Development of an

Optimization Framework
for Landing Gear Design

P. van Ginneken
Technische Universiteit Delft
 D EVELOPMENT OF AN O PTIMIZATION
F RAMEWORK FOR L ANDING G EAR D ESIGN

by

P. van Ginneken

in partial fulfillment of the requirements for the degree of

Master of Science
in Aerospace Engineering

at the Delft University of Technology,

to be defended publicly on Thursday February 18, 2016 at 10:00 AM.

Supervisor: Dr. ir. M. Voskuijl,

Thesis committee: Prof. dr. ir. L. L. M. Veldhuis, TU Delft
Dr. ir. R. Breuker de, TU Delft
Ir. P. Vergouwen, Fokker Landing Gear B.V.

An electronic version of this thesis is available at http://repository.tudelft.nl/.

Thesis registration number: 073#16#MT#FPP
Equivalent number of words = 18,800 + 43 x 200 = 27,400
 A BSTRACT
An opportunity was identified to improve the traditional landing gear design process. Especially in the con-
ceptual design phase, too many man-hours are consumed by making the same calculations over and over
again, for different concepts. An existing landing gear layout is therefore often used as starting point, to sim-
plify the design process. This results in little technical progress. Additionally, integration between the differ-
ent disciplines involved is sub-optimal, which can lead to inconsistent results. In this thesis, an optimization
framework is described that can perform the preliminary design of a landing gear fully automated. It ensures
that communication between disciplines is respected by adding a top-level optimizer which is in charge of
changing the design variables. This top-level optimizer will find the optimal landing gear solution, consid-
ering the interaction of the different landing gear disciplines and also the interaction between the landing
gear and the aircraft. For that purpose, different parametric landing gear topologies will be defined. These
different gears are all evaluated to find the optimal design solution. Two main research goals are formulated.
The first goal is to define such an optimization framework and the second goal is to create a proof of concept
tool that incorporates its principles.
The set-up of the framework can be compared to the different layers of an onion. In four steps, the best
preliminary landing gear solution is obtained with an increasing level of detail. In the first step, top-level pa-
rameters such as gear layout, strut geometry and strut position are considered based on static load cases and
the use of rigid body simulations. Top-level aircraft parameters such as the maximum take-off weight, the
cg-range, the wingspan and the touchdown velocity are needed for this step. In the second step, secondary
gear components such as lugs and pins are added to the simulation model; the gear geometry is investigated
in more detail and the cost of the gear is computed. The third step is to compute dynamic load cases instead
of the static ones used before, and to restart from the beginning. As a last step, flexible members are added to
the simulation model and more elaborate stress calculations are performed. Constraints to this framework
include that foldability of the gear inside the landing gear bay, the fact that no part of the aircraft may touch
the ground except for the landing gear, and minimum wall thicknesses for all members.
A proof of concept was developed that is able to perform all calculations defined for the first step of the
framework, and some calculations of the second step. A multi-body model is developed to investigate the
gear geometry in more detail, but the landing gear cost and the secondary components are not yet added.
Verification of the proof of concept is done by comparing the results against a previous concept study, of
which the preliminary design information was available. With similar geometries for the proof-of-concept
framework and the previous concept study, the weight computed by the framework has an error margin of
around 4 percent. If the gear geometry is allowed to change, the proof of concept tool is able to reduce the
gear weight with 7 percent compared to the original concept study.
Ultimately, the realization of this framework greatly reduces the repetitive tasks in the design phase of a
landing gear. This shortens the timespan of the design phase, opening the possibility to also evaluate non-
standard solutions that may be lighter, safer and/or cheaper.

iii
 A CKNOWLEDGMENTS
This research would not have been possible without the excellent help and working environment offered by
Fokker Landing Gear B.V. I am very thankful to have been given the opportunity of doing a final thesis at this
inspiring place.
There were quite a few people involved in helping and motivating me, but five need to be mentioned by name.
First, Peet Vergouwen of FLG and Mark Voskuijl of Delft Univerity of Technology for being my tutors. Then,
Bert Verbeek of FLG for creating the VLM model and encouraging me to try new programming methods. I also
want to thank Mike Smeets for his analytic work on the initial concept gear and helping me to understand it.
Finally, thanks to Rob Bergers for proof-reading most of my documents and for keeping spirits up.

v
 C ONTENTS

List of Figures ix

List of Tables xi

Nomenclature xiii

1 Introduction 1
1.1 Principles of landing gear design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Landing gear layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Landing gear topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Principles of engineering optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Available engineering optimization methods . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 MDO architectures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.3 The optimizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Previous research on landing gear design and engineering optimization . . . . . . . . . . . . . 6
1.3.1 Landing gear integration in aircraft conceptual design. . . . . . . . . . . . . . . . . . . 6
1.3.2 Landing gear design in an automated design environment . . . . . . . . . . . . . . . . 7
1.3.3 An approach for sizing and topology optimization integrating multibody simulation . . . 8
1.4 Research questions and project scope. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 "Landing gear in thirty days" 11

2.1 Block A: basic parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Block B: Preliminary sizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Block C: dynamic load cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Block D: Verify results by FEM calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Positioning constraints tricycle landing gear 17

3.1 No parts of the aircraft touch the ground, except the landing gear . . . . . . . . . . . . . . . . 17
3.2 Allowed locations for the centre of gravity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Load on the nose gear. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4 Operation on the airfield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4 Ground reaction loads on the wheels 23

4.1 Reference frame used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Static loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2.1 Braking loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.2 Turning loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.3 Landing with side load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3 Sub load cases dependent on tyre inflation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

5 Initial gear set-up 27

5.1 Initial starting point for the optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.1.1 Features of the original concept study . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.1.2 Limitations of the original concept study . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.2 Determining the loads in the hardpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2.1 Analytic solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2.2 Using dedicated software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

vii
viii C ONTENTS

6 Component weight estimation 31

6.1 Stresses in an H-beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.2 Stresses in cylinders 1: external forces and moments . . . . . . . . . . . . . . . . . . . . . . . 32
6.3 Stresses in circular tubes 2: internal pressures . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.4 Stresses on a socket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
7 Description of the proof of concept 35
7.1 MDO architecture chosen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
7.2 Software implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
7.3 Input handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7.4 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7.5 Output handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
8 Verification of the proof of concept 41
8.1 Analytical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.2 Multi-body model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
9 Results 45
9.1 Results of case studies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
9.1.1 Changing the gear location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
9.1.2 Changing the gear geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
9.2 Other possibilities offered by the framework . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
10 Conclusion 51
11 Recommendations 53
11.1 Expanding the tool: adding more capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
11.2 Expanding the tool: adding more detail . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
A Load cases from regulations 55
B Load cases description 57
B.1 Ground loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
B.2 Touchdown loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
C Detailed description of LIT software 63
C.1 Notes on Object-Oriented Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
C.2 Notes on fmincon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
C.3 Structure of the software folder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
C.4 Input files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
C.5 MATLAB files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
C.5.1 MATLAB files for the landing gear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
C.5.2 MATLAB files for structural members . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
C.5.3 Runfile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Bibliography 71
 L IST OF F IGURES

