Semester Project - Mechanical Engineering

Fluid Structure Interaction of a supersonic

rocket

IN COLLABORATION WITH THE EPFL ROCKET TEAM

Supervisor: Flavio Noca
ERT Supervisor: Kevin Marangi

Authors:
Adriano Cassella
Stefano Tamo

June 10, 2022

Contents
1 Definitions and abbreviations 1

2 Introduction 1
2.1 Context and goals of the study . . . . . . . . . . . . . . . . . . . . . . . . 1
2.2 Type of analysis, Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.3 Procedure of the Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

3 General Knowledge 2

I Nosecone 2D 6

4 Geometric modeling and hypotheses 6

4.1 Presentation of the geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.2 2D-computational domain . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.3 Characteristic dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.4 Symmetry / periodicity of the problem . . . . . . . . . . . . . . . . . . . . 8
4.5 Geometrical space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

5 Boundary conditions, external forces, initial conditions 8

5.1 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5.2 External forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

6 Computational mesh 9
6.1 Mesh type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
6.2 Cell type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
6.3 Size / number of cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

7 Numerical methods 13
7.1 Spatial discretization method . . . . . . . . . . . . . . . . . . . . . . . . . 13
7.2 Type of simulation/solver . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
7.3 Solution options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
7.4 Computed quantities? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

8 Results 14

9 Analysis & conclusions 17

II Nosecone 3D 18
10 Goal 18

11 Geometric modeling and hypotheses 18

11.1 Presentation of the geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 18
11.2 3D-computational domain . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
11.3 Characteristic dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
11.4 Symmetry / periodicity of the problem . . . . . . . . . . . . . . . . . . . . 20

12 Boundary conditions, external forces, initial conditions 20

12.1 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
12.2 External forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

13 Computational mesh 22
13.1 Mesh type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
13.2 Cell type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
13.3 Size / number of cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
13.4 Mesh type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
13.5 Cell type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
13.6 Size / number of cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

14 Numerical methods 29
14.1 Spatial discretization method . . . . . . . . . . . . . . . . . . . . . . . . . 29
14.2 Type of simulation/solver . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
14.3 Solution options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
14.4 Computed quantities? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

15 Results 31

16 Analysis & conclusions 36

III Rocket simulations 38

17 Geometric modeling and hypotheses 38

17.1 Presentation of the geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 38
17.2 Symmetry / periodicity of the problem . . . . . . . . . . . . . . . . . . . . 41
17.3 Geometrical space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

18 Boundary conditions, external forces, initial conditions 41

18.1 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
18.2 External forces CFD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
19 Computational mesh 46
19.1 Mesh type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
19.2 Cell type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

20 Numerical methods 47
20.1 Spatial discretization method . . . . . . . . . . . . . . . . . . . . . . . . . 47
20.2 Type of simulation/solver . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
20.3 Solution options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
20.4 Computed quantities? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

21 Mesh size / domain size convergence study CFD 48

21.1 Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
21.2 Presentation of the different CFD meshes . . . . . . . . . . . . . . . . . . . 48
21.3 CFD results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
21.4 Choice of the CFD mesh / domain . . . . . . . . . . . . . . . . . . . . . . 57

22 Mesh size / domain size convergence study FEA 57

22.1 Criterion FEA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
22.2 Presentation of the different FEA meshes . . . . . . . . . . . . . . . . . . . 58
22.3 Equivalent stress distribution results . . . . . . . . . . . . . . . . . . . . . 61

23 FEA : general results 62

24 Analysis & conclusions 62

IV Synthesis 64

25 Main conclusion 64

26 Futur improvements 64

Appendices 65

A Convergence Study CFD on the Computational Domain size 65

B CFD Tetrahedral mesh elements best residuals 70

C FEA shell study beginning 76

D DLL access procedure 86

References 88
1 Definitions and abbreviations

Table 1: Definitions and abbreviations

Abbreviations Signification
EPFL Ecole Polytechnique Fédérale de Lausanne
ERT EPFL RocketTeam
CFD Computational Fluid Dynamics
FEA Finite Elements Analysis
FSI Fluid Structure Interaction
PTFE Polytetrafluoroethylene
ACP ANSYS Composites PrepPost
3D 3 dimensions
SST Shear Stress Transport
2D 2 dimensions
AoA Angle of Attack
FVM Finite Volume Method
NACA National Advisory Committee for Aeronautics
BOI Body Of Influence
COG Center Of Gravity
DLL Discovery Learning Lab
CPU Central Processing Unit

2 Introduction
2.1 Context and goals of the study
When designing a rocket, two of the most critical values that must be computed are
the pressure exerted on the airframe of the rocket, and the resulting deformations.As a
semester project at the EPFL Rocket Team (ERT), our goal is to study the impact of the
aerodynamic loading onto the rocket’s airfarame coupling Computational Fluid Dynamics
(CFD) and Finite Elements Analysis (FEA) simulations. The combination of these two
is usually called Fluid-Structure Interaction (FSI).
There are several different ways of executing an FSI. Firsly, the most famous one that is
often called Two-way coupling, is the one where the fluid flow is affected by structural
deformation and at the same time structural deformation is affected by fluid flow. This
type of coupling provides generally a good accuracy. However, regarding the computa-
tional time needed, the Two-way coupling is very expensive. Therefore, for this project,
we decided to use the second coupling procedure, usually called the One-way coupling.
The general procedure of a One-way coupling consists of running first a CFD simulation
(or a structural FEA simulation) and then a structural FEA simulation (or a CFD simu-
lation respectively) using the results computed during the first simulation as a boundary
condition for the second one.

1
In order to continue the work that has previously been done on the CFD studies in general
at the Rocket Team, we found great help in the projects performed by other members of
the team, namely [1] and [2]. We will often use their work as reference in our projects,
for instance with some parameters used on the used software, or with some results.

2.2 Type of analysis, Methodology

We are going to perform two types of simulations, as discussed previously:

• First we are going to perform a CFD analysis, in order to obtain the pressure and
temperature fields exerted on the rocket airframe;

• Then, we are going to export the pressure and the temperature as boundary condi-
tions in our FEA analysis, in order to compute the deformation into the rocket.

To perform this study, we will use the ANSYS software, for the CFD and the FEA parts
of our project:

• ANSYS Fluent for the CFD part of the FSI study

• ANSYS Mechanical for the FEA part

A very practical feature of ANSYS is that you can export the pressure and temperature
obtained through the CFD directly as loads in the FEA software, which will be of great
help for us.

2.3 Procedure of the Project

In order to become familiar with ANSYS software’s various functions and parameters, we
are going first to perform some simulations only on the nosecone of the rocket. This would
allow us to validate our general setup, both for the CFD and the FEA simulations (for
instance, regarding boundary conditions or used models). Actually, having to run these
"validation" simulations for the whole rocket would be too much time costly, and so it is
thus preferable to focus first on the nosecone. An important to be highlighted regards the
fact that we would not perform any FEA simulations on the 2D version of the nosecone,
as the pressure computed with the CFD simulations can only be imported on a surface
(and not on an edge, which is the "surface" of the nosecone in 2D).

3 General Knowledge
Unit system
The unit system used in all the project is the metric one, namely : [mm], [N], [kg], [Pa],
[K].

2
Physical behavior
The equations that are solved in our case, are the Navier-Stokes equation coupled with the
energy equation for compressible flows (for the CFD analysis) and the thermo-mechanical
deformation equation for the FEA analysis. The Navier-Stokes equation can be written
in a dimensionless form, like the following:

∇ · u = 0
(1)
 ∂u + (u · ∇)u = −∇ p − 1 ∇2 u
∂t Re

The energy equation solved by Fluent is the following one, coming from the ANSYS
Theory Guide [3].
!
∂
(ρE) + ∇ · (⃗v(ρE + p)) = ∇ · keff ∇ T − ∑ h j⃗Jj + (τ eff · ⃗v) + Sh (2)
∂t j

where

• keff is the effective conductivity

 
• ∇ · keff ∇ T − ∑ j h j Jj + (τ eff · ⃗v) refer respectively to energy transfer due to con-
⃗
duction, species diffusion, and viscous dissipation.

• Sh includes the heat of chemical reaction, for us Sh = 0

p v2
• E = h− ρ + 2 : is the energy term

For an ideal gas formulation, h = ∑ j Yj h j , where Yj is the mass fraction of species j and
RT
h j = T c p,j dT
ref

Fluid properties
The studied fluid (air) is Newtonian, and is treated as an ideal gas. The specific heat C p
is taken as piecewise-polynomial with the default coefficients for a temperature range of
100 K to 3000 K.
The thermal conductivity is set to constant at a value of k = 0.0242 W/(mK ).
The dynamic viscosity µ( T ) follows a Sutherland model with a three coefficient method,
all the default parameters are kept.
The air molecular weight Mair is set to default value 28.966 kg/kmol.

3
Material Properties
For the FEA simulation the material properties have been entered in ANSYS. We looked
into the report of Oscar and Vinski [4] to have an idea of our rocket’s composition. We
found out that two materials were mainly used:

• PTFE (for the nosetip)

• HexPly 8552S/37%/280HS/AS4-3K (for the rest of the rocket)

We modelled both materials on ANSYS mechanical, thanks to the report of Oscar and
Vinski.
An important point must highlighted. Actually, there are many ways to model a composite
on ANSYS:

• either by using the Engineering data capability

• or with the ACP-Pre module of ANSYS

For this project, we decided to use the first option. However, to have a better representa-
tion of the composites (for instance regarding the number of layers, or their orientation...),
the ACP-Pre module should be used. The main problem here regarded the fact that the
ACP-pre module obliged us to use shell elements. Actually, the nosetip being a full 3D
solid, we could only model it with 3D solid elements, and not shell elements.

Turbulent flow calculation

In order to model turbulence in ANSYS Fluent, many models are available (SST k-ω,
Spallart-Almaras,...).
By using the work done in [1] and [2], we decided to use the Shear-Stress Transport model
(SST) k-ω using the default value for the constant, as it was this latter that had been
selected for previous CFD studies. It must be kept in mind that, since the pressure-based
solver is used, the viscous heating option must be activated. Indeed according to the
ANSYS User Guide [3], the viscous heating is negligible when the Brinkman number Br
is lower than 1. This last is defined as :

µUe2
Br = (3)
k∆T

where
kg
• µ = 2.7 × 10−5 ms : is the dynamic viscosity at T = Tnosetip = Tmax = 535 K

• Ue = 694 m/s : is the flow velocity upstream at Mach 2

• k = 0.0242 W
mK : is the thermal conductivity of the fluid

4
 • ∆T = 535 − 300 K : is the temperature difference.

With these values, in our case, Br > 1.

