Operations Based Optimisation using Simulation and CFD

John J Doherty * and Daniel P. Clifton†

QinetiQ Ltd, Farnborough, Hampshire, GU14 0LX, England

and

Mark A. Gillan ‡ and Fabio Ciampoli §

University of Surrey, Guildford, Surrey, GU2 7XH, England

An initial investigation of an optimisation based approach for design across a continuous

range of operating conditions is presented. The objective for this ‘operations based
optimisation’ approach is to avoid the need to choose critical design point conditions and
associated weighting factors by tackling the overall operational performance instead. The
approach integrates numerical optimisation, response surface modelling, CFD and
operational simulation. An optimisation test bed involving the aerodynamic optimisation of a
Champ Car rear wing assembly for reduced lap time using track simulation has been
developed to assess the new optimisation approach. Details of the operations based
optimisation approach and the Champ Car test bed are reported. Results generated using
the new approach are presented and the wider potential of the approach for aerospace
applications is discussed.

Nomenclature
CD = drag coefficient for full car
CL = downforce coefficient for full car
CDc = drag coefficient for main car chassis with contribution of rear wing assembly removed
CLc = downforce coefficient for main car chassis with contribution of rear wing assembly removed
CDrw = drag coefficient for rear wing assembly calculated in isolation with onset conditions
CLrw = downforce coefficient for rear wing assembly calculated in isolation with onset conditions
K = suspension force coefficient
LT = lap time
m = total mass of car
µ = tyre coefficient of friction
P = engine power
r = yaw rate
rh = car ride height
R = local radius of curvature along racing line
V = speed of car

I. Introduction

O ptimisation based design using computational fluid dynamics (CFD) is now widely applied within many fields
of engineering design including aerospace, maritime and motorsport. Within each of these application fields
significant progress has been made in terms of overcoming the many implementation challenges, including the
development of parameterised geometry, automatic grid generation and the ability to seamlessly exchange

*
QinetiQ Fellow, Aerodynamics & Aeromechanical Systems Group and Visiting Professor University of Surrey,
Senior Member AIAA.
†
Undergraduate industrial placement, Aerodynamics & Aeromechanical Systems Group.
‡
Sir George Edwards Chair in Aerospace Engineering, Director of University of Surrey's Advanced Vehicle
Analysis Group (SAVAG), Member AIAA.
§
Research Fellow, SAVAG.

1
information within an overall design optimisation process. However in most real world design problems issues
associated with modelling complexity and the availability of sufficient computational resources lead to some level of
simplification of the design problem. This simplification can take a variety of forms including removal of at least
small scale geometry features, simplified aerodynamic flow modelling and neglecting multi-disciplinary interactions
such as aeroelastic deformation.
The objective for the investigation presented in this paper focuses on a further category of simplification which
is usually employed within design optimisation applications. In particular a common feature across aerospace,
maritime and motorsport applications is the fact that optimisation problems are typically formulated to focus on a
small number of operating conditions, with all other ‘off-design’ operating conditions being neglected. In reality the
actual vehicle will need to operate over a continuous, possibly wide range of operating conditions, all of which
represent design conditions. Indeed the frequently used phrase ‘off-design conditions’ might be better understood as
meaning ‘design conditions that were not considered’.
Typically in optimisation applications the full operational space over which the vehicle must operate is usually
simplified by an a priori selection of one or more key design point conditions, which are then included within a
single or multipoint optimisation problem formulation. During the resulting optimisation the design performance is
analysed for these key design point conditions, with the remainder of the operational space being treated as ‘off-
design conditions’ which are not directly included in the optimisation problem formulation. For example a civil
transport aircraft optimisation may focus on drag reduction for 1 or 2 representative cruise conditions. Similarly a
racecar optimisation may focus on downforce maximisation for a single cornering condition.
Unfortunately the a priori choice of these key design conditions and their importance relative to one another is a
complex problem in itself, particularly because the vehicle may be required to be operated in a variety of different
ways. For example an aircraft will be required to perform for a number of different operational missions,
corresponding to different payloads, ranges and altitudes such that the design balance between cruise efficiency
versus low speed performance will vary. Likewise a racecar will compete for many different tracks changing the
design balance between downforce for cornering performance versus straight line speed. Indeed the key design point
conditions may differ for each of these alternative operational missions and the relative importance of each will
almost certainly differ.
Ideally for an optimisation process the objective function would consider the performance across a full
operational mission rather than purely the performance at a number of key design point conditions. In the extreme
the optimisation objective would take into account the performance for many or all possible operational missions.
To assess the feasibility of improving the optimisation problem formulation towards this idealised goal of
‘operations based optimisation’ an existing aerodynamic optimisation tool has been extended and then applied to the
design of the aerodynamics of an open wheeled racecar for full track optimisation. This paper describes how this
new optimisation formulation has been implemented. Initial results for the racecar full track optimisation are then
described. Finally the applicability of the approach to other engineering fields is discussed.

