DEGREE PROJECT IN MECHANICAL ENGINEERING,

SECOND CYCLE, 30 CREDITS

STOCKHOLM, SWEDEN 2021

Undertray Design and

Development Procedure with
CFD
An Optimization Study of Different Undertray
Designs with CFD Computations

AARON POUTIAINEN

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF ENGINEERING SCIENCES
 UNDERTRAY DESIGN AND
DEVELOPMENT PROCEDURE WITH
CFD
An Optimization Study of Different Undertray Designs with CFD
Computations

Aaron Poutiainen

A thesis submitted in fulfillment of the requirements

for the degree of Master of Science

School of Engineering Science

Department of Mechanical Engineering
KTH Royal Institute of Technology
Stockholm, Sweden 2021
Undertray Design and Development Procedure with CFD
An Optimization Study of Different Undertray Designs with CFD
Aaron Poutiainen <aaronp@kth.se>

Supervisors: Johan Jansson and Måns Andersson

Advisors: Team members from KTH Formula Student
Examinar: Shervin Bagheri

Master Thesis 2021

Host Company: KTH Formula Student
School of Engineering Science
Department of Mechanical Engineering
KTH Royal Institute of Technology
 ABSTRACT
Race car aerodynamics has played a vital part to improve lap times over the years of motor racing.
Having good road adhesion with slick tires and aerodynamic downforce will increase the vehicles
maximum lateral force and thus higher cornering speeds can be achieved. The undertray and
diffuser is the most efficient aerodynamic component on most racing vehicles and is capable of
producing six times more downforce than its contribution to drag plus, if optimized correctly, able
to significantly reduce the vehicle’s drag coefficient. The intent of this project is to optimize a
completely unique undertray design for the KTH Formula Student teams racing vehicle DeV17.
The undertray is inspired by the Aston Martin Valkyrie venturi tunnel design and is optimized
by iterative change in CAD design parameters for three diffrent chassis designs. The results are
obtained with CFD RANS simulations using the k-𝜔 (SST) turbulence model with the Siemens
Star-CCM+ software. The optimum design gave 530 N and 90 N of downforce and drag respectively
at a velocity of 80 km/h. The venturi tunnel design is proven to give a 29% downforce improvement
over a conventional flat plate design with stronger longitudinal vortices and lower, more widespread,
minimum pressure distribution. The most important aspects that affect downforce in undertray
design is concluded to be the diffuser outlet height, upsweep and vehicle ground clearance. No
specific aerodynamic advantages in having a convergent tapering of the tunnel cross-section is
observed, meaning the undertray can be represented as only consisting of an expanding diffuser.
The tunnel design is considered to give promising track testing results and be a spark for further
innovative ideas within aerodynamic design for both the automotive and racing industry.

SAMMANFATTING
Tävlingsbilars aerodynamik har spelat en viktig roll för att förbättra varvtiderna under åren inom
motorsport. Att ha god väghäftning med ’slicks’ däck och aerodynamisk nedåtkraft kommer att
öka fordonets maximala sidokraft förmåga och därmed kan högre hastigheter i kurvor uppnås.
Underredet och diffusern är den mest effektiva aerodynamiska komponenten på de flesta racer-
fordon och kan producera sex gånger mer nedåtkraft än dess bidrag till luftmotståndet och, om
den optimeras korrekt, kan den avsevärt minska fordonets luftmotståndskoefficient. Syftet med
detta projekt är att optimera en helt unik underredes design för KTH Formula Student lagets
racingfordon DeV17. Underredet är inspirerat av Aston Martin Valkyrie venturitunnel design och
optimeras av iterativ förändring av CAD designparametrar för tre olika chassidesigner. Resul-
taten erhålls med CFD RANS-simuleringar med turbulensmodellen k-𝜔 (SST) och programvaran
Siemens Star-CCM+. Den optimala designen gav 530 N och 90 N av nedåtkraft respektive luft-
motstånd under en hastighet på 80 km/h. Venturitunnel designen har visat sig ge en förbättring
på 29% nedåtkraft jämfört med en konventionell platt design, med starkare längsgående virvlar
och lägre, mer utbredd, minimitryckfördelning. De viktigaste aspekterna som påverkar nedåtkraft
i underredes designen dras som slutsats till att vara diffuser utloppets höjd, upphöjning vinkeln
och fordonets markfrigång. Inga specifika aerodynamiska fördelar med att ha en konvergerande
avsmalning av tunnelns tvärsnitt observerades, vilket innebär att underredet kan antas endast
bestå av en expanderande diffuser. Tunneldesignen anses ge lovande bantestresultat och vara en
gnista för ytterligare innovativa idéer inom aerodynamisk design för både bil- och racingindustrin.

i
 ACKNOWLEDGEMENTS
This master thesis project could not be conducted without the help and expertise from the KTH Formula Student team.
The team was crucial with providing licenses to all software, various CAD models and valuable knowledge concerning
the development of racing vehicles. A special thanks to my supervisors Johan Jansson and Måns Andersson for offering
their help and allowing me to pursue a thesis within a field I am passionate about. I would also like to express gratitude
to my examiner Shervin Bagheri for choosing to review this project.

Finally, a heartfelt thank you to my family who always show their loving support in times of need.

iii
 CONTENTS

ABSTRACT i

ACKNOWLEDGEMENTS iii

I INTRODUCTION 1

I.A DOWNFORCE AND TRACK PERFORMANCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

I.B UNDERTRAYS IN RACING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

I.B.1 Venturi-tunnels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

I.B.2 Multi-deck diffusers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

I.B.3 Fan car . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

I.B.4 Flat plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

I.C LIMITATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

II THEORY 5

II.A BASIC 2D AERODYNAMICS OF THE UNDERTRAY . . . . . . . . . . . . . . . . . . . . . . . . 5

II.B DETAILED UNDERTRAY FLOW PHYSICS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

II.C LIMITATIONS OF THE DIFFUSER ANGLE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

II.D INLET AIR COMPRESSION (FRONT DIFFUSER) . . . . . . . . . . . . . . . . . . . . . . . . . . 8

II.E COMPUTATIONAL FLUID DYNAMICS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

II.E.1 CFD Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

II.E.2 The Navier-Stokes equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

II.E.3 RANS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

II.E.4 Turbulence modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

II.E.5 Treating turbulent boundary layer flows in CFD . . . . . . . . . . . . . . . . . . . . . . . 13

II.E.6 Alternative methods of solving Navier-Stokes . . . . . . . . . . . . . . . . . . . . . . . . . 14

III METHODS 16

III.A FIRST DESIGN INSPIRATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

III.B RELEVANT RULES AND DESIGN CONSTRAINTS . . . . . . . . . . . . . . . . . . . . . . . . . 16

 III.C DESIGN ITERATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

III.C.1 First CAD design (Solid Edge) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

III.C.2 Second CAD design (NX) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

III.C.3 Final CAD design (NX) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

III.C.4 Parameterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

III.D METHODOLOGY FOR OPTIMIZATION THROUGH CFD . . . . . . . . . . . . . . . . . . . . . 22

III.E CFD SETUP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

III.E.1 Computational domain and boundary conditions . . . . . . . . . . . . . . . . . . . . . . 23

III.E.2 Mesh setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

III.E.3 Mesh independence study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

III.E.4 Physics setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

IV RESULTS 30

IV.A OPTIMISED DOWNFORCE RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

IV.A.1 Downforce results on first CAD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

IV.A.2 Downforce results on second CAD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

IV.A.3 Downforce results on final CAD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

IV.A.4 Result comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

IV.B COMPARISON WITH THE CONVENTIONAL DESIGN . . . . . . . . . . . . . . . . . . . . . . . 36

IV.C RIDE HEIGHT SENSITIVITY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

IV.D VORTEX EFFECT ON DOWNFORCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

V DISCUSSION 41

V.A LIMITATIONS AND ISSUES WITH OTHER COMPONENTS . . . . . . . . . . . . . . . . . . . . 41

V.B CFD PARAMETER SETUP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

VI CONCLUSIONS 43

VI.A FUTURE WORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

REFERENCES 45

vi
A APPENDIX 46

B APPENDIX 54
I INTRODUCTION 1(54)

I. INTRODUCTION

D
uring the years of automotive history, engineers have realized the huge potential aerodynamics has on the
performance and development of road vehicles. From making cars able to take tight corners without losing grip by
using downforce, managing airflow to prevent overheating of the engine’s components and reducing fuel consumption
by making the body streamlined to reduce drag.

In automotive racing, aerodynamics plays a key role to help reduce lap times and can be the deciding factor of winning
races on certain tracks. Approximately 380 000 USD is the cost of all aerodynamic components on a Formula 1 car
during a racing season, which is nearly 8% of the F1 cars total season cost [1]. In general, motorsport can be considered a
proving ground for the development of technologies and aerodynamic research work which can be applied to road-going
vehicles.

This Master-thesis project will focus on the design and development procedure on the undertray and diffuser of KTH
Formula Students racing vehicle DeV17. The undertray is completely designed from scratch to produce as much
downforce (negative lift) as possible without adding much drag to the vehicle. It will also need to be well integrated
with other important components such as chassis, suspension and the electric motors placed inside the wheels to
fully optimize the vehicle’s performance. CFD simulations will be performed in Siemens Star CCM+ using RANS
(Reynolds-averaged Navier Stokes). Different approaches to aerodynamic improvements of the undertray will be studied
to find out the easiest and most efficient way to go from a conceptual idea to a fully functional high downforce generating
undertray.

A. DOWNFORCE AND TRACK PERFORMANCE

A desire to increase cornering speeds has been a vital goal to improve lap times over the years of motor racing. Engineers
soon learnt that the maximum lateral force of the vehicle was improved by having good road adhesion from having
wider slick tires and more downforce, as shown by the following equation:

Maximum lateral force = 𝐾𝐶 (max) x down load. (1)

𝑘 𝑐 (max) is the cornering coefficient and ’down load’ is considered to be the contact force between road and tire, which
is the sum of the aerodynamic downforce and weight of the vehicle. This means that the maximum lateral force can be
improved by aerodynamics and thus increasing cornering speed. If all four tires were assumed to be able to achieve
the theoretical maximum side force simultaneously and the suspension set-up, weight and aerodynamic downforce
distribution were perfectly matched then the plot of the cornering speed and downforce can be illustrated as shown in
Fig. 1, this is for a corner with a 40 m radius.
I INTRODUCTION 2(54)

K C (max) = 1.0
160
K C (max) = 1.4

140

120

Cornering speed (km/h)

100

80

60

40

20

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Lift-to-weight ratio (L/N)

Fig. 1 Downforce effect on cornering speed with effective tire coefficient 𝐾𝐶 [2]

However, it is not possible to improve cornering ability by increasing the weight, even though this would increase the
’down load’ on each tire, since the vehicle would still need to withstand a larger centripetal force due to the weight
increase.

Apart from enhancing the vehicles cornering ability, aerodynamic downforce also has a great effect on improving the
vehicle’s acceleration and braking. The acceleration is improved since the downforce will allow the tires to transmit
greater thrust forces without the occurrence of wheelspin. This is very important especially for F1 cars due to their
engines being capable of producing wheelspin at speeds over 160 km/h, if no aerodynamics have been applied to the
chassis. When braking the deceleration forces, just as the thrust forces, are limited by the adhesion between road and
tires. Due to most racing vehicles not having ABS (anti-lock braking system) this means that the car will be dependant
on aerodynamic downforce to prevent the tires from locking during braking and worsen the braking distance.

B. UNDERTRAYS IN RACING
The undertray can generate approximately half of all downforce on an F1 car and is the most efficient aerodynamic
device on most racing vehicles [3][4][5], see Appendix A.1. Due to its small frontal area, it is capable of generating
downforce without contributing to any big drag increase. A vast range of undertray designs have been used during the
years of Formula 1, but due to safety regulations many have been banned. Here are some intriguing undertray designs
that have been tested and used in different F1 racing competitions.

1. Venturi-tunnels
Having convergent-divergent venturi tunnels or simply
"huge" tunnels beneath the F1 car resulted in a high
increase of downforce and remarkable levels of grip. This
design was proven to be very successful for the Lotus F1
team, and helped the team to dominate the 1978 Grand
Prix season with their Lotus 79 F1 car. The Lotus 79
used a pair of inverted stub wings enclosed by side plates
which extended down to make constant contact between
the ground and the rubber skirts attached to the car, an
ingenious design by Colin Chapman, see Figure 2. This Fig. 2 Lotus 79 using inverted wings [6] nLT
contact with the road will seal and prevent outside high-
I INTRODUCTION 3(54)

pressure air to sip through underneath the undertray where there is a low-pressure region, resulting in huge amounts of
downforce.

Though this design was very successful, it was considered extremely dangerous since there was a possibility of a sudden
lose of downforce due to the undertray being ride height sensitive when air leaked in beneath the skirts [2]. This lead
to the venturi-tunnel undertray being banned completely from the sport. There is still a considerable possibility of
this design, without the skirts, returning for the 2022 F1 cars, as it has proven to be able to help generate a lot of
downforce without causing too much turbulence in the wake of the car [7]. The means of this is therefore partly to
improve overtaking possibilities.