In the case of the 3D nosecone, since the flow is turbulent, the boundary layer region
should be considered as the most important part to be meshed properly. Indeed, a good
inflation layer is required in order to capture well the wall shear stress τw . The parameter
controlling the mesh refinement of the inflation layer is the dimensionless wall distance
noted as y+ . According to the advise of Marangi K. [1] and to [5], y+ should as close as
possible to 1. Indeed, the wall distance is defined as :

yuτ
y+ = (4)
ν

where y is the absolute distance from the wall (first height of the cell), ν is the kinematic
viscosity and uτ is the friction velocity.
q This latter is proportional to the square root of
the wall shear stress. In fact, uτ = τρw . Regarding the mesh settings other guidelines
are given later in the report.

Initial conditions
As we only performed steady simulations, no particular initial condition was set.

5
Part I

Nosecone 2D
4 Geometric modeling and hypotheses
4.1 Presentation of the geometry
As the first part of our project, we focused on the nosecone of the rocket. For this
geometry, we performed two types of analysis: a 2D axisymmetric analysis and a 3D
analysis. The 3D analysis is discussed later in the second part of this report.
In the following subsections, a presentation of the 2D case is done.

4.2 2D-computational domain

On Figure 1, a geometry of the 2D case of the nosecone is provided. On Figure 2, a more
detailed view of the nosecone is provided.

Figure 1: Full 2D nosecone geometry

6
 Figure 2: 2D nosecone geometry (zoom)

The labelled dimensions are:

• L = 512.284[mm] • D5 = 917.28[mm]

• D1 = 14 · L = 7171.976[mm] • D6 = 200[mm]

• D2 = 4 · L = 2049.136[mm] • D7 = 458.64[mm]

• D3 = 4 · L = 2049.136[mm] • D8 = 424.96[mm]

• D4 = 4185.6[mm] • D9 = 57.146[mm]

with L being the axial length of the nosecone. In order to have an idea of which
dimensions should have the upstream and downstream domain, we used the work of
Marangi K. [1]. However, after some first simulations, due to shock wave reflections, it
has been decided to increase the size of the domain, with respect to what Marangi K. [1]
used as computational domain’s dimensions.

7
4.3 Characteristic dimensions
We chose as characteristic length the axial length of the nosecone, namely L = 512.284mm.
The characteristic velocity here is the speed of our rocket, namely v = 2 · a = 686[m/s],
with a being the speed of sound at standard temperature and pressure conditions. This
velocity correspond to a Mach number of 2.

4.4 Symmetry / periodicity of the problem

To begin with, we decided to perform some analysis with a 2D axisymmetric cut of our
nosecone. Actually, the geometry of this latter is axisymmetric, and, having chosen a
null AoA (angle of attack), the boundary conditions would also be axisymmetric. Our
2D geometry is located in the XY plane. In order to model this geometry, we used the
parametrisation of the selected nosecone to obtain some points in a plane, that we linked
together in order to obtain a curve, which would represent our nosecone’s contour.

4.5 Geometrical space

As mentioned before, for the first nosecone analysis we would first consider a 2D axisym-
metric geometrical space, and then a 3D geometrical space (this latter would be dealt
with later, as mentioned previously).

5 Boundary conditions, external forces, initial conditions

5.1 Boundary conditions
In order to define the boundary conditions, we first had to create some named selections.
These latters are:

• Inlet

• Outlet

• Nosecone

• Walls

On Table 2, we can see which type of boundary condition is applied to these named
selections. For the pressure far-field boundary condition type, we applied a gauge pressure
of 101325 [ Pa], and a Mach number of 2. The turbulence parameters were let to default
values. The wall type boundary condition enables to put a no-slip condition on the
nosecone.

8
 Table 2: Boundary condition type applied to each named selection

Named selection Boundary condition type

Inlet Pressure far-field
Outlet Pressure far-field
Nosecone Wall
Walls Pressure far-field

5.2 External forces

In our case, no external force is applied and gravity is neglected.

6 Computational mesh
6.1 Mesh type

Figure 3: 2D axisymmetric nosecone mesh

9
 Figure 4: 2D axisymmetric nosecone mesh (zoom)

For the 2D case, our mesh is unstructured.

6.2 Cell type

We used triangles and quadrilateral elements in our mesh ("MultiZone Quad/Tri Method"
in ANSYS), in order to to find an acceptable trade-off between the mesh refining and the
convergence speed. The resulting mesh can be found on Figures 3 and 4.

6.3 Size / number of cells

In order to control the refinement of our mesh, we applied many edge sizings. On Figures
5 and 6, we display in colors the edges on which we applied local sizings. On Table 3, we
can find how many divisions are applied on the corresponding edges. The goal of this first
2D simulation being to validate our setup by verifying that the measured wave angle is
similar to the analytical value, we had to find a way to refine the area where the wave was
expected to appear. This would then allow us to measure it. We proceeded the following
way:

• First, in order to obtain the analytical value of our wave angle, we used the formula
[6]
µ = arcsin(1/M) (5)

Where µ is the wave angle and M is the Mach number. We calculated µ = 30°.

10
• Then, we created a body of influence around our nosecone, with one edge of it
having an angle of 30° with the axis of symmetry.

• Finally, we applied a fine sizing at this particular edge, in order to have a very
refined mesh at the wave’s predicted location.

Figure 5: Edge sizings applied

11
 Figure 6: Edge sizings applied (zoom)

Table 3: Edge sizings applied

Edge(s) color Number of divisions Figure

Dark green 200 5
Dark blue 100 5
Yellow 80 5
Orange 150 5
Red 320 6
Pink 50 6
Purple 80 6
Light green 100 6
Light blue 30 6
Brown 200 6
Beige 20 6

As a 2D case, we decided not to take into account the boundary layer, the goal of the
2D study being to validated the setup. Our final mesh is composed of 153329 nodes and
51094 elements.

12
7 Numerical methods
7.1 Spatial discretization method
For all the CFD simulations, since we used Fluent, the Finite Volume Method (FVM) is
used.

7.2 Type of simulation/solver

The CFD simulations are done with a pressure-based solver with a coupled scheme at
steady-state. Initially, we decided to use a density-solver. This latter is generally selected
for compressible flows. However a pressure-based solver is more stable in terms of residu-
als’ convergence and according to the ANSYS User’s Guide [7], the Fluent pressure-based
solver gives good results up to Mach 3. The velocity formulation is absolute.
For the 2D case, presented in this part, an axisymmetric option is used in the General
settings of Fluent.

7.3 Solution options

For the CFD simualtions, hybrid initialisation was selected. For our calculation, we de-
cided at first to use cell registers in order to refine our meshes. After some attempts, it
was noticed that convergence was in general difficult to reach, and the calculation time
regarding the simulations was at that time too important.
It was then decided not to use them anymore. Instead, to refine our meshes, we chose to
use a body of influence (or BOI, on which we will focus on later in the study).
The simulation was run for 1000 iterations. To assess convergence, we followed the evo-
lution of five different residuals:

• continuity

• x-velocity

• y-velocity

• energy

• k

• omega

The evolution of all these residuals with respect to the iterations can be found in Figure
9. The convergence criteria for all the residuals have been put to 1e-5.

13
7.4 Computed quantities?
The goal of this 2D simulation was to identify which Fluent settings are validated by
the Compressible Flow Theory. Therefore, we computed the shock wave angle using the
software ImageJ.

8 Results
As results, we computed some contours, namely:

• A pressure contour (Figure 7)

• A density contour (Figure 8)

On Figure 8, we measured the wave’s angle in order to validate our results. We used
the aforementioned formula:
µ = arcsin(1/M ) (6)

Where µ is the wave angle and M is the Mach number. As measurement, we found
µexperimental ≈ 30°. By using our reference Mach number, M = 2, we deduced that
equation 5 was respecetd.

14
Figure 7: 2D axisymmetric nosecone pressure contour

15
Figure 8: 2D axisymmetric nosecone density contour

Figure 9: 2D axisymmetric nosecone residuals

16
 On Figure 9, we can notice quite a constant decrease in all the residuals along the
whole simulation, apart from the ω residual, which oscillates quite much during the entire
duration of the simulation. This strong oscillation of ω is due to the fact that we did not
modelled the boundary layer. At the end, it can be seen that convergence is reached after
207 iterations.

9 Analysis & conclusions

As noticed above, we obtained residuals that converged quite quickly. Moreover, we vali-
dated the wave angle. Thus, we can conclude that the different models used (turbulence,
physical properties,...) and the applied boundary conditions work well for our 2D case,
and will then be used for the setup of the nosecone’s 3D simulations, and also for the
whole rocket study. Anyway, an important point to keep in mind is that for the 3D simu-
lations, it will be required, to ensure the best transition to FEA calculations, that certain
criteria are met. The most important one of them is a y+ value of 1 on the rocket wall,
which can only be obtained through a thorough modelisation of the boundary layer (this
point will be developed in further details later).

17
Part II

Nosecone 3D
10 Goal
The goal of this second part is to partly use the Fluent Setup used for the 2D nosecone
and also to compute some first basic results for the FEA.
We will also compute the shock stand-off distance which is defined later in this part. A
comparison regarding this parameter will be done;we will compare some measurements
obtained with the Schlieren effect in a Wind Tunnel with numerical results [8].

11 Geometric modeling and hypotheses

11.1 Presentation of the geometry
The geometry used in this second part is essentially the same as for the first part. In
this part we consider again only the nosecone. However, we added a sub-element of the
nosecone tube, which is comprised of the right end of the nosecone and one cylindrical
part of 10 mm height. We added this last part, in order to have a result of the total
deformation exactly at the position where the nosecone ended in the whole rocket. The
different parts of the geometry can be seen on Figure 10.

Figure 10: Nosecone 3D geometry used in this second part.

18
11.2 3D-computational domain
Regarding the 3D computational domain for the CFD simulation, since the flow is less
restricted than in the 2D case, there will be less problem with wave reflections at the
boundary of the fluid domain. Therefore we decided to take the ratio of length used by
Marangi K. [1]. The computational domain that converged with the best residuals is
presented in Figure 11. The different dimensions of this computational domain are given
in Table 4. The Body of Influence (BoI) located in the wake of the nosecone has been
built to reduce the oscillations of the k − ω residuals. Different sizes and shapes of this
BoI have been tried. The geometry presented here is the one that allows to reduce the
oscillations of the residuals due to a strong re-circulation region. The nosecone is a case
where the ratio "(solid body)/(wake region)" is low. Therefore, it is necessary to have a
well-refined mesh behind the nosecone in order to capture well the turbulence.

Table 4: Part 2 - Nosecone 3D : Size of the Computational Domain for the CFD with
L1 = Ltotal

Ratio Size [mm]

L1 - 522.3
L2 2 L1 1044.6
L3 5 L1 2611.5
BoI - (lx , ly , lz ) = (2000, 400, 400)

Figure 11: 3D computational domain with Lengths : L1, L2, L3 and with a Body of
Influence in the wake region.

19
11.3 Characteristic dimensions
The characteristic length in this part is the total length of the solid geometry presented
before. Namely, L1 = 522.3 mm. As before for the 2D nosecone, the characteristic velocity
is for a Mach number of 2. However, to compare the shock wave angle’s numerical result
with measurements with the Schlieren effect, a CFD simulation is also done at Mach 1.7.