II. Racecar Application

In order to assess how an operations based optimisation problem formulation could be achieved, it was decided
to trial a new approach through application to a racecar rather than an aerospace application. It was considered that a
racecar application would provide a good test bed for trialing this new approach because racecar operating
conditions are continuously varying during a race. Although the aerodynamics of a racecar are highly complex
compared to aircraft aerodynamics, it is still possible to select a relatively small scale application to use as the basis
for the current test bed. In particular it was decided to focus on the aerodynamic design of the rear wing assembly
for a Champ Car and to use lap time as a measure of overall operational performance. Champ Car teams make use of
the same engine, tyres and chassis, but are allowed to fine tune the front and rear wing aerodynamics. The Champ
Car World Series1 has traditionally incorporated both oval and street circuits. For this initial investigation it was
decided that a complex street circuit would provide a challenging test case.

A. Champ Car geometry

At the start of the current work it was first necessary to obtain a representative Champ Car 3D CAD geometry.
The initial geometry was based upon the Champ Car shape used up until the 2006 season. It was created using a
reverse engineering approach based upon information and photographs available in the public domain, geometry
details described within the Champ Car World Series rule book available from reference 1 and existing design
experience. Details such as struts, suspension wishbones, brake intakes, push rods and driveshaft geometry have
been neglected for simplicity. The Be153-055 racecar aerofoil section was used to define the baseline rear wing

2
main element and flap, with initial positioning and separation estimated from photos of Champ Cars. The resulting
Champ Car CAD is shown in figure 1.
Only limited information is available in the public domain so this reverse engineering approach will not provide
a fully accurate representation of the real Champ Car. In particular it is very difficult to obtain detailed information
regarding the exact geometry of the underside of the car. Hence it was necessary to use a reasonable amount of
engineering judgement to define the tunnels and diffusers which are present on the underside of a Champ Car.
However since the Champ Car is simply a test bed for evaluation of the operations based optimisation approach the
authors were happy that a reverse engineered Champ Car geometry would provided a satisfactory representation.

Figure 1. Initial Champ Car CAD geometry.

Shortly after the creation of this initial Champ Car representation it was announced that from 2007 onwards the
Champ Car World Series would be switching to a new Panoz chassis concept which would be used by all the
competing teams. It was therefore decided to adopt this new geometry within the current study. Hence a new Panoz
concept geometry was constructed and is shown in figure 2. This new geometry was based upon the initial
conceptual drawings published at the time of the announcement. It is this second Champ Car geometry that has been
as the test bed for operations based optimisation.

Figure 2. Final Champ Car CAD geometry representing proposed Panoz 2007 concept.

For completeness it is worth remarking that the actual Panoz DP01 chassis has been subsequently unveiled and
can be viewed at reference 1. Notable differences between the initial Panoz concept adopted in this study, as shown
in figure 2, and the actual chassis shown at reference 1, are that the rear wing is no longer supported by two upright
struts but is instead supported by a new horizontal wing section and larger endplates. Also the front wing is now

3
constructed with a single flap which runs from one endplate, under the nose, to the opposite side of the car. These
most recent aerodynamic changes have not been implemented into the current study.

B. Full Car and Rear Wing Assembly CFD Analysis

During the optimisation application two levels of CFD modelling have been used. Full car computations have
been used to establish aerodynamic forces for the baseline car. These full car computations have also been used to
provide onset flowfield boundary conditions which are used for the second level of CFD modelling for the rear wing
assembly in isolation. This two level CFD modelling approach has been used successfully in a past application
completed for the optimisation of a racecar rear wing2. The use of the isolated rear wing assembly, with onset
flowfield boundary conditions, provides significant computational savings, for the purposes of the required
optimisation test bed, whilst adequately modelling key performance characteristics as described later.

Figure 3. NEWPAN hybrid panelling used for Champ Car calculations.

Analysis of both the baseline Champ Car and all subsequent rear wing assemblies is achieved using the
NEWPAN3 panel method CFD software. NEWPAN is a fast and flexible low order CFD tool used widely within the
motorsport community. NEWPAN can use structured, unstructured or mixed surface panelling elements, which
allows very complex vehicle shapes to be modelled. As part of the set-up of a NEWPAN model initial wake surfaces
must be defined for lifting/downforce components. These wake surfaces are then iteratively ‘relaxed’ during the
NEWPAN computation to give a converged aerodynamic simulation. The basic NEWPAN panel calculation
provides a prediction of the inviscid aerodynamic flow, including pseudo compressibility effects.
Figure 3 shows the NEWPAN surface
panelling for the full Champ Car chassis,
using a hybrid mix of structured and
unstructured elements generated by a semi-
automatic process. Since the objective for the
optimisation study is based on design for a Max ride height
wide range of operating conditions, including
cornering, a full car panel model is used. The
requirement for modelling cornering
conditions also meant that the steering angle
for each of the front wheels must be altered
for different cornering conditions. In practice
this is achieved easily by transformation of
the baseline panel model. This full car
panelling contains just over 33000 surface
panels, including underbody panelling. Figure 4. Modeling of ride height in NEWPAN.
A ground plane is included in the
NEWPAN calculations to provide representative moving track boundary conditions. The separation between the car
geometry and the ground plane can be altered via a NEWPAN input variable in order to model the effect of varying