1.1 Tricycle, taildragger and bicycle gear layouts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Standard strut topologies in use. Figure adapted from [9]. . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Principles of IDF and MDF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Principle of CO [21]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Existance of local and global minimum on a constrained interval. . . . . . . . . . . . . . . . . . . 6
1.6 General layout of the research by Chai and Mason [1]. . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.7 Landing gear design module by Heerens [2] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.8 Research by Cumnuantip [3] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.1 Proposed set-up of the landing gear in thirty days tool. . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2 An overview of different standard landing gear configurations. . . . . . . . . . . . . . . . . . . . . 14

3.1 Definition of A, B and cg-range [2]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3.2 Geometrical constraint relating to wing tip clearance [1] . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3 Geometrical constraint relating to engine clearance [29] . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Visualization of the sideways turnover angle ψ and the triangle spanned by the NG and MLG . . 19
3.5 Visualization of the sideways turnover constraint [28] . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.6 Visualization of the turning on runway requirement [2] . . . . . . . . . . . . . . . . . . . . . . . . 21
3.7 Visualization of the turning on taxiway requirement [1] . . . . . . . . . . . . . . . . . . . . . . . . 22

4.1 Aircraft global reference frame as used throughout this thesis . . . . . . . . . . . . . . . . . . . . . 23

4.2 Free body diagram for turning conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

5.1 Isometric view of original concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5.2 Stick diagrams of the initial concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.3 The virtual VLM model: isometric and side view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

6.1 Model of an H-beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

6.2 Forces in x, y, and z direction acting on a cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.3 Allowable bending stress of cylinders made from 300M steel [37] . . . . . . . . . . . . . . . . . . . 33
6.4 Cylinder with hoop stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.5 Socket parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

7.1 Optimization structure of the POC framework. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

7.2 Classes derived from parent StructuralMember. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
7.3 Classes derived from parent LGSystem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

9.1 4-D representation of the design space. System weight represented by color as a fourth dimen-
sion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9.2 X position MLG fixed in each sub-figure. Moving aft from top left and in clockwise direction . . 46
9.3 X position NLG fixed in each sub-figure. Moving aft from top left and in clockwise direction . . . 47
9.4 Y position MLG fixed in each sub-figure. Moving outboard from top left and in clockwise direction 47
9.5 Weight of the trailing arm as a function of the main gear location . . . . . . . . . . . . . . . . . . . 48
9.6 Stick diagrams of the initial concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
9.7 Optimizing the gear by changing the locations of the hardpoints. . . . . . . . . . . . . . . . . . . . 49

B.1 Free body diagram for static load condition. Side view . . . . . . . . . . . . . . . . . . . . . . . . . 57
B.2 Free body diagram for turning conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
B.3 Free body diagram for the second nose wheel jaw condition. Top view on the left and side view
on the right. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

ix
x L IST OF F IGURES

B.4 Level landing condition [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

B.5 Tail down landing condition [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
B.6 Side load landing condition [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
 L IST OF TABLES

2.1 Inputs for routine 1 in block A of LIT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2 Inputs for routine 2 in block A of LIT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.1 Aircraft design group classification [30] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.2 Aircraft approach category classification [30] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Operation on taxiway [30] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4.1 Loads on left and right main gear strut for different turning load cases. . . . . . . . . . . . . . . . 25

5.1 Types of joint at hardpoints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

7.1 Inputs for POC tool: aircraft parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

7.2 Inputs for POC tool: shock absorber parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

8.1 Verification of analytical and multi-body model with respect to original model. . . . . . . . . . . 42

B.1 Loads on left and right main gear strut for different turning load cases. . . . . . . . . . . . . . . . 59
B.2 Load cases for aircraft towing [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

C.1 Runtime needed for the different algorithms of fmincon . . . . . . . . . . . . . . . . . . . . . . . . 65

xi
 N OMENCLATURE

L IST OF SYMBOLS
Symbol Unit Description
A m2 cross-sectional area
a m horizontal distance between nose landing gear and centre of gravity
B m wingspan
b m horizontal distance between main landing gear and centre of gravity
cg-range m the distance between the most aft and most forward possible cg position
D N drag force in global coordinate system
d m horizontal distance between main landing gear and nose landing gear
E Pa Young’s modulus
e m vertical distance between ground and centre of gravity
F N force acting on a member
f - 1) dynamic response factor
h m height
I m4 area moment of inertia
J m4 polar moment of inertia
L m length
M N*m moment
n - load factor
p Pa pressure
Q m3 first moment of area
R - Ratio
r m radius
S N side force in global axis system
T N*m Torque
t m 1) lateral distance between the main gear struts
2) thickness
u m distance between centres of main gear tyres
V N vertical force in global axis system
v m distance between centres of nose gear tyres
W N weight
w m width
x m x position with respect to reference frame aircraft
y m y position with respect to reference frame aircraft
z m z position with respect to reference frame aircraft

G REEK LETTERS
Symbol Unit Description
β rad nose-wheel steering angle
Γ rad dihedral angle of the wing
∆ rad sweep angle of the wing
δ m clearance
θ rad pitch angle
µ - friction coefficient
σ Pa normal stress
τ Pa shear stress
φ rad roll angle

xiii
xiv L IST OF TABLES

ψ rad sideways turnover angle

S UBSCRIPTS
Symbol Description
a axial
av average
b bending
br bearing
h hoop stress
i inner
L left
M main gear
max maximum
min minimum
N nose gear
o outer
R right
r radial
s shear
st torsional shear
t tension / tensile
tr transverse
u ultimate
x direction of the local x-axis
y direction of the local y-axis
z direction of the local z-axis
 A CRONYMS
AAC Aircraft Approach Category.

AC Advisory Circular.

ADG Aircraft Design Group.

BLISS Bi-Level Integrated System Synthesis.

CAD Computer Aided Design.

CG Centre of Gravity.

CO Collaborative Optimization.

CSSO Concurrent Sub-Space Optimization.

DLR German Aerospace Center.

FBD Free Body Diagram.

FEM Finite Element Method.

GA Genetic Algorithm.

IDF Individual Discipline Feasible.

KBE Knowledge Based Engineering.

LIT Landing gear In Thirty days.

MBS MultiBody Simulation.

MDF Multidisciplinary Feasible.

MDO Multidisciplinary Design Optimization.

MF Main Fitting.

MLG Main Landing Gear.

MLW Maximum Landing Weight.

M O S Margin of Safety.

MRW Maximum Ramp Weight.

MTOW Maximum Take-Off Weight.

NLG Nose landing Gear.

OEM Original Equipment Manufacturer.

OOP Object-Oriented Programming.

xv
xvi A CRONYMS

POC Proof of Concept.

RA Retract Actuator.

SA Shock Absorber.

SQP Sequential Quadratic Programming.

TA Trailing Arm.

TG Tail Gear.