11.4 Symmetry / periodicity of the problem

For the CFD simulation, generally for this type of geometry an axisymmetric simulation
is advised. However, we decided to do a full 3D CFD simulation in order to better
understand the different aspects of a 3D simulation.
Another option that could be considered would be to do an axisymmetric simulation
with the Swirl option activated in ANSYS Fluent. This last option allows to model the
3D phenomena present in a 3D flow while reducing the calculation time by using the
principle of an axisymmetric flow. Regarding the FEA simulation, since the pressure field
is symmetric, we expect to have an axisymmetric total deformation.

12 Boundary conditions, external forces, initial conditions

12.1 Boundary conditions
The following boundary conditions have been used for the CFD simulations (see Table
5). The named selections are presented on Figure 12. The named selection nosecone is
logically for the faces representing the nosecone walls inside the fluid domain.

Table 5: Part 2 - Nosecone 3D : Boundary conditions used in the CFD simulation

Named Selection inlet outlet farfield

Type Pressure-far-field Pressure-far-field Pressure-far-field
Mach number [−] 2 2 2
Gauge Pressure [ Pa] 101325 101325 101325
Thermal [K ] 300 300 300
Named Selection nosecone
Type wall
Stationary wall on
Shear Condition No slip
Roughness Models Standard
Roughness Heigth 0

20
Figure 12: Part 2 - Nosecone 3D : Different named selections for the boundary condi-
tions. The nosecone wall is inside the fluid domain.

These boundary conditions were chosen according to the study performed by Marangi
K. [1]. In addition an operating pressure of 0 Pa is set according to [9]. Indeed, in ANSYS
Fluent the equation 7 is applied1 .

p abs = poperating + p gauge (7)

where p abs is the absolute pressure of the flow and p gauge is the computed value by ANSYS.
In fact, it should be kept in mind that in Fluent all the pressures set and computed are
gauge pressures. Equation 7 is used in order to avoid the round off error done when the
Mach number is low. Since here we have a high Mach number (Mach 2 and Mach 1.7),
we can simply set that the absolute pressure is equal to the gauge pressure.
For the FEA boundary conditions, we decided to fix the back face of the nosecone
as shown in Figure 13a. The pressure field and the temperature field are then imported
from the CFD simulation and applied along all the faces of the nosecone Figures 13b and
13c. We also decided to apply a uniform internal pressure of 1 atm, since the rocket is
not pressurized. This pressure does not change a lot the results but it is good to apply it
for the sake of rigor.
1 Equation coming from https://www.afs.enea.it/project/neptunius/docs/fluent/html/ug/
node330.htm

21
 (a) Fixed back face. (b) External pressure from the CFD simulation.

(c) External temperature from the CFD simulation.

Figure 13: Part 2 - Nosecone 3D : FEA boundary conditions. Fixation, pressure and
temperature

It is important to mention that we decided to use the following additional settings for
the imported CFD pressure :

• Apply by : Direct

• Loaded on : Initial surface (Dead pressure analysis)

The "Apply by : Direct" specifies that the pressure should be applied on the surface
directly and not to compute a derived force and then apply it.
The Dead Pressure analysis considers that the surface does not deform a lot and the
normal vector of the initial surface stays normal to the final deformed surface. This
choice has been done considering that the deformation of the nosecone is not too large.

12.2 External forces

As before for the 2D case, no external forces are applied in the CFD simulation.

13 Computational mesh

For the CFD

As for the 2D nosecone, in this section the mesh selected for the CFD is presented.
It is important to mention that only the best mesh in term of residuals’ convergence is
presented here. Indeed, for this specific geometry, more than 20 types of meshes were tried

22
in order to achieve a good y+ and a good residuals’ convergence. The meshing process
has been done with the Watertight workflow meshing option in ANSYS Fluent. Two
versions of the same mesh are given in the following subsection. Actually, after running
the simulation with an initial mesh, we used an automatic mesh refinement method (Cell
registers) to refine the mesh around the shock waves.

13.1 Mesh type

Figures 14, 15 present the initial mesh corresponding to settings given in tables 6, 7 and
8. Figure 16 shows the situation after the automatic mesh refinement. We tried many
times using a pressure gradient criterion like in [1], but ANSYS crashed for running out
of memory. Therefore, the cell register presented here is based on a region criteria.

Figure 14: Part 2 - Nosecone 3D : Final mesh used for the 3D nosecone in the XY plane.
No refinement

Figure 15: Part 2 - Nosecone 3D : Final mesh used on the XY plane around the nosetip
to see the boundary layers. No refinement

23
Figure 16: Part 2 - Nosecone 3D: Final mesh after automatic mesh refinement. The
region near the nosetip where the shock waves are present is very refined.

13.2 Cell type

The cell type used in this simulation is the poly-hexcore type, which is a combination of
hexahedron and polyhedron.

13.3 Size / number of cells

Regarding the sizings of the different parts of the nosecone, they are presented in table
6. It should be noticed that all the length values are given in mm and all angles are in
degree. The labels presented in the table for the local sizing are shown on Figure 17. The
BoI is the same as the one presented in the geometry section.

Figure 17: Part 2 - Nosecone 3D : Geometry labels used for assigning the good local
sizings

24
Table 6: Part 2 - Nosecone 3D : Local sizings used in Ansys Meshing with the watertight
flow option

Local Sizing nosetip nosecone backface BoI

Growth Rate 1.2 1.2 1.2 1.2
Size Control Type Face size Face size Body size Body of Influence
Target Mesh size [mm] 2 10 15 20

With the watertight workflow, other parameters like surface mesh, boundary layer and
volume mesh need to be defined. The following table (Table 7) regards the surface mesh
and volume mesh settings.

Table 7: Part 2 - Nosecone 3D : Surface mesh and Volume mesh settings.

Surface Mesh Volume Mesh
Minimum size [mm] 50 Fill with poly-hexcore
Maximum size [mm] 4000 Peel Layers 3
Growth Rate 1.2 Min Cell Length [mm] 2
Size Functions Curvature & Proximity Max Cell Length [mm] 4096
Curvature Normal Angle 12 Use size Field ? yes
Cells per gap 2 Polyhedral Mesh Feature angle 30
Scope Proximity To faces-and-edges Avoid 1/8 octree transition ? no
Smooth Folded Faces/Repair Free Nodes Limit 10 Quality warning limit 0.05
Auto Assign Zone Types ? yes Gap Factor 0.25
Invoke Quality Improve? yes Max Aspect Ratio 25
Quality Improve Skewness Limit 0.4 Min Aspect Ratio 1
Quality Improve Max Angle 80 Keep First Boundary Layer Heigth yes
Quality Improve Collapse Skewness Limit 0.95 Adjacent Attach Angle 80
Auto Remesh to Remove Clustering ? auto Use default stair-step handling ? yes

Regarding the boundary layer settings table 8 presents the most important values. It
is important to mention that the Offset Method Type has been chosen after many trials.
According to Figure 18 one can see that last-ratio is the most appropriate option for this
case. The height of the first layer is computed via an online calculator [10]. To be sure
that the boundary layer effect is well captured by the mesh, a contour plot of the y+ is
presented later.

Table 8: Part 2 - Nosecone 3D : Mesh settings for the boundary layer

Offset Method Type last-ratio

Number of Layers 25
Transition Ratio 0.272
First Height [mm] 0.002
Ignore Boundary Layers at Acute Angles ? yes

25
Figure 18: Part 2 - Nosecone 3D : 4 types of Offset Method for defining the boundary
layer around the nosecone.

With these mesh parameters a minimum orthogonal quality of 1.55 × 10−5 is reached.
This mesh quality is very bad but it is normal since the worst elements are located in
the boundary layer region. A trade-off exists between good mesh quality and optimal y+
values.
With the initial mesh (no refinement), the numbers of nodes and cells are 446524 and
241395 respectively. The automatic mesh refinement increased the numbers of nodes and
cells up to 3761476 and 3514569 respectively. Despite the increase in time computation,
this option of automatic mesh refinement is very useful and should be used.

For the FEA

Further below we present briefly the mesh settings for the FEA simulation. More detailed
description of the global ANSYS Mechanical parameters will be done later.

13.4 Mesh type

Figures 19 and 20 present the mesh used in the FEA of this nosecone. The total number of
nodes and cells generated are 284673 and 187490 respectively. Due to the small thickness
of the nosecone (thickness of 1.73 mm), bad cell quality is generated in the thickness. One
should maybe consider to do a FEA with shell elements.

26
Figure 19: Part 2 - Nosecone 3D : Mesh used in the FEA in the X-Z plane with a cross
section plane.

Figure 20: Part 2 - Nosecone 3D : Mesh used in the FEA isometric view

13.5 Cell type

We decided to do a 3D FEA simulation using 3D solid elements. Since the geometry is
quite complex, tetrahedric cell type is used.

27
13.6 Size / number of cells
Tables 9 and 10 present the different mesh sizings used in this FEA. The different local
sizings tabulated on Table 10 can be viewed in Figures 21a, 21b and 21c. The sphere
of influence used to size the mesh is located at the nosetip point with coordinate centre
( x, y, z) = (0, 0, 0).

Table 9: Part 2 - Nosecone 3D : Global mesh settings

Elements’order quadratic
Global size [mm] 5
Use adaptative sizing? yes
Resolution 7
Mesh simplification? yes
Transition fast
Span angle center fine
Inflation no

Table 10: Part 2 - Nosecone 3D : Local mesh sizings and refinement

Type Radius Element size

edge sizing
Sphere of influence 10 mm 1 mm
Type Cell type Geometry selection
method
Patch Conforming method Tetra 2 body
Geometry Level of refinement
refinement 1
3 faces 1
Geometry Level of refinement
refinement 2
3 faces 1
Type Radius Element size
body sizing
Sphere of influence 275 mm 10 mm
Geometry Level of refinement
refinement 3
1 face 1

28
 (a) (b)

(c)

Figure 21: Part 2 - Nosecone 3D : edges and faces used for the local mesh sizings and
refinement.

14 Numerical methods
The numerical method used in this section for the CFD simulation is the one validated
by the 2D case.

14.1 Spatial discretization method

As said before in the 2D case, the spatial discretization method is the one selected by
Fluent, namely FVM for the CFD and Finite Element Method (FEM) for ANSYS Me-
chanical.

14.2 Type of simulation/solver

As mentioned before, all the simulations done in this project are performed in a steady-
state regime. The CFD simulations are done with a pressure-based solver with a coupled
scheme, as this was the case for the 2D CFD.
Regarding the FEA computation, we decided to do a static steady state analysis. It
will be interesting to do a transient structural analysis since the pressure changes with
time in the real case. But for a first basic approach, the static case can give interesting
results.

29
14.3 Solution options
As for the 2D case, a hybrid initialization has been chosen. The residuals are computed
and plotted for the 7 following variables. The convergence criterion is set at 1 × 10−6 .

• continuity • energy

• x-velocity
• k
• y-velocity

• z-velocity • ω

The number of iterations computed before starting the mesh refinement criteria is
300. Then during 100 iterations with a frequency of 50, the automatic mesh refinement is
activated. After that, another final 100 iterations are computed without mesh refinement
in order to lower again the residuals (see the plot later).