2. Multi-deck diffusers
Even though there were heavy restrictions on the size of the undertray and being prohibited to have direct contact
with the ground, F1 racing engineers were still able to find ways to further increase airflow beneath the car despite all
restrictions. One way of doing this was using a multi-deck diffuser also called double diffuser. This diffuser setup was
first successfully used in the Brawn GP F1 team and was considered to be the component that helped Jenson Button
dominate and win the 2009 Formula 1 Grand Prix [8]. This result was completely uncalled for by bookmakers who set
the low odds of 1-100 on Jenson Button winning the championship. The double diffuser worked by having an extra
channel that fed more air to an upper deck above the main diffuser, this increased the total amount of mass flow that
travelled beneath the undertray resulting in higher amounts of downforce. This technology was criticised by other teams
for not being rule compliant, but despite this, it was approved by the officials and all other teams started using the
successful double diffuser later that season as well.

In 2011 the F1 technical working group decided to ban multi-deck diffusers entirely from the sport [9].

3. Fan car
Probably one of the most daring and outrageous designs
to increase downforce and grip was seen in the Brabham
BT46B fan car. Just as the name suggests, this F1 car was
fitted with a giant fan at the rear which created suction
underneath by blowing out air at high speed, see Figure 3.

The success of the Lotus 78 and 79 drew the attention

of Brabham designer Gordon Murray. He noticed that
the Lotus F1 teams superiority over the other teams was
strongly due to the venturi tunnels and sliding skirt fitted
to the undertray. A similar design on the Brabham car was
sought after, but unfortunately their Alfa Romeo flat-12
cylinder engine took up too much space, this made it hard
for the team to fit a venturi tunnel big enough for it to be
Fig. 3 Brabham BT46B [10] nL effective [11]. Therefore, Gordon Murray came up with
the idea of fitting a big fan at the rear which would help accelerate the airflow beneath the chassis, reducing pressure and
creating downforce. The fan was driven by a series of complex clutches running from the engine, which meant the
higher rpm the engine ran, the faster the fan would spin and a greater suction would be achieved. This also meant the
Brabham could keep the downforce constant even at low speeds, which could be advantageous in slow corners. It is
undoubtedly that Gordon Murray got inspiration from the 1970 Chaparral 2J “sucker car” from the North America
Can-Am sportscar series. The Chaparral 2J had proven to be significantly faster than its opposition by having higher
cornering speeds thanks to the two fans fitted to its rear.

The Brabham BT46B fan car caused an uproar from the rivalling teams who claimed the car did not follow racing
I INTRODUCTION 4(54)

regulations by having a moving aerodynamic component; the fan. However, Brabham was able to falsely argue that the
primary purpose of the fan was to cool the engine and did not give the car an aerodynamic advantage.

The fan car was never officially banned during the time it actively raced, it was instead voluntarily withdrawn by Bernie
Ecclestone, the owner of the Brabham F1 team. The reason for the withdrawal is believed to be due to the concern of
Ecclestone not getting the support of becoming president of the Formula One Constructors Association (FOCA). This
resulted in Ecclestone negotiating a deal whereby the car would be allowed to race the following three races and later be
withdrawn completely from the sport. However, the Brabham BT46B fan car only raced once in the 1978 Swedish
Grand Prix at Anderstorp, where it stood in a class of its own and won with over a 30-second margin [11].

4. Flat plate
I general, all cars travelling at high speed which have a suspension setting the chassis moderately close to the ground will
have an underbody that produces downforce [5]. Probably one of the most simple ways of improving this downforce,
generated from the underbody, is by smoothing its surface to prevent any disturbance in the incoming airflow that travels
beneath the vehicle. This can be done by adding flat plates to the vehicle’s underside, and this is noticeable on many
racing vehicles as well as normal road vehicles such as the Audi A6 Avant. This design is also currently on today’s F1
cars, most likely due to strict regulations.

The advantages of having a simple design is that it is much easier to integrate the undertray with other components
without it obstructing the space for them, which could easily be the case for components such as the suspension or drive
shaft if the vehicles diffuser height is too large. Whether there is any huge aerodynamic disadvantage of having a simple
flat bottom design compared to larger venturi tunnels will be analyzed in this project.

C. LIMITATIONS
The undertray optimization was mainly done by iteratively changing different CAD design parameters in the CFD
program Star-CCM+ and running the simulations until the results, such as downforce and drag values, were fully
improved. Due to time limitations and lack of computational resources, the simulations can not be too computationally
expensive, meaning that some external parts were excluded, such as rotating wheels, bargeboards, frontwing, etc. The
mesh cell count is also limited to no more than 2.5 million cells. Using these limitations makes it possible for a PC with
8 GB ram, 4 cores and 8 logical processors (AMD Ryzen 5 3500U) to complete a simulation within two hours.
II THEORY 5(54)

II. THEORY

A. BASIC 2D AERODYNAMICS OF THE UNDERTRAY

The way the undertray generates downforce is by creating a low-pressure region beneath the vehicle. This will make the
above air wanting to push down on the chassis and try to come in from each side to create a pressure equilibrium and
thus, downforce is achieved. The low-pressure region beneath the car is caused by two fundamental aerodynamic laws;
mass flow continuity and Bernoulli’s principle [2].

The flow through the undertray works in similar ways as the flow through a venturi tunnel, see Figure 4 and 5. As the air
travels through a smaller cross-sectional area, the air needs to accelerate to hold the principle of mass flow continuity,

𝜌𝑣 1 𝐴1 = 𝜌𝑣 2 𝐴2 , (2)

where 𝜌 is the air density and 𝑣 and 𝐴 is the velocity and area respectively, at cross-sections 1 and 2. Due to the
increase in velocity of the air travelling beneath the vehicle, the pressure needs to decrease as shown in the principle of
conservation of mechanical energy equation, also known as Bernoulli’s principle,

1 2
𝜌𝑣 + 𝑝 = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡, (3)
2
this principle can be applied for isentropic flows i.e. when the effects of irreversible processes, like turbulence, and
non-adiabatic processes, like heat radiation, are neglected. As the velocity increases the pressure 𝑝 needs to decrease for
the equation to be constant. These formulas are all valid under the assumption of incompressible flow, which can be
applied when the airs compressible effect is negligible. This incompressible assumption is used for airflows below 0.3
Mach. For this project, the Formula Student racing vehicle DeV17 will have a top speed of approximately 120 km/h
and this speed is inaccessible during the endurance race due to the track having a lot of tight corners and few straights.
Incompressible flow is therefore used by the assumption that no airflow speeds, around the vehicle, will exceed 0.3
Mach (ca 370 km/h in sea-level conditions).

Fig. 4 Venturi effect Fig. 5 Airflow beneath vehicle

Using Bernoulli’s principle to describe undertray flow physics can be seen as a considerable simplification on realistic
flow behaviour, due to turbulence being neglected. Bernoulli’s equations are applied along the streamlines of the fluid,
for turbulent flows these streamlines are tangled up in complex eddies making it very difficult to describe the pressures
or velocities with mathematical formulas. Another problem is that the phenomena of boundary layer separation is also
neglected with Bernoulli’s equations, this phenomena is important since it will dramatically alter the aerodynamic
forces acting on the vehicle, more on this is covered in section II.C. Therefore, using solely Bernoulli’s principle is
not enough to explain the complexity of undertray flow physics but rather a simplified model to help explain how a
low-pressure region is obtained under the vehicle.
II THEORY 6(54)

In Figure 5, a simple sketch of the airflow moving beneath a car is illustrated. 𝑣 1 is the velocity when the air is
channelled through the smallest cross-sectional area, 𝐴1 (the closest surface to the ground) and 𝑣 2 is when the air exits
the underbody through the diffuser, 𝐴2 , and returns to ambient pressure. Equation 2 can be rewritten as,

𝐴2
𝑣1 = 𝑣2 . (4)
𝐴1

This means that the greater the difference is between the cross-sectional area of 𝐴2 and 𝐴1 the faster the velocity of the
air travelling underneath the car will be [3]. Therefore, having the diffuser as big as possible without causing separation
of the flow can be a reasonable approach to help increase the downforce generated from the undertray. Of course, a
larger diffuser will also slow the velocity of the airflow exiting the underbody, 𝑣 2 , but this velocity decrease is considered
small and will therefore not affect 𝑣 1 as much as the increase of 𝐴2 . A bigger exit area with the diffuser is also beneficial
for decreasing drag by helping pressure recovery. If we instead consider an underbody with a constant cross-sectional
area, this would result in the air needing to make a quick transition from a low-pressure region beneath the car to a
higher pressure region behind the car, causing turbulence and increasing drag.

B. DETAILED UNDERTRAY FLOW PHYSICS

According to renowned Formula One aerodynamicist Willem Toet, there is not much knowledge to be gained when
trying to analyse the performance of the undertray and diffuser in 2D flow, since it can’t explain what is important to
understand such as taking vortices into consideration [4]. Therefore, to get a better understanding of how undertray’s
produce downforce and what can impact their performance one needs to analyze them in the 3D domain.

In [5] it is described that three main mechanisms influence the downforce generated from the undertray, these are
mainly; "pressure recovery", "diffuser upsweep" and "ground effect" which shows the interaction between undertray and
ground. The diffuser is therefore considered to be one of the features on the undertray which will affect its downforce
performance the most, but is also acknowledge to be the least understood aerodynamical part on the vehicle [3].

The authors describe the diffuser acting like a "pump down" mechanism which helps increase downforce on the undertray
even more by enabling pressure recovery. The pressure recovery is important to determine the magnitude of the
minimum pressure under the car and the undertray’s performance can be reflected by the pressure recovery coefficient,

𝐶 𝑝2 − 𝐶 𝑝1
𝐶𝑝 = , (5)
1 − 𝐶 𝑝1

where 𝐶 𝑝1 and 𝐶 𝑝2 is the inlet and outlet pressure coefficient at the diffuser respectively, see Appendix A.5. From
this, it can be theoretically understood that having a larger high angled diffuser will help increase the pressure recovery
coefficient more and thus improve downforce by lowering the underbody pressure.

The diffuser upsweep describes the upwards slope surface of the diffuser ramp which helps change the direction of the
flow upwards. This upward motion will help to improve downforce due to the airs upward reorientation.

The ground effect phenomena occur when there is an interaction between aerodynamic components in proximity with
the ground. For undertray’s the ground effect explains the relationship between the ride height of the vehicle and the
amount of downforce it generates. In theory, the lower the ride height the more downforce will be generated due to
ground effect, however, in practice, the maximum downforce will be limited by viscous effects or blockage effects at low
ride-heights. At the reference frame of the car, the ground will move backwards at the same speed as the free-stream
and the grounds boundary layer will help ’pull’ the air above it, through the undertray, thus enhancing downforce, this is
an example of Couette flow.

To achieve a mathematical representation of the downforce and drag generated on a vehicle, from the undertray, a
simplified version of a vehicle called the Ahmed body can be aerodynamically analyzed, see Appendix A.4. The
II THEORY 7(54)

Ahmed body was originally described by S.R. Ahmed in 1984 and is a standard geometry used to study flows around
bluff body’s which has in some studies also been adapted for analyzing ground-effect diffuser flows. The downforce
component can be pressure-based with the assumption that the centerline pressure is a sufficient representation of
the average pressure across the bluff body’s width, 𝑊. Equations 6 and 7 show a representation of the aerodynamics
downforce and drag respectively, generated from the undertray using this assumption, here the forces can be described
as the surface pressure distribution integrated over the body surface which will give an estimate of the pressure force
acting on the undertray [5]. These equations are considered to show good correlations with direct force measurements
and produce identical trends in results for wind tunnel experiments of the Ahmed body [12].

  
  
 ∫ 𝐿𝐵 −𝐿𝐷 ∫ 𝐿𝐵  ∫ 𝐿𝐵 −𝐿𝐷 
(6)
  
𝐿 = −𝑊   𝑝 𝑓 (𝑥) 𝑑𝑥 + 𝑝 𝑑 (𝑥) 𝑑𝑥  − 𝑝 𝑢 (𝑥) 𝑑𝑥 
 0 𝐿 −𝐿 0
} | 𝐵 𝐷{z
 
 | {z } | {z }
  Flat underbody (floor) Uppderbody
Diffuser
 
  
  
  
∫ 𝐻 ∫ ℎ0 ∫ 𝐻 
(7)
  
𝐷 = 𝑊  𝑝 𝑛 (𝑦) 𝑑𝑦 −  𝑝 𝑑 (𝑦) 𝑑𝑦 + 𝑝 𝑏 (𝑦) 𝑑𝑦  
 0  0
} | 0 {z
ℎ 
| {z } | {z } 
Nose Diffuser Base (rear)
  
  
In Equation 6 the downforce is represented as the width of the bluff body multiplied with the pressures integrated over
the bluff body length, 𝐿 𝐵 , and over the diffuser length, 𝐿 𝐷 . Similar with Equation 7 for the drag representation, but here
the pressures are integrated over the bluff body’s height, 𝐻, and diffuser height ℎ0 , see Appendix A.5 for illustration. 𝑥
and 𝑦 are the streamwise and ground-normal directions respectively and the subscripts 𝑓 , 𝑑, 𝑢, 𝑏, 𝑛 denote the static
pressures for the flat underbody, diffuser, upper body, base and nose of the bluff body respectively.

From these equations, it is noticeable that the diffuser has a contributing factor to the undertray’s downforce performance
while also helping to decrease the overall drag of the Ahmed body. The minimum pressure point is located at the start of
the diffuser.

Longitudinal counter-rotating vortices travelling in the streamwise direction of the vehicle are formed in the diffuser,
which has a big influence on the undertray’s aerodynamic performance. The vortices have the capability of enhancing
suction by taking high kinetic energy air from the outside and beneath the vehicle and accelerate it into the diffuser
[4][5]. Depending on the width of the undertray, further downforce enhancement may be achieved with the use of vortex
generating fences incorporated into the diffuser or to prevent ’dirty’ air generated from the wake of the wheels to enter
the diffuser.