WA Wheel Axle.
 1
I NTRODUCTION
How can the time spent in the concept design phase be decreased, while also delivering a product with a
higher level of detail? That question triggered this thesis project. When Fokker Landing Gear B.V. (FLG) is
asked to do a concept study, it is the start of a hectic period. Different landing gear (LG) solutions need to be
evaluated and the customer needs to be convinced that the offered solution is better than the concepts pro-
posed by the competition. Among many other activities, preliminary stress calculations are made to size the
landing gear components, a shock absorber is defined, the kinematics are evaluated, the rolling stock (tyres,
brakes, wheels) is selected and the virtual product is defined in a CAD program.
Four main problems are identified during this concept design phase. Firstly, engineers are spending most
of their time performing repetitive work, not creative. The same calculations are made over and over to an-
swer the question "what if part X is moved to position Y?". Some simple and self-written Mathcad sheets are
available for different stages of the conceptual design process, but post-processing still takes more time than
wanted. Secondly, the initial concepts for a new aircraft are often based on an existing gear, to simplify and
speed up the design process. In the best case, this leads to a non-optimal, but acceptable, solution. In the
worst case, there is no existing gear that fulfils all requirements. Thirdly, the coordination between different
involved disciplines is not optimal. Even in concept design, there are many parameters involved and usually
several concepts are evaluated at the same time. It is critical that everyone involved uses the same parameters
for their calculation. Although this may seem trivial, previous experience shows that this is often not the case.
Lastly, the LG and the aircraft are often designed as two separate systems, without much integration1 . This
may result in an incomplete set of requirements for the landing gear. If that is discovered in a late stage of the
design process, it can in turn lead to unwanted, costly, and time-critical redesign activities.
This thesis investigates if the landing gear design process can benefit from the use of engineering optimiza-
tion methods, to overcome the above mentioned problems. For that purpose, two main research objectives
are set. First, an optimization framework for landing gear design is defined. It describes how MDO can be
used to find the best landing gear for any aircraft, within the boundaries of preliminary design. This frame-
work must be able to run with a limited amount of input parameters and help the design team to determine
the conceptual landing gear within six weeks. It is therefore named the "Landing gear In Thirty days" tool:
LIT.
LIT can ultimately be used to search a target of the best landing gear for any aircraft, while eliminating many
of the repetitive design tasks done by a designer and freeing time for creative investigations. This results in a
broader (more options considered) and more detailed analysis of possible landing gear solutions, within the
same time frame that is needed nowadays.
The second objective then is to build a proof of concept (POC) tool. This POC tool incorporates the principles
described for LIT and proves that these principles have practical purpose.

Previous research in this field has been conducted by Chai and Mason [1], Heerens [2], and Cumnuantip
[3]. These publications are described in section 1.3 This chapter first gives an overview of the two disciplines
involved in this thesis: landing gear design and engineering optimization. The research objective can be
found in section 1.4, while section 1.5 explains the layout of this thesis.
1 Interviews with Mike Smeets, Peter de Haan, Tjaard Sijpkes, Fokker Landing Gear B.V.

1
2 1. I NTRODUCTION

1.1. P RINCIPLES OF LANDING GEAR DESIGN

It is assumed that most readers of this text are already familiar with the basics of LG design and therefore it is
kept short. If that is not the case and if you are intested, Aircraft landing gear design: principles and practices
[4] is recommended. For landing gear design from the perspective of an aircraft designer, part three of the
Roskam series is an excellent reference [5].
The landing gear has three principle functions: to provide a smooth ride for passengers, to introduce the
ground reaction loads into the airframe structure and, optionally, to retract into the landing gear bay. This
section discusses two top-level design choices: the gear layout and the strut topology. The different compo-
nents of a gear strut are described in section 5.1.

1.1.1. L ANDING GEAR LAYOUT

There are two common options for the LG layout: a tricycle landing gear and a taildragger landing gear. The
tricycle configuration is seen on virtually all larger passenger aircraft, but also on many helicopters. The
tricycle layout has its Main Landing Gear (MLG) behind the Nose landing Gear (NLG). The main advantages
of a tricycle landing gear are that the aircraft floor is level to the ground during taxi, that the pilot has good
visability from the cockpit and that the steering characteristics on the ground are good compared to the other
configurations [5].
A taildragger gear is most often seen on smaller aircraft but also on helicopters. The taildragger has its MLG
in front of the Tail Gear (TG). The main advantage of a taildragger configuration is that it is usually light [5].
Disadvantages are the visibility from the cockpit due to its raised nose and the take-off procedure which is
more difficult with a taildragger configuration.
A third type that is rarely used is the bicycle gear layout of the B47. It is only used when the other two options
are not feasible, due to very specific aircraft requirements. It is both complex and heavy and will not be
considered in this thesis.
The tricycle, taildragger and bicycle gear are visualized in figure 1.1:

(a) Landing gear of the Lynx Mk9: (b) Landing gear of the Apache: tail- (c) Landing gear of the B47: bicycle
tricycle [6]. dragger [7]. [8].

Figure 1.1: Tricycle, taildragger and bicycle gear layouts

1.1.2. L ANDING GEAR TOPOLOGY

Figure 1.2, shows some of the standard landing gear strut topologies. This section explains each of them and
discuss their benefits and weaknesses.

T ELESCOPIC GEAR
Also called cantilevered gear, the telescopic gear is the least complex strut arrangement. It has three main
components: the shock absorber , a main fitting and an actuator or side brace. The cantilevered gear is
mostly used as NG or TG.
Because of its simplicity, it is the cheapest option, both in terms of aqcuiring cost, as in maintenance [9].
Disadvantages of this topology are that it requires a relatively long shock absorber length and that only simple
retraction schemes are possible: forward or aft retraction with optionally a rotation of the wheel(s).
1.2. P RINCIPLES OF ENGINEERING OPTIMIZATION 3

Figure 1.2: Standard strut topologies in use. Figure adapted from [9].

L EVERED SUSPENSION GEAR

More complicated than a telescopic gear, the levered gear offers more design freedom regarding the place-
ment of the shock absorber. This may lead to a shorter gear. Another advantage is its performance on rough
airfields, which is considered excellent [9], [4]. The third advantage of the levered gear is that it can fold in a
smaller space than the telescopic gear.
Because of its excellent rough field behaviour, the trailing arm LG is often seen on helicopters. Disadvantages
are its increased complexity, weight and cost.

S IDEWAYS LEVERED GEAR

The wing is not always suited to mount the landing gear under, although it is often the first option. Reasons
can be that the aircraft doesn’t have a fixed wing (for helicopters), that the wing is too thin (for a fighter jet) or
that the aircraft has a high wing. In these cases, a sideways levered gear becomes an option. Being attached
to the body of the aircraft, the sideways levered gear is relatively complex and heavy.
One advantage of the sideways levered gear is that it can fold in a narrow space. The more complex the
retraction scheme however, the higher its cost; both in purchase price and maintenance.