14.4 Computed quantities?

For the CFD

The quantities computed in the case of the nosecone 3D are the shock waves angle at
Mach 1.7 and the shock-stand off distance at Mach 2. These two quantities are visually
measured using the software ImageJ. However, some pressure contour and Mach number
contour are presented later in this part.
The shock stand-off distance is the adimensional distance between the shock wave and the
nosetip. This parameter is different from 0 when the nosetip is rounded or spherical. In
our case the nosetip has a radius of 6 mm. According to [11] and [12], the shock standoff
distance can be defined by Equation 8 and 9.

δ ρ
≈ 1 (8)
R ρ2

δ
= 0.143 exp(3.24/M∞ ) (9)
R
Where R is the radius of curvature of the nosetip (for our nosetip R = 6 mm), δ is the
distance between the shock wave and the nosetip, M∞ is the upstream Mach number
(in our case M∞ = 2) and ρ1 , ρ2 are the density upstream and downstream the shock
wave respectively. The ratio of density is computed using the formula given in [6] and
the theoretical shock wave angle given by NACA 1135 [13]. This formula assumes no
viscosity, ideal gas law for air and isentropic flow before and after the shock wave.

30
For the FEA
The computed quantities for the FEA simulation are the total deformation in mm and
the Equivalent Von Mises Stress.
Since this second part is only a small part of the whole project, no convergence study
has been done for the CFD and the FEA. A convergence study is done after for the full
rocket geometry.

15 Results
Figures 22 and 23 present the pressure contour and the temperature contour for the 3D
nosecone at Mach 2. Thanks to the cell register the shock waves are well refined around
the nosetip. This helped us to compute the measured shock stand-off distance. A more
enlarged picture around the nosetip is given on Figure 24, where we used the density
contour to identify the shock stand-off distance.

Figure 22: Part 2 - Nosecone 3D : Static Pressure contour at Mach 2

31
Figure 23: Part 2 - Nosecone 3D : Static Temperature contour at Mach 2 with contour
line.

Figure 24: Part 2 - Nosecone 3D : Density contour around the nosetip for Mach 2

32
 Using the formulas given in the above subsection and the software ImageJ, we found
an adimensional shock stand-off distance of Rδ = 0.156 ± 0.02. Table 11 compares the
numerical value with the two theoretical ones found in [11] and [12].

Table 11: Part 2 - Nosecone 3D : Shock Stand-off distance for Mach 2 with R = 6 mm.
Comparison of two theoretical values with the numerical one.

From Equation 8 From Equation 9 Numerical

δ
R [-] 0.675 0.722 0.156 ±0.02

Figure 25 shows the pressure contour around the nosetip at Mach 1.7 and the shock
wave angle θnum measured with ImageJ. As said before, this simulation was done to
compare the shock wave angle with the one given by Schlieren measurement θSchlieren
done in the Hydro & aero technologies and science laboratory in Geneva. This latter is
presented on Figure 26 and comes from the student’s project [8]. θnum = 47.054° and
θSchlieren = 41.378°.

Figure 25: Part 2 - Nosecone 3D : Pressure contour around the nosetip for Mach 1.7
(left figure) and shock wave angle measurement (right figure).

33
Figure 26: Part 2 - Nosecone 3D : Schock wave angle measurement on ImageJ at Mach
1.7 obtained with Schlieren effect.

To be sure that the boundary layer is well captured, we plotted also the contour map
of the y+ value on the nosecone and an histogram of this parameter. These two results
are presented on Figures 27 and 28.

Figure 27: Part 2 - Nosecone 3D : y+ Figure 28: Part 2 - Nosecone 3D : His-

contour plot for Mach 2. A maximum togram of the y+ for the wall of the
of 2 is reached at the nosetip. nosecone.

The last figure presented for the CFD simulation regards the residuals’ convergence
at Mach 2. A small discussion is done in the conclusion of this second part. What we can
see is that the variables listed in section 14.3 have for final residual errors, the ones listed
below:

34
 • Last residual continuity : 5 × 10−5 • Last residual energy : 5 × 10−8

• Last residual x-velocity : 3 × 10−9

• Last residual k : 3 × 10−6
• Last residual y-velocity : 1 × 10−8

• Last residual z-velocity : 1 × 10−8 • Last residual ω : 3 × 10−6

Figure 29: Part 2 - Nosecone 3D : Residuals of the CFD simulation at Mach 2. Automatic
mesh refinement is activated from iteration 300 to iteration 400.

Regarding the FEA results, Figures 30 to 31b show the Von Mises Stress distribution
in the Nosecone and the total deformation. We can see that the total deformation is an
axisymmetric distribution as expected and also that the maximum deformation is reached
at the nosetip, where the maximum pressure is applied.

35
Figure 30: Part 2 - Nosecone 3D : Total deformation in the nosecone in mm. Deforma-
tion is axisymmetric as expected and maximum is obtained at the nosetip.

The maximum of the stress distribution is reached at the contact region between the
two assembled bodies. This result is partly due to bad quality cells in the thickness. A
bad quality cell resulting in a tiny face, the Pressure/surface value becomes very huge.

(b) Enlarged view of the Maximum Von Mises

(a) Von Mises stress distribution
stress.

Figure 31: Part 2 - Nosecone 3D : Von Mises stress distribution. Maximum is reached
in the contact region between the two assembled bodies.

16 Analysis & conclusions

To conclude this second part, some important aspects and results should be discussed
more in details.
Firstly, regarding the CFD simulations, we see that the cell registers defined around
the nosetip could be improved by defining a cell register on the pressure gradient. We
have tried this option, but since the recirculation region is too large with respect to the
geometry, all the cells refined were the cells in the wake of the nosecone. These trials
got terminated by ANSYS because of a memory crash. The used cell register allows us
to achieve good convergence. Indeed only the continuity and the k − ω residuals did

36
not converged below 10−6 . The continuity is related to the mass in - mass out balance.
Therefore, it is possible that some wave reflections are present in the fluid domain, creating
errors in the mass balance. k − ω errors is related to turbulence. An improvement that
can maybe done is to change the shape of the wake body region.
Secondly the Boundary Layer Offset Method finally used is the best one and this last-ratio
option should be used more than the others. Indeed, we also tried to use the uniform type
and the smooth-transition, but they both require a large number of layers. As said before
the mesh’s minimum orthogonal quality is very bad but we should keep it like that. The
only point that must be verified is that the bad quality cells are the one compressed in
the boundary layer.
The last results discussed for the CFD simulations regard the accuracy of the numerical
shock stand-off at Mach 2 and the shock wave angle at Mach 1.7. We see that the shock
stand-off distance obtained numerically is different from the ones obtained by the research
papers. However, the computation of this parameter is still very discussed in the Fluid
Mechanics field and one can find more than 2 different ways to compute this distance.
Therefore having the same order of magnitude like in our case can be assumed to be
a quite good result. Regarding the shock wave deflection at Mach 1.7, we see that the
relative error between the Schlieren measurement and the numerical one is about 13 %.
This large relative error is generated by many aspects. One of them can be the quality of
the experiment. Another source of discrepancies is maybe the fact that even if a refined
mesh was applied, we could not reproduce the perfect rounded shape of the nosetip. By
comparing these two values of the deflection with the value given in the NACA 1135 [13],
we found that the Schlieren measurement underestimates the theoretical one and the
numerical value overestimates it. Indeed, the theoretical prediction (assuming isentropic
ideal gas and no viscous flow) is 45°.
Regarding the FEA results, all the relevant conclusions have been already done in the
result subsection above. What we learned from this first basic FEA study is that the
mesh should definitely be improved in the thickness of the nosecone, because bad quality
elements generated wrongly high stress or Shell elements should be used for all the hollow
bodies.

37
Part III

Rocket simulations
17 Geometric modeling and hypotheses
17.1 Presentation of the geometry
In order to get the geometry of the whole rocket, we took some parts on the GrabCAD
account of the EPFL Rocket team. We also had to simplify some aspects of our rocket. For
instance, in reality, the rocket is not hollow, and contains some elements like electronics
or even the motor. In our geometry, our rocket is mostly hollow, we did not take into
account the components inside it. At the end, our CAD model of the whole rocket , as
can be seen on Figure 32, is composed of:

• The nosecone (nosetip+tube)

• The body (represented by a hollow cylinder)

• The four (simplified) fins

• The boattail

Figure 32: Rocket geometry

38
 In order to have a closed volume for our CFD simulations, we added a wall at the
extremity of the boattail, as we did with the nosecone. Then, after the CAD of the rocket
was completed, a computational volume has been created around the rocket, for the CFD
simulations. This domain’s dimensions can be found on Figures 37 and 38. The parts
whose shape has been the most altered by our simplifications certainly are the fins. On
Figures 33, 34, 35, 36 , we provide a comparison between the original fin and the one we
simplified and used through our study. We took care to keep the same width, and the
same general dimensions. We discarded the rounded plate and the chamfers, as they may
cause problems during the meshing process.

Figure 33: Original fin Figure 34: Simplified fin

Figure 35: Original fin Figure 36: Simplified fin

39
 Figure 37: Rocket computational domain

Figure 38: Rocket computational domain (zoom)

The labelled dimensions are the following

• L2 = 2814.6[mm]

• L1 = 2 · L2 = 5629.2[mm]

• L3 = 5 · L2 = 14073[mm]

• L4 = 1.5 · L2 = 4221.9[mm]

• L5 = 0.1 · L2 = 281.46[mm]

40
17.2 Symmetry / periodicity of the problem
With the whole rocket, we have two symmetry planes, thanks to the four fins. However,
we did not use them, as we encountered some problems regarding the boundary conditions
to be imposed on the symmetry planes.

17.3 Geometrical space

In the case of the whole rocket, we were bound to consider a 3D geometry. Indeed, the
four rocket’s fins removed the axisymmetry that was here for the nosecone, even if we
consider no Angle of Attack (AoA) for the rocket.

18 Boundary conditions, external forces, initial conditions

18.1 Boundary conditions

CFD part
Regarding the CFD simulation, we used the exact same boundary conditions as for the
3D nosecone case, that have been displayed in Table 5. The only difference is that instead
of assigning a wall boundary condition to the nosecone wall, we assigned it to the rocket
wall (or, more precisely, to the different named selections that composed it, and that can
be found on section 19.1).