C. LIMITATIONS OF THE DIFFUSER ANGLE

Though it would greatly increase the aerodynamic downforce, the size of the diffuser has many limitations. One
limitation is the risk of flow separation when the angle of the diffuser increases too dramatically.

All bodies surrounded by the relative movement of fluid will have a boundary layer with viscous forces close to the
surface of the body. This boundary layer can either be laminar or turbulent and during flow separation (or boundary
layer separation) it will detach from the surface of the body and flow into the wake. When this occurs all forces acting
on the body, such as lift and drag, will change dramatically. The main phenomenon that causes separation is due to a big
pressure difference along the body, this is called an adverse pressure gradient. The adverse pressure gradient will occur
when the static pressure increases in the opposite direction of the flow. When this happens the fluids potential energy
will increase leading to the kinetic energy decrease and this will result in a deceleration of the fluid’s velocity. This
deceleration is most prominent near the wall of the geometry since the fluid’s velocity in the boundary layer is slower
compared to the free-stream velocity. If the static pressure increase is large enough the fluid’s velocity may slow down
II THEORY 8(54)

to zero or even become reversed, causing boundary layer flow separation.

It is well documented in aerodynamic observations that turbulent boundary layers tend to sustain an adverse pressure
gradient better than laminar boundary layers, this means that by forcing a transition from laminar to turbulent one can
prevent the separation phenomenon from occurring on bodies that would otherwise cause separation had the flow been
laminar. This is due to the efficient mixing turbulent boundary layers have with transporting kinetic energy from the
edge of the layer to the low momentum flow close to the surface, and thus preventing separation. One way of forcing a
transition from laminar to turbulent boundary layer flow is the use of vortex generators [13].

D. INLET AIR COMPRESSION (FRONT DIFFUSER)

According to Barnard [2] there are no particular downforce
benefits of having a compression of the air when it is
travelling from the inlet to the throat (the section with
the smallest cross-sectional area) of the undertray. In
other terms, this means that there is no benefit of having
the convergent part of the venturi tunnel present when
designing the undertray, that the divergence via the diffuser
is a design feature more effective for producing downforce.

If the inlet height is increased as seen in Figure 6c it will

mainly alter the flow and pressure distribution around
the inlet and the minimum pressure at the throat will
Fig. 6 Underbody inlet designs, front wheel downforce
be roughly the same for all three configurations. The
reduces from A to C [2]
disadvantage of having a higher inlet height is that there
will be a local lift force generated at the inlet when the air from the free-stream hits the front surface, creating a
high-pressure region. This can cause the overall downforce of the vehicle to decrease as the inlet design configuration is
creating lift. Due to this, it can be argued whether the Lotus 79 F1 car undertray design was optimal, since it used
inverted wings as the underside of the undertray and this would cause the front wheels to have less downforce compared
to if there was no inlet height to the undertray, see Figure 2.

E. COMPUTATIONAL FLUID DYNAMICS

In an idealistic world, engineers will be capable of accurately depicting advanced aerodynamics through simple computer
simulations, and the need for validation using expensive wind tunnel testing will be a mere historical glimpse of what
was once necessary for aerodynamicist. Unfortunately with today’s computer technology, we are far from a scenario
where computer simulations can give an accurate depiction of reality without some smaller, even sometimes greater,
magnitudes of inaccuracy, let alone achieving results reliable enough to replace any wind tunnel. Nevertheless, the
integration of computerized fluid dynamics into a wide range of engineering disciplines is rising dramatically due to
positive trends in computational power and affordability.

Computational fluid dynamics, in short CFD, uses numerical analysis, approximations and data structures to solve
problems and illustrate simulations involving fluid flows. Computers are used to perform the calculations required to
simulate how free-stream flow interacts with a body of surfaces defined by boundary conditions. The time to solve
certain simulations can be heavily affected by the computers processing speed which is a big limiting factor for running
simulations with complex geometries. However, using CFD has proven to be a viable analysis method for understanding
fluid mechanics and there are a variety of different software and solving methods available for engineers to choose from.

1. CFD Methodology
To perform a CFD simulation the following steps are usually made (more details on this will be covered in methods):
II THEORY 9(54)

• The geometry and its physical boundaries are defined using computer-aided design (CAD), which is imported into the
CFD program.
• The domain (or volume where the fluid is occupied) is then defined as an extruded box surrounding the CAD geometry
and the geometry is subtracted from this domain (this step may differ depending on how complicated the CAD is).
• The domain is then divided into discrete cells (the mesh) that can either consist of (or be combined with) hexahedral,
tetrahedral, polyhedral, prismatic or pyramid cell elements.
• The boundary conditions are defined, this involves specifying all bounded surfaces of the domain such as inlet, outlet,
walls etc.
• The physical modelling is defined where the equations that should be solved are chosen, in this project the Navier-Stokes
equation is used with a specific turbulence model (RANS).
• The simulation can then be started and the equations will be solved iteratively until the residuals become small and
reach a steady state.
• And finally a postprocessing tool is used to analyze and visualize the resulting solutions.

This is a summarized procedure of how a general CFD simulation is made, to understand what the software is doing one
needs to consider the Navier-Stokes equations.

2. The Navier-Stokes equations

The Navier-Stokes equations are the most general set of partial differential equations used to describe viscous fluid
motion. They are the governing equations of fluid dynamics and consists of conservation laws of mass, momentum
and energy. The Navier-Stokes equations are therefore capable of determining pressure and velocities anywhere in
the flow around any shape. The continuity equation is expressed in the differential form (Equation 8) and the energy
equation involves the interchange between different energy forms; such as kinetic and potential energy. For road vehicles
and racecars, the Navier-Stokes equations can be simplified concerning incompressible flow and disregarding the
energy equation due to the temperature gradient being negligible. This will lead to the following simplifications of the
continuity and momentum equations [14];
𝜕𝜌
+ ∇ · (𝜌u) = 0, (8)
𝜕𝑡
where u = 𝑢 𝑥 , 𝑢 𝑦 , 𝑢 𝑧 is the fluids flow velocity and for incompressible flow 𝜌 is constant so 𝜕𝜌
𝜕𝑡 = 0, this leads to,

3
Õ 𝜕𝑢 𝑖
∇ · u = 0, i.e. = 0. (9)
𝑖=1
𝜕𝑥 𝑖

The simplifications of the momentum equation (Equation 10) requires longer calculations;

𝜕 𝜌u
+ ∇ · (𝜌u ⊗ u) = −∇𝑝 + ∇ · 𝜏 + 𝜌f, (10)
𝜕𝑡
where f represents body forces acting on the continuum, for example gravity or inertial acceleration. Using ∇ · u = 0
some terms in Equation 10 can be simplified in the following way,

∇(u ⊗ u) = (u · ∇)u + (∇ · u)u = (u · ∇)u (11)

  2
𝜏 = 𝜇 ∇u + (∇u)𝑇 − 𝜇(∇ · u)I = ∇u + (∇u)𝑇 . (12)
3
Equation 12 is the the Stokes stress constitutive equation and the divergence of this stress is given by;
   
∇ · 𝜏 = 𝜇∇ · ∇u + (∇u)𝑇 = 𝜇 ∇2 u + ∇(∇ · u) = 𝜇∇2 u. (13)
II THEORY 10(54)

Now by inserting (11) and (13) into (10) the incompressible version of the momentum equation can be obtained,

𝜕u 1 𝜇
+ (u · ∇)u = − ∇𝑝 + ∇2 u + f. (14)
𝜕𝑡 𝜌 𝜌

Finally, by combining the equations (9) and (14) the set of partial differential equations known as the Navier-Stokes
equations for incompressible flow is obtained,
(
∇·u=0
(15)
𝜕u
𝜕𝑡 + (u · ∇)u = − 𝜌1 ∇𝑝 + 𝜈∇2 u + f.

𝜈 = 𝜌𝜇 is the kinematic viscousity and each term in Navier-Stokes can be categorised; 𝜕u 𝜕𝑡 is the variation-, (u · ∇)u the
convection-, 𝜈∇2 u the diffusion of velocity and 𝜌1 ∇𝑝 and f are the internal and external sources.

Though the Navier-Stokes equations are considered very useful in describing fluid motion they have not yet been proven
to have any smooth solutions in three dimensions, this means that they are infinitely differential at all points in the
domain. This is considered a great challenge for mathematicians and The Clay Mathematics Institute has called this the
Navier–Stokes existence and smoothness problem and has an offer of 1 million USD for anyone capable of presenting a
valid solution or counterexample.

The CFD program needs to use a particular discretization method of the non-linear PDE’s in Navier-Stokes to transfer
the continuous equations into discrete counterparts, this makes it easier to derive numerical solutions of the equations.
The most common discretization method used in commercial CFD software, for 3D simulations, is the finite volume
method which will recast the Navier-Stokes equations in a conservative form and will solve them over discrete control
volumes, through cells in the mesh.

The flow field is split into three-dimensional grid blocks and the values at the centre of each block such as velocity,
pressure etc. can be determined. The numerical methods used to solve the equations involve systematic prediction and
correction of the values at each point iteratively. The exact solution will never be found and the simulations results are
considered sufficient when the values cease to alter significantly after each new iteration. How much the values will
fluctuate or reach a steady-state after a number of iterations is shown via the residuals which quantify the error in the
solution of the system of equations. The residuals can never reach zero but the lower the residual value is, the more
numerically accurate the solution will be.

3. RANS
Turbulence is the most naturally occurring fluid flow which can be observed in everyday phenomena such as flowing
rivers, chimney smoke, storm clouds and more. It is caused by excessive kinetic energy in parts of the fluid which
overcome the damping effects of the fluid’s viscosity, this creates unsteady vortices, called eddies, that interact with
each other and will have a significant effect on boundary layer behaviour and thus skin friction drag. There is a great
mathematical difficulty in predicting and analysing turbulent flows due to the chaotic disorder of particles and renowned
physicists like Richard Feynman has described turbulence as the most important unsolved problem in classical physics
[15].

In CFD, problems arise when describing flow using the Navier-Stokes equations since the solutions will only accurately
depict real physical flow when the very finest details of turbulent motion are taken into account. Since the eddies
created in turbulent flows can be both large scale and microscopically small, to take account of their effects fully, the
computational grid has to be extremely fine in certain areas. It has been estimated that approximately 1018 number of
cells are needed to capture and solve the smallest turbulent motions around the structure of a car when numerically
solving the Navier-Stokes equations directly without using any turbulence models [2]. This is called direct numerical
solution (DNS) and is not considered a practical proposition for aerodynamics in any immediate future due to the
II THEORY 11(54)

immense computing power needed to do simulations. To get around this computational problem one can simplify the
terms that describe turbulent motion by substituting them with a specific turbulence model. The solving methods used
when implementing turbulence models in the Navier-Stokes equations are either Reynolds-averaged Navier-Stokes
(RANS) or Large Eddy Simulations (LES). Both these solving methods require less computational power than DNS, but
cannot be used to understand the physics of turbulence since the turbulence scales and their interactions are not fully
resolved [16]. Due to the approximations in the turbulence model, it can be argued whether the solving methods used
will show some magnitudes of inaccuracy since the simulations will not represent real physical flow phenomena with
regards to turbulence.

4. Turbulence modelling
In the RANS equations, all turbulent fluctuations are averaged and the turbulence model is used to mimic the effects of
turbulence. To derive RANS from the Navier-Stokes equations the Reynolds decomposition needs to be used, which
involves the separation of the velocity flow variable into the time-averaged (mean) component, u, and the fluctuating
component, u0, in the following way,
u(x, 𝑡) = u(x, 𝑡) + u0 (x, 𝑡). (16)
Using the Reynolds decomposition, the Navier-Stokes equations (15) can be rewritten as [14],

∇ · u = 0


(17)

h i
1
 𝜕𝑡 + (u · ∇)u = ∇ · − 𝜌 𝑝I + 𝜈∇u − u0 ⊗ u0 + f.
 𝜕u

This system of equations is known as the RANS equations and can be written in tensor formation as,

𝜕𝑢𝑖
=0


(18)
 𝜕𝑥

𝑖 h   i
𝜕𝑢 𝑗
𝜕𝑢
 𝜌𝑢 𝑗 𝜕𝑥 𝑖𝑗 =
 𝜕
𝜕𝑥 𝑗 −𝑝𝛿𝑖 𝑗 + 𝜇 𝜕𝑢 𝑖
𝜕𝑥 𝑗 + 𝜕𝑥𝑖 − 𝜌𝑢 𝑖0𝑢 0𝑗 + 𝜌 𝑓 𝑖 .


The Reynolds stress term, −u0 ⊗ u0 (or −𝜌𝑢 𝑖0𝑢 0𝑗 in tensor form) is a nonlinear term which represents the correlations
between fluctuating velocities (turbulence). This term is unknown and therefore needs to be modelled with a specific
turbulence model to write the RANS equations in closed form.

There are a variety of different turbulence models one can choose from, here are some of the most common ones used in
the CFD industry [17][18]:
II THEORY 12(54)

Table 1 Turbulence models

Model Nr of eq. Wall functions applicable Strength Weakness

• Stable with good convergence. • Solving shear flows.

Spalart- • Accurate for attached wall bounded • Underpredicitng seperation.
1 No
Allmaras flows. • Decaying turbulence.