B OGIE GEARS
When a gear strut contains three or more wheels, it is called a bogie. These topologies are usually seen as the
main gear strut of large passenger aircraft. The design of a landing gear for such an aircraft is however not
foreseen in the near future of FLG. Therefore, the bogie design is not considered further.

OTHER LANDING GEAR TOPOLOGIES

Of course, there are more than four possible landing gear topologies. One might even argue that every landing
gear has its own unique topology. Most can be categorized under one of the above mentioned types though.

1.2. P RINCIPLES OF ENGINEERING OPTIMIZATION

This section discusses the advantages of Knowledge Based Engineering (KBE) and Multidisciplinary Design
Optimization (MDO) in section 1.2.1. Section 1.2.2 then goes into more detail about different MDO architec-
tures and section 1.2.3 discusses the optimizer. Again, this section is only an introduction into engineering
optimization. Two books on this topic can be recommended for those not familiar with it: Principles of opti-
mal design: modeling and computation [10] and Practical optimization [11].

1.2.1. AVAILABLE ENGINEERING OPTIMIZATION METHODS

KBE is a merger of the fields of Object-Oriented Programming (OOP), artificial intelligence, and Computer
Aided Design (CAD) [12]. It is a method that stores information about previously solved problems to solve
similar problems faster. This is especially powerful when a problem is iterative and repetitive, as any engi-
neering design problem is. Time reductions of more than 95 percent have been claimed by automating labour
intensive, repetitive design tasks with KBE [13], [14], [15].
To optimally use KBE, a design process should be "highly rule-driven, multidisciplinary, repetitive and de-
manding geometry manipulation and product (re)configuration" [16]. It can thus be concluded that KBE is
best applied in the detailed design phase, where variations between different configurations are small, geo-
4 1. I NTRODUCTION

(a) MDF [21]. (b) IDF [21].

Figure 1.3: Principles of IDF and MDF

metric models exist, and extensive amounts of information about the product are available.
During the preliminary design process of a landing gear, which is highly search oriented and where little
knowledge about the product is available, KBE is less suited.

MDO solves engineering problems that involve different disciplines by taking into account their interac-
tions. As stated by Schönning "It should be noted that the multidisciplinary solution might not be the solution
for any one discipline analyzed separate from the other disciplines, but is the best solution accounting for the
interactions" [17]. Thus, MDO is best applied to problems that have multiple, coupled, disciplines.
The basis of any MDO technique is to minimize2 an objective function. Given a set of design variables, the
system optimizer searches their optimal value. The set of all design variables is called the design space. Typ-
ically, there are also constraints In the case of a landing at high angle of attack for example, the gear struts
should be positioned so that the aircraft cannot tip over and crash [18].
The mathematical formulation of a typical MDO problem is:

mi ni mi ze f (x, y)
sub j ec t t o : h(x, y) = 0 (1.1)
g (x, y) ≤ 0
where the objective function is f, the design variables are x and y and the constraints are given by h and g.
It is concluded that the best optimization technique for LIT, in this stage of development, is MDO.

1.2.2. MDO ARCHITECTURES

Now that it is decided to use MDO, an architecture must be chosen. A description of MDO architectures is
available from several sources, including Tedford and Martins [19], Keane and Nair [20], Martins and Carriage
[21], and Kroo [22]. The Multidisciplinary Feasible (MDF), the Individual Discipline Feasible (IDF) and the
Collaborative Optimization (CO) architecture are most interesting for the scope of this thesis.
The principle of MDF, also called fully integrated optimization or all-in-once optimization, is illustrated in
figure 1.3a. Unique for the MDF technique is that variables can be shared by disciplines. A value computed in
discipline 1 can be used by discipline 2 and vice versa. This mimics the iterative nature of the design process.
The optimizer changes variables x1-x3 and z1-z3, while the disciplines find the optimal value for y1-y3 at each
iteration.
This can result in long computational times if the step sizes for y1-y3 are small and/or if the runtime of a
certain discipline is long (e.g. if a finite element computation must be performed).
This interdisciplinary sharing of variables is not allowed in the IDF technique, see figure 1.3b. With IDF, the
optimizer is in control of changing variables. Analogous to the MDF technique, discipline 1 calculates the y1,
based on x1, z1, y2 and y3. y2 and y3 are however dummy variables; a "guess" made by the optimizer. They
are computed in disciplines 2 and 3. To allow this decomposition, constraints are added. When the optimiza-
tion is finished, dummy variable y2 should equal y2 as computed by discipline 2. The same holds for y1 and
y3.
The advantage of IDF is that every discipline is only evaluated once per iteration. Another advantage is that
the various disciplines can be evaluated in parallel, when the hardware allows so.
2 or maximize. In that case, give the negative value to the routine.
1.2. P RINCIPLES OF ENGINEERING OPTIMIZATION 5

Figure 1.4: Principle of CO [21].

The disadvantage of IDF is that the dummy variables and the computed variables y1-y3 might not converge
to the same value. That results in an infeasible outcome.
The IDF architecture is also called Optimizer-Based Decomposition.
Collaborative optimization can best be explained as IDF on two different levels, see figure 1.4. The advantage
of using this MDO architecture is that the evaluation of the different disciplines is completely decoupled.
Different software packages, different optimization routines and even different servers can be used for each
disciplinary analysis. A distinction can be made between the system-level optimizer and the disciplinary op-
timizer.

1.2.3. T HE OPTIMIZER
To find the best optimizer for a certain problem is a specialization in its own and again, this section is only
an introduction. Factors that influence the choice of optimization function are the objective function, the
type of constraints, the continuity of the design variables, the availability of gradient information, and the
existence of local minima.
A distinction can be made between single-objective functions and multi-objective functions. An example of
the first is to minimize the weight of a gear strut; an example of the second is to minimize both the weight and
the cost of a LG. Single-objective functions are easier to solve, and therefore it is recommended to rephrase
multi-objective problems into single-objective problems [11]. For the above example, the problem could be
rephrased into minimizing the weight given a maximum cost price. A trend-line can be found by running the
program for several maximum cost prices. Alternatively, an objective function can be created based on the
combination of cost and weight. A weight reduction of one kilogram is then allowed to cost an X amount of
euros. Other solutions are possible [23].
Several types of constraints can be identified, including linear constraints, bounds on the design variables
and non-linear constraints. A combination of these constraints is also possible. Every optimization function
is limited in the types of constraints it can evaluate. The MATLAB documentation is informative on this.
Design variables can be discrete or continuous. To give an example: one can be completely free to design the
tyre, in that case the design space for the tyre variables is continuous. If a standard tyre must be chosen from
a catalogue, the design space is discrete and another optimization algorithm needs to be chosen3 .
A fourth factor to consider while selecting an optimizer is the presence of local and global optima, as illus-
trated by figure 1.5:
Figure 1.5 shows the 3 local minima that are found when evaluating the function y = 2xcos(x) ∈ [-8,8]. This
function can be visualised, but when the number of dimensions increases it is in general a challenge to find
the global optimum. Optimizers that always find the global optimum exist, but usually need a long runtime.
Other strategies to find the global optimum are to start a local optimization algorithm at multiple points
spread over the design space.
3 there are ways to work around this, but to explain that would be too much detail for this introduction. For more information, [24] is a
recommended starting source.
6 1. I NTRODUCTION

Figure 1.5: Existance of local and global minimum on a constrained interval.