4
ride height rh. This is achieved in NEWPAN by raising both the height of the ground plane and the height of the
wheels whilst keeping the chassis fixed as represented in figure 4.
In previous applications of NEWPAN within the motorsport industry it has typically been the case that cornering
conditions are modelled as a sideslip condition. This is not strictly correct and in the current study it was decided to
improve the definition of the flowfield boundary conditions around the car for a cornering condition by using an
appropriate yaw rate boundary condition. The yaw rate r is derived from both the speed of the car V, which is also
input to NEWPAN, and the local radius of curvature of the racing line being followed by the car R.
Relaxed wake surfaces were included for the front and rear wing elements in order to estimate their contribution
to the overall generation of downforce. Typically 5-10 wake relaxations were found to give sufficiently converged
calculations for the full car computations, requiring approximately 200 minutes of computation on a single Pentium
IV processor.
Figure 5 shows the result of a NEWPAN analysis for the baseline Champ Car corresponding to a typical
cornering condition at a speed V=30m/s. Figure 5 shows pressure variations over the full car together with the shape
of the relaxed wakes (shown as wake filaments) for the front and rear wing assemblies. As expected for the
cornering condition the front wheels are deflected for steering and asymmetry of the rear wing wake filaments can
be clearly seen.

Figure 5. Pressure distributions and relaxed wake filaments for baseline Champ Car computed by
NEWPAN for low speed cornering condition.
The PANBL3 boundary layer tool, supplied with NEWPAN, can be used to post-process the inviscid panel
method results to provide an estimate of viscous effects for wing-type components. For the Champ Car application
estimates of total car downforce coefficient CL used NEWPAN predictions directly, while estimates of total car drag
coefficient CD were built up using a combination of NEWPAN inviscid drag predictions, PANBL viscous drag
predictions for the front and rear wing elements, empirical estimates of skin friction drag for all other wetted
surfaces and additional empirical estimates of bluff body drag, particularly for the wheels. This budgeting approach
for total drag is a gross simplification of the real complexity involved in trying to predict total aerodynamic drag for
an open wheeled race car. Indeed achieving accurate estimates of total drag is likely to remain as a significant
challenge within state of the art CFD applications for some time to come. For the purposes of the current application
this budgeting approach provides representative overall levels of downforce and drag.
In addition to obtaining full car estimates for downforce and drag, estimates of the downforce coefficient CLc and
drag coefficient CDc corresponding to only the main car chassis, without the rear wing assembly, were also derived
by simply removing the contribution due to the rear wing assembly during post-processing of the full car
calculations.
During the optimisation application the shape and relative position of the two rear wing elements are to be
controlled by the optimisation variables. Hence it was necessary to establish a fully automatic process for
regenerating the surface panelling for the rear wing assembly, including the end plates, suitable for input to
NEWPAN. In addition the initial wake surfaces originating from the trailing edge of each rear wing element and the
endplates must also be regenerated for input to NEWPAN. This automated surface panelling and wake generation
process was implemented within the NEWPAN environment using Python scripting and allows a very wide range of
rear wing shapes and relative positions to be automatically panelled. In particular the initial wake geometries
generated provide representative wake shapes, meaning that subsequent NEWPAN wake relaxation computations

5
converge in a very small number of iterations. Figure 6 shows a typical rear wing assembly for which surface
panelling and initial wake shapes have been generated automatically for subsequent input to NEWPAN.
The NEWPAN software provides a mechanism for imposing near flowfield boundary conditions to model the
effect of non-uniform onset conditions. In particular the flowfield around the rear wing assembly can be extracted
from an appropriate full car NEWPAN calculation and subsequently used to define the corresponding onset
conditions for the rear wing assembly in isolation. Further details of this approach are described in the NEWPAN
documentation3 and the authors have successfully used this approach in a previous motorsport optimisation
application2. By using this approach during an optimisation application it is only necessary to recompute the flow
for the rear wing assembly in isolation which typically requires only 1-2 minutes of compute time. However a
different full car calculation is required for each operating condition in order to generate the correct onset conditions
for the rear wing assembly.

Figure 6. Automatically generated surface panelling and initial wake geometries for rear wing
assembly.
For each operating condition and for each new rear wing geometry, estimates of the rear wing assembly
downforce coefficient CLrw and drag coefficient CDrw computed for the new rear wing assembly in isolation with
appropriate onset conditions, can be added to the main car chassis CLc and CDc results (for which baseline rear wing
assembly contribution is removed) for the same operating condition, to give updated values for the full car CL and
CD estimates. These new estimates for the full car CL and CD will hence vary as the rear wing geometry is changed.
Again this approach is a simplification of the real aerodynamics but has been shown to give useful trends in previous
work2 and as such is suitable for the purposes of the current demonstration.