• Valid mostly for fully turbulent

• Simple and easy to implement.
flows.
• Good convergence.
k-𝜖 2 Yes • Inaccurate for no-slip walls,
• Reasonable predictions for flows with
flows with strong separation and
complex geometry.
swirling/rotating flows.

• 𝜔 is easier to solve than 𝜖.

• Hard to converge.
• Similair to k-𝜖 but improved accuracy
k-𝜔 2 Yes • Sensitive to initial conditions.
for internal flows, curvatures and sep-
arated flows.

• A combination using K-𝜖 in the outer

outer region cells, outside the bound-
• May still have convergence issues.
k-𝜔 (SST) 2 Yes ary layer, and k-𝜔 in the inner bound-
ary layer.
• Better convergence than standard k-𝜔.

In this project, the turbulence model of k-𝜔 (SST) is used, due to its relatively quick convergence and reliability when
utilizing the transition between two different turbulent models. This model will therefore combine a robust low-Reynolds
turbulence model down to the viscous sublayer and use a high Reynolds turbulence version of the k-𝜖 model in the outer
portion of the boundary layer.

All turbulent models presented in Table 1 use linear eddy viscosity models which are based on the Boussinesq hypothesis.
This hypothesis states that there is a correlation between viscous stresses and the Reynolds stress term on the mean flow.
It also estimates the turbulent viscosity to be isentropic, which means that the Reynolds stresses and the mean strain rate
is equal in all directions, this can be expressed as;

𝑅𝑖 𝑗 2
= −𝑢 𝑖0𝑢 0𝑗 = 2𝜈𝑡 𝑆𝑖 𝑗 − 𝑘𝛿𝑖 𝑗 , (19)
𝜌 3

where 𝑅𝑖 𝑗 represents the Reynolds stress term, 𝜈𝑡 is the turbulent eddy viscosity, 𝑘 is the turbulent kinetic energy, 𝛿𝑖 𝑗 is
the Kronecker delta and 𝑆𝑖 𝑗 is the mean rate of strain tensor,

0
!
𝜕𝑢 𝑖0 𝜕𝑢 𝑗
𝑆𝑖 𝑗 = + (20)
𝜕𝑥 𝑗 𝜕𝑥𝑖

To obtain the Navier-Stokes equations in closed form, only containing the velocity and pressure, one needs to close the
RANS equations by modelling the stress term 𝑅𝑖 𝑗 as a function of the mean flow, shown in Equation (19), this is called
the closure problem [19].

For the Wilcox k-𝜔 model the turbulent eddy viscosity is given by, 𝜈𝑡 = 𝑘/𝜔, where 𝑘 and 𝜔 are two partial differential
equations, 𝑘 being the turbulent kinetic energy as stated before and 𝜔 is the specific rate of dissipation of the turbulent
kinetic energy into the thermal energy. The two transport equations needed for 𝑘 and 𝜔 for a turbulent flow is stated
II THEORY 13(54)

with the following equations [19][20];

   
𝜕 (𝜌𝑘) 𝜈𝑡
+ ∇ · (𝜌𝑘u) = ∇ · 𝜌 𝜈 + ∇𝑘 + 𝑃 𝑘 − 𝛽∗ 𝜌𝑘𝜔 (21)
𝜕𝑡 𝜎𝑘

2
     
𝜕 (𝜌𝜔) 𝜈𝑡 𝜕𝑢 𝑖
+ ∇ · (𝜌𝜔u) = ∇ · 𝜌 𝜈 + ∇𝑘 + 𝛾1 2𝜌𝑆𝑖 𝑗 · 𝑆𝑖 𝑗 − 𝜌𝜔 𝛿𝑖 𝑗 − 𝛽1 𝜌𝜔2 . (22)
𝜕𝑡 𝜎𝜔 3 𝜕𝑥 𝑗
Due to their complexity, equations (21) and (22) can be written in words to help get a better understanding [20];

For the Menter’s k-𝜔 SST (shear stress transport) model the 𝑘-equation is the same as for the Wilcox k-𝜔 model,
equation (21). But the 𝜖-equation is transformed into an 𝜔-equation by substituting 𝜖 = 𝑘𝜔 when moving to the cells of
the inner region of the boundary layer. This leads to the following change of equation (22);

2
     
𝜕 (𝜌𝜔) 𝜈𝑡 𝜕𝑢 𝑖 𝜌 𝜕𝑘 𝜕𝜔
+ ∇ · (𝜌𝜔u) = ∇ · 𝜌 𝜈 + ∇𝜔 + 𝛾1 2𝜌𝑆𝑖 𝑗 · 𝑆𝑖 𝑗 − 𝜌𝜔 𝛿𝑖 𝑗 − 𝛽1 𝜌𝜔2 + 2 , (23)
𝜕𝑡 𝜎𝜔,1 3 𝜕𝑥 𝑗 𝜎𝜔,2 𝜔 𝜕𝑥 𝑘 𝜕𝑥 𝑘

the extra term (on the far right side) is called the cross-diffusion term, which arises during the 𝜖 = 𝑘𝜔 transformation in
the 𝜖- equation. Specific values are denoted to the revised model constants, 𝜎𝑘 , 𝜎𝜔,1 , 𝜎𝜔,2 , 𝛾2 , 𝛽1 and 𝛽∗ in order
optimise the performance of the k-𝜔 SST model. These constants values are purely based on experience with the model
in general-purpose computations.

5. Treating turbulent boundary layer flows in CFD

According to the second law of thermodynamics, all fluids have positive viscosity, unless special cases for very
low-temperature superfluids, which can have zero viscosity. The fluid’s viscosity will result in internal frictional forces
acting between the adjacent layers of fluid that are in relative motion. This will in turn result in skin friction drag due to
the shear stresses arising in a stationary boundary wall when exposed to a fluid moving along the boundary. Equation
(24) is how the shear stress is defined for a flat plate with a fluid in relative motion to its surface, where 𝜇 is the fluids
𝜕𝑦 is the local shear velocity gradient, see Appendix A.2 for illustration.
viscosity and 𝜕𝑢

𝜕𝑢
𝜏=𝜇 (24)
𝜕𝑦

The usual way to treat wall boundaries, with no-slip wall conditions in CFD, is to create a mesh grid sufficiently small
so that sharp gradients and small eddies prevailing are resolved there. According to the law of the wall, first published
by Theodore von Kármán in 1930, the two dimensionless properties 𝑦 + and 𝑢 + are used to describe the average turbulent
flow velocity from a certain point at a distance from the wall and states that the velocity is proportional to the logarithm
of the distance from that point to the wall. The dimensionless velocity 𝑢 + is defined as the free-stream velocity 𝑢 divided
with the friction velocity 𝑢 𝜏 and is also defined as a function of 𝑦 + ;
 
𝜌𝑢 𝜏 𝑦 𝑢
+
𝑢 = 𝑓 = 𝑓 (𝑦 + ) = , where (25)
𝜇 𝑢𝜏
r
𝜏𝑤
𝑢𝜏 = . (26)
𝜌
II THEORY 14(54)

𝑦 + is the dimensionless wall coordinate defined as the distance from the wall made dimensionless by multiplying the
absolute wall distance 𝑦 with the friction velocity 𝑢 𝜏 divided with the dynamic viscosity 𝜈;
𝑦𝑢 𝜏
𝑦+ = . (27)
𝜈
𝑦 + can also be interpreted as a local Reynolds number based on the distance normal to the wall (Re 𝑦 = 𝑈𝑦/𝜈). For low
values on 𝑦 + close to the wall the viscous forces are dominating the flow, as 𝑦 + increases we move further from the wall
and here the inertia forces are dominating, making the viscous forces negligible.

To account for the fluid’s behaviour very close to the wall, which is mostly dominated by viscous effects and not affected
by free stream parameters, the viscous sub-layer is used. The viscous sub-layer is in practice so close to the wall that it
stays within the layer of 𝑦 + < 5 and is proven to have 𝑢 + = 𝑦 + within this layer, see Appendix A.3 [20]. This is valid
under the assumption that the shear stress is approximately constant and equal to the wall shear stress throughout the
layer;
1 2
𝜏(𝑦) = 𝜏𝑤 = 𝐶 𝑓 𝜌𝑈∞ . (28)
2
The 𝑦 + property is useful for calculating the first layer thickness, Δ 𝑦 1 , of the cells at the boundaries in the CFD
simulations. First, the desired value of 𝑦 + needs to be denoted, the choice of this value will be dictated by which
turbulence model is used and if the model is a high- or low-Reynolds number model. All 𝜔-based models are by
construction low-Reynolds number models that give very accurate descriptions of the boundary layers by correctly
reproducing the behaviours of different flow quantities close to the wall, i.e. at the viscous sublayer [21]. Therefore,
when using this turbulence model to solve the flow at the near-wall region, the cell centre must be placed in the viscous
sublayer, preferably at 𝑦 + = 1 [22][23]. The first cell layer thickness can then be calculated using the following formula
[24];
𝑦+ 𝜇
Δ 𝑦1 = . (29)
𝑢𝜏 𝜌
These equations are based on the flat plate boundary layer theory.

6. Alternative methods of solving Navier-Stokes

Many researchers have dedicated a lot of their work to try to find reasonable ways of treating turbulent flows with
CFD. One alternative method to the averaged approximations made with RANS is to treat the differential equations
in Navier-Stokes with a weighted least-squares stabilized Galerkin finite element method (called G2) which allows
adaptivity to be utilized in the mesh [25]. Researches behind this computational methodology refer to it as Direct FEM
Simulations or in short DFS which uses G2 adaptivity to solve the Navier-Stokes equations. The adaptivity is based on
solving a linearized dual problem and obtain sensitivity in output or quantities of interest like the residuals and the finite
element mesh size. DFS can be seen as a combination of adaptive DNS and LES where certain parts of the domain,
at critical local flow features, such as boundary layer separation and transition the DNS is used, while LES is used
in larger parts of the domain such as in the turbulent wake of the bluff body [26]. G2 adaptive DNS/LES is said to
enable simulations with mesh points in orders of magnitudes smaller than using non-adaptive mesh refinement and is
considered a breakthrough in computational simulations of turbulent flow by its developers.

The adaptive mesh refinement is based on an a posterior error estimation of certain outputs of the solutions, such as
forces, stresses, fluxes or local mean values. The a posterior error estimate for a continuous Galerkin method in space
and time, which is called cG(1)cG(1), takes the following form [27][28]:
𝑁 Õ 𝑁 Õ
ˆ ≤ ˆ 𝜑]
Õ Õ
ˆ − 𝑀 (𝑈)|
|𝑀 ( 𝑢) E 𝑛,𝐾 = ˆ − 𝑅(𝑈; 𝑈),
[𝑅(𝑢; 𝑢) ˆ 𝑛,𝐾 , (30)
𝑛=1 𝐾 ∈𝑇 ℎ 𝑛=1 𝐾 ∈𝑇 ℎ

where 𝑈ˆ = (𝑈, 𝑃) is an approximate solution to the primal problem which is continuous piecewise in space and time for
II THEORY 15(54)

a coarse mesh with maximum cell size ℎ, 𝑢ˆ = (𝑢, 𝑝) is a weak solution of the Navier-Stokes equations, 𝜑ˆ = (𝜑, 𝜃) is
a solution to the linearized adjoint (dual) problem and E 𝑛,𝐾 is a error indicator over the cells 𝐾 in a discrete space
mesh 𝑇 ℎ . The adaptive mesh algorithm starts from an initial mesh where the primal problem is computed for the
full-time interval after which the dual problem is solved backwards over the same time interval. The error indicators
E 𝑛,𝐾 are then computed for each cell 𝐾 in the mesh, finally the cells with the highest E 𝐾 are bisected to
Í𝑁
E 𝐾 = 𝑛=1
ˆ is detected. The algorithm can
create a new refined mesh and the error algorithm starts over until convergence in 𝑀 (𝑈)
be summarised with the following steps given a coarse mesh 𝑇0 , 𝑘 = 0 [27][28]:
ℎ

1) Compute a solution 𝑈ˆ to the primal problem using 𝑇𝑘ℎ .

2) Compute the solution 𝜑ˆ to the dual problem using 𝑇𝑘ℎ .
ˆ is observed, or if the estimated error Í𝐾 ∈𝑇 ℎ E 𝐾 is less than a given tolerance, then STOP,
3) If convergence in 𝑀 (𝑈)
else
4) bisect 10% of cells 𝐾 ∈ 𝑇 ℎ with highest error indicators E 𝐾 .
5) Set 𝑘 = 𝑘 + 1, then go to 1).

The DFS methodology uses the assumption of free-slip boundary conditions at high Reynolds numbers, 𝑅𝑒 ≥ 106 ,
meaning the air’s viscous effects are neglected [27]. This allows simulations without the need to resolve the viscous
boundary layers in the mesh and partly due to this the mesh cell count can be substantially lower than meshes of standard
RANS simulations. The validity of this assumption can be debated since it goes against the scientific consensus of
Ludwig Prandtl’s boundary layer resolution of d’Alembert’s paradox. The paradox proves that for an incompressible
inviscid potential flow the drag force is zero on a body moving with constant velocity relative to the fluid, which is a
direct contradiction to all experimental observation. Prandtl’s resolution to the paradox is to take the air’s viscosity into
consideration at a region close to the wall and states that thin viscous boundary layers could be the possible cause of the
drag generation on bodies [29]. Prandtl proposed an approximate model of the flow inside the boundary layers with a
no-slip boundary condition, called boundary-layer theory, but also kept the assumption of inviscid flow theory outside
of the boundaries in order to grant a mathematical solution to the non-linear Navier-Stokes equations. Prandtl’s theory
results in the arise of skin friction drag and is proven to be favourable in aeronautic applications.

