Mechanical design of a rotating

mechanism support panel

Facoltà di Ingegneria Civile e Industriale

Corso di Laurea Magistrale in Ingegneria meccanica

Candidate
Giovanni D’Onofrio
ID number 1697308

Thesis Advisor Company Tutor

Prof.ssa Annalisa Fregolent Ing. Luigi Scialanga

Academic Year 2021/2022

This thesis has been typeset by LATEX and the Sapthesis class.
 iii

Abstract
The aim of this thesis is the preliminary design of a sandwich panel, conceived
to support a rotating mechanism and connect it to an Earth observation
satellite through aluminium brackets. The rotating mechanism includes a
flywheel whose purspoe is the balancing of vibrations induced by a rotating
antenna. The design of the panel is crucial for the satellite, considering the
high flywheel mass, the reached peak accelerations and the complex vibrational
environment during launch. In order to find an optimal configuration of
the panel, different core and skins thicknesses, core densities, number of
laminate plies and laminate stacking sequences were investigated. A carbon
fibre reinforced polymer (CFRP) and aluminium 7075 T6 were the evaluated
material for the skins, while aluminium was used for the honeycomb core. The
chosen design criteria are related to the modal decoupling between the system
and the satellite, and to weight minimization. The design is therefore based
on the value of the bending stiffness of the plate. Structural resistance to
the quasi-static loads exerted by the Vega C launcher was verified. Since the
system is excited by disturbs coming from the rotating mechanism imbalances,
a fatigue life assessment was needed on the brackets. The stress cycles caused
by the disturbances were identified on appropriate nodes by means of modal
frequency and transient analyses. The ESACRACK software was used to
calculate the damage accumulated during the entire load history of the system.
The optimal sandwich configuration involved CFRP symmetrical and balanced 8
ply laminates with stacking sequence [0/90/45/-45]s, and a 50.8 mm honeycomb.
A solution involving the same core and 0.5 mm aluminium skins can be adopted
in case of heat dissipation issues. It was verified that the critical point in
terms of fatigue of the brackets is a corner point, requiring the definition of a
proper fillet in a more detailed model. It also was determined that the flywheel
imbalances do not cause any fatigue damage to the structure.
 v

Contents

Introduction 1

1 Background on satellite structures 5

1.1 Architecture of a satellite . . . . . . . . . . . . . . . . . . . . . 5
1.2 Structural loads . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Wheel disturbances . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Laminate and sandwich . . . . . . . . . . . . . . . . . . . . . . 14

2 Composite theory 17
2.1 Micromechanics . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 Constitutive relation . . . . . . . . . . . . . . . . . . . 18
2.1.2 Elastic constants of a composite ply . . . . . . . . . . . 19
2.2 Macromechanics . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.1 Off-axis stiffness of a lamina . . . . . . . . . . . . . . . 20
2.2.2 Strain of a laminate . . . . . . . . . . . . . . . . . . . . 22
2.2.3 Constitutive equations of a laminate . . . . . . . . . . 23
2.3 Relevant stacking sequences . . . . . . . . . . . . . . . . . . . 25
2.3.1 Symmetric laminate . . . . . . . . . . . . . . . . . . . 26
2.3.2 Balanced laminate . . . . . . . . . . . . . . . . . . . . 26
2.3.3 Stacking sequence notation . . . . . . . . . . . . . . . . 27
2.4 Strength and failure criteria . . . . . . . . . . . . . . . . . . . 27
2.4.1 Strength of a unidirectional composite ply . . . . . . . 28
2.4.2 Failure criteria . . . . . . . . . . . . . . . . . . . . . . 28

3 Finite element model 31

4 Design of the panel 37

4.1 Dynamic decoupling . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Quasi-static loads . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 Conditions for the static verification . . . . . . . . . . . . . . . 41
4.4 Design procedure . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.5 Modes of the panel . . . . . . . . . . . . . . . . . . . . . . . . 45
4.6 Comparative study of admissible configurations . . . . . . . . 45
vi Contents

4.7 Analysis on the optimum configuration . . . . . . . . . . . . . 53

5 Fatigue verification of the brackets 61

5.1 Frequency analysis settings . . . . . . . . . . . . . . . . . . . . 61
5.2 Results of the frequency analysis . . . . . . . . . . . . . . . . 64
5.3 Determination of the stress cycles . . . . . . . . . . . . . . . . 69
5.4 ESALOAD and ESAFATIG . . . . . . . . . . . . . . . . . . . 71
5.5 Definition of the load history . . . . . . . . . . . . . . . . . . . 74
5.6 Fatigue damage assessment . . . . . . . . . . . . . . . . . . . . 80

Conclusions 85
 1

Introduction

An artificial satellite is a spacecraft orbiting around Earth. The employment

of satellites is crucial in a wide range of different applications. The main
examples are the technological sectors of telecommunications, navigation, re-
mote sensing and meteorology [16]. A satellite is pushed on the convenient
orbit by means of a launcher. The launcher is a rocket-powered vehicle di-
vided in stages, each providing the thrust during a certain phase of the flight.
The satellite is encapsulated at the top of the launcher, within the rocket’s
nose. After the lift-off from the launch pad stages are progressively ignited,
and are cut-off as they run out of propellant. The fairing which protects the
satellite is also jettisoned during the flight, as the satellite approaches the orbit.

Before, during and after launch satellite structures are subjected to many
different types of excitation, such as static, sine, random, shock and acoustic
loads. The induced vibrations are transmitted from the launcher to the satel-
lite through their interface, or via the air inside the fairing. The satellite is
composed of a main structure and many lighter subsystems, each characterized
by its own dynamic behaviour, depending on its mass and stiffness. One
of the most important goal in the design of space structures is therefore to
avoid coupling between dynamics of the satellite and that of its subsystems,
since the resulting resonances can be destructive. During launch the load
environment is the most critical, because of the high static accelerations and
the simultaneous existence of multiple loads of different types. A second
design goal is therefore to achieve structural resistance to these loads, ensuring
that structures are stressed in their elastic field. In space applications high
mechanical properties must always be combined with low density, since weight
is associated with the power consumption necessary to put the satellite in orbit.

Composite materials have been introduced in aerospace industry in the early

1960’s replacing metal alloys. Currently they are largely employed in space
structures, since they combine excellent strength and stiffness with low weight.
In this regard, Wijker has highlighted that carbon fibre composites are up to
five times stronger than steel, yet are five times lighter [23]. A wide range
of composite materials has been developed in the last decades. Carbon fibre
reinforced polymers (CFRPs) are among the most employed in spacecraft
2 Contents

structures. CFRPs are produced in very thin plies constituted of carbon fibre
filaments embedded in a polymeric matrix. Plies are stacked with various
orientations to form a laminate, and laminates can be attached at the ends of
a honeycomb to form a sandwich structure. In a sandwich laminates are called
”skins” and the central honeycomb is called ”core”. The sandwich structure
is designed to withstand in-plane bending loads. The honeycomb is a very
light modular structure, generally made of aluminium hexagonal prisms, the
purpose of which is to allow an efficient use of the skin material.

The present work is part of the project for the development of an Earth obser-
vation satellite by Thales Alenia Space. The satellite is equipped with a large
rotating antenna, which can generate a nutation motion of the spacecraft and
vibrations which can disturb and degrade the functioning of sensitive payloads.
In order to balance its motion, an electrically driven rotating mechanism has
been designed to be mounted on the satellite. The preliminary design of the
sandwich panel on which the rotating mechanism is to be mounted is the
subject of this study. The panel is equipped with brackets for its mounting on
the satellite, and must accommodate the interfaces of the rotating mechanism.
Since this is a preliminary study, the aim is to determine the project feasibility
and its main features, taking in consideration the dynamic, static and fatigue
behaviour.

In the first two chapters the theoretical background of the study is presented.
The general architecture, the types of load, the typical disturbances and the
most common materials for satellite applications are the topics of the first
chapter, together with the definitions of laminate and sandwich. The second
chapter covers the main arguments of composite theory, with the determina-
tion of the constitutive equations of composite lamina and laminate, their
specification for symmetric and balanced laminates and the description of the
main composite failure criteria.

In the following sections the work is presented, starting with the description of
the finite element model. A model of the panel has been built with Patran
software, while the brackets have been imported from an existing model. The
flywheel has been modelled as a point mass, having already been designed to be
decoupled from the panel. The design of the panel has been carried out follow-
ing a dynamic criterion, together with mass minimization and static resistance.
The dynamic criterion deals with dynamic decoupling, traduced in the need
to have the frequencies of the first bending mode of the panel above a certain
threshold. To do this, the bending stiffness of the panel has to be sufficiently
increased with respect to its mass. Structural resistance deals with quasi-static
loads, which are the worst combinations of static and dynamic loads exerted
at any phase of the mission. Static and modal analyses were performed with
Nastran finite element solver and Patran pre- and post-processor. The use
Contents 3

of CFRP for the sandwich skins has been investigated and compared with a
high strength aluminium alloy. Despite having lower mechanical properties
and higher density with respect to CFRP, aluminium is still an alternative for
its superior thermal conductivity. In fact, the heat generated by the electric
motor driving the rotating mechanism could cause heat dissipation issues. In
addition to the different skin materials and thicknesses, cores of different thick-
nesses and densities have been considered. An optimization of the laminate
stacking sequence was necessary in the case of CFRP skins. The laminate
stacking sequence has been determined with a logic of minimization of the
maximum failure index on the most loaded ply. The most satisfying solutions
have been found, and an optimum region has been identified. The maximum
loads exchanged at the panel-bracket and panel-flywheel interfaces have been
collected for a future design of the joints.

Once in orbit the system is excited by unavoidable disturbances due to the

imbalances of the rotating mechanism, whose continuous functioning has to
be guaranteed for 12 years. The imbalances cause rotating centrifugal loads
of very small magnitude, which can still be damaging in terms of fatigue due
to the high number of cycles. For this reason a fatigue verification of the
brackets have been carried out. The stress cycles on the potentially critical
nodes have been identified through a frequency domain analysis, and have been
used to perform a fatigue verification with the ESALOAD and ESAFATIG
tools. With these tools it was possible to assess the total accumulated damage
on the selected points, by considering not only the imbalances but the whole
load history of the system. An overall estimate of the load cycles has been
performed, considering the events of handling, transportation, qualification
tests and flight. A stress spectrum, containing the informations about every
stress cycle on the potentially critical points has been generated, and the stress
cycles due to the flywheel disturbances have been added to it. With these
informations the total accumulated damage has been determined, evaluating
the conditions which allow to ensure a life of 12 years on orbit.
 5

Chapter 1

Background on satellite
structures

In this chapter some general notions, useful for the understanding of the subject
covered in this text, are given. After a brief introduction on the main satellite
subsystems, the focus is moved to the structural domain, which is central for
a mechanical design. The main load sources on a spacecraft are described
together with their main examples, with particular attention to the flight phase
of the satellite-launcher system. The centrifugal load caused by imbalances are
then explained in detail, since they are the reason for the fatigue verification
carried out on the brackets. Towards the end of the chapter the most common
materials in the airspace industry are briefly explained. A deeper view is given
for composites and the laminate and sandwich configurations in which they
find relevant applications.

1.1 Architecture of a satellite

The basic infrastructure of a satellite is called bus. It consists of the satellite
structure itself and several subsystems [23] [9]:

• propulsion system, used to move from one orbit to another through

impulses supplied by a rocket engine;

• power system, which generates, stores, convert and distribute electrical

power to the other systems through solar arrays and secondary batteries
necessary to supply energy when the satellite is shadowed by the earth;

• control systems, in particular the attitude control system necessary to

sense and correct external perturbations through sensors and actuators,
maintaining a stable trajectory and keeping the antennas pointing towards
earth;
6 1. Background on satellite structures

• onboard computers and data handling system, comprising of an onboard

computer, avionics building blocks and other devices, for data storing
and processing and for traffic management;

• thermal control, composed of foil-like insulation blankets and radiators

which maintain the temperatures within the optimum range of the equip-
ment and prevent damages caused by thermal expansion or contraction;

• telemetry, tracking and command to exchange information with ground

stations, sending data about the status of the satellite and its position,
and receiving control signals from the ground.

The bus contains one or several payloads, namely the instruments or equipment
which performs the user mission [7]. An illustrative example of a satellite
exploded view is given in Figure 1.1 [8].

Figure 1.1. Exploded view of a spacecraft

The satellite structure is examined more in detail. It can be divided into

primary and secondary structure. The primary structure is composed of a
central cone or cylinder, which supports shear walls and equipment panels.
This central element is attached at its base to the last stage of the launcher,
interfacing it with the satellite and providing the basic load path during
prelaunch and launch phases. Propellant tanks, batteries, communication
equipment and other components of considerable masses must be attached
to sufficiently strong panels or trusses, which are in turn connected to the
primary structure. The secondary structure is made of side walls and enclosure
1.2 Structural loads 7

panels, which do not carry significant loads but provide support for smaller
components [5] [24].
The structure is the backbone of the satellite, providing:

• the required strength in order to bear high loads during the launch phase
and throughout the whole satellite life;

• sufficiently high stiffness to endure launch loads without excessive de-

formation, and in orbit where the presence of highly directive or flimsy
equipment impose stringent requirements on structural deformation;

• lightness and compactness, to put the satellite in orbit without excessive

effort;

• low thermal expansion, to minimize induced stresses due to heating from

solar exposure.

The shape of a satellite structure can depend on the type of attitude control.
The two main types of attitude control are satellite spin stabilization or three-
axis stabilization. The first one applies when the satellite axis has to be fixed
due to some mission requirements. The satellite attitude is be controlled with
a spin around this axis without additional effort. The structure in this case
has a cylindrical symmetry in order to balance forces and moments around
the spin axis. In the three-axis stabilization active attitude control is required.
Reaction wheels can be used to modify the orientation, each controlling the
rotation around one axis. For this reason the structure doesn’t need to have a
specific shape, and is often box-like.

1.2 Structural loads

Informations and guidelines for load characterization in space applications
are given in the ECSS handbook for spacecraft mechanical load analysis [19].
Structural loads characterize every stage of the life of a satellite before, during
and after launch. The main possible excitations on a spacecraft are:

• loads from handling, transportation and loads induced by the vibration

tests for qualification of the structure;

• loads from the coupling with the launcher;

• loads in orbit from the unfolding of elements, temperature gradients,

micro-meteorites or debris and pressure gradients.

The most critical environment takes place that during launch, due to the high
accelerations in lateral as well as in axial direction. Those excitation sources
are transmitted from the launcher to the spacecraft. The transmission can
8 1. Background on satellite structures

be structural via their interface, or acoustic via the air which surrounds the
satellite inside the fairing of the launcher. Once on orbit the acceleration levels
reduce drastically and, due to the decreasing air density, loads become purely
structural. Computing the micro-vibration environment which takes place on
orbit can be still useful in terms of performance requirements of high precision
pointing devices. Each launch phase causes a certain flight environment, which
is a source of load for the spacecraft. The intensity of the environments is
measured in terms of acceleration, and load distributions can be analysed
in the time or frequency domain, based on the convenience. The possible
environments generated during a spaceflight are:

• static or quasi-static acceleration, generated by static or quasi-static

load1 , measured at the centre of gravity of the system;

• low-frequency dynamic response (from 0 Hz to 100 Hz) to harmonic or

transient loads, expressed in terms acceleration response to a sinusoidal
input;

• high-frequency random vibration environment (from 20 Hz to 2000 Hz)

specified as a power spectral density function (PSD)2 , usually of the
acceleration;

• high-frequency acoustic pressure environment (from 20 Hz to 8000 Hz)

inside the launcher fairing, similar to a random environment since struc-
tures vibrate randomly in response to acoustics;

• shock events (above 500 Hz) expressed in terms of the shock response
spectrum function (SRS)3 , usually of the acceleration.