C. Track Geometry
For the current study the Road America race course in Elkhart Lake, Wisconsin was chosen. This is one of the
official Champ Car World Series race tracks and was chosen because it includes a wide mix of straights, together
with both tight and shallow corners. This leads to a wide range of operating conditions for the design of the Champ
Car and hence represents a demanding test case for assessing the operations based optimisation approach. The
telemetric map of the Road America track, as published at reference 1, is shown in figure 7.
Using the layout of the Road America track shown in figure 7 a racing line was defined with the help of an
experienced motorsport racing driver. This racing line represents the mean path that the Champ Car is likely to
follow during a race. In particular this racing line took the form of a continuous 2D curve overlaid on the Road
America track layout. Using this 2D curve a value for the local radius of curvature of the racing line R is extracted
for 1m intervals along the racing line (about 6500 points since the track is about 6.5km long). This distance-
curvature distribution is required for input to the track simulation method described in the next section. For the
current study this racing line is fixed, although in practice the racing line itself could also be designed as part of a
more comprehensive optimisation application.

6
D. FROST Track Simulation Method
The Fast Race Optimisation Simulation
Tool (FROST) from the University of Surrey
has been used within the current study. This
is a parameterised, quasi-static simulation
method which takes as input the definition of
the track geometry and critical factors
determining the potential performance of the
race car. FROST effectively assumes that the
race car is operating at the limits of its
envelope, such as at the point at which the
grip from the tyres is only just sufficient to
prevent the car from spinning off the track
during a corner. FROST predicts the speed,
acceleration and braking performance of the
car over a complete lap within a few seconds
of compute time and also gives a prediction
of the lap time.
As described earlier the track geometry is
input as a distance-curvature distribution. A Figure 7. World Champ Car Series telemetric map for Road
number of characteristics of the car must be America race track.
defined, including the tyre coefficient of
friction µ, the total mass of the car m and engine power P. The aerodynamic downforce coefficient CL and the
aerodynamic efficiency ratio CL/CD for the car, which in the original version are assumed to be fixed, are also input
to FROST.
Prior to the current optimisation study FROST has been validated using information released by Jaguar F1 for a
past lap of the Silverstone track in England by Eddie Irvine. Figure 8 shows a diagram representing the track layout,
accompanied by vectors of longitudinal acceleration at 10m intervals along the track. The blue vectors are from
onboard measurements during Irvine’s qualifying lap and the red vectors show the corresponding predictions from
FROST. Vectors pointing inside the track depict braking and outside the track depict acceleration. Figure 8 clearly
shows close correlation between simulation and the actual measured track data.

Figure 8. Longitudinal acceleration and braking forces predicted by FROST simulation (Sim)
compared to measured data (Track) around Silverstone race track.

7
 Similarly figure 9, which shows the speed of the car against the distance along the track for both the FROST
simulation and the actual measured track data, further demonstrates the validity of the FROST track simulation.
Finally the predicted lap time from the simulation is within 0.2s of Irvine’s actual qualifying lap time.

Figure 9. Car speed against distance along race line predicted by FROST simulation (Sim)
compared to measured data (Track) for Silverstone race track.
A sensitivity study was also completed using the Silverstone validation case to assess the typical impact of
changes to the critical performance characteristics for the car. In particular the incremental effect on the lap time
∆LT due to a 1% improvement in the tyre coefficient of friction µ, the total mass of the car m, the engine power P,
the aerodynamic downforce coefficient CL and the aerodynamic efficiency ratio CL/CD was investigated. The results
are shown in figure 10, which clearly shows the importance of the grip of the tyres for lap time. In order to fully
understand the implications for the Champ Car study of the sensitivity results shown in figure 10 it is worth noting
that the World Champ Car Series rules dictate that all teams make use of the same chassis, tyres and engines and a
minimum total mass constraint is also imposed. Hence the teams must search for lap time reductions by other means
so focus switches to improving aerodynamic performance by means of front and rear wing fine tuning.
0.6

0.5
∆LT with 1% change in parameter (s)

0.4

0.3

0.2

0.1

0
µ m P CL CL/CD
Parameter

Figure 10. Sensitivity of lap time to various racecar performance parameters.