Researchers behind DFS propose in [30] a new resolution to d’Alembert’s paradox where they motivate the use of a
full-slip boundary condition which is a direct contradiction to Prandtl’s boundary layer theory. They state that the fluids
viscous effects diminish with increasing Reynolds number and suggest that the result of the drag is mainly from the
turbulent nature of the flow with only a small drag contribution coming from the viscous boundary layer. This is proved
with a turbulent computational solution of the Euler equations with the adaptive finite element method and the results
are backed by various analytical and computational evidence [30].
III METHODS 16(54)

III. METHODS

A. FIRST DESIGN INSPIRATIONS

In the autumn of 2019, the KTH Formula Student (KTHFS) team decided to take a different approach in the development
of their new racing car DeV17. There was no longer going to be deadline compromises of the extremely important track
testing. To solve this they decided to dedicate 2 years to develop a completely new car from scratch and spend the last
6 months mainly for track testing before the main racing competition, FS Germany. This meant that new innovative
engineering ideas that had not yet been tested could be presented and developed for the new car. Some of these ideas
included using four separate motors for each wheel, torque vectoring and a completely redesigned aero package.

All the different undertray designs covered in

the Introduction were presented to the team.
The first ideas that were considered to be
implemented were inspired by the successful
and ingenious Brabham BT46B fancar, since
no rules banning the fan design were notable
in the FS rule book for 2020. But this idea was
scrutinized by KTHFS advising engineers and
was quickly scrapped thereafter. The second
design was to use venturi shaped tunnels,
inspired by equally successful Lotus 79 and
the modern road car Aston Martin Valkyrie,
Fig. 7 Aston Martin Valkyrie [31] nLT
designed by acclaimed aerodynamicist Adrian
Newey. This design was considered reasonable enough concerning KTHFS resources and would be a completely new
innovative idea of achieving high downforce from the undertray, a design that had never been used on earlier KTHFS
vehicles or noticed on any other FS team in Europe.

B. RELEVANT RULES AND DESIGN CONSTRAINTS

Many rules need compliance for the team to be accepted to attend the competition. The 2020 FS Germany rule book
consists of over one thousand different rules governing nearly all parts of the vehicle, approximately ten of these rules
concern aerodynamic components. The most relevant rules limiting the undertray are;

• D 2.3 "Sliding skirts or other aerodynamic devices that by design, fabrication or as a consequence of moving, contact
the track surface are prohibited."
• T 2.3.2 "The minimum static ground clearance of any portion of the vehicle, other than the tires, including a driver,
must be a minimum of 30 mm."
• T 2.1.3 "No part of the vehicle may enter a keep-out-zone defined by two lines extending vertically from positions
75mm in front of and 75mm behind the outer diameter of the front and rear tires." (see Figure 8)
• T 8.2.4 "All aerodynamic devices must not extend further rearward than 250mm from the rearmost part of the rear
tires."
III METHODS 17(54)

Fig. 8 Excluded zone around wheels Fig. 9 Maximum dimensions and limits for aero

These rules prevented many different design ideas, too restrictive rules is a common drawback in many motorsport
industries today, but are nevertheless necessary to improve safety and prevent racing fatalities. But the rules regarding
the undertray are still considered flexible enough to allow for innovative designs that had not yet been tested in KTHFS
history.

C. DESIGN ITERATIONS
The CAD programs used to model the undertray were the Siemens owned programs Solid Edge, NX and Teamcenter
for interaction with the other groups (chassis, suspension, wheels, etc.) CAD models.

1. First CAD design (Solid Edge)

The first design was purely based on intuition by analyzing the Aston Martin Valkyrie and the convergent-divergent
venturi tunnel. It was assumed that compressing the air, by making the inlets cross-sectional area taper towards the
waist and expand quite dramatically towards the outlet, would accelerate the air the most and thus the lowest pressure
and highest downforce would be achieved, see Figure 10.
III METHODS 18(54)

Fig. 10 First CAD design

This design was made using the CAD program Solid Edge and used arbitrary dimensions for inlet, waist and outlet, but
within the limits for what the rules allowed. The undertray had an extensive dependency on many other components
such as the positioning of the wheels, the underside shape of the chassis and the positioning of the vehicles’ batteries.
Since this design was created during an early stage many of these components were not finished nor decided. This
meant that no sufficient aerodynamic conclusions could be drawn from running simulations with this specific design
and some approximate assumptions of the chassis design were made and implemented on the CAD model.

In Figure 11 the undertray using a simple block to represent the chassis is shown, this is useful to get a more realistic
drag value and to prevent exaggerated downforce values due to the top of the undertray being exposed to the free-stream
air. However, it is still not a realistic representation of the chassis (unless it is purposely designed as an extruded
rectangle) and due to this designs strict underbody shape it is restricting many different designs for the chassis.
III METHODS 19(54)

Fig. 11 Undertray with a block representing the chassis

2. Second CAD design (NX)

Due to the limitations the first CAD design had with being arbitary designed and not easily integrable with other vehicle
components, a new parameterized CAD, tailored to fit the chassis perfectly was created in NX.

The first CAD design was created in Solid Edge and exported as a parasolid (.x_t) part to later be imported into
Star-CCM+ in order to run the CFD simulations. However, this meant that if any design change was desired, one needed
to manually edit the CAD model in Solid Edge and export and import the parasolid file into Star-CCM+ to run the
simulation again with the new design. This way of editing and improving the undertray was extremely time inefficient
and was unfavourable due to the already time-consuming CFD simulations. It was realised that time could be saved by
having a parameterized CAD model with a variety of different design variables which could easily be edited directly
inside the CFD program, this is done by using Star-NX CAD Client Models integration.

Fig. 12 Second CAD design

In Figure 12 the second design is presented. The undertray shape in this design was made to resemble the first CAD
from Solid Edge as closely as possible but differed in having a parameterized design that was made in conjunction with
the chassis, this meant that any change in the geometry of the chassis would automatically change the geometry of the
III METHODS 20(54)

undertray without causing any intersecting surfaces. The new design also provided a lot of space in the middle of the
chassis, see the bottom right view in Figure 12. This space is important since it is the initial positioning of the vehicle’s
batteries, which are used to power the electric motors. This design prevents the undertray diffuser to expand as much as
it did in the first design, which will likely have a negative effect on downforce. However, due to rules regarding the easy
accessibility and removal of the batteries, meant that this space could not be compromised.

3. Final CAD design (NX)

There were geometrical limitations with the chassis that further prevented the CFD optimization of the undertray and
the size of the batteries needed to be bigger than first expected. So a new chassis design was made with a lower nose and
bigger battery compartment, see Figure 13.

Fig. 13 Final CAD design

This chassis design was created at a much later stage in the development of the vehicle and through CFD analysis of the
old chassis and research from [2], it was assumed that a chassis closer to the ground would utilize ground effect and
therefore improve downforce even more. The new chassis was bigger to help with spacing for the batteries and other
important components fitted both inside and around the chassis. A bigger chassis also meant less space along the width
for the undertray which could harm the undertray’s performance.

4. Parameterization
The two latest CAD models presented had over 100 variable parameters where 11 of these parameters were mainly used
during the optimization study. The biggest changes in performance were noticed when altering the height and width
of the inlet, waist and outlet (called Expansion in the parameters). Some parameters also involved the angles of the
undertray surfaces, see 𝑝272 in Figure 14. However, no major difference in performance was noticed when changing
these angles.
III METHODS 21(54)

Fig. 14 Inlet and waist design parameters

Fig. 15 Outlet (Expansion) design parameters

III METHODS 22(54)

Fig. 16 Lengthwise design parameters

The main parameters that had a large effect on performance are presented in Figure 14, 15 and 16. These parameters
were both incorporated in the second and final CAD designs with the same name expressions presented in Table 2.

Table 2 CAD parameters

Heights Widths Lengthwise
• WC2Inlet
• InletHeight (IH) • InletWidth (IH)
• Inlet2Waist
• WaistHeight (WH) • WaistWidth (WW)
• WaistLength
• ExpansionHeight (EH) • ExpansionWidth (EW)
• ExpansionDistance

Some issues could occur when the parameter values were changed to too extreme values, for instance, CAD features
such as the Studio surface or Trim became defect and parameters could become over-constrained making them fixed
to one value. The only way to solve this was by editing the tolerance settings in the CAD or redoing some features
completely. This was one drawback that limited the parametrization to a certain extent, therefore the CFD optimization
study will only incorporate designs that the CAD models were valid for.

D. METHODOLOGY FOR OPTIMIZATION THROUGH CFD

To optimize and improve a design through CFD, one must get converging results that are independent of other parameters
in the CFD program such as the mesh or domain size. It is known that many parameters in CFD can influence the
simulations results and there is currently no fundamental understanding of how these parameters need to be changed to
prevent inaccurate or varying results [32]. They are only some general rules of thumb on how one can prevent this by
having a certain parameter setup, more on this is covered in section V.B CFD Parameter Setup.

Having a good setup is very important to prevent disproportionate results that could falsely convince the user that the
design is performing better than what it actually is. Analyzing when the setup is sufficient enough can be done by
testing different domain sizes, bigger often tends to be better since it reduces the domains influence on the solution [33],
and performing a mesh convergence analysis. In Figure 17 a block diagram for how the methodology of improving
undertray designs by first achieving a good CFD parameter setup is presented.
III METHODS 23(54)

Fig. 17 Block diagram of CFD methodology

Before any CFD simulation is performed for a new CAD design, a mesh convergence analysis is made by computing
CFD simulations with a varying number of cells in the mesh until the results converge to a steady-state, more about this
is covered in section III.E.3 Mesh independence study. Some CAD models were created with variable design parameters
which could quickly be edited in the CFD program Star-CCM+, this enhanced the design optimization procedure for the
undertray greatly. Any "good" results, such as an increase in downforce or reduction in drag, will be followed by edit in
that specific design parameter, any "bad" result will instead be followed by a revert to the previous design parameter and
an edit of a new design parameter. This methodology is used for all designs and CFD computations in order to get
consistent simulations with comparable results to improve the undertray optimization.

E. CFD SETUP

1. Computational domain and boundary conditions

Knowing which domain size is "correct" in order to have a CFD simulation to depict reality as accurately as possible
can be difficult. Many CFD simulations that involve wind-tunnel validation, use a domain size that is similar to the test
section in the tunnel and a good size in the automotive industry is considered as; 5∼20 car lengths downstream the
vehicle and 2∼3 car lengths upstream and vertical [33][34][35].

The length of the CAD geometry is approx-

imately 2.7 m, to keep a low number of cells
to avoid high computational costs the domain
size of 14 m downstream, 3 m upstream, 3 m
vertically and 2.5 m on each side is chosen. In
Figure 18 the computational domain with the
region boundaries, used for all simulations is
presented.

Fig. 18 Computational domain

III METHODS 24(54)

Table 3 Boundary conditions

Boundary Boundary condition Value
Inlet Constant velocity 22 m/s
Outlet Constant pressure 0 GPa
No slip wall,
Floor 22 m/s
tangential velocity vector
Geometry No slip wall –
Symmetry & Top Symmetry –

In Table 3 all relevant boundary conditions and their values used inside the computational domain are presented. The
air velocity of 22 m/s (ca 80 km/h) was chosen as a starting point for the CFD computation and was used as a standard
value for all upcoming simulations in order to keep the comparison as consistent as possible. Both floor and geometry
used the boundary conditions of ’no-slip wall’ where the floor also needed to have a tangential velocity vector, pointed
in the same direction as the free-stream velocity. This was important to simulate a moving ground under the vehicle,
since it may have a great influence on ground effect and downforce [3]. The symmetry boundary condition mirrors the
features in the domain which will allow the flow to move along the boundary and preventing the flow from crossing the
boundary, making all velocity variables in the normal component set to zero.

2. Mesh setup
The meshing of the domain was set up using Star-CCM+ own mesh generator. With Star-CCM+ the user is capable of
generating fine meshes on advanced geometries while using a graphical user interface. The program also offers an
extensive amount of meshing options, see Appendix A.6. The mesh setup chosen for all simulations in this project is;

Surface Remesher → Trimmed Cell Mesher → Prism Layer Mesher .

| {z } | {z }
Surface mesh Volume mesh

The CAD model and domain first need an initial surface mesh before the volume mesh can be applied, this is done by
using a Surface Remesher. If the geometry is too complex there can be a risk of errors due to self-intersecting surfaces
in the mesh, to prevent this, STAR-CCM+ has a Surface Wrapper feature which ensures a closed and manifold geometry
and wraps the complex CAD to a sufficient quality for a volume mesh to be generated around it. The volumetric mesh
chosen is the Trimmed Cell Mesher which generates hexahedron (topical cubes) cells. This mesher was chosen due to it
being computationally efficient when solving the simulations. The Trimmer also has the ability to refine wake regions
which can be helpful when simulating unsteady and turbulent flows. The Prism Layer Mesher is finally applied to add
prismatic cell layers next to the wall boundaries, as mentioned in chapter II.E.5, this is necessary to capture the sharp
velocity gradients and eddies prevailing from the wall.
III METHODS 25(54)

Fig. 19 Volumetric mesh on undertray and chassis

The first layer thickness of the cells at the wall boundaries are calculated using Equation 29. But first, the skin friction
drag coefficient is calculated from the Prandtl one-seven-power law,

0.027
𝐶𝑓 = , (31)
𝑅𝑒 1/7
𝑥

which is considered a reasonable approximation for low-Reynolds turbulent boundary layers and is therefore relevant
when using the k-𝜔 turbulence model [24]. The Reynolds number is calculated by approximating the undertray and
chassis as a flat plate geometry and simply using; 𝑅𝑒 𝑥 = 𝜌𝑈∞ 𝑥/𝜇, where 𝑥 = 𝐿 = 2.7m is the streamwise length of
the geometry. This will give a skin friction drag coefficient of; 𝐶 𝑓 = 0.00308, this can then be used to calculate the
wall shear stress 𝜏𝑤 and the friction velocity 𝑢 𝜏 from equations (26) and (28). Then finally by using equation (29) the
first layer thickness is calculated; Δ 𝑦 1 = 0.0000177 m = 0.0177 mm and used as input parameter in Star-CCM+. The
prismatic cell layers on the geometry is presented in Figure 21.