FEA part
When studying a rocket, one common boundary condition to impose is to fix the rocket
boattail. In our study, we tried to find a way to be closer to reality. To do so, we simulated
a fixation on the Center Of Gravity (COG).
First, we tried to use a coupling between the COG and internal faces of the rocketwall.
The problem with this approach is that a face coupled with a fixed point will also be
fixed. An other way had to be found. The idea was to create a Y-Z plane at the same
X-coordinate as the COG, and then to intersect this plane with the rocket shell.
This would result in two circles, and we would fix the inner circle. The final result can be
seen with the purple tag on Figure 43, alongside with the imported pressure and the in-
ternal atmospheric pressure. The imported pressure field and temperature field are shown
respectively in Figures 39 and 40. The precision of these imported values depending on
the mesh refinement, we chose to display only the ones related to the refined mesh, in
order to have the most accurate imported pressure and temperature.
When importing the pressure, a special care must be taken when choosing the import
options. On Figure 41, a screenshot of the parameters giving the most relevant results is

41
provided. When applying the internal pressure, the same care must be taken, as can be
seen on Figure 42.
Regarding the high temperature at the fins’ roots, one may ask himself why we have
such results. This elevation in temperature can be explained through the viscous heating
caused by the existing vortices. Moreover, as the regime is supersonic, this effect is even
more important, this is illustrated by Equation 10. To be able to have the occurence
of this phenomenon, we activated the Viscous heating option in the turbulence model.
In the following equation, f ( Pr ) is a term proportional to the Prandlt number, which is
defined as Pr = αν(TT ) .

Twall ( γ − 1) 2
= 1 + f ( Pr ) M (10)
T f luid 2
Remark: the options are in French because the ANSYS version on the DLL computers
has been installed in French.

Figure 39: Imported pressure field

42
Figure 40: Imported body temperature

43
Figure 41: Imported pressure options

44
 Figure 42: Applied pressure options

18.2 External forces CFD

As external forces, we applied an internal pressure to our inner rocket wall, in order to
simulate atmospheric pressurisation.

Figure 43: Boundary conditions used regarding the FEA analysis

45
19 Computational mesh
19.1 Mesh type
As for the 3D CFD simulations performed with the nosecone, we once again used the
watertight meshing capability of ANSYS. As a help to mesh the computational domain of
the whole rocket, we used the same type of name selection as [1], as well as the same local
sizing types (that we can find below, alongside with their used designations), namely:

• nosecone tip ("conetip")

• nosecone ("cone")

• body ("body")

• boattail ("enginemount")

• fins area ("tailsarea")

• fins leading edges ("tailsleading")

• fins trailing edges ("tailstrailingedges")

• fins extremities ("tailsedge")

Indeed, the different rockets having the same components (nosecone, body, fins, boattail),
the refinements applied to one rocket could be used as a generic refinement method for
other rockets. A view of these different refinements can be seen in Figures 44 and 45. We
took care to apply a really fine meshing at the nosetip and different areas of the fins, as
they are the ones which will sustain the heavier load from the fluid.

Figure 44: Named selection details for the rocket (1)

46
 Figure 45: Named selection details for the rocket (2)

19.2 Cell type

Regarding the computational domain for the CFD simulations, many types of cells are
available in the watertight meshing. Three of them were used in this study, namely:

• Tetrahedras

• Polyhedras

• Polyhexcore

20 Numerical methods
20.1 Spatial discretization method
Finite volume method is the spatial dicretization method used in Fluent.

20.2 Type of simulation/solver

Same as nosecone

20.3 Solution options

Same as nosecone

20.4 Computed quantities?

Here, it was decided to compute the generated drag by taking as reference surface the
whole rocket surface.

47
21 Mesh size / domain size convergence study CFD
To validate our results, we need to perform a convergence study regarding our whole
rocket. More precisely, we would do many convergence studies, each one analysing the
impact of a parameter’s variation on the behaviour of our results. The different related
parameters are:

• the body of influence’s size;

• the computational domain’s size

For the sake of simplicity, only the body of influence related convergence study would
be analysed here, and the computational domain size related one would be put in the
appendix. On the firstly mentioned convergence study, to confirm that our meshes have
converged, we would look for a value of 1e-5 for all the residuals for both the coarse mesh
and the refined mesh.

21.1 Criterion
In order to assess convergence, we would use three different parameters, namely:

• the drag force

• the maximum pressure Pmax

• the maximum temperature Tmax

21.2 Presentation of the different CFD meshes

Figure 46: Coarse mesh cross section X-Y

plane Figure 47: Coarse mesh cross section Y-Z
plane

48
 Figure 48: Coarse mesh cross section X-Y plane (zoom)

Figure 49: Coarse mesh cross section Y-Z plane (zoom)

With the watertight workflow, other parameters like the surface mesh, boundary layer and
volume mesh need to be defined. The following table (Table 12) is for the surface mesh
and volume mesh setting. The detailed parameters input on the watertight meshing can
be found this table. Moreover, the options used regarding the boudary layer modelisation
are shown in Table 13.

49
Table 12: Rocket : Surface mesh and Volume mesh settings (both for coarse BOI and
refined BOI).
Surface Mesh Volume Mesh
Minimum size [mm] 50 Fill with poly-hexcore
Maximum size [mm] 3000 Peel Layers 3
Growth Rate 1.2 Min Cell Length [mm] 2
Size Functions Curvature & Proximity Max Cell Length [mm] 2048
Curvature Normal Angle 12 Use size Field ? yes
Cells per gap 3 Polyhedral Mesh Feature angle 30
Scope Proximity To faces-and-edges Avoid 1/8 octree transition ? no
Smooth Folded Faces/Repair Free Nodes Limit 10 Quality warning limit 0.05
Auto Assign Zone Types ? yes Gap Factor 0.25
Invoke Quality Improve? yes Max Aspect Ratio 25
Quality Improve Skewness Limit 0.4 Min Aspect Ratio 1
Quality Improve Max Angle 160 Keep First Boundary Layer Heigth yes
Quality Improve Collapse Skewness Limit 0.85 Adjacent Attach Angle 80
Auto Remesh to Remove Clustering ? auto Use default stair-step handling ? yes

Table 13: Rocket : Mesh settings for the boundary layer (for both coarse and refined
BOI)

Offset Method Type last-ratio

Number of Layers 25
Transition Ratio 0.272
First Height [mm] 0.004
Ignore Boundary Layers at Acute Angles ? yes

Figure 50: Refined mesh cross section X-Y

plane Figure 51: Refined mesh cross section Y-Z
plane

50
 Figure 52: Refined mesh cross section X-Y plane (zoom)

Figure 53: Refined mesh cross section Y-Z plane (zoom)

Table 14: Coarse BOI local sizings

Named selection Size control type Target mesh size[mm]

conetip Face Size 2
cone Face Size 20
body Face Size 20
enginemount Face Size 20
tailsarea Face Size 10
tailsleading Face Size 2
tailsedge Face Size 2
tailstrailingedges Face Size 2
body of influence BoI 20

51
 Table 15: Refined BOI local sizings

Named selection Size control type Target mesh size[mm]

conetip Face Size 2
cone Face Size 10
body Face Size 10
enginemount Face Size 10
tailsarea Face Size 10
tailsleading Face Size 2
tailsedge Face Size 2
tailstrailingedges Face Size 2
body of influence BoI 10

We used two meshes in order to perform the convergence study. In order to model the
best way possible the surroundings of the rocket, we created a body of influence (BOI)
around a rocket, as can be seen on Figures 37 and 38. On the coarse mesh we used coarse
elements in the BOI whereas we used refined elements in the refined one. For each mesh,
we collected the number of nodes and the number of cells. As residuals criterion, we
selected 1e-5, as it is approximately the threshold at which some residuals seem not to
converge in some situations.

21.3 CFD results

The results regarding the different meshes can be found in tables 16, 17, 18. For each type
of 3D elements, we used two types of BOI: one refined and one coarse. As can be seen, for
all the types of elements, for the refined BOI, we have a drag value of approximately 1201
[N]. We can thus assume that is a converged value. However, for the max. temperature
Tmax and for the max. pressure Pmax , polyhedra and polyhexcore elements approximately
have the same value, respectively around 533000 [Pa] and 544 [K] , whereas tetrahedras
have quite distant values, respectively 519792.3 [Pa] and 584.8613 [K]. An analysis of this
discrepancies in the results is provided just below. Regarding convergence, we can notice
the different points:

• for polyhexcore, both simulations have converged

• for tetrahedra, none of the simulation has converged

• for polyhedra, only the refined BOI simulation has converged.

As a conclusion, we decided to use polyhexcore as being the elements to proceed with,

with the refined BOI this latter presents more accurate results.
Indeed, as can be seen on Figure 56, the related residuals all undergo a smooth decrease
until the convergence is reached. Moreover, in order to have the most precise pressure
distribution on the rocket wall, the y+ value on this latter’s surface must be around 1.

52
For the chosen CFD results, this is the case on the main part of the rocket, as can be seen
on Figure 54. The highest value of y+ can be found on the nosetip and on the fins’ edges.
At last but not least, an important remark must be done regarding the tetrahedra ele-
ments. Actually, they possess less nodes per elements than polyhexcore and polyhedra
cells. Thus, a higher number of tetrahedral elements is required in order to achieve the
same accuracy as polyhedra and polyhexcore elements. A study has been performed to
assess this effect in the appendix B.

Figure 54: y+ value on the rocket wall

Figure 55: Residuals for polyhexcore,

Figure 56: Residuals for polyhexcore,
coarse BOI
refined BOI

53
Figure 57: Residuals for tetrahedra, Figure 58: Residuals for tetrahedra, re-
coarse BOI fined BOI

Figure 59: Residuals for polyhedra, Figure 60: Residuals for polyherda, re-
coarse BOI fined BOI

Figure 61: Pressure contour polyhex- Figure 62: Pressure contour polyhex-
core coarse BOI core refined BOI

54
Figure 63: Pressure contour polyhedra Figure 64: Pressure contour polyhedra
coarse BOI refined BOI

Figure 65: Pressure contour tetrahedra Figure 66: Pressure contour tetrahedra
coarse BOI refined BOI

Figure 67: Temperature contour poly- Figure 68: Temperature contour poly-
hexcore coarse BOI hexcore refined BOI

55
Figure 69: Temperature contour poly- Figure 70: Temperature contour poly-
hedra coarse BOI hedra refined BOI

Figure 71: Temperature contour tetra- Figure 72: Temperature contour tetra-
hedra coarse BOI hedra refined BOI

Table 16: Polyhexcore meshes’ CFD results

Type of mesh coarse BOI refined BOI

Elements size in BOI [mm] 20 10
Nbr.of cells 963165 3293423
Nbr.of Node 1845600 4583194
Drag Force [N] 1272.6 1201.2
Pmax [Pa] 533176.6 533694
Tmax [K] 542.4205 543.3211
Residual criterion 1,00E-05 1,00E-05
Iters to converge 175 223

56
 Table 17: Polyhedra meshes’ CFD results

Type of mesh coarse BOI refined BOI

Elements size in BOI [mm] 20 10
Nbr.of cells 894211 2857912
Nbr.of Node 3263671 13749624
Drag Force [N] 1272.9 1203.6
Pmax [Pa] 533084.6 533222.4
Tmax . [K] 542.6066 545.3249
Residual criterion 1,00E-05 1,00E-05
not converged,
Iters to converge 186
omega at 7.2e-5

Table 18: Tetrahedra meshes’ CFD results

Type of mesh coarse BOI refined BOI

Elements size in BOI [mm] 20 10
Nbr.of cells 3317670 14258071
Nbr.of Node 824713 2748885
Drag Force [N] 1255.2 1201.1
Pmax [Pa] 520735.7 519792.3
Tmax [K] 592.7551 584.8613
Residual criterion 1,00E-05 1,00E-05
no convergence of k, no convergence of
Iters to converge
continuity and omega continuity and k

21.4 Choice of the CFD mesh / domain

For the rest of the study (namely the FEA study), we chose to use the refined mesh.