The last decision factor while choosing an optimizer is the availability of gradient information. If gradient
information is available or can be computed, it should be used. This will decrease the runtime [11]. An al-
gorithm that does not use gradient information, for instance Genetic Algorithm (GA), will typically start at
many points. For each point, a few design variables are changed and the objective function is evaluated.
Based on the changes in objective value, new points are initiated. This is often compared to evolutionary
changes, where pieces of DNA are swapped to make organisms perform better. If an optimizer can compute
the gradient information, for example Sequential Quadratic Programming (SQP), the evaluation consists of
changing the design variables one by one and computing the partial derivatives. The next point is chosen in
the direction of the negative gradient. A local optimum has been found if the gradient in a point equals zero.
In this thesis, the fmincon method of the MATLAB optimization toolbox is used. It is a gradient based, lo-
cal optimizer that is advised for non-linear single objective functions and non-linear constraints. All design
variables will be treated as being continuous.

1.3. P REVIOUS RESEARCH ON LANDING GEAR DESIGN AND ENGINEERING OP-

TIMIZATION
With the basic principles of landing gear design and MDO covered, it is now time to look at previous research
to merge the two fields. Three subsections will each describe a reference that incorporates the use of MDO in
LG design process.

1.3.1. L ANDING GEAR INTEGRATION IN AIRCRAFT CONCEPTUAL DESIGN

Chai and Mason detail the development of an MDO routine for the sizing of a landing gear [1]. This rou-
tine was implemented in an aircraft design tool under development at NASA [25]. The general layout of this
routine is repeated in figure 1.6.
The program contains four modules that together find the optimal landing gear. The module CONFIG defines
a baseline landing gear model that is used by the other modules. This module calculates internal variables
such as the load on the tires and the brake energy based on input parameters such as aircraft weight, number
of struts and number of tires, using physical equations [1]. From these calculations, standard parts such as
tires, wheels and brakes are selected from a database.The lengths of the MLG components are also estimated
by the CONFIG module.
The module LIMIT calculates requirement as the turnover angle, the ground clearance, the maximum allow-
able pitch and roll angles and the turn radius. Kinematic characteristics are also evaluated by this module:
via what scheme is the gear folded and into which space?
The module PAVE determines all airfields the aircraft can operate from, by determining the flotation charac-
teristics - the pressure imposed on the pavement by the tyres, and the behaviour of the runway as a result of
that.
If the previous three modules have defined a configuration that fulfils all requirements, the module GEARWEI
calculates the gear weight. It first estimates the weights of the different structural components, using their
loads. It also estimates the weight of every non-structural component from a database.
One limitation of this program is that a first estimate is needed from the user. The program is not capable
1.3. P REVIOUS RESEARCH ON LANDING GEAR DESIGN AND ENGINEERING OPTIMIZATION 7

Figure 1.6: General layout of the research by Chai and Mason [1].

Figure 1.7: Landing gear design module by Heerens [2]

of deriving a feasible starting point. Another limitation is that it was only tested for a specific aircraft design,
without considering its applicability to other aircraft concepts. The multidisciplinary architecture used is
however a good starting point for this thesis.

1.3.2. L ANDING GEAR DESIGN IN AN AUTOMATED DESIGN ENVIRONMENT

Other research is described by Heerens [2]. His work is also part of an aircraft design tool and fits into the
Initiator that is being developed at Delft University of Technology [26]. The Initiator is a software program
that can perform a conceptual sizing for existing and new aircraft configurations, such as a blended wing
body and the Prandlt plane. It exists out of several modules that cooperate to find the optimal aircraft config-
uration, using multidisciplinary analysis techniques. Example modules are those for the component sizing,
aerodynamic sizing and weight estimation. Heerens added a module to the Initiator that sizes the landing
gear. An overview of this module is given in figure 1.7.
The module of Heerens has three routines, that are similar to those used by Chai and Mason [1]. The or-
der workflow is however different. In this program, the limitations of the design space are investigated first.
Then, the loads on the landing gear components are determined for several pre-defined bogie layouts. This
includes the analysis of flotation requirements. Finally, the landing gear weight is computed using a similar
method as the program by Chai and Mason [1].
8 1. I NTRODUCTION

A limitation to this program is that there is limited optimization: the process flow doesn’t have feedback loops.
It is more a search, based on existing configurations, for a landing gear layout that is acceptable. Another lim-
itation is that, although this landing gear design module has been verified to give feasible results for different
aircraft. This program is only applicable to CS-25 aircraft; test cases included the A380-800, B777-300ER,
A320-200 and B737-200.

1.3.3. A N APPROACH FOR SIZING AND TOPOLOGY OPTIMIZATION INTEGRATING MULTIBODY

SIMULATION
The research by Cumnuantip [3], [27] goes in more detail than the previous two methods. The aim of this
work is namely to find the best landing gear and landing gear bay combination, see figure 1.8a. The case
study for this work is that of a blended wing body, but the method is more generally applicable.

(a) General layout. (b) Implementation in the multibody software

Figure 1.8: Research by Cumnuantip [3]

In the first two blocks of the program an initial position, length and shock strut characteristics of each MLG
are determined, based on analytical equations and positioning constraints. This first set-up is the start point
for the MultiBody Simulation (MBS) model in block three. The landing gear parts of this MBS model are as-
sumed to be rigid; the tyres are modelled as linear springs. Dynamic loads are evaluated for three load cases.
From these dynamic loads, the components are sized. The landing gear model and its degrees of freedom can
be seen in figure 1.8b. In the last step, each component is sized based on the acting loads.
This process can be ran independent of the number of gears. It is concluded that 8 main landing gears struts
yield the lowest total aircraft weight [27]. Other investigated options were 4, 6 and 12 MLG struts.
The results of this research look very promising. To look at the landing gear in this much detail however,
detailed information from the aircraft such as a structural layout of the wing is needed. The framework devel-
oped in this thesis is aimed at finding the optimal landing gear for an aircraft in the preliminary design phase,
when this information is not available. In that respect, the method presented by Cumnuantip is not a good
starting point.
Another distinction between this research and the one by Chai and Mason, is that Cumnuantip uses a GA
to find a solution and not a gradient-based method. The given argumentation is that the design space is
discrete, and not continuous.