These environments can occur simultaneously especially during the flight, and
their occurrence is variable depending on the flight conditions. P. Berlin [5] gives
a detailed description of the circumstances in which these load environments
take place during launch.
Quasi-static loads are expressed in terms of load factors, which are dimensionless
multiple of the gravitational acceleration representing the inertial force acting
on a structure. Those factors are provided in the launcher vehicle user’s
manual, and are employed in the preliminary design of a spacecraft structure
[19]. A launcher is subject to both lateral and longitudinal static accelerations
during launch, reaching values above 4g (39 m/s2 ). The loads associated to
1
QSL are considered equivalent to static loads, but they take into account both static
loads and dynamic loads with frequencies sufficiently below the first natural frequency of
the structure
2
Power Spectral Density expresses the power distribution of a signal as a function of
frequency
3
Shock Response Spectrum is defined as the maximum response of a singe degree of
freedom system as a function of its natural frequency and for a given damping ratio
1.2 Structural loads 9

this accelerations, considered in their most severe combination and measured

at the spacecraft-to-launcher interface, constitute the QSL. Figure 1.2 shows
a typical longitudinal acceleration time history during the launch of Vega C
launcher vehicle, which considers the effect of wind and gusts [3]. The three

Figure 1.2. Longitudinal static acceleration profile of Vega C launcher

peaks in figure are associated to the time intervals between ignition and cut-off
of each solid propellant stage. The acceleration time history is not constant
in these three phases because of propellant consumption, which reduces the
system mass. When the first three stages have been jettisoned and the vehicle
is already out in space, the AVUM liquid propellant stage allows to place the
payload in the prescribed orbit with very small acceleration.
An example of sinusoidal low frequency vibration is the POGO effect4 , which
takes place during launch. The vibrations due to this effect derive from
resonances in the rocket fuel lines and can induce critical stresses on the
satellite. Acoustic noise is experienced in the first moments of lift-off and when
the system reaches transonic speed. In the first case pressure waves bounce
back from the ground in the launcher, while in the second case the acoustic
noise is generated in the boundary layer on the external fairing surfaces. In
addition to the random vibrations generated by acoustic phenomena mechanical
random vibrations can take place, such as that those coming from propellant
combustion in the engines. Shock loads result from very short and violent
transients, such as stage ignition and cut-off, and fairing jettison. A situation
similar to a pressure vessel can happen in the fairing as the external pressure
decreases, requiring the presence of venting to prevent high pressure differences.
4
PrOpulsion Generated Oscillations
10 1. Background on satellite structures

Out in space structural loads can be induced for example by thermal gradients
caused by solar radiation. This phenomenon has to be taken into account also
since in space no heat exchange through convection exists. Small meteorites
and orbital debris are abundant in space and can hit the satellite surface. Even
very minute solids can cause significant damage due to their high velocity (up
to 7.5 m/s). Moving parts can cause shock loads on the structure, like solar
panels during their unfolding. Other components like reaction wheels and
thrusters are sources of micro-vibrations which have to be taken into account
because they are severe disturbs for high precision instruments.
The frequency range of excitation is of primary importance in that it affects the
applicability of the analysis methodologies. The term ”low frequency” doesn’t
refer to a single and fixed range, but rather to the interval between 0 Hz (static
excitation) and the first natural frequencies of the analysed structure. The high
frequency range contains a great number of modes thus it is characterized by
high modal density. In the high frequency domain the contribution of a mode
to the total response is drastically reduced. For low frequencies of excitation,
applying a modal approach calculating each mode in the frequency interval
of interest is an efficient way to predict the structural response accurately.
That’s why in this cases Finite Element Analyses (FEA) are widely used.
At higher frequencies the use of FEA becomes progressively prohibitive in
terms of computational cost needed to reach sufficient accuracy. To study the
high frequencies other models such as the Statistical Energy Analysis (SEA)
stochastic model have been developed.

1.3 Wheel disturbances

In satellites the most relevant rotating devices are reaction wheels, needed for
attitude control. Every rotating mechanism on a satellites behaves in similar
way: the loads induced by their rotation are the of the same type than those
induced by reaction wheels. The typical disturbance sources of these rotating
mechanisms are [19]:
• mass imbalances of the rotating body;
• bearings imperfections;
• electric motor imperfections.
Mass imbalances are the most relevant sources of vibration for the spacecraft.
The main reasons are the induced micro-vibration environment interacting
with the dynamics of the rotating mechanism, and the possibility of fatigue
damage in structural components which have to work continuously in orbit
for many years. The focus here is on the second aspect, and the structural
components to be verified in terms of damage are the brackets.
A mass imbalance can be either static or dynamic. The names come from
1.3 Wheel disturbances 11

the fact that a static imbalance can be detected without putting the wheel in
rotation, while a dynamic imbalance shows up only on a rotating system. They
are both due to constructive asymmetries with respect to the axis of rotation,
and usually they can be minimized but not avoided during manufacturing.
The existing models for these disturbances are showed in Figure 1.3 and are
briefly explained [19]. An imbalance is said to be static when the rotor CoG is

Figure 1.3. Physical models of static and dynamic imbalance

eccentric of a quantity rs with respect to the spin axis [1]. This situation is
modelled with a mass ms representing the imbalance, placed at a distance rs
from the rotation axis. The rest of the rotor mass is perfectly balanced. This
mass put under rotation generates a centrifugal force rotating at the same
frequency as the flywheel. The force magnitude is:

Fs = ms rs ω 2 = Us ω 2 (1.1)

where Us is the static imbalance, measured in g · cm, and ω is the modulus of

the angular velocity vector.
In a condition of dynamic imbalance the CoG lies on the spin axis, but the
wheel is slightly tilted so that the principal axis of inertia is misaligned with
respect to the spin axis. Two masses md , separated radially by a 180◦ angle
and axially by a quantity d, are added to model this disturbance. The addition
of the masses does not affect the position of the CoG, but when put in rotation
they generate two centrifugal forces with an arm d. The resulting moment
vector Md rotates at the same frequency as the wheel. Its modulus is given in
Equation 1.2.
Md = md rd dω 2 = Ud ω 2 (1.2)
The term rd is the radial distance of the masses from the spin axis, and Ud
is the dynamic imbalance, expressed in g · cm2 . When a dynamic imbalance
12 1. Background on satellite structures

exists, the rotational axis of is no more a principal axis of inertia, and the
inertia tensor I is no longer diagonal. Its non diagonal terms are equal to Ud
[2].
The rotating centrifugal force and moment generally occur together, in the
plane orthogonal to the rotation axis. The moment vector is in quadrature
with respect to the force because of the vector product. The magnitudes are
both proportional to the square of the frequency, so in high speed devices
the centrifugal forces may become dangerous even with a small value of the
imbalances Us and Ud .

1.4 Materials
To meet the requirements specified in Section 1.1, several types of materials
have been developed, which in principle can be classified as metal alloys and
composites. Common metal alloys today used in spacecraft structures are [23]:
• aluminium alloys, widely employed for their moderate cost, light weight,
ductility, high specific stiffness and specific strength, good fatigue re-
sistance and high thermal conductivity. They are however sensitive to
corrosion and moisture, and their temperature range of applicability is
low, below 150°C;
• titanium alloys, with moderate density, good resistance to corrosion, good
mechanical properties which hold at high temperatures (up to 500°C)
and long fatigue life. The disadvantages are related to their high cost
and low ductility;
• steels, characterized by good mechanical properties and employed in
heavily loaded structure, where the main requirement is strength. Steels
are not widely used in aerospace applications due to their high density,
which is up to 2.5 times higher than aluminium alloys.
A variety of composite materials has been developed in the past decades to
overcome the mechanical behaviour of metal alloys. Composite materials are
systems usually composed of a filler imbedded in a matrix. These two main
elements together form a ply, which is generally very thin, below 1 mm. The
filler is constituted of fibres disposed along one direction. The consequent
anisotropic mechanical behaviour allows to obtain the best performances when
fibres are oriented in the load direction. Fibres provide strength, stiffness and
fatigue resistance and are extremely light. The most employed fibre material is
carbon (graphite), but also fibreglass or aramid (Kevlar) find applications. The
matrix is an homogeneous and ductile material. Its function is to hold fibres
together ensuring their simultaneous work under loading. Matrices can be
made of metal and ceramic, but polymeric materials are by far the most used.
Epoxy, a thermosetting polymer, is the most affirmed material for composite
1.4 Materials 13

matrices in spacecraft applications.

Fibre-Reinforced Polymers (FRP) constitute a particularly important class of
composites. CFRP, AFRP and GFRP belong to this class, their fibres being
respectively made of carbon, aramid and glass, with an epoxy matrix. Carbon
Fibre Reinforced Polymers are extremely stiff, strong and light, have small
expansion coefficient and are extremely resistant to corrosion and fatigue. For
those reasons they are widely used in spacecraft structures. Carbon-epoxy
composites do not soften or melt at high temperatures, but they loose strength
after oxidation. Depending on the temperatures reached during manufacturing,
carbon fibres are categorized as high-modulus or high-strength 5 . GFRPs are
cheaper but have lower elastic modulus, thus they are used whenever stiffness is
not a main design factor. Fibreglass is divided in E-glass and S-glass, the latter
being stronger and more expensive. AFRPs are instead used in applications
where impact resistance to high speed collisions (debris or micro-meteorites) is
required.
A comparison of tensile strength properties of the fibres used in the mentioned
composites [13] is given in Figure 1.4 and Table 1.1.

Material Tensile strength [MPa] Failure strain [%]

High-modulus carbon 3500–5500 0.1–1.0

High-strength carbon 3500–4800 1.5–2.0

S-glass 3500 4.5

E-glass 4600 5.0

Kevlar-49 3000 2.8

Table 1.1. Tensile properties of the main fibres used in aerospace applications

The failure stress of a composite is determined by fibres, since their strength is

up to 100 times higher than the matrix strength. From Figure 1.4 the brittle
behaviour of the materials is evident. No plastic behaviour is visible and the
curves are linear until failure, with the only exception of Kevlar. This means
that in composites no yielding occurs, and they are characterized only by their
failure (or ultimate) stresses and strains. The high Young’s modulus of carbon
fibres cause failure at low strain, while fibreglass can buckle much more before
breaking. The brittle behaviour of carbon fibres has an important drawback,
visible in Table 1.1. Their failure strength is variable in a wide range, because
this value is very sensitive to the presence of defects such as small voids and
5
The obtained elastic tensile modulus increases with the final processing temperature,
penalizing strength
14 1. Background on satellite structures

Figure 1.4. Tensile stress-strain curves of fibres used in aerospace applications

cracks. These imperfections are created when the thin filaments are joined into
bundles before being imbedded in the matrix. It is very difficult to prevent
this process.

1.5 Laminate and sandwich

Composites can be built in two configurations: laminate and sandwich [13].
A laminate is basically an overlapping of plies. This disposition allows to
efficiently withstand multidirectional loads through a tailoring of the mechanical
properties. This fundamental degree of freedom is given by the possibility of
stacking plies with different in-plane orientations, which correspond to the
longitudinal fibre directions. The task of the designer is to find the number,
order, thickness and orientations of the plies which optimally satisfy the
mechanical requirements. Obviously in the years some particular configurations
have been adopted in the majority of the applications, following criteria of
simplicity and effectiveness. The laminate is manufactured to be loaded in the
plane where the fibres lie, since its resistance to out-of-plane loads is much
lower.
The plies composing a laminate can be composites or metal sheets, and many
combinations find application. One peculiar configuration used in spacecrafts
is obtained stacking both metal sheets and fibre-polymer composite plies,
obtaining a Fibre-metal laminate (FML). The advantages of FMLs consist
in the opportunity to obtain a tailorable combination metal and composite
properties, reducing and balancing the disadvantages of both [21].
The typical structure of laminate and sandwich is reported in Figure 1.5.
1.5 Laminate and sandwich 15

An angle θ ranging from 0° to 90° is associated to each ply in the laminate,

representing the fibre orientation with respect to a reference in-plane axis
(x-axis in figure).

Figure 1.5. General structure of a laminate (a) and a sandwich (b)

The sandwich structure is built attaching two skins to the sides a central core,
through two adhesive films. The core is often an honeycomb, a low-density
modular structure composed of hexagonal prisms, with axes perpendicular
to the skins. The materials used for the core are mainly aluminium and
Nomex6 , but also fibreglass and CFRP are used. Aluminium is preferred when
low weight is a primary constraint and the component is not heavily loaded,
or when good thermal conduction is required between the two faces of the
sandwich. The skins can be anisotropic laminates or single plies of isotropic
metal alloys (mainly aluminium), while for adhesives epoxy or other polymers
are used. The peculiar properties of sandwich structures are the low weight
and buckling resistance provided by the honeycomb, and a good stiffness to
weight ratio. Designing it properly, a sandwich structure can ensure higher
6
material composed of aramid fibres in a phenolic resin matrix
16 1. Background on satellite structures

specific stiffness and strength than a FRP laminate alone.

A sandwich can be seen as the two dimensional version of an I-beam, therefore
the effectiveness of a sandwich configuration can be visible through this analogy.
The honeycomb core performs the function of the web in the I-beam, spacing
the skins from the bending neutral axis. The bending stiffness of the sandwich
is increased by the honeycomb, as the web does in the beam. This is possible
since, distancing the flanges from the neutral axis, the height of the resistant
section increases, together with the area moment of inertia of the section. Since
bending stiffness depends on the product of the Young’s modulus and the area
moment of inertia of the section, it increases increasing the thickness of the
honeycomb.
The use of a sandwich structure is convenient when the main constraint is to
reduce weight where bending loads are present, but also to ensure acoustic
damping, thermal insulation or conduction (employing composite or aluminium
honeycomb) and impact absorption. The skins can bear in-plane tension,
compression and shear stresses, depending on the plies orientation if they are
laminates. The core can withstand in-plane transverse loads, transmitting
them from a skin to the other, and low out-of-plane loads. The shear strength
of the core is fundamental to avoid independent bending of the two skins. The
core must also have sufficient compressive strength being the only element
which can withstand loads in the thickness direction [25]. The core is chosen
mainly for its low density, and its mechanical properties are much inferior to
those of the skins. That’s why sandwiches are designed for bending and not
for loads perpendicular to the skins. When subjected to bending moment, the
strength and bucking resistance of a sandwich increases with the core thickness,
with a very small increase in mass.
Failure in a sandwich structure [6] can be the classic mechanical failure of
the faces in tension or compression and of the core in shear, but can also
occur following other modes. Some examples are: global buckling, with a
sudden change of shape under a critical load; core crushing, when the core has
insufficient compression strength; local indentation due to concentrated loads;
face dimpling of local portions of the skin into the correspondent core cells; face
wrinkling, when only the skin buckles and debonding, with the detachment of
skin and core.
 17

Chapter 2

Composite theory

A large part of this chapter is taken from Prof. Gaudenzi’s lectures [11],
and from R. M. Jones’ book [12]. In this chapter only the most important
equations will be derived, and the fundamental path which allows to derive the
remaining ones will be explained. The mechanics of composites includes two
complementary theories, micromechanics and macromechanics. The first one
is the study of the properties of a single ply through those of its component
materials, while the second is about the stacking of plies to form a laminate and
the influence of the properties and disposition of the plies with the laminate
structural response.

2.1 Micromechanics
The composite ply is a continuous system composed of matrix and fibres, so
the mechanical properties of the lamina are combinations of those of these two
phases. The realistic hypotheses that each phase is isotropic, and homogeneous
and perfect bonding exists between the two phases are made. Assuming this,
to characterize the behaviour of the constituents, only Young’s modulus, shear
modulus and Poisson’s ratio are needed.

Ef
Ef , ν f , G f =
1 + νf

Em
Em , νm , Gm =
1 + νm

neglecting for simplicity the presence of small imperfections such as voids,

created during manufacturing, the percentages by volume of matrix and fibres
vm , vf are to be chosen, their sum being equal to the total volume of the ply
vtot .
18 2. Composite theory

2.1.1 Constitutive relation

The parameters so far mentioned have to be properly combined to form the
equivalent elastic parameters of the anisotropic lamina. These coefficients bind
stress and strain in the constitutive equations of the ply.
We define a (x1 , x2 , x3 ) coordinate system integral with the body, with the x3
axis normal to the lamina. The general relations for an isothermal situation
[18] in matrix form are:
σi = Cij · εj (2.1)
εi = Fij · σj (2.2)
where σi and εj are the stress and strain vector with their 6 components,
Cij is the stiffness matrix and Fij is the flexibility matrix, both 6x6 and
symmetric . In the general situation of anisotropic material the matrices are
characterized by 36 independent coefficients, which reduce to 21 due to inherent
symmetries. If the material is orthotropic, namely if it is characterized by
three mutually perpendicular planes of symmetry, the number of independent
stiffness parameters further reduces to 9.
The constitutive relation with strains as a function if stresses will be taken as a
reference, because in practical terms it is easier to impose stresses on the lamina
and measure its deformations. Matrix C can be found simply inverting F if
needed. The two matrices can be found through uniaxial tension or shear tests
on specimens, making sure that the principal axes of the lamina coincide with
the (x1 , x2 , x3 ) coordinate system. Equation 2.2 for an orthotropic material is:
1
− νE212 − νE313
 

ε1

E1
0 0 0 σ

 1 
− νE12 1
− νE323 0 0 0

 ε2    σ2 
 
1 E2
ν ν 1
 
 ε3 
  − 13 − 23 0 0 0  σ 
= E1 E2 E3  3 

1
 (2.3)
γ   0 0 0 0 0   τ23 
 
 23   G23  
γ31  1  τ 
 0 0 0 0 0
  
G31  31 
γ12 1 τ12
0 0 0 0 0 G12

It is clear that an orthotropic material stressed along its principal coordinates

does not exhibit shear-extension or shear-shear couplings, but only simple
extension and simple shear (terms on diagonal), and extension-extension
coupling (non-zero terms off diagonal). Note that the pairs of correspondent
non-diagonal terms must be equal.
The plies have the thickness dimension (along x3 ) much shorter than the other
2, and they are designed to bear in-plane loads. For these reasons, it is possible
to consider the ply in a state of plane stress. With this assumption we can
make some simplifications on Equation 2.3, that are written in Equation 2.4.
σ3 = τ13 = τ23 = 0 (2.4)
The stiffness matrix C for a plane stress state is called reduced stiffness matrix
Q. The flexibility and stiffness matrices for the orthotropic lamina under plane
2.1 Micromechanics 19

stress become:
− νE121
 1
0

E
 ν112 1
F = − E1 E2
0 
 (2.5)
1
0 0 G12
 
E1 ν12 E2 0
1
Q= ν12 E2

E2 0 
(2.6)
1 − ν12 ν21

0 0 G12 (1 − ν12 ν21 )
where only the first, second and sixth rows have been extracted from Equation
2.3. The remaining out-of-plane strains γ23 and γ31 are zero, while ε3 can be
found from the third row of Equation 2.3 applying the plane stress condition.
The direction of fibres x is be called longitudinal direction L, and y is the
transverse direction T . The ply lies on the LT plane, and the rewritten
constitutive relation are:
 
1
− νETTL 0
  
εL  EνLLT
σ
 L 
1
− EL
 εT  =  0   σT  (2.7)
  
ET
1
γLT 0 0 G
τLT
LT

in which, since the flexibility matrix is symmetric, νELTL = νETTL . Because of

the presence of longitudinal fibres, it is also true that EL >> ET , therefore
νLT >> νT L .