22 Mesh size / domain size convergence study FEA

To perform the FEA convergence study, as mentioned before, the following boundary
condition has been used:

• we simulated the fixation of the center of gravity by creating a circle around this
later, and by fixing it

In the appendixC, an attempt to perform an FEA analysis using shell elements and a fixed
boattail has been performed. However, we were not able to obtain conclusive results.

22.1 Criterion FEA

In order to assess convergence, we would use two different parameters, namely:

57
 • the maximum equivalent stress σmax

• the maximum displacement

22.2 Presentation of the different FEA meshes

Regarding the FEA simulations in our studies, we had to use different meshes. A very
important point that we have to highlight here is the fact that most of our rocket is in fact
represented as a hollow shell (apart from the nosecone and the fins). A direct consequence
of this is an imposed constraint in the dimension of our elements in their depth (their
max depth length would be the shell thickness). Thus, the coarsest mesh would already
have to be at a certain point refined enough, in order to have acceptable elements quality.
Indeed, the depth of a cell being quite small, large dimensions of the cell in the Y-Z plane
would result in a distorted cell, and thus in a bad quality. As we have quite a complex
geometry, we used many local sizings to impose as many constraints possible to our mesh,
as can be seen on Table 19. The refered figures are Figures 73, 74, 75, 76.
We also kept some global mesh parameters constant for all the meshes. These parameters
are displayed on Figure 77. From one mesh to the other, some sizings were kept constant.
An important point to highlight here is the fact that on ANSYS MECHANICAL, we have
the possibility to impose a general element size. At the end, maybe we could have had
the possibility to only put sizings that would be different to the general sizing, in order
to avoid redundancy.
At last, regarding the sizings, we also used:

• a face sizing for the fins (Figures 78 and 79)

• a face sizing for the rocket body (Figures 80 and 81)

• a vertex sizing on the nosetip (Figures 82 and 83)

For the fins, we used a face sizing so that the rectangular faces of our fins could present
a regular mesh. A face sizing was also used in order to impose the use of two elements
in the depths, in order to have acceptable elements quality, as mentioned before. For the
nosetip, a vertex sizing was used (only for the coarse mesh) in order to model as accurately
as possible the exported pressure at the nosetip. The four different meshes used for the
FEA convergence study are displayed in figure 84, 85, 86, and 87. From the coarsest mesh
to the most refined one, we have:

• Coarse

• Middle

• Pre-refined

58
• Refined

Table 19: Local sizings applied to the rocket

Coarse Normal Middle Refined

mesh mesh mesh mesh Color Figure(s)
Edge size [mm]
General size 2,50 2,00 1,80 1,75
Sizing 1 2,50 2,00 1,80 1,75 orange 73 & 75
Sizing 2 1,00 1,00 1,00 1,00 green 74
Sizing 3 2,00 2,00 2,00 2,00 red 74
Sizing 4 3,00 3,00 3,00 3,00 yellow 76
Sizing 5 3,00 2,00 1,80 1,75 purple 76
Sizing 6 3,00 3,00 3,00 3,00 pink 76
Sizing 7 2,00 1,50 1,25 1,00 brown 76

Figure 73: Local sizings (1)

Figure 74: Local sizings (2)

Figure 76: Local sizings (4)

Figure 75: Local sizings (3)

59
 Figure 77: Some global mesh parameters

Figure 79: Face meshing n°1 parame-

Figure 78: Face meshing n°1 ters

Figure 81: Face meshing n°2 parame-

ters
Figure 80: Face meshing n°2

Figure 83: Vertex sizing parameters

Figure 82: Vertex sizing

60
 Figure 84: Coarse FEA mesh Figure 85: Middle FEA mesh

Figure 86: Pre-refined FEA mesh Figure 87: Refined FEA mesh

22.3 Equivalent stress distribution results

Figure 88: Stress distribution for the Figure 89: Stress distribution for the
coarse mesh middle mesh

Figure 90: Stress distribution for the Figure 91: Stress distribution for the re-
pre-refined mesh fined mesh

61
23 FEA : general results
For the FEA convergence study, we found out that the maximum equivalent stress was
located at the fins’ roots, as can be seen on pictures 88, 89, 90, 91. This may due to the
fact that the max exported temperatures are also located on the fins roots, and have quite
high values. As shown in table 20, the max stress rises once from the coarse to the middle
mesh, then drops twice, first between the middle and the pre-refined meshes, and then
between the pre-refined and refined meshes. As the difference of the values between the
two most refined meshes is quite small, we can assume that the max stress has converged.
We may have done one study on an even more refined mesh, but we were limited by
the DLL (Discovery Learning Lab) CPU (Central Processing Unit) capacity (The DLL
procedure can be found in appendix D).
Regarding the total deformation, it is always located on the fin tips and have quite
insignificant values, as shown in table 20. Moreover, our main focus is the maximum
stress (to assess the integrity of the rocket), so we will not get into more details regarding
the total deformation in our convergence study.

Table 20: Results regarding the different meshes used in the FEA analysis
Coarse mesh Middle Pre-refined mesh Refined mesh
Nb of nodes 2023781 3187177 3683990 3950644
Nb of cells 1055246 1671132 1913184 2057495
Max Equivalent stress [MPA] 2,1945 2,6518 2,2455 2.2343
Max total deformation [mm] 0,045849 0,035196 0,039604 0,059081

24 Analysis & conclusions

When performing CFD simulations regarding a rocket, the first important task to consider
is the simplification of the rocket. Indeed, some details can be ignored in order to make
easier for instance the meshing of the rocket’s computational domain.
Regarding the meshing, the very practical Watertight Meshing tool is used, and enable the
user to have a precise control of the mesh refining, and of the boundary layer representation
(as mentioned thouroughly during the nosecone 3D CFD study). This last aspect is
crucial, as it will allow us to tune the y+ value at the rocket wall, which must be ideally
1, to have the best representation of the boundary layer around the rocket, and thus,
the best pressure distribution around this latter. In our particular case, as polyhexcore
elements give us the best results in terms of convergence during the CFD convergence
study, their output pressure and temperature fields were selected to be used as boundary
conditions for the FEA analysis.
For this second step, the most critical points to take into consideration regard the mesh
and the boundary condition. Actually, as rockets usually are shells filled with components

62
(motors, avionics,..), one must be really careful when creating the mesh. Indeed, the
elements must have acceptable quality, which can be possible only if their lengths in the
Y-Z plane (the rocket having X as axis), are similar to the depth of the rocket. This
results in the coarsest mesh being already refined enough.
Once this step is accomplished, an appropriate boundary condition must be selected,
alongside with the imported pressure and temperature (from the CFD simulation) and
the internal pressure applied. Here, we decided to artificially fix the COG, in order
to have a study which would be closer to reality. At the end, we discovered that the
maximum equivalent stress is located at the fins’ roots. In reality, these stresses should
be smaller, as the faces located transversely to the flow have been exaggerated during the
fins simplification.

63
Part IV

Synthesis
25 Main conclusion
Through this semester project, we have been able to follow the required steps to conduct
a FSI analysis regarding a rocket subject to a supersonic flow, namely:
• run CFD simulations to obtain the pressure and the temperature fields applied on
the rocket airframe;

• export these two fields as boundary conditions for the FEA analysis;

• run FEA simulations to obtain the stress distribution within the rocket
Before being able to perform these steps, we had to validate our results first by conducting
a 2D axisymmetric simulation on the nosecone only, then a 3D simulation still on the
nosecone. We used the wave angle to validate our results.
During this study, some critical points have been highlighted, namely:
• the importance of having a well refined mesh in the wake region to have the best
residuals’convergence;

• the parameters tuning in the Watertight workflow in ANSYS, where we should find
the best trade-off between mesh quality and y+ value on the nosecone/rocket wall;

• the particular care that has to be applied when choosing the importation options
regarding the imported pressure from ANSYS Fluent to ANSYS Mechanical;

• the relative superiority of the polyhexcore elements in the CFD simulations

It is also important to underline the luck that we had regarding the wave angle, as we
have been provided with some experimental results.

26 Futur improvements
As future improvements, some tasks that could be possibly performed to improve the
accuracy of our rocket behavior’s representation:
• simulate the temperature from the motor regarding the FEA part;

• simulate the rocket wall’s roughness;

• when some experimental results obtained with pressure taps would be available,
perform a comparison between these latters and numerical results from ANSYS.

64
Appendices
A Convergence Study CFD on the Computational Do-
main size
Geometry
The geometry used here for the rocket is the same as the one presented on Figure 32 at
page 38. However the computational domain has changed. Indeed, the convergence study
presented here considers 2 different Computational Domains. Theses latters are tabulated
on Table 21. The different named selections for the size settings are presented in Figure
92 and Figure 93

Figure 92: Rocket and Body of Influence used for the Convergence study.

65
 Figure 93: Lengths of the Geometry, tabulated in Table 21

Table 21: Two different Domain sizes used in the Convergence study
Length L1 [mm] L2 [mm] L3 [mm] L4 [mm] V1 : (lx , ly , lz ) [mm]
Domain 1 2814.6 2xL1 = 5629.2 5xL1 = 14073 2xL1 = 5629.2 (3000, 400, 400)
Domain 2 2814.6 4xL1 = 11258.4 10xL1 = 28146 4xL1 = 11258.4 (3000, 400, 400)

Physical modelling and hypotheses

All the fluid behaviour and turbulence models used before are also used in this study. We
can find some details in the General Knowledge section on page 2.

Boundary conditions, initial conditions and external forces

The boundary conditions are the same as the one used for the convergence study presented
in the main core of the report. Namely, all the rocket walls are defined as "stationary no
slip wall" and the inlet, outlet and far field are assigned as "pressure far-field", with a
Mach number of 2, a Gauge pressure of 101325 Pa and a constant temperature of 300 K.
As before, no external forces are applied, because we neglected the gravity.

Mesh
The mesh settings are the same for the two computational domain (Domain 1 and Domain
2). Actually in order to have representative results, only one parameter of the simulation
should change. The mesh settings used for both domains are presented in Tables 22, 23.
The named selections used for the local sizings are the ones defined in the main core of

66
this report on page 46. All the non-mentioned parameters are kept to default values.
The mesh generated with these settings are very similar, since the only thing that changes
is the size of the domain. The pictures of both meshes are presented on Figures 94 to 97.