1.4. R ESEARCH QUESTIONS AND PROJECT SCOPE

The above sections show the previous research into integration of MDO and landing gear design. The three
studies all have clear limitations, however. The program by Chai and Mason is especially developed for a
1.5. T HESIS STRUCTURE 9

specific aircraft and cannot be used for other aircraft. The program by Heerens is more generic than that of
Chai and Mason, but it does not use true optimization methods. The work by Cumnuantip has more detail
than the other studies and uses optimization methods, but it uses discrete variables and requires detailed
information about the aircraft. That information is typically not available in the concept design phase.
The objective of this thesis is to redefine the landing gear design process and to make it suitable for engineer-
ing optimization techniques; without making any assumption on the aircraft type. As such, this thesis fits
within a larger project at FLG. The ultimate goal of that project is to have a software tool that can help in the
design a landing gear concept for any kind of aircraft between 3,000 and 50,000 kg, fixed-wing or rotary-wing.
The time-frame for this phase of the design is similar to the current turn-around time of a concept design: six
weeks. The primary goal is thus not to shorten this design phase, but to do a better job in the given time. A
trade-off should be performed between different landing gears, that are all evaluated in more detail and with
more consistency than what is currently done for a single gear solution.
Work was already started by generalizing specific software, such that is can be used for other projects. That
work follows a bottom-up approach: create all the blocks that are needed and consider their integration after-
wards. This research takes a different approach and looks at the landing gear design process from a top-level
perspective to define the needed framework.
A proof of concept will be created to prove that these principles have practical purpose. The following re-
search questions were defined:

• Which inputs are needed for the landing gear design process?

1. Which regulations need to be followed?

2. Which customer requirements are needed?
3. Are there other requirements that can drive the design?
4. What are typical landing gear performance requirements?

• What does the current landing gear design process look like?

1. Which disciplines are involved?

2. What are the standard landing gear architectures?
3. Which software programs are used during the design process?

• How can the landing gear design process benefit from engineering optimization principles?

1. Which optimization toolset is best suitable for the problem of landing gear design?
2. What is the most efficient order to evaluate the various disciplines?
3. How can all these disciplines and tools be integrated in a single software tool?
4. What is the minimum set of variables to derive a feasible landing gear layout?
5. What is the output from this process?

1.5. T HESIS STRUCTURE

This thesis has 11 chapters. Chapter 2 gives a top-level description of LIT. The rest of this thesis is dedicated
to the development of a proof of concept for LIT. It follows a similar approach as described in chapter 2, but
it is less general and aimed at a single gear: the POC gear. The specifics of that landing gear are the subject of
chapter 5.
Before that, chapters 3 and 4 describe the system-level positioning constraints for a tricycle landing gear and
the most important load cases. Chapters 6 describes how the components are sized, based on these ground
reaction loads.
The resulting POC framework is the subject of chapter 7; the results that were obtained with it follow in chap-
ter 9. A verification of these results is given in chapter 8.
Finally, the conclusions and recommendations are given in two separate chapters: 10 and 11.
 2
"L ANDING GEAR IN THIRTY DAYS "
This chapter explains the final outline for the Landing gear In Thirty days (LIT) tool, see also figure 2.1. The
final objective for this software tool is to help in the conceptual design phase of a landing gear. This tool
should not be of the "push a button and get your answer" type, but a tool-set that assists the LG designer in
the evaluation of different configurations. Both fixed-wing and rotary-wing aircraft are included. Not only
the landing gear should be optimized, also the interaction with the aircraft structure is of special importance.
The best LG solution takes this integration into account. Excluded from the scope are gears of large passenger
aircraft that carry more than 100 passengers; these are not in the market niche of FLG.
For the purpose of conceptual design, the preliminary sizing of all load-carrying components is included,
as well as the preliminary sizing of the shock absorber, the evaluation of kinematic characteristics, the de-
termination of flotation properties and the preliminary design of the rolling stock. Excluded are hydraulics,
electrical harnesses, seals, sensors and other non-loaded components, although space reservations must be
made for these parts.
The final outcome of LIT is a breakdown of weight and cost for every landing gear component, as well as a
basic CAD model of the gear struts. Per gear component, there is additional information in a text file such as
cross-sectional information, the critical load case and the results of a Finite Element Method (FEM) calcula-
tion on that part.
LIT can be explained by making an analogy with an onion. Layer by layer, the design is analysed in more
detail. This is sketched in figure 2.1. Each step is explained in more detail below.
Note that the description in this chapter does not mention any specific software programs. For each dedi-
cated task, multiple software packages are suitable. As optimization routine, the use of for example iSight,
MATLAB, and Python can be considered. For multi-body calculations, there is the choice between for ex-
ample SimMechanics, VLM and ADAMS. The construction of 3D drawings can be done with CATIA, but
Pro-Engineer and solidworks are alternatives. Also for FEM calculations, there are multiple options, such
as Abaqus or NASTRAN. To not exclude any options in advance, it will just be stated that, for example, a
multi-body package is needed.

2.1. B LOCK A: BASIC PARAMETERS

The first top-level parameters are the overall gear layout (either tricycle or taildragger), and the topology of
the gear struts (trailing arm, telescopic, sideways levered, bogie, other). The objective of the first block is to
investigate all different possibilities and to determine the most promising combinations. This computation
is possible with around 20 design variables and is rather quick: below 5 minutes per possibility. Block A con-
tains three different routines. These are described below.

The first routine in block A, Find load per gear strut, calculates the ground reaction forces on each strut
(either main, nose or tail), based on its position in the global reference frame and the gear layout chosen. The
considered load cases are static and come from the CS requirements; the user can select whether CS-23, 25,
27, 29, or a combination of the four is applicable. On a project-by-project basis, additional customer load
cases are added.
The required inputs for this routine are specified in table 2.1. They are treated as constants and must be

11
12 2. "L ANDING GEAR IN THIRTY DAYS "

Figure 2.1: Proposed set-up of the landing gear in thirty days tool.

obtained from the aircraft manufacturer.

The output of this routine is a list of all load cases, with corresponding forces on the tyres and at the CG, the
location of the tyres and the gear layout considered.
There are 12 positioning constraints. These are treated in a separate chapter: 3

Table 2.1: Inputs for routine 1 in block A of LIT.

Parameter Unit Parameter Unit

Aircraft weight(s) N Turnover angle4 deg
CG location fwd (x,y,z) m wing span m
CG location aft (x,y,z) m tail height m
Regulatory set - max pitch angle deg
Max approach speed m/s steering angle nose wheel deg
Max roll angle @ TD deg wing sweep angle deg
Optional; to be added after the first iteration:
distance tyres secondary gear m radius sec tyres m
distance between main gear tyres m radius MG tyres m

The second routine in block A is Initiate stick models. The objective of this routine is to get feasible initial
designs, for all different combinations of gear layout and topology. Not every combination will be possible for
any aircraft, however. A sideways levered gear is normally only seen on helicopters and fighters - and there
are good reasons for that. The user of LIT should eliminate infeasible combinations in advance.
The initial design is based on the preliminary sizing of the shock absorber, and a flotation analysis. Based
on these calculations, the preliminary size of the other components can be estimated. There are two shock
absorber types: a single acting shock strut and a double acting strut. Both options must be considered. The
4 see explanation in section 3.2
2.2. B LOCK B: P RELIMINARY SIZING 13

Table 2.2: Inputs for routine 2 in block A of LIT.