2.1.2 Elastic constants of a composite ply

Micromechanics relations can now be written. They express the coefficients of
the matrix F and C as linear combinations of those of fibres and matrix. This
is performed writing compatibility equations on strains1 (εL , εT , γLT ), and
relations on the stresses (σL , σT , τLT ) coming from the equilibrium of forces
in the decoupled cases of longitudinal traction, transverse traction and shear
force. In doing this, matrix and fibres are treated as springs. In the case of
longitudinal traction the springs are in parallel, while for the transverse traction
and shear cases they are in in series. Substituting each equilibrium equation
into the correspondent cinematic compatibility equation, the Equations 2.8 are
obtained.
EL = Ef vf + Em (1 − vf )

Ef Em
ET = (2.8)
Ef vm + Em vf
1
A compatibility equation is written on the displacements, and displacements are related
to strains through the definition of strain. The compatibility equation expresses the total
displacement in a certain direction as a function of the displacements of fibres an matrix
20 2. Composite theory

Gf Gm
GLT =
Gf vm + Gm vf

Note that in the first two equations the term containing Em is much smaller
than the one with Ef . In the first equation it means that EL is fibre dominated,
while in the second it means that ET is matrix dominated. In the third relation
GL T is matrix dominated, because Gm is much smaller than Gf . This means
that an orthotropic ply behaves mainly as its fibres with respect to longitudinal
stresses, and as its matrix with respect to transverse and shear stresses. For
the Poisson’s coefficients, their definitions and the compatibility equation on
the transverse direction allow to obtain the poisson equivalent coefficient of
the ply.
νLT = νf vf + νm vm (2.9)
With these four relations one can derive the elastic constants of the ply knowing
those of the constituents, and then write the stiffness and flexibility matrices
under the made assumptions.

2.2 Macromechanics
In this section the fundamental equations which govern the elastic behaviour
of a laminate will be derived under some simplifying assumptions on stresses
and strains. The hypothesis allow to reduce a three-dimensional problem to a
simpler two-dimensional one, with good confidence on the obtained model.
The basic assumption is of perfectly bonding in the laminate, so that the
laminae work perfectly together when stacked on the laminate. This can
be true if the adhesives are strong enough. A second hypothesis is that the
displacement distribution is continuous at the plies interfaces, so that no
relative slip is allowed. Under these assumptions the laminate acts as a single
plate.
The aim is to derive the stress-strain relation for the single layers for a general
in-plane orientation, and define the forces and moments on a laminate through
those on the individual plies. Combining together these 2 systems of equations
we will be able to derive the stiffnesses of a laminate.

2.2.1 Off-axis stiffness of a lamina

Until now only the x1 axis has been considered to be in the fibres direction.
In engineering applications however plies are stacked with different fibre ori-
entations in order to minimize the material usage and optimize the laminate
strength in the required directions, depending on the applied loads. The
principal reference system (x1 , x2 , x3 ) used until now for a single lamina, is
usually rotated by an angle θ with respect to to the (x, y, z) coordinate system
2.2 Macromechanics 21

taken as representative of the entire laminate. We want to determine the

constitutive relation of a rotated lamina in the laminate coordinate system.
The transformation matrix T relates the stresses on the (x1 , x2 , x3 ) reference
system to those in (x, y, z).

cos2 θ sin2 θ
 
2cosθsinθ
2
T =  sin θ

cos2 θ −2cosθsinθ   (2.10)
−cosθsinθ cosθsinθ cos2 θ − sin2 θ

The subscripts 12 and xy will be used to distinguish among quantities in the

two defined coordinate systems. Equations 2.11 and 2.12 express stresses or
strains from one reference system to the other.
   
σ1 σx
 σ2  = T  σy  (2.11)
   

τ12 τxy
   
ε1 εx
 ε2  = T  εy  (2.12)
   

ε12 εxy

Recall the definition of stiffness matrix in plane stress, which is Equation 2.6.
    
σ1 Q11 Q12 0 ε1
σ 12 =  σ2  = Q12 Q22 0   ε2  = Q ε12 (2.13)
    
12
τ12 0 0 Q66 γ12

Shear strains are defined as γij = 2εij . The matrix R allows to pass from
the shear vector containing ε12 (or εxy ) to that containing γ12 (or γxy ). This
matrix allows to write Equation 2.14.
     
ε1 ε1 εx
−1  −1
ε12  ε2  = R  ε2  = RT R  εy  = RT R ε xy
= (2.14)
   

γ12 ε12 γxy

Equation 2.11 is inverted, and Equations 2.13 and 2.14 are substituted into it.

σ xy = T −1 σ 12 = T −1 Q ε12 = T −1 Q RT R−1 εxy (2.15)

12 12

σ xy = Q εxy (2.16)
xy

We are dealing with lamina whose principal coordinates are rotated of an angle
θ with respect to a chosen reference (x, y, z). The matrix Q contains the
xy
22 2. Composite theory

transformed reduced stiffnesses, and allows to write Equation 2.16, which links
stresses and strains of the plate in the (x, y, z) coordinate system. The matrix
is in general full.
 
Q11 Q12 Q13
Q = Q21 Q22 Q26  (2.17)
 
xy
Q61 Q62 Q66

The Qij are made explicit in Jones’ book [12], and they have the properties:

Q16 (−θ) = −Q16 (θ) Q26 (−θ) = −Q26 (θ) (2.18)

while for the other coefficients Equation 2.19 holds.

Qij (−θ) = Qij (θ) (2.19)

Knowing Q for any ply, it is possible to calculate the stiffness matrix of the
xy
laminate. Omitting for simplicity the (xy) subscripts, the stress-strain relation
for the k th layer of a composite laminate, oriented of an angle θk , is given by
Equation 2.20.
σk = Q εk (2.20)
k

It has to be noticed that while for the hypothesis of perfect bonding the strain
distribution through the laminate thickness is a line, the stress distribution is
not, since the Qij are in general different for each lamina. This means that
in the xz or yz plane the strain distribution will be a line, while the stress
distribution will be a broken line.

2.2.2 Strain of a laminate

Since the laminate is a thin plate loaded in the elastic field, it is reasonable
to introduce the Kirchhoff hypotheses: during deformation each straight line
perpendicular to the middle surface remains straight and perpendicular, and
thickness doesn’t change. A plate is considered to be thin when its length and
width are greater than 10 times the thickness. The Kirchhoff’s assumptions
are put in mathematical form in Equations 2.21.

γxz = γyz = 0 εz = σz = 0 (2.21)

Let (u, v, w) be the components of the laminate displacement along the (x, y, z)
directions, and mark with the 0 subscript the quantities referring to the middle
plane. As a result of the Kirchhoff’s hypotheses, the displacements at any
point of the plate is a linear function of the z coordinate.
∂w0
u(x, y, z) = u0 (x, y) − z
∂x
2.2 Macromechanics 23

∂w0
v(x, y, z) = v0 (x, y) − z (2.22)
∂y
w(x, y, z) = w0 (x, y)
By derivation of the displacements, strains are obtained.

∂u0 ∂ 2 w0
εx = −z
∂x ∂x2
∂v0 ∂ 2 w0
εy = −z (2.23)
∂y ∂y 2
∂u0 ∂v0 ∂ 2 w0
γxy = + − 2z
∂y ∂x ∂x∂y

The non-zero strains of the middle surface (ε0x , ε0y , γxy

0
) and its curvatures
(χx , χy , χxy ) can be recognized in previous equations2 , and the system of
Equation 2.23 can be written in matrix form.

ε0x
     
εx χx
 0
 εy  =  εy  + z  χ y  (2.24)
   
0
γxy γxy χxy

2.2.3 Constitutive equations of a laminate

The only missing relation needed characterize of a laminate is between resultant
forces and moments on the mid plane and the integrals of the corresponding
induced stresses. For a single ply, the possible loads under the Kirchhoff’s
hypotheses are showed in Figure 2.1.

Figure 2.1. Forces and moments on the mid plane of a lamina, general case
∂uj 2 2
2
εi = ∂ui
∂xi ; γij = ∂ui
∂xj + ∂xi ; χi = − ∂∂xu2k ; χij = −2 ∂x
∂ uk
i ∂xj
i
24 2. Composite theory

These loads are forces and moments per unit of width. The integrals of the
induced stresses along thickness, that is between the positions zk−1 and zk ,
must be equal to the resultant forces and moments. Recalling Equations 2.20
and 2.24 we can write:
       
Nx Z zk σx Z zk Q11 Q12 Q16 εx
 Ny  =  σy  dz = Q12 Q22 Q26   εy  dz = (2.25)
       
zk−1 zk−1
Nxy k τxy k Q16 Q26 Q66 k γxy

Q11 Q12 Q16  Z zk ε0x

     
χx

Z zk
 0
= Q12 Q22 Q26   εy  dz +  χy  zdz 
    
zk−1 0 zk−1
Q16 Q26 Q66 k γxy χxy

       
Mx Z zk σx Z zk Q11 Q12 Q16 εx

M
 y

= σ
 
 y zdz =

Q
 12 Q22 Q26   εy  zdz =
  
(2.26)
zk−1 zk−1
Mxy k τxy k Q16 Q26 Q66 k γxy

Q11 Q12 Q16  Z zk ε0x

     
χx

Z zk
 0  2
Q12 Q22 Q26  
=  εy  zdz +  χy  z dz 
 
zk−1 0 z k−1
Q16 Q26 Q66 k γxy χxy

where it has been assumed that the ply stiffness matrix is constant within the
ply itself. This is true unless a temperature or moisture gradient exists along
its thickness.
For the entire laminate of total thickness h, composed of N plies of thickness
hk = zk − zk−1 , forces and moments are expressed in Equations 2.27 and 2.28.

Q11 Q12 Q16  Z zk ε0x

       
Nx χx

XN Z zk
 0
 Ny  = Q12 Q22 Q26   εy  dz +  χy  zdz  (2.27)
      
z k−1 0 z k−1
Nxy k=1 Q
16 Q26 Q66 k γxy χxy

Q11 Q12 Q16  Z zk ε0x

       
Mx χx

N Z zk
X  0  2
 My  = Q12 Q22 Q26    εy  zdz +  χy  z dz  (2.28)
    
z k−1 0 z k−1
Mxy k=1 Q
16 Q26 Q66 k γxy χxy

The middle plane strains and curvatures are constant with z, so they can be
taken outside the integral. The product of Q and the integrals gives the
k
stiffness matrices of the laminate.

ε0x
       
Nx A11 A12 A16 B11 B12 B16 χx
 0  
 Ny  = A12 A22 A26   εy  + B12 B22 B26   χy  (2.29)
    
0
Nxy A16 A26 A66 γxy B16 B26 B66 χxy
2.3 Relevant stacking sequences 25

ε0x
       
Mx B11 B12 B16 D11 D12 D16 χx
 0  
 My  = B12 B22 B26   εy  + D12 D22 D26   χy  (2.30)
    
0
Mxy B16 B26 B66 γxy D16 D26 D66 χxy

The coefficients of the 4 stiffness matrices are explicitly stated in Equations

2.31.
N
X
Aij = (Qij )k (zk − zk−1 )
k=1
N
1X
Bij = (Q )k (zk2 − zk−1
2
) (2.31)
2 k=1 ij
N
1X
Dij = (Q )k ((zk3 − zk−1
3
)
3 k=1 ij

The Aij terms are the extensional stiffnesses of the laminate. A11 and A22 are
related to extension and A66 to shear. The term A12 is the extension-extension
coupling term, while A16 and A26 represent the shear-extension coupling. The
Bij coefficients couple extension and bending, and they constitute the principal
difference in the behaviour of a laminate with respect to the lamina. Their
presence means that loading a laminate with an extensional or shear force
can result in bending (B11 ,B12 ,B22 ) and twisting (B16 , B26 ,B66 ), and also that
a laminate subjected to moment will deform on its plane by extension and
shearing strain. The Dij are bending stiffnesses. The bending terms are D11
and D22 , while D66 is related to twisting. D12 is the bending-bending coupling,
while D16 and D26 represent the bending-twisting coupling.
Equations 2.29 and 2.30 are the constitutive equations for a composite laminate.
They express its mechanical behaviour, which appears in general to be very
complex due to the presence of many coupling terms. That’s why in practice
engineers search for laminate configurations which allow to delete the coupling
terms, reducing the number of stiffness coefficients and giving some practical
advantages in the use of laminate and sandwich structures.

2.3 Relevant stacking sequences

Equations 2.29 and 2.30 allow to calculate strains and curvatures knowing
forces and moments, and vice versa. Then we can use Equations 2.24 and
2.20 to find the stress vector of each lamina and Equation 2.11 to refer it to
the principal coordinates of the lamina. We can then use one of the failure
criteria present in the next section to verify whether failure occurs or not. In
design the goal is to find the values for the laminate stiffnesses which satisfy
Equations 2.29 and 2.30 with known loads and allowable strains and curvatures
given by the chosen materials and geometry. This task can be complex in the
26 2. Composite theory

general case where all the coupling terms exist, but in space applications some
particular configurations are employed, which simplify the design.

2.3.1 Symmetric laminate

A laminate is said to be symmetric if plies are stacked with orientations
symmetric with respect to the mid plane. The material used for any of the
pairs of symmetric plies must be the same. Opposite plies have the same
in-plane orientation θ, same thickness and material. The mid ply will be ideally
cut in 2 symmetric halves by the mid plane. Opposite laminae have the same
(Qij )k , and it can be showed that when added up for the N plies, the second
integral of Equation 2.27 and the first integral of Equation 2.28 become null.
The Bij are zero, and Equations 2.29 and 2.30 can be rewritten.

ε0x
    
Nx A11 A12 A16
 0 
 Ny  = A12 A22 A26   εy  (2.32)
  
0
Nxy A16 A26 A66 γxy
    
Mx D11 D12 D16 χx
 My  = D12 D22 D26   χy  (2.33)
    

Mxy D16 D26 D66 χxy

The bending-extension coupling is no more present in the laminate, with two

important implications. The first is that equations have been simplified and
the complexity of the laminate design is significantly reduced. The second and
more important is that these laminates are simpler to manufacture. Symmetric
laminates do not bend or twist when subjected to thermal loads. In particular
during the cooling of the laminate, needed after the curing process3 , a flat
symmetric laminate will remain flat, while an asymmetric one will bend or
twist, coming out curved from the manufacturing process. For these reasons
symmetric laminates are normally employed, unless there are some special
requirements which prescribe to use an asymmetric laminate. It has to be
specified that symmetric laminates can present extension-bending coupling
when subjected to a thermal gradient. About that, remember that laminate
theory has been derived under the hypothesis of absence of any thermal
gradient.

2.3.2 Balanced laminate

Laminates are balanced if they have a ply oriented at −θ for each lamina at
an angle θ which is not 0° or 90°. The ±θ pairs must have equal thickness,
3
Curing is the process in which the polymeric matrix is consolidated and the composite
mechanical properties are improved. The matrix already impregnates the fibres, and the plies
are stacked together. During the cure cycle the laminate is heated and pressurized following
suitable cycles. The polymerization reaction taking place in the matrix is exothermic[4].
2.4 Strength and failure criteria 27

and the 2 plies must not necessarily be adjacent. If a balanced laminate is

also symmetric, 2 pairs of ±θ oriented plies must exist for each lamina whose
in-plane angle is not 0° or 90°. From Equations 2.18 follows that a symmetric
and balanced laminate will have A16 = A26 = 0, but generally D16 ̸= 0 and
D26 ̸= 0[25]. Symmetric and balanced laminates are widely used in composite
structures, and for them Equations 2.29 and 2.30 become:
ε0x
    
Nx A11 A12 0
 0 
 Ny  = A12 A22
  
0   εy  (2.34)
0
Nxy 0 0 A66 γxy
    
Mx D11 D12 D16 χx

M
 y

=

D
 12 D22 D26   χy 
 
(2.35)
Mxy D16 D26 D66 χxy

2.3.3 Stacking sequence notation

The laminate coordinate system used to define the plies orientations, is in
principle chosen arbitrarily. However to simplify the communication and
understanding of the composition of a laminate, some guidelines are adopted
[15]. The x-axis is generally chosen to be the longitudinal axis, with the y-axis
in the transverse direction. The x-axis is in the direction of primary loading,
that is the direction in which the fibres of the principal load bearing plies
lie. Those plies are characterized then by θ = 0°. The stacking sequence
nomenclature consists in a series of numbers between square brackets. The
numbers represent the value of the angle θ between the reference system and
the fibres direction. For convention clockwise rotations on the xy plane are
positive angles. The order of the θ values reflects the stacking sequence from
the top to the bottom of the laminate. For a symmetric laminate only half
of the sequence can be written, and an s subscript is added at the end of the
sequence. The notation ±θ indicates that the +θ ply is at the top of the one
with −θ, or for a symmetric laminate that +θ plies are external with respect to
the mid plane, while −θ plies are internally adjacent to them. Any repeating
subsequence can be included between round squares, and a subscript is added
to indicate the number of repetitions of the subsequence.