Fig. 20 Trimetric view of volumetric mesh Fig. 21 Prism layer mesh

All equations used to calculate the first layer thickness are based on the flat plate boundary layer theory, meaning that
they are used under the assumption that the geometry is modelled as a flat plate. This is obviously not the case for
these undertray and chassis geometries. Due to the geometries complex shapes the 𝑦 + value can tend to differ quite
dramatically from the value it was denoted; 𝑦 + = 1, after a simulation run. Therefore, it is important to monitor the 𝑦 +
values to make sure it stays within the desired range or there can be a risk of simulation inaccuracies in the turbulent
III METHODS 26(54)

boundary layer flow.

Fig. 22 Bottom-view scalar plot of wall 𝑦 +

In Figure 22 a scalar plot of the 𝑦 + values are plotted for one of the simulation runs. It is notable that the 𝑦 + value on the
geometry stays predominantly below 1 within the viscous sub-layer, represented by the blue color. Some perturbations
in the 𝑦 + values are noticed at the edges between the chassis and undertray, but these perturbations are considered small
and negligible.

The chosen setup for the mesh is a mesh strategy based on using volumetric control, refining internal boxes created
around the geometry and in its wake. This is necessary in order to accurately predict the flow around the complex shapes
of the geometry that may cause fluctuations and eddies in the wake.

Fig. 23 Volumetric control on mesh setup

In Figure 23 the percentage values of the base cell size for the volumetric controls are presented. The values used for the
mesh setup are presented in table 4, all percentages are of the base size value.
III METHODS 27(54)

Table 4 Mesh setup

Mesh properties Values
Base size 0.2 m (variable for mesh convergence)
Target surface size 80%
Minimum surface size 15%
Maximum cell size 200%
Volume growth rate Medium
Surface growth rate 1.2
Block 1 10%
Wake refinement 25%
First layer thickness 0.0176mm (from 𝑦 + calculations)
Number of prism layers 15
Prism layer total thickness 15%

3. Mesh independence study

As mentioned previously in section III.D, mesh independence (mesh convergence) is important to ensure the results
consistency during the optimization process, meaning that the downforce values will be independent of the mesh quality
if the mesh is sufficiently fine. Due to a large number of controllable mesh properties, it is not a reasonable approach to
vary each property individually to see their effect on the simulation results, since this approach is very time demanding
and in itself a project of its own. Therefore, the mesh convergence is performed by only changing the base size values
since mostly all other properties are a percentage of the base size these will change within a reasonable magnitude. This
also gives the user considerably more control of the mesh cell count, rather than changing the properties individually.

Fig. 24 Corse mesh with 200K cells Fig. 25 Fine mesh with 2.2M cells

By changing the base size from 0.2 m to 0.5 m the mesh cell count will decrease from 2.2 million cells to 200 thousand
cells, seen in Figure 24 and 25. By iteratively running simulations for each base size value and saving the results, a
mesh convergence graph can be computed.
III METHODS 28(54)

Mesh convergence
600
Main mesh setup
10% prism layer thickness
Full-slip walls
580 Avarage converged value

560
Downforce [N]

540

520

500

480

460
0 0.5 1 1.5 2 2.5 3
Number of cells 10 6

Fig. 26 Mesh convergence graph

The mesh convergence analysis for one of the design iterations of the second CAD design is presented in Figure 26. In
this graph, it is noticed that the simulations show substantially more volatile downforce values for a low number of cells,
especially for full-slip simulations where viscous effects at the wall are neglected, hence no prism layers being used.
Due to convergence issues, some simulations with a certain number of cells could only be computed if the prism layer
mesh total thickness was reduced from 15% to 10%. When the cell count exceeds 2 · 106 the results start to settle at an
average value of 485 N with fluctuation of ±1 N. The convergence does not settle perfectly at a single downforce value
but running simulations with at least 2 · 106 cells is considered sufficient to get results with small fluctuations for the
chassis and undertray.

4. Physics setup
The physical models used to solve the RANS equations are under the assumptions that the flow is incompressible, which
can be applied for a Mach number is below 0.3. The segregated flow algorithm is used since it is considered sufficient
for low to medium speed external aerodynamics simulations, the coupled flow solver is more relevant for transonic flows
since it solves the equations with regard to density change. The k-𝜔 SST turbulence model is used due to its advantages
with modelling separated flows. Each option for the simulations physical properties are presented in Table 5.
III METHODS 29(54)

Table 5 Physics setup

Physical properties Selected options
Space Three Dimensional
Time Steady
Material Gas
Flow Segregated flow
Equations of State Constant Density
Viscous Regime Turbulent
Turbulence Model RANS
Reynolds-Average Turbulence K-Omega Turbulence
K-omega Turbulence Models SST (Menter) K-Omega
Wall All 𝑦 + Wall Treatment
IV RESULTS 30(54)

IV. RESULTS

A. OPTIMISED DOWNFORCE RESULTS

The following results of the three CAD designs will include a table with the downforce and drag values for different
parameter setups. The first row in the tables are the results with the initial parameter setup and for each result
improvement, the change of the specific parameters are stated. All downforce results stated in the tables are for a
velocity of 80 km/h, but in order to have a more general downforce result to compare with other simulations computed
at different velocities, this value can be normalized by dividing it with the density and velocity,

𝐿 2𝐿
𝐶𝐿 𝐴 = = . (32)
𝑞 ∞ 𝜌𝑢 2

This formula is convenient since it excludes the planform area, which is difficult to obtain for a complicated undertray
geometry and will vary depending on the parameter setup.

Observe that the downforce values presented in the results are positive, which means that all values in terms of lift will
be negative, hence a positive downforce value will result in a negative 𝐶 𝐿 𝐴 and lift-to-drag ratio value.

The static pressure in the free-stream is assumed to be constant and equal to zero for each simulation, then the pressure
recovery coefficient, equation 5, can be rewritten as,
𝑝2 − 𝑝1
𝐶𝑝 = , (33)
𝑞∞ − 𝑝1

where 𝑝 1 and 𝑝 2 is the static pressure at what is considered to be the inlet and outlet of the diffuser respectively.

1. Downforce results on first CAD

Table 6 Results of the first CAD design at 80 km/h

Parameters Results
All parameters Downforce: 206 N
were fixed Drag: 93 N

The final results for the first CAD design, after a mesh convergence, are presented in Table 6. The residuals and
convergence of the results is shown in Appendix A.7 and A.8.

The static pressure distribution in the streamwise length of the undertray is shown in Figure 27 and 28. This is for a side
plane cutout at the highest surface of the tunnels.
IV RESULTS 31(54)

Static pressure distribution

300

200

100

Static pressure [Pa]

0

-100

-200

-300

-400

-500
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Undertray length [m]

Fig. 27 Static pressure scalar on side cutout Fig. 28 Static pressure distribution

Fig. 29 Trimetric view of static pressure scalar plot

This undertray design has a downforce value of 𝐶 𝐿 𝐴 = −0.695 m2 , a lift to drag ratio of 𝐿/𝐷 = −2.22 and a pressure
recovery coefficient of 𝐶 𝑝 = 0.465.
IV RESULTS 32(54)

2. Downforce results on second CAD

Table 7 Results of the second CAD design at 80 km/h

Parameters Results
Initial parameter
Downforce: 391 N
setup:
Drag: 73 N
See Appendix B.1
Altered parameters 1: Downforce: 394 N
WaistWidth = 150 mm Drag: - N
Altered parameters 2:
WaistWidth = 120 mm Downforce: 404 N
InletHeight = 85 mm
Altered parameters 3:
WaistWidth = 225 mm
Downforce: 414 N
InletHeight = 85mm
InletWidth = 225 mm
Altered parameters 4:
WaistWidth = 200 mm
WaistHeight = 50 mm
InletHeight = 70 mm Downforce: 444 N
InletWidth = 200 mm
ExpansionHeight = 400 mm
ExpansionWidth = 350 mm
Altered parameters 5:
(Same as Altered parameters 4) Downforce: 458 N
WaistLength = 100 mm
Altered parameters 6:
(Same as Altered parameters 4)
Downforce: 482 N
WaistLength = 10 mm
Inlet2Waist = 300 mm
Optimal parameter
Downforce: 485 N
setup:
Drag: 73 N
See Appendix B.2

The final results for the second CAD design, after a mesh convergence, are presented in Table 7. This undertray design
was optimized with a downforce increase of approximately 94 N from its initial design. Downforce performance
improvements are noticed when the parameters controlling the inlet height and width are reduced to allow a more
smooth transition from the inlet to the waist. Further improvements are noticed when there is a quick transition from the
inlet to the diffusers cross-sectional area, this is done by decreasing the WaistLength and Inlet2Waist parameters. The
last performance contributing parameter is the diffuser height, this height is increased as much as possible using the
ExpansionHeight parameter.

The static pressure distribution in the streamwise length of the undertray is shown in Figure 30 and 31. This is for a side
plane cutout at the highest surface of the tunnels. The residuals and convergence of the results is shown in Appendix
A.9 and A.10.
IV RESULTS 33(54)

Static pressure distribution

200

100

Static pressure [Pa]

-100

-200

-300

-400

-500
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Undertray length [m]

Fig. 30 Static pressure scalar on side cutout Fig. 31 Static pressure distribution

Fig. 32 Trimetric view of static pressure scalar plot

This undertray design has a downforce value of 𝐶 𝐿 𝐴 = −1.64 m2 , a lift to drag ratio of 𝐿/𝐷 = −6.64 and a pressure
recovery coefficient of 𝐶 𝑝 = 0.530.
IV RESULTS 34(54)

3. Downforce results on final CAD

Table 8 Results of the final CAD design at 80 km/h

Parameters Results
Initial parameter setup: Downforce: 501 N
See Appendix B.3 Drag: 90 N
Altered parameters 1: Downforce: 504 N
InletHeight = 60 mm Drag: 88 N
Altered parameters 2:
InletHeight = 50 mm Downforce: 511 N
WaistHeight = 40 mm Drag: 88 N
Inlet2Waist = 250 mm
Altered parameters 3:
InletHeight = 80 mm Downforce: 514 N
WaistHeight = 70 mm Drag: 90 N
Inlet2Waist = 200 mm
Altered parameters 4:
InletHeight = 70 mm Downforce: 519 N
WaistHeight = 50 mm Drag: 90 N
Inlet2Waist = 250 mm
Altered parameters 5:
InletHeight = 70 mm
Downforce: 530 N
WaistHeight = 70 mm
Drag: 90 N
Inlet2Waist = 200 mm
(Or see Appendix B.4)

The final results for the second CAD design, after a mesh convergence, are presented in Table 8. This undertray design
was optimized with a downforce increase of approximately 29 N from its initial design. The downforce performance was
increased by lowering the inlet height equal to the waist height, making the waist completely unessential to the design.
This prevented a positive static pressure at the inlet, which was noticed in the previous design, and the pressure stayed
negative all along the undertray bottom surface. Big downforce improvements are also noticed when the divergence of
the tunnels are straightened (quicker transition from inlet to outlet) which will give the tunnels a higher profile, this is
controlled by shortening the WaistLength and Inlet2Waist parameters while increasing the WaistHeight.

Static pressure distribution

0

-100

-200
Static pressure [Pa]

-300

-400

-500

-600
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Undertray length [m]

Fig. 33 Static pressure scalar on side cutout Fig. 34 Static pressure distribution
IV RESULTS 35(54)

Fig. 35 Trimetric view of static pressure scalar plot

The final CAD design has a downforce value of 𝐶 𝐿 𝐴 = −1.79 m2 , lift to drag ratio of 𝐿/𝐷 = −5.89 and a pressure
recovery coefficient of 𝐶 𝑝 = 0.550.

The residuals and convergence of the results is shown in Appendix A.11 and A.12.

4. Result comparison
Comparing the results of the three different designs it is observed that the highest downforce achieved is from the final
CAD design. This may be due to the chassis contribution to the downforce, which was both larger and used a nosecone
closer to the ground which could utilize ground effect. However, the larger chassis also had the penalty of increasing the
drag compared to the second design. Therefore, the second design had the best lift-to-drag ratio due to the chassis
having the least drag contribution. The first design had the worst performance in both downforce and drag, this was
partially due to the fact that no optimization process could be performed since all parameters were fixed, but mainly
because of the rectangular chassis and that the inlet produced a local area with lift.