Parameter unit Parameter Unit

Efficiency SA % Efficiency tyre %
Load factor at TD - vertical TD velocity m/s
Static SA pressure Pa ratio static over fully extended pressure -
ratio fully compressed over static pressure - Tyre stroke m

flotation analysis determines the number of wheel and their separation.

Inputs of this routine are listed in table 2.2; these are again treated as constants. In some cases, tyre infor-
mation is provided by the Original Equipment Manufacturer (OEM), in other cases is not. Thus, it must be
possible to design the tyre characteristics, as well as to use a pre-determined (and fixed) set.

The output of this step is a stick diagram of the initial gear design, for every combination of gear topology,
gear layout and Shock Absorber (SA) type. The total number of possibilities is substantial and does not stop
with the distinction between telescopic gear, trailing arm gear, sideways levered gear and bogie gear. If only
the trailing arm gear is considered, several sub-types can be identified that all fall under the category trailing
arm gear. For example; a gear that retracts forward is different than a gear that retracts aft or to the side. And
even then, retraction is possible by folding a member, or by sliding it in like an actuator. This has been illus-
trated in figure 2.2. Not every branch is fully worked out, however. The reader of this report should be able to
fill in the blanks for him/herself.
Constraints to this routine are that the minimum pressure in the shock absorber cannot be too low and that
the maximum pressure cannot be too high. Often quoted values for these pressures are 60 and 6000 psi
respectively [4]. Furthermore, folding of the gear must be possible within the LG bay and without the com-
ponents touching each other.

The third routine, preliminary component sizing, calculates the cross-sectional parameters for the main
LG components. This calculation is based on the loads determined in the first routine, the landing gear
layouts determined in the second routine, and simple stress calculations. These calculations are performed
for every load case, as it is not known in advance which load case is critical. Typically, there will not be one
critical load case for the whole gear strut, but a different one for every component. For that reason, it is highly
relevant output for the designer. Every part is modelled as a cylinder or as an I-beam and has a constant cross
section. Only the main load-carrying components are evaluated; lugs, pins, bushes and other smaller parts
are excluded for this step.
The input for this routine is created in the previous two routines; outputs include a list of component and
system weights for all different gear combinations considered, and the optimal gear locations. Additional
outputs are the margin of safety5 and the critical load case for each part.
One constraint is again the kinematic properties of the gear: it must fit inside the LG bay. A second constraint
is related to the forces introduced in the airframe structure. They are limited by the strength of this structure.
After all, minimizing the weight of the landing gear plus the weight of the surrounding aircraft structure is the
final objective. The last, obvious, constraint is that the parts are not allowed to fail due to the forces acting on
them.

2.2. B LOCK B: P RELIMINARY SIZING

In the second block, the more promising design solutions from block A are further analysed. More detail is
added by allowing the position of the main components to vary, by adding secondary components such as
lugs, pins and bushes, and by evaluating the cost of the LG system. Before continuing, there is one manual
step. The designer must evaluate the different results of block A and determine which concepts are worth
pursuing in more detail. In the first block, a typical evaluation takes between one and five minutes. In the
second block, one evaluation may take anywhere between an hour and several hours. This step should thus
be done for a handful of concepts only, not for dozens of them.

Block B follows a similar method as block A, but it is more elaborate and shifts from a single-objective
5 the concept of margin of safety will be explained in chapter6
2. "L ANDING GEAR IN THIRTY DAYS "

Figure 2.2: An overview of different standard landing gear configurations.

14
2.3. B LOCK C: DYNAMIC LOAD CASES 15

optimization routine to a multiple-objective optimization: the predicted cost of the landing gear is also com-
puted in this block. These cost can be estimated from the knowledge of the manufacturing process. The
number of design variables increases; besides the contact points of the gear struts with the ground, the posi-
tions of the so-called hardpoints6 can be varied and the secondary components must be sized. To allow this
analysis, a parametric multi-body package is used that derives the loads at the specified hardpoints.
The inputs for the multi-body models are the loads per landing gear strut and the hardpoint locations. For
every load case, the loads in the hardpoints are calculated. From these loads; the strut weight is calculated,
using the same routines as in block A.
The principle output of this routine, for every concept, is a stick diagram in the multi-body package and a text
file with cross-sectional parameters of the different components. This is not very intuitive to interpret for the
designer and should thus be converted into a CAD drawing. The designer operating LIT must inspect these
CAD drawings to evaluate if a physically achievable result is obtained. Additional outputs are information
about the component sizing: the margin of safety and the critical load case.
Besides the constraints of block A, there are additional constraints in this block. Parts cannot overlap in the
physical space, all thicknesses must be positive, pins cannot be larger than their corresponding lugs, etc.
These constraints are rather trivial, but they must be specified as the optimization routine is purely mathe-
matical and does not know them.

2.3. B LOCK C: DYNAMIC LOAD CASES

In block C, the dynamic capabilities of the multi-body program are used to analyse the gear in more detail
than in block B. At this point, the number of possible gear concepts should be further reduced to a maximum
of 4-5 per gear group (main and secondary) as the runtime now becomes a major factor to consider. Using
multiple processors this step can still be performed overnight, but the design freedom must be restricted at
this point.
Some load cases have been neglected so far due to a lack of information. These are computed in this block.
Included are the wheel spin-up load case, the drop test, and the spring back load case. Furthermore, there
are several load cases that were approximated with a static method in block A, but which are in fact dynamic.
Examples are turning while taxiing on the ground, the sudden application of the brakes by the pilot, and all
touchdown load cases. These must be studied in more detail as well. Dynamic load cases defined by the
customer should also be evaluated at this point.

To allow this dynamic analysis, the rigid multi-body model must be expanded with a shock absorber, tyre
models, and for heaver aircraft also with pavement models. When the load peaks of these dynamic load cases
are found, the tool can be restarted with this additional set of loads. These load peaks of the dynamic analysis
are then added as static load cases. To speed up the process, static load cases that were not critical can be
deleted from the old set.
The number of design variables is not changed when proceeding from block B to block C. The system-
level design variables are still the connection points of the LG parts and the position of the tyres, while the
component-level design variables remain those that define the cross sections of the different parts. The con-
straints remain the same as well.
The output of block C is similar to the output of block B, only updated. There is again a stick diagram in the
multi-body package, and a collection of text files with information about the cross sections of the compo-
nents and their critical load cases. This is then used to create a 3D product in a CAD package.

2.4. B LOCK D: V ERIFY RESULTS BY FEM CALCULATION

The final step in block D is to verify the simple computations by means of a (static) FEM calculation. This is
not a true optimization; to have a FEM analysis incorporated in an optimization routine would simply take
too long. To allow this step, modal parts must be used for the LG components, instead of the strongly sim-
plified I-beams and cylinders. This allows for a more detailed analysis of the LG components. A modal part
is a flexible 3D CAD part, on which linearised stress calculations can be performed via FEM software. They
take local bending, deformation and stress peaks into account. During the optimization, the modal parts are
modified until they do not fail due to the acting loads. As the modal parts are defined in the CAD environ-
ment, they can be implemented in the multi-body model. They can also be used to visualize the results. That

6 A point where the two LG parts connect with each other

16 2. "L ANDING GEAR IN THIRTY DAYS "

is another advantage of using modal parts.