2.4 Strength and failure criteria

The different load conditions to which a composite laminate is subjected during
its life can cause a variety of effects, which determine a permanent reduction
of strength and stiffness of the single layers and therefore of the laminate.
These effects are called damages. Damages are formed at a micro scale and
can coalesce and grow, resulting in the macroscopic failure of the ply. The
principal failure modes are matrix crack, fibre fracture or buckling, fibre-matrix
28 2. Composite theory

debonding, fibre pull-out and delamination.

Models to predict microcracks and other damage sources at the micro scale
are prohibitively complex. The focus is then on the global response of the
structure, that is on the failure at ply level[18]. Failure criteria just verify if
allowable stresses are exceeded, ignoring the microscopic effects which caused
it. Failure criteria provide a simple tool to predict the complex phenomenon
of composite failure.

2.4.1 Strength of a unidirectional composite ply

For isotropic materials principal stresses and strains are the largest values
regardless of material orientation. For orthotropic materials instead they
depend on load direction, so the actual strength offered by the composite might
not be the highest one, depending on fibre and load directions. If a traction
load is applied on fibre direction the allowable stress will be much higher than
if it is applied transversely, because in the second case the strength of the
lamina is matrix dominated. This behaviour imposes to choose a reference
coordinate system with respect to which calculate the strengths. The ply
principal material coordinates (x1 , x2 , x3 ) are the natural choice, and stresses
in a rotated reference system can be found with Equation 2.11.
To define the allowable stresses, it must be noticed that composite materials
have in general different behaviour in tension and compression. For a lamina in
a plane stress state 5 allowable stresses have to be determined in the principal
coordinate system. They are:
• Longitudinal strengths in tension or compression, Xt and Xc ;
• Transverse strengths in tension and compression, Yt and Yc ;
• Shear strength S.
To find these allowable stresses several tests are performed, with independent
application of the stresses ±σ1 , ±σ2 and τ12 . Note that shear strength is
independent on the sign of shear stress, but this is true only in the principal
coordinate system. The same tests are used to find the stiffness characteristics
(E1 , E2 , ν12 , ν21 , G12 ). Each of the five listed values represent a failure
mode, which is independent from the others. A composite fails if any of the
independent stresses induced by the loads exceed the correspondent strength.
To define the composites failure criteria all these strengths must be taken in
consideration, unlike the criteria for isotropic materials, where just the material
yield or ultimate strength is considered.

2.4.2 Failure criteria

Since in applications the stress state is biaxial or triaxial, the uniaxial strengths
obtained in the principal coordinates have to be combined in some way and
2.4 Strength and failure criteria 29

must be referred to a general lamina orientation. A failure criteria consist in a

analytical definition of a failure surface on the principal stress space (σ1 , σ2 , τ12 )
within which failure does not occur with a certain confidence. The equation
for this surface is obtained through a curve-fitting of failure experimental data.
The failure surface intersections with the 3 axes are the strength values of the
material, therefore failure surfaces are not symmetric with respect to the σ1 and
σ2 axes. All the criteria are based on the stress state except for the maximum
strain criterion. Some criteria are more suitable for certain composites, as
testified by comparison of failure surfaces with failure data. In all the criteria
the orthotropic material is regarded as homogeneous. The principal failure
criteria are listed.

• Maximum stress criterion: to avoid failure each of the principal stresses

must be consistent with the respective strengths both in tension and
compression.

Xc < σ1 < Xt Yc < σ2 < Yt |τ12 | < S (2.36)

Each of the five inequalities is related to a failure mode. This criterion

does not match correctly experimental data, mainly because of the
absence of any interaction between the stress components, that is between
failure modes.

• Maximum strain criterion: each of the principal strains must be consistent

with the ultimate strains in tension and compression.

Xε,c < ε1 < Xε,t Yε,c < ε1 < Yε,t |γ12 | < Sε (2.37)

The discrepancies of this criterion with experimental data are similar to

those of the maximum stress criterion.

• Tsai-Hill criterion: a quadratic polynomial criterion which is an extension

for orthotropic materials of Von Mises yield criterion. It has been
proposed for a general triaxial stress state and, but is here presented in
the form specialized for plane-stress.

σ12 σ1 σ2 σ22 2
τ12
− + + <1 (2.38)
X2 X2 Y2 S2

The material is supposed to fail if in the three-dimensional stress space

the point identified by the actual stresses is outside the surface described
by the previous equation. The values for the principal strengths X and
Y to be used depend on the signs of the normal stresses. This criterion
fits with experimental data better than the previous two. The principal
limitation of this model is that it does not properly account for the
different behaviours in tension and compression of the material.
30 2. Composite theory

• Hoffman criterion: develops the Tsai-Hill quadratic model adding linear

terms to account for the different properties in tension and compression.
The criterion is presented for a plane stress situation.

σ12 σ1 σ2 σ2 Xc + Xt Yc + Yt τ2
− + − 2 + σ1 + σ2 + 122 < 1 (2.39)
Xc X t Xc Xt Yc Yt Xc Xt Yc Yt S

The presence in the equation of different strengths for tension and com-
pression implicate a difference in the stress space representation with
respect to Tsai-Hill. The limit surfaces into which the criteria are satisfied
are both ellipsoids, but Hoffman is represented by a single ellipsoid while
in Tsai-Hill 4 different portions of ellipsoid are traced in each of the
quadrants of the σ1 − σ2 plane. Those portions together form a unique
surface, but they are different in slope. For equal strengths in tension
and compression the two criteria become equivalent.

• Tsai-Wu criterion: a tensor polynomial failure criterion, which accounts

for differences in tension and compression. The surface is represented by
a complete quadratic tensor polynomial, with an increased number of
terms with respect to the previous criteria, therefore improved adherence
to experimental data. to avoid failure. As for the previous, the criterion
is written for plane stress.
! !
1 1 1 1 1 1 2 1 2
+ σ1 + + σ2 − σ12 − σ + τ +2F12 σ1 σ2 < 1
Xt X c Yt Yc X t Xc Yt Yc 2 S 2 12
(2.40)
The coefficient F12 has a particular expression:
" ! ! #
1 1 1 1 1 1 1
F12 = 2 1− + + + σ+ + σ2 (2.41)
2σ Xt Xc Yt Yc Xt Xc Yt Yc

Where σ is dependent on the material and is obtained through a biaxial

test with σ1 = σ2 = σ. This is a complication with respect to the the
previous criteria, in which only uniaxial tests were needed. In many
cases the presence of F12 has small influence on the results and can be
neglected, avoiding the need of biaxial tests.
 31

Chapter 3

Finite element model

This chapter contains a description of how the finite element model of the system
was built, of the employed materials and of the selected element properties.
The finite element solver and the pre-and post- processor used for the design
are respectively MSC Nastran and Patran. The panel is a circular plate meant
to be mounted on the bottom side of a satellite, around its longitudinal axis. A
number of brackets have to be positioned circumferentially on the panel. The
brackets have the function to constrain the plate to the satellite on their upper
surface. The mounting of the system on the satellite is guaranteed on the top
faces of each bracket. A heavy rotating mechanism has to be mounted on the
panel through 3 equally spaced rectangular interfaces of known dimensions.
The FEM model has been built starting from an already existing one, which
has been modified to be consistent with the present application. Figure 3.1
shows this initial model. The panel lies on the xy plane and has a radius of

Figure 3.1. Initial finite element model

0.5 m. Its thickness is a design variable. Eight equispaced brackets, placed at

32 3. Finite element model

a distance of 0.45 m from the plate centre, are connected to the panel. Each
connection is modelled with 2 Multi-Point Constraints (MPC) of type RBE2
and a BAR2 element. The BAR2 element connects two coincident nodes,
placed between the lower surface of the bracket and the plate, and a scalar
bush property is assigned to it. The RBE2 is a rigid multi-point constraint,
which in this case binds all the 6 DoFs of the dependent nodes to those of an
independent node. The dependent nodes are placed on the plate and on the
bracket. The bar element with the bush property is in practice a spring with
very high stiffness, used to extract the forces and moments exchanged between
panel and the bracket. Knowing this loads it is possible to dimension the bolts.
Another similar system is used to represent the interfaces between the upper
surface of each bracket and the satellite, where the constraints are placed.
In each of the upper springs, the node belonging to the satellite has been
linked with a bar element and a bush property to the correspondent node of
the bracket. The system is constrained to the satellite in many nodes with
springs of proper rigidity, to ensure the solidity of the mounting.
In the original model the plate was constituted of both QUAD4 and HEX8

Figure 3.2. Finite element model

elements. The elements belonging to the plate have been deleted, and a new
mesh was created for the plate using QUAD4 and TRIA3 elements. The mesh
has been built setting the dependent nodes of the RBE2 constraints as hard
points. The dimension of the element have been reduced to 6.5 mm, while that
of the QUAD4 elements used for the brackets is 4 mm.
Shell properties are defined on the elements of both plate and brackets. The
thicknesses of the bracket surfaces vary from 3 mm to 10 mm. Given the low
thickness, it is possible to model the elements as shells. This is possible also
for the panel, whose thickness is much smaller with respect to the diameter.
A non structural mass has been added to the panel to take into account non
modelled items, which are mainly the 2 sandwich adhesives and the inserts,
 33

necessary to join the panel with the other components. The value for the non
structural mass is estimated with the inverse of the plate surface, and is equal
to 1.273 kg.
The rotating mechanism has already been designed, so the all relevant data
for its modelling are known. Its most relevant characteristics is the 99 kg
mass and its moments of inertia, which are are Ix = Iy = 4.77 kg · m2 and
Iz = 9.54 kg · m2 . The moments of inertia are calculated with respect to
the principal axes, centred on its CoG. To model the rotating mechanism, a
lumped mass property has been added to a point element, placed at a vertical
distance of 10 cm from the panel. This simple modelling is possible because
it is known that the flywheel is dynamically decoupled from the panel. In
fact the natural frequencies of the rotating mechanism fall into a frequency
range above 140 Hz, while the main modes of our system are surely below 100
Hz. The rotating mechanism has been designed to have this modal behaviour
since the vibrational decoupling between mechanically interacting systems is
crucial in the design of space structures. This phenomenon will be explained
further in detail in the next chapter. The radius of its flywheel is 0.44 m, so
it can fit between the brackets. The wheel is connected to the plate through
3 equispaced radial interfaces, placed at a distance of 19 mm from the edge
of the panel. Each interface models a bolted joint with 4 bolts placed at the
corners of a rectangle. The interface dimensions are 97 mm radially and 85
mm circumferentially. Each bolt is modelled with a BAR2 element with a
bush property and an RBE2 constraint as it was done for the interfaces of the
brackets. It is important to notice that each of the springs at the rotating
mechanism interface model 1 bolt, while those at the bracket interfaces model
the entire bolted joint. This mean that the loads at exchanged at the panel-
bracket interfaces have to be divided by the number of bolts constituting the
joint.
To place the wheel on the panel, it was necessary to eliminate 2 brackets. The
two brackets, symmetrical with respect to the y axis, were deleted in order to
make room for the interfaces of the rotating mechanism. The obtained model
is showed in Figure 3.2.
For the core of the sandwich panel, 2 types of aluminium honeycombs are
available. The materials used is aluminium 7075 T6 for both, but one is denser,
having smaller hexagonal cells. Honeycomb materials have been defined in
Nastran as orthotropic materials. To distinguish among them they have been
named Honeycomb 50 and Honeycomb 70, referring to their density values. The
relevant elastic properties of the cores are listed in Table 3.1. Since honeycombs
are not designed to bear in-plane normal stresses, the in-plane elastic moduli
and Poisson’s ratio have not been reported and very small values have been
set for them. The possible core thicknesses are 25.4 mm (1 inch) and 50.8
mm (2 inches). For the reasons cited in Section 1.5 a thicker honeycomb will
guarantee better flexural and vibrational performances, at the price of a very
small weight increase. The skins are designed in two different configurations:
34 3. Finite element model

Honeycomb 50 Honeycomb 70

Shear modulus YZ [MPa] 138 257

Shear modulus XZ [MPa] 310 354

kg
Density[ m3] 50 73.4

Thermal exp. coef. X 0.289 · 10− 6 0.672 · 10− 6

Thermal exp. coef. Y 6.97 · 10− 6 4.5 · 10− 6

Table 3.1. Elastic properties of the honeycombs

• single layers of a high strength aluminium 7075 T6;

• CFRP laminates.

Thickness of the aluminium skins has to be chosen, starting from a minimum of

0.3 mm.The standard thickness of a CFRP ply is 0.075 mm, thus the laminate
thickness will be a multiple of it. The elastic properties of the two materials
are visible in Tables 3.2 and 3.3. With those values constitutive models are
built.
The carbon fibre reinforced polymer has superior mechanical properties and

Elastic modulus [GPa] 73

Poisson’s ratio 0.33

Shear modulus [GPa] 27.58

kg
Density [ m 3] 2768

Thermal exp. coef. 0.232 · 10− 6

Table 3.2. Elastic properties of aluminium 7075

allows a weight saving around 40%. The aluminium choice is still valid due
to lower cost and excellent thermal conductivity. This last aspect is relevant
since the electric actuator which drives the rotating mechanism generates heat.
A part of this heat may have to be dissipated by the panel, so the aluminium
option is taken into account for future design.
Besides constitutive models, strength values have been set for orthotropic
materials to be able to apply a failure theory. For the two orthotropic core
materials only the shear stress limit has a physic value, which is of 1.0 MPa
 35

Elastic modulus X [GPa] 325

Elastic modulus Y [GPa] 5.6

Poisson’s ratio XY 0.22

Shear modulus XY [GPa] 4.17

Shear modulus YZ [GPa] 4.17

Shear modulus XZ [GPa] 4.17

kg
Density [ m 3] 1600

Thermal exp. coef. X −1.03 · 10− 6

Thermal exp. coef. Y 32 · 10− 6

Table 3.3. Elastic properties of CFRP

for Honeycomb 50, and 2.2 MPa for Honeycomb 70. The values of other stress
limits have been set very high, since the correspondent failure modes are not
allowed for the core. The only relevant failure property of the 7000 series
aluminium is the yield strength of 430 MPa. No plastic behaviour is possible
for honeycomb and skins. Stress limits of CFRP are reported in Table 3.4. For

Xt Yt Xc Yc S

1776 30 715 95 56

Table 3.4. Stress limits of CFRP [MPa]

honeycombs and CFRP, a value of 15 MPa has been set for the bonding shear
stress limit, given by the adhesive plies which bind the skins to the core. The
material of the bracket is the same aluminium used for the panel skins. The
elastic properties of brackets are the same as those listed in Table 3.2.
Material properties are used build a laminated composite material. It contains
informations about the number, thickness, material and orientation of skin plies
and core. The laminated composite material is the main input given to the shell
property of the panel. The core is modelled into the laminated composite table
as a ply of appropriate thickness. The operative part of the design is to find
laminated composite configurations which optimally satisfies the design criteria.
 37

Chapter 4

Design of the panel

In this chapter informations about applied loads, criteria of the design and
relevant parameters are given, in order to justify the choices made in the choice
of the optimal sandwich configuration, and to explain the settings used for
the modal and static analyses. After an explanation of the applied dynamic
criterion, fundamental in the design of space structures, a description of the
quasi-static loads used for the static verification is given, together with the
chosen thresholds. For what concerns the results, the bending modes of the
panel are showed, and then the actual comparative study for the derivation of
an optimal panel configuration is explained. On the optimal configurations
further static analyses, necessary for the complete static characterization of
the panel, have been carried on. At the end of the chapter some further
considerations on the interface loads and on the stresses on the brackets are
made.
It is important to point out that the main design criterion which has to be
satisfied is related to the dynamics of the system. A comparison between the
dynamics of the system which includes panel, brackets and flywheel with those
of the the rest of the satellite is crucial for a good design of a space structure.
This is why this design is based on the fulfillment of a dynamic requirement,
verified through modal analyses. The following static analyses are just needed
for a necessary verification of the material resistance to the most severe loads
applied before and during launch.

4.1 Dynamic decoupling

The dynamic (or modal) requirement refers to an aspect of fundamental im-
portance in the design of a satellite and its subsystems. This is the eventual
vibrational coupling between launcher and satellite, or between the satellite
and its subsystems. This dynamic coupling has to be absolutely avoided, so a
modal criterion has to be satisfied for a proper design of a space structure. To
examine the phenomenon in detail, the concept of main (or primary) natural
38 4. Design of the panel

modes have to be defined.