Table 9 Optimal results summary

Best results 𝐶 𝐿 𝐴 - value Lift-to-drag ratio 𝐶 𝑝 - value
Downforce: 206 N
First CAD design −0.695 m2 −2.22 0.465
Drag: 93 N
Downforce: 485 N
Second CAD design −1.64 m2 −6.64 0.530
Drag: 73 N
Downforce: 530 N
Final CAD design −1.79 m2 −5.89 0.550
Drag: 90 N
IV RESULTS 36(54)

Static pressure distribution

300
Final design
Second design
200 First design

100

0
Static pressure [Pa]

-100

-200

-300

-400

-500

-600
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Undertray length [m]

Fig. 36 Comparison of the pressure distribution plots

Comparison of each designs static pressure distributions are presented in Figure 36. Here it is observable that the inlet
design of the first CAD produced positive static pressure and thus creating a region with lift. Both the second and
final design performed better here with having negative static pressures throughout the tunnels bottom surface. The
second design had the lowest pressure of −495 Pa at a distance of 0.6 m, just aft of the waist, while the final design had
approximately −430 Pa at this point. However, the inlet pressure of the final design was far lower than the other two, at
562 Pa, since the waist and inlet height was change to be equal. This resulted in the final design having the lowest
pressure recovery coefficient value since the expansion started from the inlet hence the exclusion of the waist.

B. COMPARISON WITH THE CONVENTIONAL DESIGN

To justify the use of a larger undertray which could potentially disrupt space for other components on the vehicle, a
comparison with a more conventional flat plate design is relevant. The flat plate undertray was generated directly in
Star-CCM+ using its built-in 3D-CAD modeler, where it was extruded and attached to the second design’s chassis, see
Figure 37. The results from the conventional flat plate design are compared with the venturi tunnel undertray from the
second CAD design.
IV RESULTS 37(54)

Fig. 37 Flat plate CAD design

This undertray was designed to be in close proximity to the ground and use the advantage of the ground effect. The
front of the undertray also includes a splitter, due to the chassis being tapered towards the nose, this generates additional
downforce since a high-pressure region will emerge over the splitter. The diffuser of the undertray starts at approximately
1 m aft of its leading-edge with the diffuser height designed to be as high as possible without disrupting the space for the
rear wheels lower a-arms.

A static pressure distribution comparison between the tunnel and flat plate design is presented in Figure 38. The
low-pressure region is more concentrated with a minimum value just at the beginning of the diffuser for the flat plate
design, while the tunnels low-pressure distribution is more spread out along the length of the undertray, as shown in
Figure 39. Both undertray’s produce a low static pressure due to the airs pressure recovery through the diffuser, however,
since the transition between the horizontal surface and the diffuser in the tunnel design is more gradual, the pressure
distribution will be considerably smoother compared to the flat plate design. The tunnels static pressure, across a
large portion of the undertray’s length, is considerably lower than the flat plate with a minimum pressure of −495 Pa
compared to the flat plate’s −363 Pa.
IV RESULTS 38(54)

Static pressure distribution

200
Tunnel design
Flat plate design

100

0
Static pressure [Pa]

-100

-200

-300

-400

-500
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Undertray length [m]

Fig. 38 Pressure comparison of flat plate and tunnel design

Fig. 39 Side pressure scalar plot comparison

Due to the lower and more spread out distribution of the static pressure along the undertray length, the tunnels produce
approximately 29% more downforce than the flat plate design, but with a slightly higher drag penalty, caused by its
pressure drag. However, the tunnels can still be considered more aerodynamically efficient with their superior lift-to-drag
ratio and pressure recovery coefficient.

Table 10 Tunnel and flat plate results comparison

Results 𝐶 𝐿 𝐴 - value Lift-to-drag ratio 𝐶 𝑝 - value
Downforce: 486 N
Tunnel design −1.64 m2 −6.64 0.530
Drag: 73 N
Downforce: 378 N
Flat plate design −1.28 m2 −5.91 0.377
Drag: 64 N
IV RESULTS 39(54)

C. RIDE HEIGHT SENSITIVITY

In [4] Toet explains how the diffuser angle has a large effect on the undertray’s performance due to ride height sensitivity.
A high angled diffuser will not lose as much downforce when the ride height is increased as a lower angled diffuser. The
performance of a high angled diffuser is improved for ride heights going from 5 to 30 mm and a slight decrease in
performance from 30 to 60 mm, this is observable from the results of [4]. The lowest angled diffuser will on the other
hand have a dramatic decrease in performance from 10 to 65 mm. A similar experiment is conducted for the tunnel and
flat plate design, but for the minimum ride height of 30 mm in accordance with T 2.3.2 FS rule, see Figure 40.

Ride height sensitivity

500
Tunnel design
Flat plate design
480

460

440

420
Downforce [N]

400

380

360

340

320

300
20 40 60 80 100 120 140 160
Ground clearence [mm]

Fig. 40 Effect on downforce when increasing ride height

No dramatic difference in the performance loss for increased ride height is noticed when comparing the tunnel and flat
plate designs and the downforce stays at approximately 80-120 N higher with the tunnels. The reason why a similar
result as Toet was not obtained could be due to the large differences in the geometry or since higher ride heights were
analyzed.

D. VORTEX EFFECT ON DOWNFORCE

The formation of longitudinal vortices in the diffuser, that originate from the lengthwise sides of the undertray, are
also influential of the undertray flow physics and its aerodynamic performance. Vortices can both have favourable or
adverse effects on the flow. Streamwise vortices are favourable to the undertray’s performance in the sense that they
are associated with low pressure at the vortex core due to its rotational velocity, this low pressure at the vortex core is
desirable as it helps lower the overall underbody pressure of the vehicle. The adverse effect is the vortex breakdown
phenomenon, which causes an abrupt change in the vortex structure with the occurrence of flow retardation along the
vortex axis. For the undertray, a vortex breakdown will cause a loss in downforce [5].

Figure 41 and 42 show the line integrated convolution in a cross-section plane at the diffuser, these help display the
IV RESULTS 40(54)

airflow’s direction and velocity. It can be seen that the vortices in the tunnels diffuser are much more prominent with
higher velocity and kinetic energy compared to the flat plate diffuser. The more round cross-sectional diffuser of the
tunnels can be considered aiding the longitudinal vortices more than the squared design of the flat plate diffuser. For a
streamlined illustration of airflow through the tunnel undertray see Appendix A.13.

Fig. 41 Line integral convolution of tunnel diffuser

Fig. 42 Line integral convolution of flat plate diffuser

V DISCUSSION 41(54)

V. DISCUSSION
As with all computer simulations trying to depict real physical phenomenons such as vehicle aerodynamics, the results
accuracy can always be debated. Typical CFD projects often include wind tunnel tests to validate the results accuracy
and to obtain best practice guidelines for the CFD setup. As mentioned earlier the team did not have access to a
wind tunnel so no validation of that sort could be performed. However, the access of a general wind tunnel does not
necessarily mean more accurate simulations since there still could be inaccuracies between wind tunnel tests and track
testing. Some wind tunnels fall short of meeting specific flow quality requirements and wind tunnels that are currently
in use today have not been adequately or consistently documented, meaning that the flow quality assessment on model
test results have largely been ignored [36]. Due to this, wind tunnel tests may not be 100% accurate and therefore using
them specifically as a tool for validating CFD results can be debated.

Apart from the risk of having inaccuracies in the wind tunnel, there is also a great cost of time and resources to run the
experiments. Depending on the size of the wind tunnel test section, the vehicle geometry might need to be scaled down
which would require the extra manufacturing of new aerodynamic components apart from the ones that will be fitted to
the full-scale vehicle. Components near the ground could have their aerodynamic properties altered due to ground effect
and under real road conditions in still air, the air’s relative speed to the ground is zero, meaning there is no boundary
layer developed on the road. Most wind tunnels have stationary floors with a developed boundary layer, this means that
a more advanced automotive wind tunnel with moving belts or some other correction to the floors boundary layer might
be necessary for accurate experiments. No such wind tunnel is known to exist in the vehicle engineering department at
KTH.

The demand for wind tunnel usage may be reduced over the years due to the big advances in computers and CFD.
Nevertheless, wind tunnels still have an advantage when the computers computational power falls short, as in running
complex multi-element full car simulations.

A. LIMITATIONS AND ISSUES WITH OTHER COMPONENTS

The problem with a large diffuser is that it may disrupt
space for important components at the rear of the chassis.
One of these components is the rear wheels a-arms (control
arms), which are used as the main link to hold the wheels
attached to the chassis. It was concluded by the team
that the positioning of the a-arms could not be altered too
much and that they, therefore, needed to protrude through
the undertray, as shown in Figure 43, the a-arms are
four rods connecting the wheel with the chassis, which is
represented in brown. The a-arms are in a fixed position on
both the wheels and the chassis but are capable of moving
vertically due to the wheels vertical movement via the
suspension on bumpy roads. The a-arms could become
problematic with an adverse effect on the undertray’s Fig. 43 A-arms
aerodynamic properties and issues with its mounting to the chassis. One of the problems with the a-arms protrusion
through the chassis is the risk of air leakage that prevents a good seal from the high-pressure region outside the tunnels
to the low pressure inside. Another problem is how the hole in the undertray should be formed since a simple circular
hole will create issues with the mountings of the components due to the ends of the a-arms being welded together in a
v-shape. One solution, that is considered to be implemented, which will allow the components to be mounted easier is
to drill two long vertical holes from the bottom of the undertray. This makes it possible for the lower a-arms to be
inserted through the undertray from the bottom and will allow them free vertical movement. However, the larger vertical
holes need to be thoroughly sealed to prevent any leakage.
V DISCUSSION 42(54)

The parametrization of the undertray was limited due to some variable parameters becoming over constrained when
exceeding a specific value. This was the case for the waist and diffuser parameters, where the diffuser could not exceed
the value of 405 mm and the waist a value of 70 mm. This prevented a thorough analysis and optimization of the
undertray but the undertray was optimised within the values the CAD geometry allowed.

Due to the limited computational resources, rotating wheels and all other aerodynamic devices apart from the chassis and
undertray were excluded from the simulations. A few more computationally demanding simulations were co-produced
by a colleague in the team using a better performing desktop computer, these simulations give the possibility to observe
the effect of one rotating wheel and extra bargeboard devices. Rotating wheels have a huge impact on the vehicle’s
aerodynamics, especially for opened wheel race cars. Apart from causing a mess of chaotic air at its wake, rotating
wheels will also generating lift and a high amount of drag [2]. The top of a rotating wheel will move in the opposite
direction of the free stream air and pull the air against the stream, this will cause the flow to separate much earlier
and give the presence of strong trailing vortices at the wake of the wheel. The vortices in the wheel wake will have a
tendency to be sucked in beneath the undertray which will affect its performance dramatically by introducing undesirable
slow-moving air into the tunnels, see Appendix A.14. The team was able to solve this issue by using specially optimised
bargeboards that helped seal the side of the undertray, see Appendix A.15. Due to the wheels large impact on the
vehicles aerodynamic performance it can be argued that by excluding them, the simulations results will not give a
good depiction of reality when the vehicle is driven on the track. The same argument can be used for excluding other
aerodynamic components that have a risk of influencing or interacting with each other, such as how the front wing will
affect the incoming flow to the undertray. Therefore, full car simulations can always be considered necessary to validate
how good each component is when they all interact with each other, once attached to the vehicle.

B. CFD PARAMETER SETUP

Choosing a correct CFD setup can be essential for the simulations to return reasonably accurate results. However, what
is considered to be a "correct" CFD setup is questionable since many parameters, such as the mesh and domain size, are
quite arbitrarily chosen.

The general rule of thumb for mesh generation is to produce a mesh with local refinement near the geometry surface and
at its wake while having a coarser mesh at the far-field fluid domain. This is mainly done to prevent the computational
costs and time from exceeding the user’s limited resources. If no such limitations existed one could simply use an
extremely fine mesh throughout the entire domain to capture all small eddies and turbulent air, the simulations would
typically return more accurate results as well. But since no machine has unlimited computational power one needs to
carefully choose which areas of the domain that need a refined mesh and where the mesh can stay coarse. This choice is
done arbitrarily by the user and can therefore lead to inaccuracies in the simulations.

Another problem is also how fast the volumetric cells will grow from a fine to coarse mesh, which is called growth rate.
Some CFD engineers state that the maximum growth rate should not exceed 20% (meaning a maximum growth rate of
1.2) [37]. However, for hexahedral cells the volumetric mesh can only grow in factors of 2, 3, 4 etc. and Star-CCM+
does not allow the user to choose quantifiable parameters of the growth rate, the only options are fast, medium or slow,
but the surface growth rate parameter can, on the other hand, be set to 1.2. A too-aggressive change in cell size could
cause poorly resolved pressure and velocity gradients in certain areas around the geometry. Due to this, the user needs
to assume a good value for the number of prism layers and set the growth rate to either fast, medium or slow to get what
seems to be a smooth transition from fine to coarse mesh, with criteria mainly specified by the users own personal
preferences.

Literature covering CFD parameter setup is scattered and quite project-specific. Due to the occurrence of largely
differing or complex geometry designs, it may not be viable to follow the same parameter strategy for all simulations
since the strategy is solely based on the user experience from CFD and wind tunnel validation experiments. The DFS
methodology aims to improve this mesh setup approach by introducing adaptive mesh refinement.
VI CONCLUSIONS 43(54)

VI. CONCLUSIONS
It is concluded that the venturi tunnel undertray design, with mainly a divergent tunnel, is capable of significant
improvements in downforce performance without any dramatic drag penalty. The best performing undertray is the final
CAD design which produced 530 N of downforce with 90 N in drag, it can not be excluded that part of the downforce
improvements was caused by the larger chassis. The venturi tunnel design proved to give a 29% downforce improvement
over a flat plate conventional undertray design with the same chassis. No dramatic differences in downforce results for
different ride heights were noticed when comparing the venturi tunnel with the flat plate design.