8
E. Use of Aerodynamic Maps within FROST
As discussed earlier the original version of FROST assumed fixed values for the full car CL and CL/CD for all
operating conditions around the track. This is generally a reasonable assumption for cases where the front and rear
wing aerodynamic flow is well behaved and primarily attached. However at the extremes of operation, such as at
high speed, it is known that rear wing flow separation ensues which leads to large increases in drag and
corresponding reductions in the associated rear wing contribution to overall downforce. Hence it was required to
improve the fidelity of the aerodynamic modelling within FROST.
To address this requirement FROST has been extended as part of this study to use representations for full car CL
and CL/CD which vary with operating condition. These parametric representations of CL and CL/CD are referred to as
‘aerodynamic maps’. The parameters which dictate the value of CL and CL/CD which FROST should use at any point
during the lap correspond to the instantaneous operating conditions experienced by the car.
For the current Champ Car application it was decided to focus upon three key operating parameters which vary
during a lap. These are the speed of the car V, the local radius of curvature of the racing line R and the ride height of
the car rh. As the car races around the track each of these operating parameters is continuously changing and hence
the aerodynamic characteristics are likewise changing. An aerodynamic map is required for input to FROST which
relates the aerodynamic characteristics of the Champ Car to changes in these operating parameters. Effectively these
three operating parameters define a 3-dimensional (V, R, rh) space of operating conditions i.e. the operating
envelope of the Champ Car. There are other operating parameters which could also be used in subsequent studies
e.g. pitch angle and pitch rate, roll angle and roll rate but for the purposes of this initial demonstration the selected
operating parameters should suffice.
In order to generate the required aerodynamic maps a combined Design of Experiments (DoE) and subsequent
Response Surface Modelling (RSM) approach is used. The DoE approach used to efficiently select appropriate
sample points across the 3-dimensional operating envelope is based upon a methodology developed within QinetiQ
which is not described here. A total of 27 (V, R, rh) sample points, corresponding to 27 different non-uniformly
spaced operating conditions, were chosen based upon appropriate coverage of the Champ Car operating envelope.
Having selected the DoE sample set of 27 operating conditions, a set of 27 corresponding full car NEWPAN
calculations have been computed for the baseline Champ Car. As described earlier appropriate rear wing assembly
onset conditions are then extracted from each full car NEWPAN calculation giving 27 different onset conditions that
can be used within isolated rear wing assembly NEWPAN calculations. These onset conditions are fixed during the
optimisation process meaning that no further full car NEWPAN calculations are calculated.
In addition to the generation of the onset conditions, the full car NEWPAN calculations are post-processed to
generate values for CLc and CDc for the main car chassis without the contribution of the baseline rear wing assembly.
As discussed earlier the rear wing assembly contributions will be added from the isolated rear wing assembly
calculations. Using these 27 values for CLc and CDc RSM equations for CLc and CDc in terms of (V, R, rh) are
generated for subsequent use in the optimisation application.

F. Suspension and Ride Height Modelling in FROST

A further extension to FROST for basic modelling of the car suspension was also completed as part of the
development of the optimisation test bed. In reality the ride height of the Champ Car will continuously vary along
the race line as a consequence of a changing balance between the aerodynamic downforce and the opposing
suspension force. The downforce calculated by NEWPAN is already dependent upon the ride height. Hence it was
necessary to introduce an opposing suspension force in FROST which is also dependent upon ride height.
A suspension force coefficient K was defined to relate the suspension force linearly to the ride height. This linear
relationship is such that when the downforce is zero (car is stationary) the ride height takes it maximum value and
the suspension force is equal and opposite to the weight of the car. Likewise when the downforce reaches a large
limiting value the ride height drops to its minimum value and the suspension force is equal and opposite to the sum
of the weight of the car and this limiting value of downforce. For values of ride height between its minimum and
maximum the suspension force varies linearly with ride height. Further increases in downforce beyond this limiting
value would have no further impact on reducing ride height.
At each point along the race line within FROST the downforce and suspension force are required to be in
equilibrium. Again this is a simplification compared to the real dynamic behaviour of the car but it was considered
that this equilibrium state would be closer to reality than simply having a fixed ride height. This equilibrium
requirement means that because both downforce and suspension force are dependent upon ride it is necessary to
iteratively solve for the correct value of ride height at each point along the race line in FROST. This is however

9
straight forward within FROST because of the availability of the aerodynamic maps which define equations relating
the downforce to any combination of speed, corner radius and ride height. In particular within FROST for any point
on the racing line the speed of the car and the radius of the racing line is known so the downforce can be predicted
for any value of ride height.

III. Optimisation Problem Formulation

A. Existing Optimisation Method

The QinetiQ Constrained Optimisation Design for Aerodynamic Shapes (CODAS) tool was initially developed
for air-vehicle design in the early 1990s and has subsequently been extensively used by QinetiQ and industry for a
wide variety of conventional single-point and multi-point aerodynamic shape optimisation applications, many of
which have been validated via wind-tunnel testing4-7. CODAS combines numerical constrained optimisation and
generalised parametric surface design, with a number of computational aerodynamic prediction methods including
NEWPAN. The existing CODAS method uses the QinetiQ RQPMIN quadratic gradient-search optimisation tool. In
addition the existing CODAS design method provides a means for modifying a rear wing assembly according to the
design variables supplied by RQPMIN. Following changes to the rear wing assembly the process described earlier
for the automatic regeneration of the NEWPAN surface panelling and the initial wake surfaces then enables the
effect of each geometry change to be predicted with NEWPAN for one or more operating conditions.
Reference 3 describes a past application of CODAS for the aerodynamic optimisation of a rear wing assembly
for a Formula 1 car. This previous application employed a conventional single point optimisation approach by which
results from NEWPAN predictions are fed back directly to the RQPMIN optimisation tool in the form of an
aerodynamic objective function (e.g. downforce maximisation), together with other aerodynamic and geometric
constraint functions. During this past application it became apparent that the choice of a suitable single operating
condition is very difficult, considering the wide range of conditions for which the resulting design is required to
operate. In addition whilst the past application was successful in improving the downforce for the chosen single
operating condition, no allowance was made for off-design performance. In particular it is likely that a rear wing
assembly optimised purely for downforce could suffer from significant flow separation during high speed
conditions, leading to drag penalties which could erode the overall improvement in lap time that might be expected
from consideration of the improvements in downforce alone.

