
l
a
p
0
o
s
l
a
t
2
0
o
f
l
a
p
1
0
o
s
l
a
t
2
5
o
f
l
a
p
2
0
o
References
1. Obayashi, S., Takahashi, S. and Fejtek, I.,
Transonic Wing Design by Inverse Optimization
Using MOGA, CFD98, 6th Annual Conference of
the CFD Society of Canada, Quebec City, June
1998.
2. D. Jones, I. Fejtek, and D. Leblond, Coupling of a
Wing Inverse Design Code to a Transonic Small
Disturbance Flow Solver, CASI 48th Annual
Conference Proceedings, April 2001.
3. F. Kafyeke, P. Piperni and S. Robin, Applications
of KTRAN Transonic Small Disturbance code to
the Challenger Business Jet Configuration with
Winglets, SAE Paper 881483, October 1988.
4. Laurendeau, E., Zhu, Z. and Mokhtarian, F.,
Development of the FANSC Full Aircraft Navier
Stokes Code, Proceedings of the 46th Annual
Conference of the Canadian Aeronautics and
Space Institute, May 1999.
5. Van der Velden, "Aeropointer, a commercial CFD
shape optimization tool", CASI 48th Annual
Conference Proceedings, April 2001.
6. Laurendeau, E. and Boudreau, J., Drag
Prediction using the Euler/NavierStokes Code
FANSC, SAE Paper 2003013022, 2003.
7. Cebeci, T., Besnard, E. and Messie, S.
A Stability / Transition Interactive Boundary
Layer Approach to MultiElement Wings at High
Lift. AIAA Paper 940292 (1994).
8. AIAA CFD Drag Prediction Workshop Website.
http://www/aiaa.org/tc/apa/dragpredworkshop/dpw
.html
9. Laflin, K.R., et al, Summary of Data from the
Second AIAA CFD Drag Prediction Workshop,
AIAA Paper 20040555, 2004.
10. Piperni, P. and Boudreau, J., The Evolution of
Structured Grid Generation at Bombardier
Aerospace, CASI 50
th
Annual Conference
Proceedings, April 2003.
11. Abdo, M., Kafyeke, F., Ppin, F., Borowiec, Z.
and Marleau, A., "Transonic Aerodynamics of
Flexible Wings", CASI 48th Annual Conference
Proceedings, April 2001.
12. Kafyeke, F., Ppin, F. and Kho, C., Development
of HighLift Systems for the Bombardier CRJ
700, ICAS 20023.10.1, Toronto, Canada,
September 913, 2002.
13. Drela, M. Newton Solution of Coupled
Viscous/Inviscid MultiElement Airfoil Flows.
AIAA Paper 901470 (1970).
14. Mavriplis, D.J. and Jameson, A. Multigrid
Solution of the NavierStokes Equations on
Triangular Meshes. AIAA Journal, vol. 28, no. 8,
pp. 14151425 (1990).
15. Langlois, M., Kho, C., Kafyeke F., Mokhtarian,
F.and Jones, D. "Investigation of a MultiElement
Airfoil HighLift Characteristics at High Reynolds
Numbers", Proc. 8th CASI Aerodynamics
Symposium, Toronto, April 2001.
16. Nelson, T. E., Zingg, D. W. and Johnston, G. W.
Compressible NavierStokes Computations of
Multielement Airfoil Flows Using Multiblock
Grids. AIAA Journal, vol. 32, no.3 (1994).
17. Valarezo, W.O., and Chin V.D., Maximum Lift
Prediction for MultiElement Wings, AIAA92
0401, January 1992.
18. Kafyeke, F. Boyce, F., Kho, C.,Investigation of
Airfoil and Wing Performance with Leading Edge
Contamination, proceedings of the CASI 8th
Aerodynamics Symposium, Toronto, Canada,
April 30th to May 2nd, 2001.
19. Cebeci, T., and Chang, K. C., Calculation of
Incompressible RoughWall Boundary Layer
Flows, AIAA Journal, Vol. 16, No. 7, 1978, pp.
730735.
20. Langlois, M., Mokhtarian, F., Kafyeke, F.,
NavierStokes Prediction of Aircraft HighLift
Characteristics, Proceedings of the CASI 9th
Aerodynamics Symposium, Montral, Canada,
April 2003
21. Mavriplis, D. J., Turbulent Flow Calculations us
ing Unstructured and Adaptive Meshes, Interna
tional Journal for Numerical Methods in Fluids,
Vol. 13, No. 9, pp. 11311152, Nov. 1991