It is observed that the main aspects that affect undertray downforce performance are the diffuser outlet height, upsweep
and vehicle ground clearance. Longitudinal trailing vortices originating from lengthwise sides of the undertray are also
acknowledged to be vital in maintaining low-pressure air at the diffuser and its exit. All results obtained for this specific
undertray and chassis geometry designs supported the theory presented in [2], that an inlet height tapering towards the
waist will cause a local area of high pressure and lift, thus reducing the overall downforce performance. It is observed
that the minimum pressure is located just at the beginning of the diffuser inlet for a conventionally designed undertray.
For the venturi tunnel undertray there is no clear lengthwise position of where the diffuser starts, this minimum pressure
will therefore be located at the lowest point of the undertray, so either at the waist or inlet depending on which is closest
to the ground. The tunnels can diverge much slower than the flat plate design allowing a more spread-out minimum
pressure region instead of at a point, as observed from the pressure distribution plots, see Figure 38. The larger diffuser
also gives the air a greater capacity for pressure recovery making it possible for achieving higher downforce values.

The venturi tunnel design can be considered only consisting of a large diffuser that starts at the inlet with a constantly
diverging cross-sectional area to its outlet. Apart from increasing the diffuser height downforce improvements were
also noticed when making a quicker, straighter expanding transition from inlet to outlet. Due to geometric issues with
errors in the CAD model, a straighter expanding undertray design than the ones presented could not be aerodynamically
analysed.

It is concluded that the CFD turbulence model k-𝜔 (SST) is sufficient for undertray flow simulations [3][5]. The
model is advantageous in giving decent accuracy for flows around sharp curvatures and predicting separation. The
RANS method is considered adequate with allowing well-converging simulations for limited computational resources.
Although the DFS methodology is a new innovative approach in CFD, no comparable results were obtained due to
inefficient supervising and limitations in the setup environment. The CFD methodology chosen for this project is also
motivated in being a well documented and standard approach for analysing external vehicle aerodynamics.

A. FUTURE WORK
Apart from being a vital component for aerodynamic performance in racing vehicles, undertray design is also useful in
the automotive industry in reducing drag to minimize fuel consumption. Drag reduction can be obtained by smoothing
the vehicle’s underbody to allow a clean uninterrupted airflow, but it is also covered in numerous scientifical papers that
the diffuser ramp angle can be used to reduce drag if designed correctly. In [2] the ramp angle can be seen to be capable
of creating either an adverse or favourable effect on the drag coefficient. One explanation for this fluctuation in drag
coefficient is that pressure drag will decrease at certain ramp angles due to the diffuser allowing a smooth transition for
low-pressure air going to higher pressure thus minimizing the separation region at the rear of the vehicle. However, this
is not always the case for some ramp angles, and the explanation for an increased drag is related to the longitudinal
vortices developing from the diffuser exit. These trailing vortices have the tendency of keeping the flow attached to
surfaces nearby, this will result in the formation of standing vortices at the rear surface of the vehicle and thus decrease
the pressure at this region, even more, this is followed by an increased drag [2].

The variation of drag with diffuser length and angle is considered complicated, and it needs to be acknowledged that the
upper and lower surfaces of the vehicle can not be optimised independently for drag reduction. That the rear rather
needs to be treated as a single three-dimensional shape in order to minimize drag most effectively.
REFERENCES 44(54)

REFERENCES
[1] How much does it cost to build an F1 car? Accessed: 2020-05-05. url: https://lordofthef1.wordpress.com/2012/
11/25/how-much-does-it-cost-to-build-an-f1-car/.
[2] RH Barnard. Road Vehicle Aerodynamic Design. 2nd ed. Mechaero Publishing, 2001.
[3] Xin Zhang, Willem Toet, and Jonathan Zerihan. “Ground Effect Aerodynamics for Race cars”. In: (2006). doi: 10.1115/1.
2110263.
[4] Willem Toet explains. . . .motorsport diffusers. Accessed: 2020-10-28. url: https://www.racetechmag.com/2017/08/
willem-toet-explains-motorsport-diffusers/.
[5] Kevin Knowles and Alistair J. Saddington. “A Review of Ground Effect Diffuser Aerodynamics”. In: (2018). Journal of Fluids
Engineering.
[6] Lotus Modell 79 wing-profile. by MagentaGreen can be reused under the CC BY-SA 3.0 license. url: https://commons.
wikimedia.org/wiki/File:Lotus_Modell_79_wing-profile.svg.
[7] 2021 car revealed as FIA and F1 present regulations for the future. Accessed: 2020-09-05. url: https://www.formula1.
com/en/latest/article.2021- car- revealed- as- fia- and- f1- present- regulations- for- the- future.
3ZkeTsu1sNCrFGP1HI9KVy.html.
[8] Behind 2009’s double-diffuser dominator. Accessed: 2020-07-25. url: https : / / www . youtube . com / watch ? v =
cGrFN2jZeY8.
[9] Double diffusers to be banned for 2011. Accessed: 2020-07-25. url: https://web.archive.org/web/20100127185330/
http://www.itv-f1.com/news_article.aspx?id=47659.
[10] 2001 Goodwood Festival of Speed Brabham BT46B Fan car. by edvvc can be reused under the CC BY-SA 2.0 license. url: https:
//commons.wikimedia.org/wiki/File:2001_Goodwood_Festival_of_Speed_Brabham_BT46B_Fan_car.jpg.
[11] Allan Henry. Brabham: The Grand Prix Cars. Hazleton Publishing, 1985.
[12] Kevin R. Cooper et al. “The Aerodynamic Performance of Automotive Underbody Diffusers”. In: (1998). SAE Technical
Report Paper.
[13] David Eller. On Airfoils and Boundary Layers. KTH Aeronautical and Vehicle Engineering, 2010.
[14] Martin Kronbichler. Computational Fluid Dynamics. TUM Department of Mechanical Engineering, 2010.
[15] J.B Flor I. Eamse. “New Development in Undertanding Interfacial Processes in Turbulent Flows”. In: (2011). doi: https:
//doi.org/10.1098/rsta.2010.0332.
[16] D. Brian Spalding. 50 Years of CFD in Engineering Science: a commemorative volume in memory of D. Brian Spalding.
Springer Nature Singapore, 2020.
[17] Lecture- Turbulence Models by PhD. Andre Bakker. Accessed: 2020-11-13. url: https://www.bakker.org/dartmouth06/
engs150/10-rans.pdf.
[18] Turbulence: Which Model Should I Select for My CFD Analysis? Accessed: 2020-11-12. url: https://www.simscale.
com/blog/2017/12/turbulence-cfd-analysis/.
[19] k-omega turbulence model. Accessed: 2020-11-18. url: https://en.wikipedia.org/wiki/K-omega_turbulence_
model.
[20] HK Versteeg and W Malalasekera. An Introduction to Computational Fluid Dynamics. 2nd ed. Pearson Education Limited,
2007.
[21] Which Turbulence Model Should I Choose for My CFD Application? Accessed: 2021-02-02. url: https://www.comsol.
com/blogs/which-turbulence-model-should-choose-cfd-application/.
[22] What is y+ (yplus)? Accessed: 2021-02-02. url: https://www.simscale.com/forum/t/what-is-y-yplus/82394.
[23] Fangqing Liu. “A Thorough Description Of How Wall Functions Are Implemented In OpenFOAM”. In: (2017). Student
project work at Chalmers University.
[24] Frank M White. Fluid Mechanics. 7th ed. McGraw-Hill, 2011.
[25] Course material Advanced Computation in Fluid Mechanics DD2365 KTH. Accessed: 2021-06-07.
REFERENCES 45(54)

[26] Johan Hoffman and Claes Johnson. Computational Turbulent Incompressible Flow. Springer, 2007.
[27] Johan Hoffman et al. “Towards a parameter-free method for high Reynolds number turbulent flow simulation based on adaptive
finite element approximation”. In: (2014). doi: http://dx.doi.org/10.1016/j.cma.2014.12.004.
[28] Niclas Jansson, Johan Hoffman, and Murtazo Nazarov. “Adaptive simulation of turbulent flow past a full car model”. In: (2014).
doi: 10.1145/2063348.2063375. url: https://www.researchgate.net/publication/229034225_Adaptive_
simulation_of_turbulent_flow_past_a_full_car_model.
[29] John D. Anderson Jr. “Ludwig Prandtl’s Boundary Layer”. In: (2005).
[30] Johan Hoffman and Claes Johnsson. “Resolution of d’Alembert’s Paradox”. In: (2010). doi: 10.1007/s00021-008-0290-1.
[31] 2019 Aston Martin Valkyrie AMR Pro 6.5 Rear. by Vauxford can be reused under the CC BY-SA 4.0 license. url:
https://commons.wikimedia.org/wiki/File:2019_Aston_Martin_Valkyrie_AMR_Pro_6.5_Rear.jpg.
[32] J. Franke, C. Hirsch, and A.G Jensen. “Recomendations On The Use of CFD In Wind Engineering”. In: (2004).
[33] Marco Lanfrit. “Best practice guidelines for handling Automotive External Aerodynamics with FLUENT”. In: (2005).
[34] N.E Ahmad, E. Abo-Serie, and A. Gaylard. “Mesh Optimization for Ground Vehicle Aerodynamics”. In: (2010). CFD Letters.
[35] Computational Domain Selection for CFD Simulation. Accessed: 2020-04-19. url: http://www.cfdyna.com/CFDHT/
GeoCleaning.pdf.
[36] F. Kevin Owen and Andrew K. Owen. “Measurement and assessment of wind tunnel flow quality”. In: (2008). doi:
https://doi.org/10.1016/j.paerosci.2008.04.002.
[37] How can I understand mesh quality and simulation results in CFD? Accessed: 2021-04-16. url: https : / / www .
researchgate.net/post/How_can_I_understand_mesh_quality_and_simulation_results_in_CFD.
[38] Law of the wall (English). by aokomoriuta can be reused under the CC BY-NC 3.0 license. url: https://commons.
wikimedia.org/wiki/File:Law_of_the_wall_(English).svg.
[39] Mesh models in STAR-CCM+. Accessed: 2021-01-29. url: https://theansweris27.com/mesh-models-in-star-
ccm/.
A APPENDIX 46(54)

A. APPENDIX

Fig. A.1 Aerodynamic components downforce contributions on 2009 F1 car [5]

Fig. A.2 Boundary velocity profile

A APPENDIX 47(54)

Fig. A.3 Law of the wall [38] nLTM

Fig. A.4 The Ahmed body

A APPENDIX 48(54)

Fig. A.5 Inverted Ahmed body and centerline pressure distribution along underbody and diffuser [5]
A APPENDIX 49(54)

Fig. A.6 Interdependencies between mesh models in STAR-CCM+ [39]

Fig. A.7 Residuals for the first CAD design

A APPENDIX 50(54)

Fig. A.8 Downforce convergence results for the first CAD design

Fig. A.9 Residuals for the second CAD design

A APPENDIX 51(54)

Fig. A.10 Downforce convergence results for the second CAD design

Fig. A.11 Residuals for the final CAD design

A APPENDIX 52(54)

Fig. A.12 Downforce convergence results for the final CAD design

Fig. A.13 Streamlines illustrating airflow through tunnels

A APPENDIX 53(54)

Fig. A.14 Wake of rear wheel vortex entering undertray tunnel

Fig. A.15 Improved performance by sealing undertray bottom

B APPENDIX 54(54)

B. APPENDIX

Table B.1 Initial parameter setup for second CAD design

Heights Widths Lengthwise
• WC2Inlet = 220 mm
• InletHeight = 90 mm • InletWidth = 150 mm
• Inlet2Waist = 350 mm
• WaistHeight = 55 mm • WaistWidth = 120 mm
• WaistLength = 300 mm
• ExpansionHeight = 350 mm • ExpansionWidth = 350 mm
• ExpansionDistance = Variable

Table B.2 Optimized parameter setup for second CAD design

Heights Widths Lengthwise
• WC2Inlet = 220 mm
• InletHeight = 70 mm • InletWidth = 200 mm
• Inlet2Waist = 250 mm
• WaistHeight = 50 mm • WaistWidth = 200 mm
• WaistLength = 10 mm
• ExpansionHeight = 400 mm • ExpansionWidth = 350 mm
• ExpansionDistance = Variable

Table B.3 Initial parameter setup for final CAD design

Heights Widths Lengthwise
• WC2Inlet = 220 mm
• InletHeight = 70 mm • InletWidth = 200 mm
• Inlet2Waist = 250 mm
• WaistHeight = 50 mm • WaistWidth = 200 mm
• WaistLength = 10 mm
• ExpansionHeight = 400 mm • ExpansionWidth = 350 mm
• ExpansionDistance = Variable

Table B.4 Optimized parameter setup for final CAD design

Heights Widths Lengthwise
• WC2Inlet = 220 mm
• InletHeight = 70 mm • InletWidth = 200 mm
• Inlet2Waist = 200 mm
• WaistHeight = 70 mm • WaistWidth = 200 mm
• WaistLength = 10 mm
• ExpansionHeight = 405 mm • ExpansionWidth = 350 mm
• ExpansionDistance = Variable
TRITA-SCI-GRU 2021:290

www.kth.se