B. Operations Based Optimisation Approach

In order to establish the optimisation test bed to assess the operations based optimisation approach it was
necessary to extend the CODAS method. Much of the basic structure of the existing CODAS method could be
reused, particularly the RQPMIN optimisation tool, the rear wing assembly geometry generation, but the interfaces
to the NEWPAN CFD method had to be completely replaced and the incorporation of the FROST track simulation
method together with an automated approach for the generation of RSM equations represented new functionality.
Figure 11 provides a flow chart representation of the new CODAS method that has been set up to provide the test
bed to assess the new operations based optimisation approach for the Champ Car application.
As described previously the full car NEWPAN calculations (red box in figure 11) are run once only for the 27
different operating conditions defined by DoE sample set. Since each is an independent calculation all 27
calculations are run simultaneously in parallel across 27 processors. These calculations provide the 27 onset flow
conditions for use within the rear wing optimisation process and also provide the RSM equations describing how the
aerodynamic forces for main car chassis varies with operating conditions. The core optimisation process, which is
fully automated, is shown by the green boxes in figure 11 and is driven by the RQPMIN optimisation method. For
each optimisation cycle CODAS generates a new rear wing assembly geometry, dictated by the current values of the
optimisation variables, and subsequently runs 27 rear wing assembly NEWPAN calculations in parallel. The values
of CLrw and CDrw for each of the 27 operating conditions are collected and RSM equations describing how CLrw and
CDrw varies with operating conditions are automatically generated. Both the main car chassis and rear wing RSM
equations for downforce and drag components are input to FROST which subsequently generates full car CL and CD
at each point on the track, including balancing the downforce and suspension forces by solving for ride height. As a
result the overall lap time can be derived for the current rear wing geometry. The value of this lap time is returned to
RQPMIN for use as the overall optimisation objective.

10
 Baseline
CODAS
RQPMIN Rear Wing
Framework
Geometry

Updated Full Car

Race NEWPAN
LAPTIME Rear Wing
Circuit Model
Geometry

FROST Rear Wing Full Car

Track Assembly NEWPAN CFD
Simulation Paneling & (x 27)
Wakes Varying Speed,
Ride Height &
Rear Wing Rear Wing Yaw rate
CDrw RSM CLrw RSM
Aero Map Aero Map Rear Wing Onset Flow
Assembly To Rear Wing
NEWPAN Assembly
Rear Wing Rear Wing CFD (x 27) (x 27) Chassis Chassis
CDrw Look-up CLrw Look-up CLc RSM CDc RSM
Table Table Aero Map Aero Map

Figure 11. CODAS method structure for operations based optimisation application. Green boxes represent
core optimisation method. Red box is calculated once. Blue boxes are inputs. Purple boundary indicates
calculation for multiple operating conditions computed in parallel.

IV. Results

A. Baseline Champ Car Analysis

For the current Champ Car application appropriate values for the total weight of the car and the engine power
have been taken from the rules for the World Champ Car Series1. Suitable values for other parameters defining the
performance of the car, including tyre coefficients etc. have been suggested by one of the authors of this paper,
Mark Gillan, based upon his previous experience of Formula 1 car design.
To assess the validity of the overall modelling the optimisation process was run for a single cycle for the baseline
Champ Car rear wing assembly. This gave an initial lap time for the Road America circuit of LT=105.03s which
compared surprisingly well with typical best lap times achieved during the 2006 Champ Car qualification rounds at
the Road America track. This single cycle case involves the generation of 27 separate rear wing assembly NEWPAN
calculations which are all run simultaneously (across 27 processors) followed by a single track simulation. The
overall elapse time of the analysis is about 220s. The extended version of FROST employing aerodynamic maps and
ride height iteration requires approximately 40 seconds of compute time for a single track simulation, whilst the
longest rear wing NEWPAN analysis requires approximately 120s of compute time. The remaining compute time is
required for generation of geometry, panelling, wakes and RSMs etc.
To provide an indication of the typical shape of the Champ Car operating envelope for the Road America track
figure 12 presents the variation in the speed of the car V and the log of the local radius of curvature of the racing line
R output from the FROST simulation for this initial analysis result. It should be noted that figure 12 represents a 2-
dimensional projection of the full 3-dimensional space since the 3rd dimension, corresponding ride height which
varies continuously around the track, is not shown. Within figure 12 low values of log (|R|) correspond to tight
corners on the racing line track whilst high values correspond to a near straight racing line. The peaky behaviour of
the operating envelope that can be seen in figure 12 demonstrates that the car is able to sustain its speed in regions of

11
the track where the race line radius of curvature is rapidly changing in magnitude. There are approximately 8
separate peaks which in practice will likely correspond to the shallower corners of the Road America track shown in
figure 7.
8

5
log |R|

0
0 10 20 30 40 50 60 70 80 90
V (m/s)

Figure 12. Variation of log of radius of curvature of the racing line and the speed of the car for a
full lap of the Road America race track, as predicted by FROST simulation based on CFD
aerodynamic maps generated for baseline Champ Car geometry.