Table 22: Local sizings and Boundary layer definition used for Domain 1’s mesh and
Domain 2’s mesh

Local sizing [mm] Boundary layers

conetip 2 Offset Type last ratio
cone 20 Nbr. of Layers 25
body 20 Transition ratio 0.272
enginemount 20 First Heigth [mm] 0.004
tailsarea 10 Geometry All body
tailsleading 2
tailsedge 2
tailstrailingedges 2
V1 15

Table 23: Surface mesh and Volume mesh settings used for Domain 1 and Domain 2

Surface mesh Volume mesh

Min. Size [mm] 50 Type poly-hexcore

Max. Size [mm] 3000 Peel layers 3

Size Functions Curv.&Prox Min. cell length [mm] 2

Curvature Normal angle 12 Max. cell length [mm] 2048

Cells per gap 3 Keep first boundary layer heigth ? YES

Scope Proximity to Faces-and-edges Default stair-step handling ? YES

Quality Improve Skewness Limit 0.4

Improve Surface Mesh

Face Quality Limit 0.5

Quality Improve Max Angle 160

Quality Improve iterations 10

Quality Improve Collapse SkewnessLimit 0.5

On the following figures presented just below, it is difficult to notice a difference

between the two mesh generated. This is logical because by increasing the size of the
Domain, only more cells are added but the structure of the mesh and the sizing around
the rocket are the same. This is also the reason why this convergence study is not very
interesting in terms of new results and perspectives.

67
Figure 94: Mesh on Domain 1 with a Figure 95: Mesh on Domain 2 with a
cross sectional X-Y plane. cross sectional X-Y plane.

Figure 96: Mesh on Domain 1 with a Figure 97: Mesh on Domain 2 with a
cross sectional plane around the fins. cross sectional plane around the fins.

Criterion
In order to assess convergence, we would use as before three different paramaters, namely:

• The drag force

• The maximum pressure Pmax

• The maximum temperature Tmax

Results
The different results of the convergence study are tabulated in Table 24. The residual
criterion is the pre-defined residual control that is setted in the Fluent settings. Iters to
converge corresponds to the number of iterations needed to stop the simulation automat-
ically.

68
Table 24: Results of the convergence study on the Computational domain size with
two Domains.

Type of mesh Fluid Domain 1 Fluid Domain 2

Nbr.of cells 847459 851190
Nbr.of Node 1669969 1672694
Drag Force [ N ] 1253.8 1259.5
Pmax [ Pa] 533350.9 532907.1
Tmax [K ] 543.7353 545.4802
Residual criterion 1 × 10−5 1 × 10−5
Iters to converge 220 136
Predefined iteration 1000 1000

We can see that the number of cells and the number of nodes increased logically due
to the increase in the size of the Domain. However, the computed quantities (Drag Force,
Max. Pressure and Max. Temperature) did not change a lot. This is also explained by
the fact that, the same mesh type and structure are generated around the nosecone for
both domains.
What changed a lot is the number of iterations needed to achieve residuals convergence.
Indeed, the bigger domain (Domain 2) converged faster than the Domain 1, as can be
seen on Figures 98 and 99. This phenomenon can be explained by the absence of wave
reflection. As we have seen with the Nosecone, the larger the computational domain is, the
lower probability of having wave reflection on the boundaries of the domain. Therefore,
one should consider to use maybe the Domain 2.

Figure 98: Residuals obtained with the Figure 99: Residuals obtained with the
Domain 1. Domain 2.

69
B CFD Tetrahedral mesh elements best residuals
In this appendix, we present briefly one of the simulations done with tetrahedral elements.
As said before, many type of mesh settings, computational domain dimension and body
of influences have been tried with the tetrahedral cell type. The reason was that we
absolutely wanted to reach a better residuals convergence. However, as explained before
in the main core of the report, tetrahedral cells provide lower convergence than the two
other types of elements (poly-hexcore and polyhedral).
The study presented here is the one that achieves the best residuals convergence with the
tetrahedral type.

Geometry
The geometry used for the rocket is the same as the one used in the whole report, and
can be found at page 38. The named selection for the rocket are the same as before. The
only thing that changes is the computational domain.

Computational Domain
The computational domain is presented in Figures 100, 101 and Figure 102. Due to some
particularities specifically present with the tetrahedral cell type, two bodies of influence
have been used. One just after the boattail to refine the mesh in the re-circulation region
and one bigger than the first one also downstream the rocket to reproduce the effect of
the Peel layers parameter. The values of the different dimensions are given in Table 25

Figure 100: Size definition of the computational domain. The 2 bodies of influence are
shown in the downstream of the Rocket. The smallest one is inside the biggest one.

70
Figure 101: Size definition of the first body of influence, used to reproduce the Peel
layers effect.

Figure 102: Size definition of the second body of influence, used to refine the re-
circulation region.

Table 25: Size of the computational domain used for the Tetrahedral simulation

Domain size [mm] Ratio to L1

L1 = 2814.6 1
L2 = 11258.4 4
L3 = 28146 10
L4 = 11258.4 4
V1 : (lx , ly , lz ) = (11000, 3000, 3000) -
V2 : (lx , ly , lz ) = (6000, 600, 600) -

71
Physical modeling and hypotheses
Like in the other CFD simulations, the fluid behavior and the turbulence model are
the same. More details are described in the other simulations and also in the General
Knowledge section on page 2

Boundary conditions, initial conditions and external forces

The boundary conditions used in this simulation are described in the main core of this
report, since they are similar. To briefly recap, these boundary conditions were validated
by the nosecone 2D case and the nosecone 3D case.

Mesh
The different mesh settings used in this simulation are presented in Table 26, Table 27
and Table 28. The named selections for the rocket are the same as the one used in the
core of the report for the other CFD Rocket simulation. The only named selection added
are wake, far-wake, inlet-outlet-farfield.

• wake : is the body V2 created to refine the high turbulence region downstream the
rocket.

• far-wake : is the body V1 used as body of influence to mimic the effect of the Peel
layer settings.

• inlet-outlet-farfield : is simply the named selection created to assign the body sizing
on the 3 following surfaces : inlet, outlet and farfield.

Table 26: Local sizing used for the tetrahedral mesh type

Named selection Growth Rate Size Control Type Mesh size[mm]

conetip 1.2 Face size 2
cone 1.2 Face size 20
body 1.2 Face size 40
enginemount 1.2 Face size 20
tailsarea 1.2 Face size 10
tailsleading 1.2 Face size 2
tailsedge 1.2 Face size 2
tailstrailingedges 1.2 Face size 2
wake 1.2 Body of Influence 15
far-wake 1.2 Body of Influence 200
inlet-outlet-farfield 1.2 Body size 1000

72
Table 27: Surface Mesh settings and Improve surface mesh settings used for the tetra-
hedral cell type

Surface Mesh
Minimum size [mm] 50
Maximum size [mm] 4000
Growth Rate 1.2
Size Functions Curvature & Proximity
Curvature Normal Angle 12
Cells per gap 3
Scope Proximity To faces-and-edges
Smooth Folded Faces/Repair Free Nodes Limit 10
Auto Assign Zone Types ? yes
Invoke Quality Improve? yes
Quality Improve Skewness Limit 0.4
Quality Improve Max Angle 160
Quality Improve Collapse Skewness Limit 0.85
Auto Remesh to Remove Clustering ? auto
Improve Surface Mesh
Face Quality Limit 0.45
Quality Improve Max Angle 160
Quality Improve Iterations 10
Quality Improve Collapse Skewness Limit 0.5

Table 28: Volume mesh settings and Boundary layers settings for the tetrahedral cell
type. The absence of the settings Peel layer justifies the use of the Body of Influence V1.
Volume Mesh Boundary Layers
Fill with tetrahedral Offset Method Type last-ratio
Max Cell length [mm] 6000 Number of Layers 25
Use size field ? yes Transition Ratio 0.272
Gap Factor 0.25 First Height [mm] 0.004
Max Aspect Ratio 25 Ignore Boundary Layers at Acute Angles ? yes
Min Aspect Ratio 1
Keep First Boundary Layer Height ? yes
Adjacent Attach Angle 80
Use default stair-step handling ? yes

The Peel layer setting is used in order to have a smoother transition between small
cells (around the rocket) and big cells (near the farfield). Without this option the mesh
quality is lower than the one obtained with poly-hexcore and also the computed quantities
(pressure, temperature,...) are not well estimated. The resulting errors will generate
slower residuals convergence. The mesh generated with the above settings is presented in
Figures 103 and 104.

73
Figure 103: Mesh created with tetra Figure 104: Mesh created with tetra
cells type. Around the Fins. cells type. X-Y plane

Results
On Figure 105 the residuals evolution is shown. We decided to not show here images of
others quantities (pressure, temperature,...) like before, since the tetrahedral type was
not selected for the FEA later. This appendix was done only to illustrate our decision
explained before during the mesh convergence study.
We can see that the continuity and the two parameters concerning the turbulence (k
and ω) did not achieve the convergence criterion of 1 × 10−5 . As the tetrahedral mesh
does not capture the pressure and velocity gradient well (due to the absence of a smooth
transition of the mesh), the error generated on the turbulence intensity is large in the
wake region.
The pre-defined number of iterations was 1000, but we have seen that after 300 iterations,
the residuals started to oscillate around a constant value. This can be seen on Figure 106.
Given the high residuals reached comparing to the ones obtained with poly-hexcore, and
knowing that the number of tetrahedral cells used here was 9773025, this explained also
why we decided not to use them in our simulations.

Figure 105: Residuals evolution for the Tetrahedral cells type. Number of iterations
predefined : 1000.

74
Figure 106: Last residuals values before we stopped manually the simulation, since no
improvement was achieved.

75
C FEA shell study beginning
This appendix was done to present briefly the beginning of a different type of FEA study,
namely representing the rocket as shell elements. Unfortunately, due to a lack of time and
some issues, we were not able to obtain sufficiently conclusive results to be presented in
the body of the report. However, many options were investigated and tried, so this can
be used as a starting point for possible future FEA simulations.
All the variables, parameters and settings related to the CFD used to compute the pressure
and the temperature fields, are not described in this appendix. The reason is that the
pressure and temperature fields exported to ANSYS Mechanical are the ones computed
with the CFD Domain 2 presented before in appendix, on page 70.

Geometry
Since we used shell elements, the geometry must be changed before starting ANSYS
Mechanical. Indeed, all the bodies should be represented as a thin surface with a thickness
of 1.73 mm, except for the fins which have a thickness of 7 mm. We also decided to use the
same simplified shape for the fins as it was done in other FEA simulation. The geometry
used is presented on Figure 107. The total length of the Rocket did not change logically.
The rocket is for this part, composed by 3 shell bodies, namely:

• The nosetip : This part corresponds to the nosetip solid part used before. However,
we decided to completely remove the filling inside in order to have a perfect surface
body.

• The fins : These 4 bodies are similar to the one used before. There are only modeled
as shell body.

• The body : This part is a combination of 3 sub-parts. It is composed of the nosecone

(see the left of the Figure), the cylinder body (which was already a shell) and the
boattail.

The geometrical transformations done in order to have this geometry are composed of :
Joining, Boolean unit, Welding, midsurface extracting. All these are in Design Modeler.