The vibration of a mode happens in some of the 6 spatial DoFs, which are
translations (T1, T2, T3) and rotations (R1, R2, R3). A portion of the system
mass participates to the mode. This mass is called modal effective mass fraction
(MEMF) and it is expressed as a percentage of the total mass. The MEMF is
related to the single mode, and within the mode it assumes different values for
each DoF. A mode is defined by its eigenvalue or natural frequency, and by its
eigenvector that is the mode shape. The particular mode shape influences the
values of the MEMF associated with each DoF: the DoFs in which the motion
activates when the mode is excited are characterized by a higher MEMFs.
A mode is a main mode if the MEMF associated to at least one of the DoFs is
greater than a threshold value. Generally this value is set to 10% [10]. The
main modes of a structure dominate its vibration in the frequency range of
interest, since when excited they move a high portion of the total mass, and
lead to high reaction forces at the interface of the system [23].
The external loads generated during launch are characterized by a certain
frequency range. The main modes of the launcher that fall into this frequency
range are excited by the external loads. The consequent vibrations generate
dynamic loads on the satellite through the launcher-satellite interface. If the
main natural frequencies of the satellite coincide or are sufficiently near to
those of the launcher, the satellite will resonate as a response to the loads
generated by launcher vibration, causing stresses on the structure. The same
phenomenon takes place between the satellite and its payload and equipment.
Another negative effect of dynamic coupling is the decrease of the first natural
frequency of launcher and satellite [10].
To avoid dynamic coupling between launcher and satellite, the launcher author-
ity specifies the minimum longitudinal and lateral main frequencies required
for the satellite, allowing to avoid dynamic coupling. For the Vega C launcher,
taken as a reference in this work, the minimum lateral and longitudinal fre-
quencies are respectively 12 Hz and 20 Hz [3]. Normally, to have a sufficient
confidence in the absence of any modal coupling, the minimum frequencies
are increased by 15%. In the Vega C case the minimum natural frequencies of
the satellite have to be higher than 13.8 Hz in the lateral direction and 23 Hz
longitudinally. To achieve decoupling between satellite and its subsystems, the
minimum natural frequencies of √ the subsystem (in our case the panel) have to
be increased of an additional 2 factor. The minimum requirement becomes
19.5 Hz for lateral modes and 32 Hz for longitudinal modes.
Since the design is preliminary, we want to give a conservative estimate of
sandwich configurations which satisfy the lateral and longitudinal dynamic
criteria. For this reason we consider a minimum frequency of 40 Hz for both
lateral and longitudinal primary modes. To know if a mode is a main mode,
modal effective masses and modal effective mass fractions are requested as a
Nastran output, in addition to the eigenfrequencies and mode shapes. The
cited criterion of MEMF greater than 10% is used.
4.2 Quasi-static loads 39

The main interest in the study of the panel vibrational behaviour is on the
frequency of the principal out-of-plane bending mode. In fact, due to its geom-
etry and constraints, the most relevant mode for the panel is the one which
causes the displacement of the majority of the panel mass in the z direction,
that is the out-of-plane direction. Regardless of the particular configuration of
the skins and core of the sandwich, this mode is always the first main mode of
the system, and it is characterized by a high value of the modal effective mass
fraction associated to the z translation.

4.2 Quasi-static loads

The loads and load cases considered for the panel design are described in this
section. The good configurations with respect to the modal criterion have then
to satisfy a static requirement, described in the next section, under several
combinations of these loads.
During pre-flight operations and flight, a launch vehicle is subjected to static
and dynamic loads, caused for example by in ground operations, aerodynamics
and propulsion. Quasi-static loads (QSL) are the most severe combinations
of static and low frequency dynamic loads, exerted on the spacecraft at any
moment of the mission, including ground and flight operations [19]. The differ-
ent types of excitation occurring simultaneously during flight are expressed by
the launcher authority as equivalent accelerations applied at the satellite CoG.
The structure behaves rigidly with respect to those accelerations. One of the
main contributions to the QSL is the flight static acceleration, reported for
the Vega C launcher in Section 1.2.
In design and dimensioning we are dealing with limit levels of QSLs. Limit load
levels are expressed as multiples of the gravitational acceleration g, called load
factors. In a preliminary design limit load factors are taken from the launch
vehicle user’s manual. Those QSL limit levels, exerted on the satellite during a
single launch of Vega C [3], are reported in Figure 4.1. It is important to point
out that limit load factors are applicable only if the system is dynamically
decoupled from the launcher vehicle, that is if satisfies the dynamic require-
ment seen in the previous section: only in this case the dynamic component
of the QSL can be equivalent to a static acceleration for the satellite and its
subsystems. The 6 load events cover the whole flight, from the detaching from
earth to the placement of the satellite in orbit operated by the AVUM stage.
The spacecraft mass is set to 2000 kg. The minus sign in the longitudinal loads
indicates compression, and the plus tension. The longitudinal direction of the
launcher coincides with the out-of-plane direction (z) of the support panel.
Lateral quasi-static loads can exist in any direction in the plane of the panel
(xy), since during flight the launcher continuously changes its orientation.
Limit load factors are expressed in m/s2 by multiplying each value for the
acceleration of gravity g. According to ECSS [19], the obtained accelerations
40 4. Design of the panel

Figure 4.1. Limit load factors for a single Vega C launch

have to be multiplied again by a design factor, which for satellites is the product
of three different factors: qualification factor KQ; model factor KM and project
factor KP . The qualification factor KQ is used to determine the qualification
loads, which are loads used for the dimensioning or test qualification of a
structure. The project factor KP is related to the maturity of the program,
and its value is progressively reduced during the development of the project.
The model factor KM takes into account the uncertainties of the mathematical
model. Design limit loads are obtained multiplying the limit load factors by a
total coefficient 9.81 · 1.25 · 1.1 · 1.2. In this preliminary design loads are being
strongly increased, since we want to give a very conservative design solution
to compensate the uncertainties present at this level of analysis. In future
designs the model can be refined and the safety factors decreased. The obtained
loads are tabled in Table 4.1. Those design loads are inertial accelerations,
applied at the system CoG. Note that some load events are much more critical
than others, and that longitudinal compression accelerations are higher than
longitudinal tension accelerations. This means that compression load cases are
more dangerous, and among them the worst cases are lift-off, 1st stage flight
with maximum acceleration and tail off, and most of all 2nd stage ignition and
flight, 3rd stage ignition.
Lateral loads can have any direction on the xy plane. It is common practice
to verify 8 in-plane directions at regular intervals of 45°, covering the entire
circumference from 0° to 315°. The 0° direction is the x and the 90° direction
is y. For each of the 6 flight phases 2 load cases are examined, coupling the
4.3 Conditions for the static verification 41

QSL [m/s2 ]
Load Event Longitudinal Lateral

Compress. Tension

Lift-off -72.8 48.6 ±21.9

Flight with max. dynamic press. -64.7 24.3 ±14.6

1st stage flight with max. acc. and tail off -80.9 16.2 ±11.3

2nd stage ign. and flight, 3rd stage ign. -80.9 48.6 ±21

3rd stage maximal acceleration -80.9 N/A ±3.2

AVUM flight -16.2 8.1 ±11.3

Table 4.1. Design limit loads

lateral acceleration with both compression and tension longitudinal values. A

total of 12 load cases have to be analysed for each in-plane direction of the
lateral load.

4.3 Conditions for the static verification

In static analyses Von Mises stresses are plotted in the case of aluminium skins,
and a limit yield stress of 300 MPa has been considered. No plastic deformation
is allowed, then aluminium yield stress is also its ultimate limit stress. For
the composite laminate the Failure Index (FI) distribution is plotted for each
ply to verify failure. Failure index is a useful tool used to check failure on
composites, where strength is dependent on the direction of the stresses. It is
defined for the single element as:
σactual
FI = (4.1)
σallowable
where the σactual is calculated in the static analysis, and the σallowable has to
be properly defined through a failure criterion. One of the failure criteria
of Subsection 2.4.2 has to be employed to define a failure surface in the
(σ1 , σ2 , σ3 ) stress space. The Hoffman criterion is chosen. Giving in input the
failure parameters, Nastran is able to build a failure model in addition to the
constitutive model already defined, and calculate σallowable for the current load
case. The stress state in principal coordinates leading to failure lies on the edge
42 4. Design of the panel

of the Hoffman failure surface, while the actual stress has to be sufficiently
inside it. The software calculates the principal stresses on each element of each
ply, and uses them to find failure indexes. If F I > 1 ply failure has occurred.
In this work we will consider laminate failure to take place when just one ply
has failed, adopting a first ply failure criterion. This is not always true, since
a redistribution of stresses could lead to a non-critical stress state after the
failure of a single ply. The approximation is justified by the fact that it is in
favour of safety and leads to a simple elaboration of the results.
A Factor of Safety (FoS) is considered to define the Margin of Safety (MoS),
which has the expression [23]:
σallowable
M oS = −1 (4.2)
σactual · F oS
The static criterion used to assess structural integrity is simply:

M oS > 0 (4.3)

where the FoS value is 1.2 for aluminium and 2 for CFRP. Very high factors of
safety are being used especially for CFRP skins, because we are preliminarily
ensuring the feasibility and robustness of the project. For aluminium, since
σallowable = 300MPa, from Equations 4.2 and 4.3 an equivalent expression for
the static criterion can be found.

σactual < 250M P a (4.4)

For CFRP the equivalent static criterion is found recalling also the definition
of failure index.
F I < 0.5 (4.5)
The inequalities of Equations 4.4 and 4.5 are the conditions checked in the
static analyses under several combinations of QSL, in order to asses structural
integrity.

4.4 Design procedure

Since one of the design requirements in space applications is to minimize mass,
the thicknesses of both aluminium and CFRP skins must be as low as possible.
The minimum thickness generally employed for an aluminium skin is 0.3 mm,
while the standard thickness of a CFRP ply is 0.075 mm. The minimum
number of plies stacked to form a laminate is 8, since the laminate must be
balanced and symmetric. The constraints on the stacking sequence and the
typical shape of laminate stiffness matrices A, B, C and D are showed in
Section 2.3. Only if very particular requirements exist it is possible to avoid
a balanced and symmetric configuration, but this is not the case. For these
reasons, a minimum height of 0.6 mm (8 plies) has to be employed for CFRP
4.4 Design procedure 43

laminates.
A common 8 ply balanced and symmetric stacking sequence used in sandwich
structures is [0\90\45\ − 45]s, where the 0◦ external ply is the farthest from
the bending neutral axis of the sandwich, while the −45◦ one is the closest.
The θ = 0◦ direction coincides with the first principal direction of the laminate,
and with the x axis of the model. The θ = 0◦ and θ = 90◦ plies carry most of
the in-plane normal stresses, while the θ = ±45◦ layers are inserted to sustain
in-plane shear stresses. For this reason, a CFRP laminate has to include at
least one ply oriented in each of these 4 directions (0◦ , 90◦ , +45◦ , −45◦ ). It is
good practice to not put ±45◦ plies externally in a laminate.
An example of sandwich setup in Patran with [0\90\45\ − 45]s stacking
sequence is given in Figure 4.2. Skins and core have been grouped into black
rectangles. CFRP and honeycomb materials are respectively mat.2000001 and

Figure 4.2. Example of stacking sequence of a sandwich in Patran

mat.2000004 in figure. The layers, numbered from 1 to 17, correspond to a

vertical stacking for increasing values of the coordinate z. This means that ply
17 will be directly in contact with brackets and rotating mechanism interfaces.
With the informations in Figure 4.2 and the material constitutive models, the
software calculates the laminate stiffness matrices and builds the constitutive
model for the entire sandwich, as it was done for laminates in Equations 2.29
and 2.30.
If the dynamic or static requirements is not satisfied for a selected sandwich
configuration, that is if the first principal natural frequency is below 40 Hz or
44 4. Design of the panel

the margin of safety is negative, some design changes are necessary. Those
modifications include:
• increasing aluminium skin thickness, starting from a value of 0.3 mm at
intervals of at least 0.2 mm;
• increasing the number of CFRP plies from a minimum of 8, adding 2
plies per skin if their orientation is 0◦ or 90◦ , and 4 plies per skin if it is
±45◦ to maintain a balanced and symmetric configuration;
• changing the core thickness from 25.4 mm to 50.8 mm, which allows to
consistently increase the natural frequencies and the flexural behaviour
of the skins, at the cost of a consistent mass increase;
• changing the core material, since Honeycomb 70 is denser and more
performing than Honeycomb 50, leading to better results both in dynamic
and static terms with an increase in mass.
With each of these design changes the sandwich stiffness rises. Also the
system mass rises. Considering that the main contribution to the mass of
the entire system is not given by the panel but by the 99 kg of the rotating
mechanism, and that the stiffness is only given by the panel, it’s clear how
these modifications increase the natural frequencies of the system.
s
k
fnat ∝ (4.6)
m
Some of the extracted outputs depend on whether skin material is isotropic or
orthotropic. Von Mises stresses are plotted in the first case, failure indexes in
the second. The margin of safety positivity is then assessed. This is equivalent
to consider the inequalities of Equations 4.4 and 4.5.
Forces and moments on the bush elements are extracted to have an estimate
of the loads exchanged with brackets and rotating mechanism. We need these
loads for a future design of the bolted joints connecting the panel to brackets
and wheel. The equilibrium of forces and moments for the brackets leads to
the equality of the loads exchanged from a bracket to the panel and to the
satellite.
The point mass representing the rotating flywheel is by far the heaviest part
of the model, therefore it is the main source inertial forces. In the model loads
are transmitted from the wheel to the panel through an RBE2 multi-point
constraint. This constraint rigidly transfers the displacements to the panel,
as if the rotating mechanism supports were infinitely rigid. The highest loads
are therefore on the sandwich regions around the RBE2 dependent nodes. We
expect these interfaces to be the most critical regions. Von Mises stresses on
the brackets are monitored in this analysis, but the aluminium brackets are not
expected to be critical since they are not in direct contact with the rotating
mechanism.
4.5 Modes of the panel 45

4.5 Modes of the panel

The mode shapes (eigenvalues) of the entire system are presented, focusing
only on the plate and not on the brackets, for a reason of expository clarity.
This is the first result presented, since changing the sandwich thicknesses
and materials the mode shapes nearly do not change. Comparing the modal
effective mass fractions for different sandwich panels (see next section) the
MEMF values remain almost unchanged, as also the mode shape plots. The
only modal quantity significantly changing is obviously the natural frequency,
since with design changes we increase stiffness with a little change in mass.
This is evident especially changing the thickness of the honeycomb.
The first 3 modes are plotted in Figures 4.3. Only the first 3 mode shapes
have been showed, since they are the ones which develop out of the xy plane.
This out-of-plane flexural behaviour is of particular interest for a sandwich
panel. The undeformed plate shape is represented by the blue element grid.
The coordinate system is visible in the bottom left corner. The shown modes
are all principal, but the most relevant mode is surely the first, whose natural
frequency is the central parameter of the design. This out of plane motion
is the typical vibrational behaviour of a circular membrane. The flywheel is
by far the heaviest component, so the highest nodal displacements will take
place around the flywheel-panel interfaces. The brackets have their own mode
shapes that are not displayed in the figures. Their effect remains visible in
correspondence of the bracket-panel interfaces, where displacements are small
but not always zero.

4.6 Comparative study of admissible configu-

rations
The possible QSL combinations are composed of 2 longitudinal acceleration
values (tension and compression), 8 values for the lateral acceleration depend-
ing on its in-plane direction, and 6 flight phases. The total possible load cases
are 96. The optimization is therefore performed focusing the attention on
the most severe load event. The highest inertial forces are exerted on the
satellite in the time interval between second and third stage ignitions (2nd
stage ign., 3rd stage ign. in Table 4.1). The loads defined to establish the
optimal configuration are the 8 cases with fixed longitudinal (out-of-plane)
acceleration of -80.9 m/s2 , and lateral (in-plane ) acceleration of 21 m/s2 with
variable direction.
After a modal analysis, a configuration satisfying the modal criterion is anal-
ysed statically, and if they satisfy also the static requirement it is considered
a good configuration. Between the two requirements, the most important to
be satisfied is the dynamic one, since dynamic coupling between structures is
a destructive phenomenon and has to be avoided. The main bending mode
46 4. Design of the panel

(a)

(b)

(c)

Figure 4.3. Aluminium skins, shapes of the panel modes: 1st (a), 2nd (b)
and 3rd (c)

of the panel is always the first in terms of natural frequency. In practice, the
situations with the frequency of the main bending mode above 40 Hz and
4.6 Comparative study of admissible configurations 47

satisfying the static verification, have been considered good. Only the force
components of the interface loads have been reported, to avoid the presence
of too many outputs. The interface moments are in any case 2 orders of
magnitude lower than forces, since the applied inertial loads are translational
accelerations.
The results have been tabulated in Excel, and the most significant and ex-
planatory data are here reported. It is important to remark that the panel
mass values include 1.27 kg of non structural mass, representing non-modelled
items.
For the aluminium skins, each of the 4 possible honeycomb combinations was
examined, which hare obtained changing the material and thickness of the
core. For every core configuration, the minimum thickness of the skins which
allows to satisfy the design criteria was found, with the objective of weight
minimization. The results are showed in Figure 4.4. The first 3 columns define

Figure 4.4. Comparative table, aluminium 7075 skin

the sandwich, the 4th and 6th are related to the design criteria, the 5th concerns
the static verification, and the last 3 are needed for optimization. Each row
defines a tested configuration, and some adjacent rows containing the same val-
ues have been joined together to give a clear view. The aluminium sandwiches
with the 25.4 mm core need very thick skins, thus their mass is excessive with
respect to other good solutions. That’s why they have been discarded without
further analyses. With a 50.8 mm core of the lighter material (Honeycomb 50)
we obtain the best performances for the aluminium case. Skins are only 0.5
mm thick, and maximum Von Mises stresses are not critical (MoS=1.45). This
is a first relevant result of this study: in general the use of a thicker honeycomb
raises the natural frequencies allowing to use thinner skins, with a good saving
in mass. The Von Mises stress distribution on the critical skin, which is the
lower one, is given in Figure 4.5. The components of the inertial load are (14.9,
-14.9, -80.9) m/s2 , where the x and y components correspond to a in-plane
load of magnitude 21 m/s2 . The three interfaces with the rotating mechanisms
48 4. Design of the panel