B. Champ Car Optimisation Case

The baseline rear wing assembly geometry is shown in figure 13. The initial shape and locations of the main
wing section and the flap section are shown together with the endplate geometry. The wing sectional shapes are
constant across the full span of the rear wing assembly for the baseline case. Figure 13 also shows the box constraint
imposed by the World Champ Car Series rules1 which dictates that all portions of all rear wing elements must
remain within a defined 3-dimensional box. In order to ensure that this requirement was satisfied during the
optimisation process a geometric constraint was added into the optimisation problem formulation for each wing
element. The geometric constraints calculate the minimum distance between the each 3-dimensional wing element
and the constraining box and provided the distance remains non-zero for each wing element then the constraints are
satisfied.
A number of optimisation cases have been completed using the Champ Car optimisation test bed for varying
numbers of optimisation variables controlling the geometry of the rear wing. As expected the overall design
improvements are best for the case with most design freedom so only this case will be reported here. For this design
case 8 design variables are used to change the geometry of the rear wing elements. 2 camber variables are used to
change the aerofoil section shape across the full span of the main wing element. Similarly 2 camber variables are
used to change the aerofoil section shape across the full span of the flap. 1 variable is used to vary the main wing
twist between the centre line and the end plates. Similarly 1 twist variable is applied to the flap. Finally the flap is
allowed to translate horizontally and vertically controlled by 2 further variables.
As for the single cycle analysis case reported earlier this optimisation design case has 27 processors available to
allow all the rear wing assembly NEWPAN calculations to be run simultaneously. Figure 14 presents the history of
the optimisation of the lap time against the number of complete full track simulations (1 full track simulation
includes 27 simultaneous NEWPAN calculations). It can be seen that the optimisation process quickly identifies a
means to improve the design, resulting in a reduction in the lap time from 105.03s to 103.73s, or about 1.3s, after 46
cycles. The total elapse time for the optimisation was 3 hours based on use of up to 27 Pentium IV processors.

12
 Figure 15 shows a comparison of the baseline rear wing sectional shapes and the optimised sectional shapes after
46 full track simulations. The sections on the left hand side of figure 15 are located at the endplates, whilst the
sections on the right hand side correspond to the rear wing centre line. In each case the sections are scaled to more
clearly highlight the camber changes. Comparison of the sections at the endplates and the centerline indicate the
presence of twist for both elements, particularly the main wing.

Figure 13. Baseline Champ Car rear wing assembly, showing initial multi-element aerofoil
section, endplate geometry and bounding box dictated by World Champ Car Series rules.
105.2

105

104.8

104.6
Lap Time (s)

104.4

104.2

104

103.8

103.6
0 5 10 15 20 25 30 35 40 45 50
Number of Full Track Simulations

Figure 14. Convergence of optimisation process showing reduction in lap time versus the number of
full track simulations, each of which requires 27 NEWPAN rear wing assembly calculations.

More detailed analysis of the results has indicated that the rear wing design result has successfully increased the
downforce by improving the interaction between the flap and the main wing element through improved flap
positioning relative to the main wing and also cambering and twisting of both elements. This improvement in
downforce is seen across all 27 operating conditions although the improvement is better for some conditions than
others. Small amounts of flow separation are also predicted for the flap element for some of the operating
conditions. Overall it appears that a well balanced design has been achieved.
This design studies completed to date suggest the operations based optimisation approach has been successfully
implemented and demonstrated for the Champ Car full track optimisation test bed. In practice further studies would
be required to quantify the benefit for this application compared to, for example, a single or multi-point optimisation
and it is hoped that such studies can be completed at some later stage. However since the objective for the study was

13
to demonstrate the operations based optimisation approach rather than to derive the best Champ Car design priority
for further work has switched to aerospace applications.

Baseline Main Wing

Baseline Flap

Optimised Main Wing (Cycle 46)

Optimised Flap (Cycle 46)

Figure 15. Baseline and final designed rear wing sectional shapes for a) adjacent to endplate b) at rear
wing centerline (sections not shown to scale).