76
Figure 107: The 3 types of bodies that compose the rocket modeled in shell elements.

The last thing that should be mentioned concerning the geometry conversion from 3D
solid to 3D shell, regards the settings defined in the geometry Details windows in ANSYS
Mechanical. These latters are shown in Figure 108.

Figure 108: Geometry details windows in ANSYS Mechanical

Dimension :
This setting controls the dimension of each imported body. Here we use 3D even if
we model the rocket with shell elements.

Type de modèle :
This setting defines the model type. It can be a solid type or a shell type (in french
: coque). This setting is generally well controlled by the program.

77
Comportement de raideur :
The translation in english is Stiffness behavior. Therefore, if we consider a body as
undeformable with an infinite stiffness, we have to set this setting to rigid. Here we
deformed all our bodies, therefore we defined it as Flexible.

Option de rigidité :
This setting controls the type of shell analysis used. Indeed, there are many physical
models to describe a shell deformation behavior. ANSYS Mechanical proposes three
options : Membrane and Flexion / Membrane only / Stress computation only. Here
we selected the first option that couples shearing effect and flexion effect. If only
shearing effect are modeled, the second option must be selected.

Système de coordonnées :
This simply controls the coordinates system used.

Temperature de référence
This setting defines the reference temperature at which material properties are de-
fined. For example in our case the material deforms under the temperature field,
so we selected the option : By environnement (in French : par environnement), to
define the material mechanical properties with respect to the temperature imported
from the CFD.

Epaisseur :
This setting controls the thickness of the selected body. Generally, there is no
problem with this setting since the geometry is imported from Design Modeler.

Type de décalage :
This setting controls the offset for defining the shell body. In other words, if the
desired surface to be modelled is the mid-plane of the solid, the option should be
chosen similar to what can be found in Figure 108.

Physical modelling, hypotheses and external forces

For this simulation we decided to use the same material properties described before for
the rocket(see General Knowledge section for details : page 2).
The hypotheses used in this FEA are the same as the ones used before, namely : static
elastic steady-state analysis. The Mechanical solver uses FEM also with shell elements.
One relevant parameter to mention is that the option low-stiffness spring present in the
analysis settings windows of ANSYS Mechanical should be turned to Program controlled.
If not the simulation would crash.
As before, the computed quantities are the total deformation and also the Von Mises
stress. The unit system used is the one defined in General Knowledge section.

78
Boundary conditions
The boundary conditions applied in this study are quite similar to the ones used for the
Nosecone 3D FEA on page 20. We applied the Pressure field coming from the CFD
simulation and fixation. Due to some issues, the temperature field was not applied. This
should not be forgotten since it has an important impact on the fins. The Figure 109
presents the two boundary conditions used here.

Figure 109: Boundary conditions applied on the rocket modelled with shell elements.

The pressure imported from the CFD was applied by Direct method on the initial
surface (Dead Pressure analysis) like before.

Mesh
The mesh generation is done in two main steps. The first one is the classic one, namely
global mesh settings, local sizing, face meshing and method are done. Then the second
part, which is new with respect to what was done with the 3D solid elements, is called
node sharing at contact regions. This second part is explained later. Figures 110 to 119
show where the local sizing are applied. These local sizing are tabulated in Table 30. The
Global mesh settings are also described in Table 29.

Figure 110: Mesh ID : A Figure 111: Mesh ID : B

79
Figure 112: Mesh ID : C Figure 113: Mesh ID : D

Figure 114: Mesh ID : E Figure 115: Mesh ID : F

Figure 116: Mesh ID : G Figure 117: Mesh ID : H

Figure 118: Mesh ID : I Figure 119: Mesh ID : J

80
 Table 29: Global mesh settings

Physic Mechanics Mesh simplification yes

Element order Quadratic simplification size default
Element size 10 mm Transition slow
Adaptative sizing ? yes Span angle center smooth
Resolution 5 Inflation no

Table 30: Local sizings applied on the Rocket with the ID corresponding to the Pictures

Name Applied on Type size Behavior ID

NoseconeBack-edgesize 1 edge element size 8 mm soft A
JunctionBody-edgesize 2 edges element size 8 mm soft B
JunctionBody-Facesize 1 face element size 8 mm soft C
Boatail 2 faces element size 8 mm soft D
FinsLongEdges-edgesize 8 edges element size 5 mm soft E
FinsUpperEdges-edgesize 4 edges element size 5 mm soft F
FinsFaces-FaceSize 4 faces element size 5 mm soft G
Conetip-FaceSize 1 face element size 2 mm soft H
ConetipBack-edgesize 2 edges element size 8 mm soft I
ConetipBack-FaceSize 1 face element size 6 mm soft J

In addition to these settings, we also added the followings :

• Meshing method : Automatic Quadrilateral method applied on all the bodies with
quadratic element order and Quad/Tri free surface meshing type

• Face meshing : Face meshing applied on the face of the body (between nosecone
and fins) with a quadrilateral oriented method, to regularise more the mesh.

The mesh obtained is presented on Figure 120 for the global view and on Figure 121 for
the mesh on fins and on nosetip. A minimal orthogonal quality of 0.1398 was obtained
with this basic coarse mesh. The number of cells and nodes generated are 15896 and
47664 respectively. It must be kept in mind that the mesh presented in the following
figures is the mesh generated after the Mesh edition procedure that is explained in the
following subsection.

81
Figure 120: Mesh generated on the rocket modeled with shell elements - Global View.

Figure 121: Mesh generated on fins and on the nosetip - Enlarged view.

Contact definition
This subsection does not provide all the necessary and complete informations to have
a good rocket shell simulation. As mentioned before, due to lack of time the issues
encountered in this simulation were not resolved. What is written and explained in this
subsection provides a basic understanding of which parameters imply which effect on the
results.
This subsection investigates more the aspect of the definition of contact between bodies.
In the 3D solid elements simulation, we kept the contact settings by default, since the
simulations did not crash. However, it is better to have an understanding of what this
type of setting implies.
Contact between bodies can be managed in two different ways in ANSYS Mechanical.
The first way is to use the option called connection (just before mesh settings in the tree).
The second way is using the option mesh edition. The second way is what we recommend

82
to use, since this way provides contact definition and node sharing between bodies. This
latter increases the quality of the mesh at contact regions. The mesh edition settings are
presented in Figure 122. Since the screenshot was done on a DLL computer, the language
used is French.

Figure 122: Mesh edition settings details.

With these settings and applying an automatic connection detection, the connection
between bodies are automatically defined and the nodes between mesh are shared with
default settings. IT must be kept in mind that these settings still provide some issues,
because the geometry defined in Design Modeler is maybe corrupted. This needs to still
be investigated.

83
Results & Issues
Figure 123 shows the Equivalent Von Mises Stress field in the Rocket. We can see that
the maximum values are at the root of the fins with a max value of 6.5 MPa. However
this result should not be taken as correct for now for 2 reasons.
The first reason is that the temperature field when applied did not change the maximum
stress position and this is an illogical result. Indeed, we have seen in the body of this report
on page 43 that the temperature generated by viscous flow is high and can induce more
deformation in the material. The value should maybe not vary a lot since temperature
implies deformations in a material but does not generate more constraint.
The second reason is simply because the main body of the rocket disappeared. This is
an illogical result since no warning was generated during the computation. This means
that the defined settings provides a solution but it is not the one we expected. Therefore,
the settings and also the geometry definition in Design Modeler should be reviewed more
thoroughly.
Figure 124 provides the total deformation of the rocket with a symmetric deformation on
the nosetip and also a disappeared main body.

Figure 123: Equivalent Von Mises stress. Issue is that the main body disappeared from
the display window.

84
Figure 124: Total deformation in the rocket. Issue is that the main body disappeared
from the display window.

Conclusion on this FEA

The main advantage of doing an FEA in shell elements is that it requires much lower
memory than 3D solid elements. In addition, since only one element is added in the
thickness of each body, the orthogonal quality will be higher than for a 3D solid element
meshing. Indeed, the number of cells and nodes generated(15896 and 47664 respectively)
are very small compared to the ones created with the 3D solid elements hypothesis (see
Table 20 on page 62).
However, we have seen that doing an FEA with shell elements is not quite easy and it is
highly time consuming to understand each settings. With more time this study could be
compared to the one done with 3D solid elements.
To conclude, one futur possible improvement regarding this FEA, could be to keep the
nosetip as full solid body and also the fins, in order to have the possibility to define more
cells in the thickness.

85
D DLL access procedure
This appendix regards the access permission to the DLL computers. It must be kept in
mind that this procedure is important for the simulations we have done on the rocket,
since EPFL DLL computers have 16 logical processors.
One should first do the online formation accessible at the following link : https://
dllstisrv1.epfl.ch/remote-pc
Then a form has to be filled and signed by the supervisor and finally sent to ansys.
[email protected]. The following form is the one needed.

86
87
References
[1] Marangi K, Noca F. Mesh Optimisation for CFD of a Sounding Rocket [Semester
project for the EPFL Rocket Team.]; 2020.

[2] Grosgurin T, Noca F. An Introduction to CFD for Supersonic Flight [Semester

project for the EPFL Rocket Team.]; 2019.

[3] ANSYS FLUENT 12.0 Theory Guide - 5.2.1 Heat Transfer Theory;. Avail-
able from: https://www.afs.enea.it/project/neptunius/docs/fluent/html/
th/node107.htm.

[4] Fischer O, Voutilainen V. Optimization and Manufacturing of a Composite, Super-

sonic, Rocket Airframe [Semester project for the EPFL Rocket Team.]; 2022.

[5] Fluid Mechanics 101, editor. [CFD] How Fine Should My CFD Mesh Be?; 2019.
Available from: https://www.youtube.com/watch?v=60fDz2cVdy8.

[6] Noca F. Mécanique des fuides compressibles; 2019.

[7] ANSYS FLUENT 12.0 User’s Guide - 26.1 Overview of Using the Solver;. Avail-
able from: https://www.afs.enea.it/project/neptunius/docs/fluent/html/
ug/node776.htm.

[8] Andrieux E, Maitre E, Gauthier R, Miguel Dias Gomes Marques P, Noca F. Small
Scale Supersonic Wind Tunnel [Bachelor Project for the EPFL Rocket Team.]; 2022.

[9] ANSYS FLUENT 12.0 User’s Guide - 8.14.3 Setting the Operating Pressure;. Avail-
able from: https://www.afs.enea.it/project/neptunius/docs/fluent/html/
ug/node331.htm.

[10] Fluid Mechanics 101;. Available from: https://www.fluidmechanics101.com/

pages/tools.html.

[11] Determination of Shock Standoff Distance for Wedge at Supersonic Flow. Interna-
tional Journal of Engineering. 2019;32(7). Available from: http://www.ijeir.info/
article_91339.html.

[12] Eghlima Z, Mansour K. Effect of Nose Shape on the Shock Standoff Distance at
Nearsonic Flows. Thermophysics and Aeromechanics. 2016;23(4):499-512. Available
from: http://link.springer.com/10.1134/S086986431604003X.

[13] Staff AR. Equations, Tables, and Charts for Compressible Flow. NACA Report.
1953;1135:660.

88