Figure 4.5. Von Mises stress distribution, aluminium skin, lower skin, inertial
load (14.9, -14.9, -80.9) m/s2

are visible, since in those points the stresses reach the highest values. The
most critical region corresponds an interface which lies in the direction of the
lateral load. The rest of the plate is substantially unloaded. This situation,
with the plate unloaded except for the regions around some interface nodes,
happens with every plate shape and load case.
In the CFRP skin case a further degree of freedom is given by the choice of
the stacking sequence. The stacking sequence affects the stress distribution in
each ply, then it changes the value of the maximum failure index. The other
parameters are not affected, since no mass and stiffness are added or subtracted
to the system. Figure 4.6 shows the relevant cases for a CFRP skin with 8 plies.
The analyses with a 25.4 mm core have not been reported, because in that case

Figure 4.6. Comparative table, CFRP skin with 8 plies

the modal design requirement can not be satisfied. With the thicker core good
configurations are found, with a lower mass with respect to the aluminium
skin, and with almost the same MoS. The principal out-of-plane mode has a
natural frequency of 55.4 Hz, which is 15.4 Hz higher than the threshold. This
4.6 Comparative study of admissible configurations 49

is a very good dynamic result and it is combined with low weight. The total
skin thickness is 0.6 mm. Figure 4.7 shows the failure index distribution in the
most critical load case and on the critical ply, which is the 45◦ external ply of
the upper skin. The anisotropic behaviour of the carbon fibre composite is

Figure 4.7. Failure index distribution, CFRP 8 ply laminate, upper skin,
inertial load (0, -21, -80.9) m/s2

visible in figure: although the total load vector lies on the yz plane, the failure
index distribution is asymmetric with respect to the y axis, since the fibres of
the concerned ply are oriented at θ = 45◦ . Knowing the most loaded ply, an
optimization of the stacking sequence can be performed. The most loaded ply
is moved towards the centre of the symmetric laminate, that is towards the
right in the stacking sequence notation. With this change the distance from the
bending neutral axis is reduced by small amounts, multiples of the thickness of
the lamina. Employing the thicker honeycomb this optimization is not much
visible, and very little gain on the maximum failure index is obtained.
To have a better comprehension of the behaviour of the sandwich, a 10 ply
laminate is analysed. The best 8 ply configuration just found will probably
be the overall optimum choice. The principal attempted configurations for a
10 ply CFRP laminate are visible in Figure 4.8. A group of 2 layers can be
inserted symmetrically in the laminate, with orientations of 0◦ or 90◦ . The
±45◦ plies instead can be added only in groups of four to maintain the laminate
balanced and symmetric. Adding 0◦ plies is not useful, since those plies are
always the least stressed. Adding 90◦ plies good results are obtained, with the
use of the thinner and denser honeycomb. The dynamic requirement is hardly
satisfied. With a stacking sequence optimization, the maximum failure index
can be brought under the 0.5 threshold of the static criterion, but remains very
close to this limit. The logic for stacking optimization is always to put the
most loaded plies in at the centre of the laminate, lowering their distance from
50 4. Design of the panel

Figure 4.8. Comparative table, CFRP skin with 10 plies

Figure 4.9. Failure index distribution, CFRP 10 ply laminate, upper skin,
inertial load (0, -21, -80.9) m/s2

the neutral axis. The optimum 10 ply configurations allow a small mass saving
of 0.14 kg with respect to the 8 ply situation. Every stacking sequence in
which the 45◦ layers are in the middle and the 0◦ plies are external is optimal.
The critical layer in the most severe load case is visible in Figure 4.9. This is
again the external 45◦ ply of the upper skin. The 10 ply laminate is a very
border-line solution for both the dynamic and static design requirements. To
complete the possible cases, some 12 ply configurations have been tested. The
reason is that with a 12 ply skin, 45◦ and -45◦ plies can be added maintaining
a balanced and symmetric laminate. Adding layers of the same orientation as
the critical one the stresses will be conveniently redistributed, reducing the
maximum failure index. The results are showed in 4.10. The first 3 rows of
the table show that adding 0◦ and 90◦ plies is not useful, as it was expected.
4.6 Comparative study of admissible configurations 51

Figure 4.10. Comparative table, CFRP skin with 12 plies

The fourth row shows that employing the lighter honeycomb doesn’t lead to
a good configuration from the vibrational point of view. With the denser
honeycomb good configurations are found. Optimizing the stacking sequence
some satisfying modal and static parameters are obtained (MoS=0.24). Weight
naturally increases, but its still lower than that of the aluminium sandwich.
The stresses on the oblique plies are redistributed, so the most loaded plies
are now the 90◦ ones. Figure 4.11 shows the failure index distribution on the
critical ply. In this plot both loads and composite fibres are symmetrical with

Figure 4.11. Failure index distribution, CFRP 10 ply laminate, upper skin,
inertial load (0, -21, -80.9) m/s2

respect to the y axis, resulting in a symmetric plot.

At this point an optimal configuration has been found for the case of aluminium
skins, and for the cases of laminates with 8, 10 and 12 plies. These 4 solutions
are put into 2 graphs, relating the panel mass to the value of the natural
frequency of interest, and to the margin of safety. The graphs are showed
52 4. Design of the panel

in Figures 4.12 and 4.13. The points lying in the top-left region of both

Figure 4.12. Good configurations, comparison of mass and dynamic behaviour

Figure 4.13. Good configurations, comparison mass and margin of safety

graphs correspond to good situations, in terms of modal behaviour and static

resistance. The first graph is the most relevant, since to be in safe condition in
terms of dynamic decoupling with the satellite is more important than further
lowering the stresses. Among the CFRP skin configurations, the 8 ply with
50.8 mm honeycomb is recognized as optimal since it satisfies well the dynamic
criterion and it assures low weight. The aluminium configuration (Figure 4.4)
4.7 Analysis on the optimum configuration 53

can’t be discarded, despite its worse vibrational performances. In fact the

electric engine driving the rotating mechanism is a heat source. If in future
studies issues of heat dissipation will emerge, aluminium skins will be the
forced choice for their high thermal conductivity.
From these resuming graphs it emerges the crucial role of the honeycomb: the
use of a core of double thickness is more efficient than the addiction of plies to
the skins. A thicker honeycomb allows to optimize the use of the high strength
laminates, since it increases the area moment of inertia of the resistant section,
increasing its stiffness. At the same time, since the honeycomb has a very low
density, its thickening introduces a very small amount of mass in the system.
For completeness, the values of the MEMF of the first modes for aluminium
and 8 ply CFRP cases are extracted from Nastran output file and showed in
Figures 4.14 and 4.15. The MEMF values for the 1s t natural frequency, the
one related to the out-of-plane bending mode, are highlighted. The 1st mode is
characterized by a modal effective mass fraction of 0.59 for translations along
the z axis, and of 0.39 for rotations around the x axis. This confirms that the
mode is principal, and that it is the one that when excited leads to the bending
stresses on the sandwich. The first 3 modes mode are be the most excited
by external off-plane load, which induces off-plane translations and rotations
(T3, R1, R2). The off-plane loads are the longitudinal QSL, which are the
highest in value. The MEMF values of the 3rd mode are very to those of the
1st , while the 2nd mode moves almost the 95% of the mass in a rotation around
the y axis. Those modes, showed in Section 4.5, represent the out-of-plane
vibrational behaviour of the panel, therefore they are the most important for
this application.

4.7 Analysis on the optimum configuration

The sandwich with the light 50.0 mm core and with 8 ply laminates is the
optimal solution in terms of both dynamics and static resistance. In this
section the required static analyses are performed on this configuration, and
the relevant data are collected into Excel tables. Loads cases are the 96
combinations taken from Table 4.1. The highest values for the tabled outputs
are recognised, and further considerations are made for them. The results are
presented in Figures 4.16,4.17,4.18 and 4.19. The number in degrees on
the top-left of each table refers to the direction of the in-plane QSL, which
in each table is rotated in the xy plane with respect to the previous table,
starting from 0◦ when it is along the x axis. The rows correspond to Vega
C load events, each divided in compression (-) and tension (+), referring to
the sign of the out-of-plane load. The most critical load case was recognized
in advance and analysed in the previous section for the choice of the optimal
configuration. The load case in question is the 2nd stage ignition and flight,
3rd stage ignition in compression (-), with lateral load in negative y direction
54 4. Design of the panel

(270◦ ). With this load, the failure index reaches the highest value, highlighted
with a red box. The worst case happens when the lateral load points towards
the negative y direction because two brackets have been deleted on this side
of the panel to make room for the flywheel interfaces. Not only this, but all
the load events become more severe if lateral load points towards this region
(225◦ , 270◦ , 315◦ ). This is testified in the reported tables, where in the last 3
tables (225◦ , 270◦ , 315◦ ) each failure index is higher than the corresponding
value of the first 5 tables.
The magnitudes of forces and moment vectors have been collected at the bracket
and wheel interfaces. The components (x,y,z) are showed only for the highest
loads (red boxes in the tables). Forces and moments at the panel-bracket
interface have to be divided by the unknown number of bolt constituting the
joint while, since the rotating mechanism interfaces are known, the loads are
here referred to the single bolt. Forces and moments at the interfaces are
useful values for the dimensioning of the joints. Moments are much smaller
than forces. For this reasons they have not been considered in the comparative
tables. To how the values of the interface load components, the vector plots
showing the components of the maximum interface forces are given in Figures
4.20 and 4.21. The colors in figure green for the x component, yellow for the y
and red for the z. From Figure 4.20 the single node connecting the bottom of
the bracket with the plate is visible. Another node on the top of the bracket
connects it to the satellite. The force components have to be distributed over
a unknown number of bolts. Figure 4.21 shows the interface force components
on the four nodes of the panel-wheel interface. The most loaded node in figure
is the one on the left. The maximum values of the interface forces are resumed

F⃗bracket = (21, 1820, 1800)N F⃗wheel = (225, 245, 1390)N

The moment vectors are not reported since as their values are extremely small.
The values of the force components suggest that normal and shear forces on
the bolted joints allow to design the bolted joints with standard diameters and
materials.
Figure 4.22 shows the Von Mises stress distribution over the brackets, for the
optimum 8 ply configuration and the worst load case. The load case is the
same that causes the maximum forces exchanged between plate and brackets.
The maximum Von Mises stress value is of 74.9 MPa, much lower than the
limit of 250 MPa (MoS=2.3). This confirms that the brackets are not critical
in terms of quasi-static loads exerted during launch. The thicknesses of the
brackets are 3 mm for the vertical faces, 5mm for the upper base and 10 mm
for the lower base. The design of the brackets is not the aim of this work, but
from this analysis it is clear that an reduction of the thicknesses is possible, in
order to optimize the material usage.
4.7 Analysis on the optimum configuration 55

Figure 4.14. Modal effective mass fractions, aluminium skins

Figure 4.15. Modal effective mass fractions, CFRP 8 ply skins

56 4. Design of the panel

Figure 4.16. Optimum configuration, static analyses with in-plane load in

the 0° (x) and 45° directions
4.7 Analysis on the optimum configuration 57

Figure 4.17. Optimum configuration, static analyses with in-plane load in

the 90° (y) and 135° directions
58 4. Design of the panel

Figure 4.18. Optimum configuration, static analyses with in-plane load in

the 180° (negative x) and 225° directions
4.7 Analysis on the optimum configuration 59

Figure 4.19. Optimum configuration, static analyses with in-plane in the

270° (negative y) and 315° directions
60 4. Design of the panel

Figure 4.20. Maximum force components on the interfaces of the brackets

Figure 4.21. Maximum force components on the panel-flywheel interfaces

Figure 4.22. Maximum Von Mises stresses on the brackets [Pa]

 61

Chapter 5

Fatigue verification of the

brackets

The presence of a rotating flywheel mounted on the plate leads to further

considerations about the possible fatigue effects on the aluminium brackets
due to the flywheel disturbances, due to its continuous rotation. Under this
condition the life of the system has to be ensured for 12 years. The fatigue
stress cycles are considered to be less critical for the plate, in which the
sandwich structure ensures better fatigue behaviour than the brackets, which
are simple aluminium laminae. Fatigue is the phenomenon of crack propagation
into the material when a cyclic load is applied on it. In our system one the
sources of cyclic load are the unavoidable flywheel disturbances, transmitted
to the plate and the brackets through their interfaces. These load cycles are
very small in amplitude with respect to the QSL, and are not relevant for
the static verification. The disturbance can instead be damaging in terms of
fatigue and of induced vibrations. The disturbance of a rotating mechanism
cause a micro-vibration environment on the entire satellite, disturbing the
vibration-sensitive instruments and eventually causing resonance, when the
flywheel rotates at some critical speeds where the imbalance frequency equals
one of the natural frequencies of the wheel. In this chapter the focus is only
on the fatigue damage induced by the disturbances with a very high number
of cycles.
For the brackets, the stress cycle due to the rotating mechanisms imbalances is
found, and a fatigue verification is performed with ESALOAD and ESAFATIG.
These tools allow to consider not only the imbalances, but the load history of
the structure during its whole service life.

5.1 Frequency analysis settings

The frequency analysis has been performed without the use of a preprocessor,
writing the analysis settings in the Nastran text input file. The analysis in
62 5. Fatigue verification of the brackets

frequency domain to examine many possible flywheel angular speeds in its

typical range of functioning, which is [0,30] Hz. Nastran allows 2 types of
analysis in the frequency domain one modal and the other direct. The modal
approach has been chosen over the direct one. The reason is related to the fact
that direct method solves a set of coupled equations over the physical DoFs
of the system, which are proportional to the number of nodes. The modal
method instead solves a number of uncoupled equations equal to the number
of modes selected by the analyst [14]. Enabling to solve uncoupled equations,
the modal frequency analysis is much less computationally heavy especially for
large models and low frequency excitations. The drawbacks are the necessity
to define modal damping bi for each ith mode, and the lower accuracy of the
solution due to modal truncation.
In the modal frequency analysis the harmonic solution is searched in terms of
modal coordinates. The modal expansion in the frequency domain is written
in Equation 5.1.
x = Φξ(ω)eiωt (5.1)
Where Φ is the matrix of the mode shapes (eigenvectors) which are known,
and ξ(ω) are the modal coordinates, the unknowns of the problem. Exploiting
the orthogonality property of the eigenvectors, the equations can be decoupled.
Consider a damped system and an external force P (ω). We can write the
equation of motion of the ith mode:

− ω 2 mi ξi (ω) + iωbi ξi (ω) + ki ξi (ω) = pi (ω) (5.2)

where mi , ki , bi and pi (ω) are respectively the mass, stiffness, damping and
force associated to the ith mode. Solving the equation for ξi (ω), each individual
modal response can be computed. The values for the n analysed modes are
then summed up to find the physical solution x.
The damping has not been defined directly as a modal damping bi , but through
the quality factor Qi .
1 mi ωi
Qi = = (5.3)
2ζi bi
This quantity is defined from the more known damping ratio ζi , and it is
inversely proportional to the actual damping. Its distribution is defined in the
frequency range of interest for the modal extraction, which in our case is up to
800 Hz. In this range a constant value Q = 100 has been set.
The force and moment vectors have been defined through their x and y
components, assigning magnitude distributions and phases. The application
region is the point mass modelling the rotating mechanism. The the solution is
computed in the interval [0,30] Hz, every 0.1 Hz. Since the in-plane load vectors
rotate at the same angular speed, they are defined in the same frequency range.
The loads defined in this analysis have the form of Equation 5.4 [14].

P (f ) = C(f )eiθ (5.4)

5.1 Frequency analysis settings 63

C(f ) is the magnitude distribution and θ is the phase angle. Magnitude

distributions in frequency are included in the Nastran input file, importing
tables defined in Excel from the definitions of Equations 1.1 and 1.2. The
values of the imbalances are not known for this specific rotating mechanism,
because their evaluation is quite difficult and can be performed only with
specific tests. The standard values for rotating devices in satellite have been
used, and are showed in Equation 5.5.