V. Discussion
Aerodynamic design problems are rarely limited to operation at a single condition. The Champ Car application
that has been presented clearly demonstrates the difficulty faced by the designer if he is required to choose one or
two representative design conditions which must best characterise typical operating conditions when in reality the
racecar must perform well for a wide range of conditions. In reality the designer is looking for an overall
improvement in terms of the lap time rather than improvements at specific operating conditions. Hence the ability
for the designer to use this overall objective within a design optimisation application without the need to select and
prioritise specific operating conditions removes both effort and reliance upon prior design knowledge.
The situation faced by the designer of the Champ Car is not specific to race car design. Practically all vehicles
whether on the ground or in the air are required to perform at a range of operating conditions. These operating
conditions do not simply relate to speed, altitude, cornering condition etc. They also include other factors such as
payload, fuel load and particular geometry configuration (e.g. aeroelastic shape, external weapons carriage). In fact
each vehicle will have an operating envelope and the overall objective is to design a vehicle which delivers at least
acceptable performance everywhere within this operating envelope. In practice there is a need to prioritise
performance for certain regions within this operating envelope, such as those regions where the vehicle will spend
most of its operational time. For example for a civil aircraft which spends most its time in cruise it would be
expected that priority be given to cruise performance. However these regions where a vehicle may spend most of its
operational time are regions of the operating envelope rather than unique points. For instance any vehicle that uses
fuel which is gradually exhausted will have a varying mass and hence a varying operating condition.
The natural mechanism by which the designer can assign priority to how the performance of the vehicle should
change across the operating envelope is through the definition of operational mission requirements. If there is only
one such operational mission requirement then identification of priority regions for design is probably reasonably
straight forward. However the majority of vehicle types are more likely to have many different operational mission
requirements and the priority for how the performance at different regions of the operating envelope should change
will be dependent upon the cumulative priorities associated with each operational mission requirement. It is also
worth noting that for aerospace customer trends towards demanding greater multi-role flexibility from vehicles
implies that the number of operational mission requirements is likely to increase rather than decrease. This is
potentially just as likely for civil aircraft as for military aircraft since operational flexibility is critical to a customer’s
business model. Relying upon a designer’s experience to select how to define priorities across the operating
envelope may not be appropriate if the range of operational mission requirements is novel.
The long term vision for vehicle optimisation referred to here as operations based optimisation would involve
incorporating operational simulation directly into a vehicle optimisation process. In additional the overall
performance of each design alternative would be assessed, via mission simulation, against multiple operational
mission requirements. It would be this overall performance metric that would be used as a single objective function
within the operations based optimisation problem formulation.
A small scale demonstration of this operations based optimisation approach has been presented within this paper.
In principal it is reasonably straight forward to extend the Champ Car application to allow optimisation for the

14
overall race time, taking into account degradation of tyres, varying fuel loads, refueling strategies etc. and indeed
some of the required extensions to the track simulation capability have already been made. However it is likewise
feasible to trial the approach for an aerospace application in line with the initial priorities for assessing the approach.
As computing resources continue to improve a designer must decide on how best to exploit the new
opportunities this brings. Spare computing resource can be dedicated to increasing the fidelity of the geometry
modelling or increasing the fidelity of the flow physics modelling or incorporating multi-disciplinary interactions
(MDO). The work presented here highlights that the designer can also consider expanding the modelling to better
address the full operational design space.

VI. Conclusion
An improved approach for formulating an aerodynamic optimisation problem directly in terms of an overarching
design driver has been demonstrated. A key benefit of the operations based optimisation approach is to remove the
requirement for the designer to choose specific key operating conditions and weighting factors by directly
incorporating operational simulation and an associated mission requirement into optimisation. This also removes
some reliance upon the designers past experience.
The current Champ Car application serves to explain the process but the general approach has been shown to be
more widely applicable.
Many simplifications have been made within the Champ Car application for the purpose of focusing on an
assessment of the operations based optimisation approach. Whilst these simplifications will ultimately influence the
absolute accuracy of the Champ Car modelling the trends have been used by the optimisation process to both
establish a sensible design and to establish confidence in the benefits of the operations based optimisation approach.

Acknowledgments
J. J. Doherty would like to particularly thank D. P. Clifton for the excellent work he completed during his
undergraduate industrial placement at QinetiQ. Thanks to Dr M. A. Gillan for his input to the current work, drawing
upon his immense knowledge of motorsport design and for providing access to the FROST simulation software.
Thanks to F. Ciampoli for his help with reverse engineering of the Champ Car geometry. Thanks also go to Chris
Burkett at Flow Solutions Ltd. for his support in the use of the NEWPAN method and for kindly providing access to
additional software licenses to enable the optimisation applications to be run in parallel. Finally thanks go to
QinetiQ Ltd. for internally funding the demonstration presented here.

References
1
Official World Champ Car Series website (www.champcarworldseries.com)
2
Doherty, J. J., Domoney, A., Mitchell, R. M. and Burkett, C., “Aerodynamic Design Optimisation Applied to a Formula One
Car”, MIRA International Vehicle Aerodynamics Conference, October 2002.
3
NEWPAN User Guide, v3.4.73, June 2005, Flow Solutions Ltd. (www.flowsol.co.uk)
4
Doherty J. J. and Parker N. T., “Dual Point Design of a Supersonic Transport Wing using a Constrained Optimisation
Method”, 7th European Aerospace Conference - ‘The supersonic transport of second generation’, Toulouse, 1994.
5
Rolston S. C., Doherty J. J., Evans T. P., Grenon R. and Averardo M. A., “Constrained Aerodynamic Optimisation of a
Supersonic Transport Wing: A European Collaborative Study”, Paper AIAA-98-2516, 1998.
6
Hackett K. C., Rees P. H. and Chu J. K., “Aerodynamic Design Optimisation Applied to Civil Transports with Underwing
Mounted Engines”, ICAS proceedings, Melbourne, 1998.
7
Doherty, J. J. and McParlin, S. C., “Generic Process for Air Vehicle Concept Design and Assessment”, AIAA Paper 2004-
895, Invited Paper, 42nd Aerospace Sciences Meeting and Exhibit, Reno, January 2004.

15