Us = 1.5 g · cm = 1.5 · 10−5 kg · m (5.5)

Ud = 50 g · cm2 = 5 · 10−6 kg · m2
The imbalances are very small, and the range of wheel rotating frequencies is
not high too. For these reasons the frequency distributions of the centrifugal
force and moment modelling the disturbances have low values, as showed in
Figures 5.1 and 5.2. The functions are parabolic, and their maximum values

Figure 5.1. Frequency distribution of the centrifugal force magnitude

verify at 30 Hz and are 0.53 N and 0.18 Nm. The fatigue analysis is pursued
anyway, since for a service life of 12 years and a frequency of 30 Hz, the number
of stress cycles on the bracket is very high, in the order of 1010 cycles. This
high number of cycles can be dangerous for the brackets, even with small loads.
In addition, the real disturbances could be much higher than the typical values
employed in this analysis.
In defining the load phases, the quadrature of the moment with respect to
the force has been respected. The initial direction of the force is along the
positive x axis, and that of the moment is the positive y axis. Phase values
are θF x = θM y = 0 and θF y = θM x = π4 . A scale factor of -1 is set for Mx , since
in the first 180° of the rotation it must assume negative values. To avoid the
computation of the rigid body modes and their contribution to the solution,
the lowest frequency of interest for the mode extraction has been set to 0.1 Hz.
64 5. Fatigue verification of the brackets

Figure 5.2. Frequency distribution of the centrifugal moment magnitude

5.2 Results of the frequency analysis

Since the first vibration mode of the system is well above 30 Hz, no resonances
due to the defined forcing functions can take place in the [0,30] Hz interval.
For this reason the highest displacements, and therefore strains and stresses,
occur at 30 Hz where load magnitudes are maximum. The distribution of the
displacement magnitudes in frequency of any node of the model is parabolic
as the load magnitudes of Figure 5.1 and 5.2. The displacement is maximum
at the maximum frequency. For this reason in this section only the 30 Hz
rotation frequency will be analysed.
Brackets are composed of QUAD4 elements with a thin shell property. The
thicknesses vary between different faces of the bracket. The thickness is 10 mm
at the lower base, 5 mm at the upper base and 3 mm in all the vertical faces.
The non zero components of the stress tensor on a shell element are only those
in the plane of the shell. In differently oriented faces of the brackets, different
stress components are present in the global coordinate system. The stress
distributions on the 6 brackets have been compared to find the critical region.
The most stressed nodes are placed at the lower bases of 2 opposite brackets.
Normal stresses at the tense and compressed fibres between these 2 nodes are
equal in modulus and opposite in sign, and also shear stresses are equal. In
general the stress distribution on the whole model is symmetrical with respect
to the y axis. This is due to a symmetry in the model and in the load directions
corresponding to a situation of peak stress on symmetrical brackets. To view
the position of the critical nodes, Von Mises stress is plotted in Figure 5.3.
The maximum values are written in correspondence of the critical nodes. Since
5.2 Results of the frequency analysis 65

Figure 5.3. Von Mises stress distribution, maximum values highlighted

the stress distribution on the critical brackets are equal, the top-left one in
Figure 5.3 is taken as a reference.
Near the bases of each bracket the 3 parallel vertical faces form as many angles.
In those regions the fatigue life of the component is reduced by the presence
of a fatigue stress concentration factor, since geometrical discontinuities are
present. For this reason the most loaded node on an angle point of the bracket
is the second point to verify in terms of fatigue. The node with the highest
Von Mises stress is called Node 1, while the corner node with highest stresses
is called Node 2. Node 1 and Node 2 are located in the same bracket. The
positions of the 2 critical regions are visible in Figure 5.4, where Von Mises
stress values are displayed. Note that the stress at Node 2 is nearly a half of
that at Node 1. The fact that the overall most loaded region of the bracket is
the lower base, is coherent with its thickness (10 mm) that is the highest. The
most external positions along the shell thickness, where the most tense and
compressed fibres lie, are called Z1 and Z2. Tables 5.1 and 5.2 resume the stress
components on Node 1 and Node 2 in these two positions. Only the non-zero
stress components are reported, in the global coordinate system. The stress
components are extremely small as expected, since load magnitudes are small.
On Node 1 normal stresses are slightly higher on the lower element face and
they change sign through the thickness, going from the most compressed fibres
to the most tense. This is coherent with the typical linear distribution of the
bending stresses along the thickness. There will be a neutral bending xy plane
in the shell elements close to their middle plane. All the stress components are
different from zero for Node 2, because the shell elements to which it belongs
are on a generically oriented plane. The values are nearly constant between
the lower and upper element face. Stresses on Node 2 are slightly higher at
the upper element face. A more detailed view of some stress components on
the critical bracket is given in Figures 5.5, 5.6, 5.8 and5.7, where some of the
previously tabled values on Node 1 and Node 2 are evidenced. The plots refer
only to the lower element face (Z1) and are viewed in the global coordinate
66 5. Fatigue verification of the brackets

Figure 5.4. Von Mises stress on the critical bracket, values on Node 1 (92.7
Pa) and Node 2 (50.9 Pa) highlighted

σx σy τxy
Lower (at Z1) -90.8 -32.8 24.3
Upper (at Z2) 84.5 29.8 -21.5
Table 5.1. Stress components [Pa] on lower and upper element face, Node 1

σx σy σz τxy τyz τxz

Lower (at Z1) 38.5 6.6 28.2 -16.0 6.1 -14,8
Upper (at Z2) 40.3 6.9 33.3 -16.7 7.8 -18.8
Table 5.2. Stress components [Pa] on lower and upper element face, Node 2

system. The global coordinate system coincides with the local element system
for the element to which Node 1 belongs, but this is not valid for Node 2.
This is testified that the shell element containing Node 1 has a null σz value.
The distributions of τxz and τyz are omitted. The labels on the right refer to
all brackets, and the figures are just a zoom on the examined bracket. The
vertical faces are almost unloaded, with exception of the regions where the free
5.2 Results of the frequency analysis 67

Figure 5.5. Distribution of σx , values on Node 1 and Node 2 evidenced

Figure 5.6. Distribution of σy , values on Node 1 and Node 2 evidenced

edges form an angle, like at Node 2.

68 5. Fatigue verification of the brackets

Figure 5.7. Distribution of τxy , values on Node 1 and Node 2 evidenced

Figure 5.8. Distribution of σz , values on Node 1 and Node 2 evidenced

5.3 Determination of the stress cycles 69

5.3 Determination of the stress cycles

To perform a fatigue analysis the stress cycle have to be determined. In this
case the cycle is expected to have zero mean value, since the centrifugal force
and moment rotate at the same frequency. Each bracket sees a peak of the
stress cycle for a certain position of the loads. The stress are inverted in sign
when the loads rotate of an angle π with respect to the position related to
peak stresses. To plot the stress cycles on the critical nodes, a study on the
time domain has been performed. A modal transient analysis has been carried
out, considering the frequency of 30 Hz. The imbalances have been defined
through their components. For each component, its analytical function of time
has been defined.
• Fx = 0.53 cos(2π · 30 · t);

• Fy = 0.53 sin(2π · 30 · t);

• Mx = −0.18 sin(2π · 30 · t);

• My = 0.18 cos(2π · 30 · t).

The parameters for the modal extraction are the same as those used in the
modal frequency analysis. The time interval has been set to 1.2 seconds
(36 cycles), sufficiently high to let the transient phase exhaust and show the
steady-state cycle. In this interval 720 time steps have been defined (20 steps
per cycle), to have a smooth representation of the output. The same number
of points have been used to build the discrete representation of the external
loads. Since the computational cost for this analysis would be too high, the
output has been requested for the critical nodes only, drastically reducing the
analysis time expense.
The sinusoidal stress cycles on Node 1 are visible in Figures 5.9 and 5.10. The
shear stress cycle on Node 1 and the analogous graphs on Node 2 have not
been reported, but their shape is qualitatively the same as in Figures 5.9 and
5.10. The initial transient phase slowly extinguishes, and after 1 second the
steady-state behaviour starts to be visible. The distribution of the peaks has a
decreasing exponential behaviour as time increases. After a sufficient time the
stress cycle will repeat with almost unchanged amplitude. At steady-state the
stress cycle can be considered to have zero mean value. For this reason the
stress cycle parameters can be derived from their maximum values.
To perform the fatigue analyses, the stress amplitudes are needed in the
element coordinate system. In this reference system the only non-zero stress
components are σx , σx and τxy . The maximum values for Node 1, tabled in the
first row of Table 5.1, remain unchanged since the global and local coordinate
system coincide. For Node 2 the stress components have been viewed in the
element coordinate system. The parameters defining the stress cycles of both
nodes are found doubling the maximum values in the element reference system.
70 5. Fatigue verification of the brackets

Figure 5.9. Time distribution of σx on Node 1

Figure 5.10. Time distribution of σy on Node 1

The mean values and amplitudes are listed in Table 5.3. The number of cycles
is equal to the number of wheel rotations. Considering an uninterrupted service
life of 12 years and 30 rotations per second N = 11353 · 106 cycles.
5.4 ESALOAD and ESAFATIG 71

σm,x σm,y τm,xy σa,x σa,y τa,xy

Node 1 0 0 0 90.8 32.8 24.3
Node 2 0 0 0 33.5 47.2 6.3
Table 5.3. Amplitude and medium stress [Pa] on the potentially critical nodes

5.4 ESALOAD and ESAFATIG

The damage accumulated on the potentially critical nodes has been evaluated
considering the whole service life of the bracket, from its production to the
on orbit functioning. This is possible with the use of the ESALOAD and
ESAFATIG packages, employed by the European Space Agency for fatigue
damage evaluation. ESALOAD allows to generate a stress spectrum from
multiple load curves associated to different load events, while ESAFATIG
enables to perform a fatigue analysis knowing the stress spectrum [22].
The flow chart of the steps necessary to derive a stress spectrum with ESALOAD
software is showed in Figure 5.11. With the ESALOAD tool, a good estimate of

Figure 5.11. ESALOAD flow chart

72 5. Fatigue verification of the brackets

the real load history of the system has been assumed, defining an event file which
contains a series of load generating events. Each event must be associated to a
load curve, which contains informations about the number of cycles versus mean
and alternating accelerations (but also pressure or temperature), expressed
as multiples of the gravitational acceleration. Load curves can be taken from
an existing database containing standard pre-calculated curves, which are
modifiable. ESALOAD also gives the possibility to define load curves, giving
informations about the load input at the base of the structure and about the
dynamics of the system through a transmissibility function. Putting together
the informations from all the events and the load curves associated to each
event, the software creates the load spectrum. This file contains load cycles
versus number of cycles for every defined load event. The software is then able
to calculate the stress spectrum, which is the same as the load spectrum but
with stresses in place of accelerations, knowing the unit stresses. Unit stresses
are the stress components induced by a unitary load (1g acceleration) in the
point of interest for the fatigue analysis. Knowing the stress components due
to a unitary load on a specific point, and the complete history of the load
cycles, ESALOAD gives in output the complete set of stress cycles (maximum
and minimum stress values) to which the structure is subject during its service
life. This output is the stress spectrum file.
The stress spectrum is the input for ESAFATIG, together with the S-N curve
of the material chosen from the software material database [20]. From stress
spectrum file, the minimum and maximum stresses Smin and Smax are derived
by combining the stress components into equivalent uniaxial stresses. The
stress ratio R is found dividing Smin by Smax . The S-N curves are represented
analytically by Equation 5.6.

logNf = A − B log(Seq − C) (5.6)

Equation 5.7 allows to calculate the equivalent alternating stress of a cycle

with non-zero mean stress.

Seq = Smax (1 − R)P (5.7)

The coefficients A, B, C and P are constants associated to the material. A

fatigue stress concentration factor Kf is applicable when the selected region has
peculiar geometries leading to local stress concentrations. With these inputs
a fatigue damage assessment can be done. The fatigue damage is evaluated
through the Palmgren-Miner linear damage accumulation rule:
ni
Di = (5.8)
Ni
where ni is the actual number of cycles of a particular stress amplitude and
Ni is the cycles to failure at that same amplitude. The values of ni are taken
from each row of the stress spectrum file, while the Ni are evaluated from the
5.4 ESALOAD and ESAFATIG 73

S-N curve of the material. The final output of ESALOAD is total damage
accumulated Φ. To find the actual fatigue damage f , the value of Φ has to be
multiplied by a factor of safety, equal to 4. The design is said to be acceptable
if the fatigue damage is lower than 1.

f = 4Φ < 1 (5.9)

The ESAFATIG program flow chart is showed in Figure 5.12. The term PFCI

Figure 5.12. ESAFATIG flow chart

in figure stands for potential fracture critical item. The PFCI is a part of the
74 5. Fatigue verification of the brackets

system which for some reason is considered critical in terms of fatigue. In the
present analysis the PCFI is the most loaded bracket, in particular the areas
of Node1 and Node 2.

5.5 Definition of the load history

A fatigue analysis which includes the entire load history of the bracket needs
the definition of the load cycles associated to every defined load generating
event. This is possible in the ESALOAD tool. These informations allow to
calculate a stress spectrum, that contains in its rows the number of cycles
related to an event, and the associated minimum and maximum stress values.
Some load curves have been selected from an existing database, while others
have been generated through the definition of the applied load and of the
system dynamics. The set of events is an estimate of the possible events in
the service life of a payload. The possible load generating events for a space
structure are resumed in Figure 5.13 [19]. Being completely unknown, the load

Figure 5.13. Events during the service life of a payload

cycles due to the production and assembly phases are not considered. The load
curves related to the horizontal, vertical and tilting handling are taken from
an existing load curve database, together with the road and air transportation
curves. The curves are defined in ESALOAD for careful handling and transport.
The transport curves are available per km travelled. The handling is defined
between each phase of Figure 5.13. The assumed transport events are:

• road transport of 10 km from production site to assembly site;

• road transport of 1 km from assembly site to test location;

5.5 Definition of the load history 75

• air transport of 5000 km from test location to launch site.

The effects of the vibrations induced by ground vehicles and by manoeuvres,

gust and landing loads during air transportation are taken in consideration.
The system is subjected to qualification tests to verify its compliance with the
specifications regarding various types of load environment such as sinusoidal,
random, and shock. The test are meant to reproduce the actual load envi-
ronment during flight (See Section 1.2). During qualification tests loads are
increased by the qualification factor KQ=1.25 to introduce a safety margin
which accounts for uncertainties related to mission variability and differences
between the tested article and the actual flight article. The load curves related
to the qualification tests have to be generated through the definition of the
load input seen at the base of the system and of the transmissibility function.
The base is in practice the upper face of the bracket, where it is mounted on
the satellite.
The load inputs consider a sine and a random test. The different nature of
the inputs require different ways of defining them. The sinusoidal load is
given as couples of acceleration amplitude and frequency, while the random
environment is given in terms of power spectral density (PSD) values and
associated frequencies. With these couples of inputs the program defines a
curve in frequency for each type of load input. Additional informations are
the duration of 60 s for both tests, and a sweep rate of 4 for the sine test.
The load inputs are and estimate of simple tests, and have been provided
externally. The sine qualification input is tabled in 5.4, while the random test
input is showed in Table 5.5. Both tests executed in x, y and z with the same

Frequency [Hz] Amplitude [g]

5 10
110 10
Table 5.4. Load input for the sine qualification test

Frequency [Hz] PSD [g2 /Hz]

20 0.001
100 0.02
500 0.02
2000 0.001
Table 5.5. Load input for the random qualification test

load input values, for a total of 6 qualification tests. These loads are applied
76 5. Fatigue verification of the brackets

on the bracket in the constraint condition of the test. Loads on a satellite

subsystem are typically derived from loads applied on the entire satellite, for
example at the satellite-launcher interface. Those load inputs are tabled in
the launcher user’s manual [3] for every different load environment. A transfer
function has to be applied to these loads to consider the amplification seen
by the subsystem of interest, due to the dynamic behaviour of the satellite
body. The high acceleration value of Table 5.4 are due to this amplification.
The analysis needed to calculate the exact load input at the bracket interface
is complex and expensive, and its informations are not yet available in this
context. For this reason the tabled load inputs are an estimate of the real
qualification test loads.
The sine and random loads during flight have to be taken in consideration as
well. Their values are derived from the load inputs qualification tests, divided
by the qualification factor KQ=1.25. Those events consider the sine, sine equiv-
alent and random vibrations seen at the bracket base. Sine and random flight
loads are needed to consider sinusoidal loads during the atmospheric flight,
some transients loads expressed as sine-equivalent excitations, and acoustic
(random) pressure fluctuations under the fairing, generated by the plumes of
the engines during lift-off.
The transmissibility function T (ω) contains informations about the dynamic
behaviour of the analysed system. It is defined as the absolute value of the
dynamic transmissibility. The dynamic transmissibility Tp (ω) is a frequency re-
sponse function which relates the applied force (or acceleration) to the response
due to that input. Each kth mode, describing a part of the total dynamics of the
system in some DoFs, contributes also to the definition of the transmissibility
function. In ESALOAD a transfer function can be defined in each spatial
direction p, as a superposition of the functions related to each mode [22].
N
ω2
Cp
X
Tp (ω) = 1 + 2 k
(5.10)
k=1 1 − ω + 2iζk ωωk + ωk
2

In Equation 5.10, N is the number of modes and Ckp is called acceleration

coefficient, and is modal effective parameter related to the participation of the
kth mode in the direction p. The sum of the acceleration coefficients must be
equal to 1. The transmissibility function of the bracket in each direction has
been derived following the definition on Equation 5.10. A modal analysis on the
single hard-mounted bracket has been performed to extract the needed modal
parameters. The output of the modal analysis is visible in Figure 5.14. The
T1,T2 and T3 directions corresponds to translations in x, y, and z. Rotations
are not considered, as prescribed by the ESALOAD manual [22]. The x and
y axes defined in this modal analysis lie in the plane of the lower base of the
bracket. The modal coefficients Ckp are fractions related to the importance of
the kth mode in the vibration along a certain direction. The modal effective
mass fractions a analogous coefficients, expressing the same informations. The
5.5 Definition of the load history 77

Figure 5.14. Modal parameters of the single bracket

modal parameters needed to define the transmissibility function are the natural
frequency ωk , the damping ratio ζk and the MEMF values. The first 4 modes
have been taken as representative of the whole system. For the damping ratio
ζk a small value 0.01 is set for the selected modes, to consider a very critical
situation in terms of amplification of the load at the junction.
Given a load input and a transmissibility function, the program calculates the
response ẍ of the system and an equivalent number of cycles. The different
load types require different mathematical treatments, being defined each with
its proper type of input. A detailed view on the calculation of response and
number of cycles for every type of load is given in [19]. For sine (harmonic)
loads, expressed as acceleration amplitudes q̈r specified at each frequency, the
response at the point of interest is found simply applying Equation 5.11.

ẍ = T (ω)q̈r (5.11)

For the bracket a different transfer function exists for each spatial direction,
while the defined sine test inputs are the same in every direction. The cal-
culation of the number of cycles changes with the load type as well. For a
sinusoidal load the frequency range is divided in small ranges of amplitude
fi − fi−1 in which the response ẍ is assumed constant. The cycles Ncyc are
calculated as
60(fi − fi−1 )
Ncyc = (5.12)
voct ln(2)
where voct is the sweep rate, measured in octaves per minute, that is the
derivative in time of the harmonic load frequency.
The random load environment refers for example to the acoustic pressure
generated inside the fairing during lift-off. Due to the probabilistic approach
needed to characterize the random excitation, the input must be given through
its power spectral density function Sr (ω). The PSD of the forcing signal q̈r is a
78 5. Fatigue verification of the brackets

real and positive function which represents the distribution in frequency of the
root of the mean square of the signal q̈rms . The PSD contains informations of
the energy (power) content of a signal in frequency. Knowing the PSD of the
input and the transmissibility T (ω), the PSD of the response in the direction
p is calculated as:
S p (ω) = T (ω)T ∗ (ω)Sr (ω) (5.13)
where T ∗ (ω) is the complex conjugate of T (ω). The knowledge of S p (ω) allows
to reconstruct the root mean square of the response in each direction ẍp . If
the mean of the signal is zero, the definition of root mean square (or standard
deviation) is given by Equation 5.14.
sZ
∞
ẍprms = S p (f )df (5.14)
0

Since any random process has the tendency to follow a Gaussian distribution,
taking a value of the maximum response equal to 3 times its root mean square
value allows to have a very good confidence that this maximum acceleration is
not exceeded. The number of cycles for random acoustic inputs is calculate
under the additional assumption that the excitation is stationary. The analo-
gous to the number of cycles for a random signal is the rate of upcrossings of a
certain acceleration level. The expected rate of upcrossings of an acceleration
level b is:
−b2
p
vbp = v0p e 2M0 (5.15)
where v0p is the number of zero crossing per second of an excitation in the
direction p. M0p is the variance of the motion in the same direction, when the
mean value is assumed to be null. The relations for the calculation of v0p and
the generic Mip are showed respectively in Equation 5.16 and 5.17.
v
u p
u M2
v0p = t (5.16)
M0p
sZ
∞
M0p = S p (f )df (5.17)
0

The last load curve has defined from the already existing curve related to
the ascent phase of the Ariane 5 launcher. It is based on low frequency
measurements at the launcher-satellite interface with many different payloads.
This load curve, together with the sine and random flight load curves already
defined, allows a description of the most important loads occurring during flight.
Two corrections are applied, scaling acceleration amplitudes with appropriate
coefficients. This is done under the realistic hypothesis that, changing the
structure to which ascent loads are applied, the only changing parameter are
acceleration amplitudes. The first correction is related to the fact that the
5.5 Definition of the load history 79

limit accelerations must be referred to the Vega C launcher. To do this, the

maximum limit loads of Figure 4.1 in the lateral (x and y) and longitudinal
(z) directions are considered. Since the standard Ariane 5 curve is normalized
to 100g, acceleration amplitudes must be scaled according to Equation 5.18.
LL
Pa,act = Pa,std · (5.18)
100
Pact and Pstd are respectively the scaled and standard acceleration amplitudes,
while LL is the maximum value of the Vega C limit load in the selected
direction. The second correction refers to the fact that the load curve is
derived at the launcher-satellite interface and not at the bracket location. The
sinusoidal acceleration at the bracket location is amplified by the dynamics of
the satellite. This correction is applied defining for each direction a scaling
factor SF , found applying Equation 5.18.
DLL − Pm
SF = (5.19)
Pa
Pm is the mean value of the acceleration (Pm = 0), Pa is the acceleration
amplitude and DLL are the design limit loads of the structure. The DLL values
are taken from the maximum accelerations to which the structure is subject
during the sine qualification test (first row of Table ??). The coefficients are
resumed in Table 5.6. The nodes of interest to perform the fatigue analysis are

Coeff. Acc. ampl. X Acc. ampl. Y Acc. ampl. Z

LL
100
0.0135 0.0135 0.05
SF 7.4 7.4 2
Table 5.6. Coefficients for the correction of the ascent phase load curve

known from the result of the frequency analysis. The selected nodes are the
one with the highest peak stress value (Node 1) and the most stressed node
between those where the faces of the bracket form an angle (Node 2). The
unit stresses of Node 1 and Node 2 are obtained through a static analysis on
the bracket. The unit stress values are defined for each DoF, so an inertial
acceleration of magnitude 1g has been applied along each individual DoF.
The in-plane stress components have been extracted, and are tabled in Table
5.7. ESALOAD requests the stress components in the local coordinate system
of the element, therefore only the 3 in-plane stress components are different
from zero. The unit stresses have been associated to every load curve defined
in this section. The outputs of this association are the stress spectra of the
two potentially critical points. The stress spectrum contains each existing
stress cycle, defined by the maximum and minimum stress components and
80 5. Fatigue verification of the brackets

Node 1 Node 2
DoF σx σy τxy σx σy τxy
Trasl. X 7.8 2.5 2 3 4.3 0.6
Trasl. Y 1.7 1.0 0.1 2.1 2.9 0.4
Trasl. Z 0.6 0.5 0.7 1.9 2.6 0.3
Rot. X 4.5 2.2 0.3 3.7 5.2 0.7
Rot. Y 11.8 4.4 2.2 2.0 2.9 0.4
Rot. Z 1.5 0.7 0.2 0.8 1.0 0.2
Table 5.7. Unit stresses of the potentially critical nodes [MPa]

the associated number of cycles. The informations in Table 5.3 have been
expressed as maximum and minimum values, and added to the stress spectrum
file of the respective node. This complete stress spectrum is an input of the
ESAFATIG package, together with the S-N curve of the 7075 aluminium and
the value of the stress concentration factor. The final output is the linear
damage accumulated on Node 1 and Node 2.

5.6 Fatigue damage assessment

In this section the accumulated damage is evaluated on the potentially critical
nodes of the bracket, and some considerations are made the most critical
load events. The design philosophy for the brackets is safe-life. A continuous
functioning of the rotating mechanism in orbit has to be guaranteed for 12
years. This information is represented by the number of cycles related to the
on orbit imbalances of the flywheel. The S-N curve (Wöhler curve) of the
7075-T6 aluminium alloy has been selected from the ESAFATIG database.
The curve refers to a bare aluminium sheet. The stress concentration factor
is different for Node 1 and Node 2. Its value is reflected on the shape of the
S-N curve. The curves (see Section 5.4) are showed in Figure 5.15 and 5.16.
The theoretical stress concentration factor is defined as the ratio of the actual
maximum stress over the a reference stress (the theoretical stress in a situation
where the discontinuity does not exist).
σmax
Kt = (5.20)
σref

For ductile material under dynamic loading a fatigue stress concentration factor
has to be calculated. It is surely smaller than Kt = 1, since plasticization in
5.6 Fatigue damage assessment 81

Figure 5.15. S-N curve for Node 1, Kf = Kt = 1

Figure 5.16. S-N curve for Node 2, Kf = Kt = 1.5

82 5. Fatigue verification of the brackets

Figure 5.17. Header of the ESAFATIG output file

ductile materials makes less critical the stress concentration around a notch. It
will be conservatively assumed that Kt = Kf , since to know the exact Kf the
effective fillet radius is needed. Von Mises stresses due to the imbalances of
Figure 5.4 may suggest that the stress concentration on Node 2 results in an
equivalent stress around 3 times the reference value. This would be misleading,
since in the model does not include any fillet in correspondence of its corner
points. A fillet radius should be designed in order to lower the concentration
of stresses on Node 2, employing a specialistic text [17].
The ESAFATIG output files are generated. Their heading is showed in Figure
5.17. The file contains a series of rows representing the maximum, minimum
and equivalent stresses, filtered according to a minimum threshold. This
threshold is inversely proportional to the number of cycles. A section extracted
from the ESAFATIG output file, regarding Node 1, is showed in Figure 5.18.
For every row in figure, the program checks if the equivalent stress Seq is below
the yield stress Syld and the experimental stress Sexp . Sexp is the value used in
the experiments needed for the definition of the S-N curve. The last column
is the Miner linear damage accumulated in consequence of the corresponding
stress cycles. From line number 28 on the damages are the highest, in the order
of 10− 2. This is true for both Node 1 and Node 2. It has been verified that
the correspondent lines are due to the random qualification test and random
flight loads. The random events are the only causing wide band loads, in the
[20,2000] Hz range. As was visible in Figure 5.14, the first natural frequency
of the bracket is relatively high. The transmissibility functions Tp (ω) have
been set according to the modal parameters of the bracket, therefore they
5.6 Fatigue damage assessment 83

Figure 5.18. ESAFATIG output file, stress cycles for Node 1

reach high values in correspondence of its natural frequencies. Only random

loads can generate high responses and high stresses, exciting the bracket in
a frequency range that includes some of its eigenfrequencies. In addition, a
critical damping ζ = 0.01 has been set for all modes, so the resonance peaks
of the transmissibility function will not be damped sufficiently. According to
the modal coefficients of Figure 5.14, the most critical accelerations due to a
random event are in the y direction.
The on orbit imbalances do not affect the accumulated damage. These cycles
are filtered by the program, being lower than the fatigue threshold although
the number of cycles is very high N = 11353 · 106 . It is known that scaling the
values of the imbalances of Equation 5.5, the disturbance loads and the induced
stresses scale linearly. The stress cycles related to the disturbances in both
Node 1 and Node 2 stress spectrum files, have been scaled of a factor 10. The
accumulated damage has resulted to be insensible to the scaled values too. It is
concluded that, considered the low rotation frequency, the imbalances induced
by the rotating mechanism are by far not sufficient to lead to a propagation of
the fatigue crack.
The total damages accumulated on the locations of Node 1 and Node 2 are
84 5. Fatigue verification of the brackets

Figure 5.19. Total damage and life coefficient, Node 1

Figure 5.20. Total damage and life coefficient, Node 2

showed respectively in Figure 5.19 and 5.20. A safety factor equal to 4 is

already included into the life coefficient, while it has to be applied on the total
damage. Applying the safety factor on Node 1, the most stressed bracket node,
a total damage of 0.1170 is obtained. This node is considered to be verified
without any further considerations. Applying the safety factor to the total
damage on Node 2, which is the most stressed corner node, the value 0.4876
is obtained. Node 2 is the critical node of the brackets in terms of fatigue.
The damage is not far from the limit value of 1. It is important to notice that
the damage for Node 2 has been calculated setting the stress concentration
factor to 1.5. A further analysis on Node 2 with a factor equal to 2 showed
that in this case the damage exceeds the threshold value of 1. It is concluded
that a more detailed model is needed for the refinement of the fatigue design.
Fillets have to be added to the brackets in such a model, to have more precise
informations on the exact value of the total damage on the critical point. The
fatigue analysis can be considered to have have a positive result only if the
radius of those fillets is sufficient to ensure that Kf ≤ 1.5.
 85

Conclusions

The design of a sandwich panel for the mounting of a 99 kg rotating mechanism

has been performed. The panel houses 3 interfaces with the rotating mechanism
and 6 aluminium brackets guarantee its mounting on the satellite. The bracket
disposition is symmetrical with respect to the y axis, and is chosen to make
room for the interfaces of the wheel. An optimal sandwich configuration has
been determined for both cases of CFRP and aluminium skins. The best con-
figuration in terms of compliance with the design requirements involves the use
of CFRP balanced and symmetric laminates for the skins, and an aluminium
honeycomb core. The laminates are made up of 8 plies for a total thickness of
0.6 mm. The honeycomb is 50.8 mm (2 inches) thick. The mass of the panel
is 4.49 kg. The two equivalent optimal stacking sequences are [0/90/-45/45]s
and [0/-45/90/45]s. With this configuration, the first main bending mode
of the system involves the 57 % of the total mass in an out-of-plane motion,
with a natural frequency of 55.4 Hz. This frequency is far above the threshold
of 40 Hz needed for the dynamic decoupling with the satellite. It has been
observed that a thicker honeycomb is more efficient in increasing the panel
bending stiffness with respect to a thickening of the skins. This reflects in
a high increase in the relevant eigenfrequency with the use of the 50.8 mm
core. The resistance to the quasi-static loads exerted during the Vega C ascent
phase was ensured, with safety factors sufficiently high for a preliminary study.
The maximum failure index is found considering only the most critical QSL
combination. The failure index distribution reaches its maximum value in the
-45° external ply of the laminates. The static verification gave positive results,
with a maximum failure index of 0.294 on the CFRP skins. Further static
analyses were performed to consider every possible combination of the lateral
and longitudinal quasi-static loads. The analyses allowed to find the conditions
of maximum interface loads and of maximum stresses on the brackets, which
were below the elastic limit. The components of the maximum interface forces
and moments were extracted for a future design of the bolted joints. An
optimal sandwich configuration was found also in the case of aluminium skins
due to the high thermal conductivity of aluminium, which allows to better
dissipate the heat produced by the electric motor. This solution involves the
use of the 50.8 mm honeycomb and 0.5 mm aluminium 7075 plies. The mass
of the panel in this case is 5.15 kg.
86 5. Fatigue verification of the brackets

A modal frequency analysis allowed to derive the stress cycles on 2 poten-

tially critical nodes of the brackets, caused by the imbalances of the rotating
mechanism. A fatigue life assessment was performed on these points, with
the evaluation of the linear damage accumulated during handling, transports,
tests, ascent phase and on orbit disturbances. A first result of the analysis
is that the most damaging environments for the bracket are those covering
high frequencies, such as random loads, since the first natural frequency of
the bracket is 630 Hz. The most important result is that the disturbance
loads coming from the flywheel imbalances do not cause any damage, since
their amplitude is not sufficient to cause a propagation of the crack within the
material. This remains true even with higher values of the disturbances. The
total damage values that the most critical node is the one located in a point
where a face of the bracket form an angle. The damage on this point takes
into account a stress concentration factor. The values Kf = 1.5 and Kf = 2
were evaluated, and the safe-life is verified only in the first case. This result
suggest that a refined model is needed for the brackets, with proper fillets on
the points of geometric discontinuity. The fillets must have a radius sufficient
to ensure that Kf ≤ 1.5 in the corner point.
For the development of the project the brackets have to be redesigned to
optimize the material usage, reducing thicknesses or changing the material. It
has been verified that such a design can be done, since even the most loaded
regions of the brackets are far below the yield strength of the material. A
more detailed finite element model of the system is needed in a further study.
Some suggested refinements are the 3-D modelling of the honeycomb and
the definition of fillets on the brackets for a more precise definition of the
accumulated damage on the points of geometric discontinuity.
 87

Bibliography

[1] M. L. Adams. Rotating machinery vibration: from analysis to troubleshoot-

ing. CRC Press, 2000.

[2] J. Alcorn, C. Allard, and H. Schaub. Fully coupled reaction wheel static
and dynamic imbalance for spacecraft jitter modeling. Journal of Guidance,
Control, and Dynamics, 41(6):1380–1388, 2018.

[3] Arianespace. Vega C User’s Manual. https://

www.arianespace.com/wp-content/uploads/2018/07/
Vega-C-user-manual-Issue-0-Revision-0_20180705.pdf, 2018.

[4] J. Bai. Advanced fibre-reinforced polymer (FRP) composites for structural

applications. Elsevier, 2013.

[5] P. Berlin. The geostationary applications satellite. Number 2. Cambridge

University Press, 1988.

[6] L. A. Carlsson and G. A. Kardomateas. Structural and failure mechanics

of sandwich composites, volume 121. Springer Science & Business Media,
2011.

[7] ECSS. Glossary. https://ecss.nl/glossary/, 2022.

[8] EDN. Satellite anatomy. https://www.edn.com/

satellite-anatomy-101/, 2015.

[9] European Space Agency . ESA Website. https://www.esa.int, 2022.

[10] A. García-Pérez, Á. Sanz-Andrés, G. Alonso, and M. C. Manguán. Dy-

namic coupling on the design of space structures. Aerospace Science and
Technology, 84:1035–1048, 2019.

[11] P. Gaudenzi. Lectures of space constructions. Sapienza University of

Rome, 2020.

[12] R. M. Jones. Mechanics of composite materials. CRC press, 2018.

[13] A. P. Mouritz. Introduction to aerospace materials. Elsevier, 2012.

88 Bibliography

[14] M. Nastran. Dynamic analysis user’s guide. MSC Software, 2013.

[15] A. T. Nettles. Basic mechanics of laminated composite plates. 1994.

[16] J. N. Pelton, S. Madry, and S. Camacho-Lara. Handbook of satellite

applications. Springer, 2017.

[17] W. D. Pilkey, D. F. Pilkey, and Z. Bi. Peterson’s stress concentration

factors. John Wiley & Sons, 2020.

[18] J. N. Reddy and A. Miravete. Practical analysis of composite laminates.

CRC press, 2018.

[19] E. Secretariat. Spacecraft mechanical loads analysis handbook. European

Cooperation for Space Standardization, 2013.

[20] Sinnema, B. Henriksen, and Castel. ESFATIG User’s Manual, version

5.2.0.a. ESCRACK, 2018.

[21] Z. S. Toor. Space applications of composite materials. Journal of Space

Technology, 8(1):20, 2018.

[22] Veul, S. Brunetti, and C. Henriksen. ESALOAD User’s Manual, version

4.3.0.a. ESACRACK, 2018.

[23] J. J. Wijker. Spacecraft structures. Springer Science & Business Media,

2008.

[24] T. Yasaka and J. Onoda. Spacecraft structures. 2003.

[25] X.-S. Yi, S. Du, and L. Zhang. Composite materials engineering, volume
2. 2018.