In block D, the focus shifts from finding the best solution to verifying that the found solution is indeed feasi-
ble. There is no additional output, and the constraints are also similar to the previous blocks. The output of
this step is again a 3D CAD product, such that pictures of the design can be shared with the design team and
the customer. A detailed weight breakdown per component must also be available, as well as dimensions of
all parts and the critical load cases per component.
 3
P OSITIONING CONSTRAINTS TRICYCLE
LANDING GEAR

The previous chapter explained the overall set-up of LIT. The first step in the development of the Proof of
Concept (POC) tool, is to identify the allowed gear location. That is done in this chapter. The constraints
of this chapter are applicable to all aircraft with a tricycle gear. Specific constraints for the POC tool are
postponed until section 7.4.
Torenbeek states the different geometrical limitations for tricycle landing gears [28]. This section describes
their implementation. In total, 12 positioning constraints are identified; divided over four topics. Each topic
has its own dedicated section.
This chapter needs the lengths between the MLG struts and the CG, the NLG strut and the CG, and the MLG
and the NLG. The used variables are introduced in figure 3.1.

Figure 3.1: Definition of A, B and cg-range [2].

3.1. N O PARTS OF THE AIRCRAFT TOUCH THE GROUND, EXCEPT THE LANDING
GEAR
During landing and take-off, the tail, wingtip, or engines are not allowed to hit the ground. There are 3 con-
straints that prevent this: the tip clearance constraint, the nacelle clearance constraint and the tail clearance
constraint.

T IP CLEARANCE CONSTRAINT
The wing tip of an aircraft may never touch the ground. There are two relevant performance parameters for
this condition: the maximum pitch angle θ and the maximum roll angle φ that will both be provided by the
OEM. When the aircraft is both pitching and rolling, for example during a landing with cross-winds at high
angle of attack, there is a risk that the wing tips touch the ground. With equation 3.1, the roll angle at which
the wing tip touches the ground can be calculated [28, p. 350]:

17
18 3. P OSITIONING CONSTRAINTS TRICYCLE LANDING GEAR

Figure 3.2: Geometrical constraint relating to wing tip clearance [1]

Figure 3.3: Geometrical constraint relating to engine clearance [29]

2h m
tan φ = tan Γ + − tan θ tan Λ + ∂ (3.1)
B −t
where Γ is the dihedral angle of the wing, h m is the main gear height, B is the wingspan, t is the gear track, Λ
is the wing sweep angle and δ is the tip clearance. See also figure 3.2.
It is noted that some references on landing gear design (e.g. [1] and [2]) try to calculate the pitch angle re-
quired for take-off, based on preliminary aerodynamic parameters. Then, they assume that the pitch angle
during touchdown equals the pitch angle at take-off. In this thesis, the maximum allowable pitch angle is
assumed to be an input, provided by the aircraft manufacturer.

N ACELLE CLEARANCE CONSTRAINT

The second constraint is that the engine is not allowed to touch the ground during take-off or landing. It is
not relevant whether the engine is a propeller, a jet engine, or an open rotor. The minimum height above the
ground and its location are important, not the engine type.
Still, two cases can be distinguished: an engine that is positioned below the wing or an engine that is posi-
tioned on the tail, as is seen on the Fokker 1007 .
When the engines are positioned under the wing, the roll angle at which the nacelle touches the ground is
given by equation 3.2. The pitch angle is assumed to be 0.

h nacel l e
tan φ = (3.2)
y nacel l e − 2t

where h nacel l e is the lowest nacelle height and y nacel l e is the y position where the nacelle is closest to the
ground. Note that a pitch angle of 0 degrees actually represents the worst-case scenario as the engines will
move upwards when the aircraft is under a positive pitch angle.
If the engine is positioned on the tail, figure 3.3 should be considered. For this scenario, the maximum pitch
angle can be expressed as equation 3.3:

h nacel l e
tan θ = (3.3)
x nacel l e − x mai n
7 Yet other positions are imaginable, for example the Honda HA-420 business aircraft that has the engines on top of the wing. In that
case, this constraint will be inactive.
3.2. A LLOWED LOCATIONS FOR THE CENTRE OF GRAVITY 19

with x nacel l e the x location where the nacelle height is minimal and x mai n the x distance between the MLG
and the origin. The effect of a roll angle is not considered as a survey of aircraft with a tail engine shows that
they all have a main gear track wider than the engine track. This suggests that the roll angle has no effect on
when the engine touches the ground.
This constraint is not relevant for helicopters.

TAIL CLEARANCE CONSTRAINT

A third constraint is that the tail cannot touch the ground. This can be checked with a method similar to
the engine-on-tail constraint. The difficulty here is to determine which point of the tail is the most likely
candidate to touch the ground.
It is assumed that this point is fixed by the OEM. It is called the location of the tail bumper, with x t ai l bumper
and h t ai l bumper as its x coordinate and height above the ground respectively. This point will be treated as a
constant value. The constraint for this criterion is given by equation 3.4

h t ai l bumper
θmax − tan−1 ( )≤0 (3.4)
x t ai l bumper − x mai n

3.2. A LLOWED LOCATIONS FOR THE CENTRE OF GRAVITY

As indicated in figure 3.4, a triangle can be drawn that connects the main and nose gear struts. It should
be prevented that the CG moves outside that triangle during ground operations. In total, 4 constraints need
to be checked in this section: the ground stability constraint, two sideways turnover constraints, and the
touchdown constraint.

Figure 3.4: Visualization of the sideways turnover angle ψ and the triangle spanned by the NG and MLG

T OUCHDOWN CONSTRAINT
The landing gear must be positioned behind the Centre of Gravity (CG), or the aircraft would rotate and fall
on its tail. The worst-case scenario for this constraint is a landing at the highest and most aft CG position.
The distance between the MLG and the aft position of the main landing gear is given by equation 3.5 [28, p.
352].

b ≥ (h m + e s ) tan θ (3.5)

Where b is indicated in figure 3.1 and e s is the total static wheel travel8 .

G ROUND STABILITY CONSTRAINT

The so-called sideways turnover angle ψ prevents the CG from falling outside the triangle during a landing
on one wheel. It is visualised in figure 3.4 and expressed in equations 3.6 and 3.7 [28, p. 354]
e
tan ψ = (3.6)
a sin δ
with δ:
8 The deflection of the tyre plus shock absorber.
20 3. P OSITIONING CONSTRAINTS TRICYCLE LANDING GEAR

Figure 3.5: Visualization of the sideways turnover constraint [28]

t
tan δ = (3.7)
2(e + a)
with e the height of the CG above the ground and a and b as in figure 3.1. The typical maximum value for the
turnover angle varies between aircraft types [4].

S IDEWAYS TURNOVER CONSTRAINT 1

The allowed position of the main gears is dependent on the forward CG location and the position of the nose
gear. A positioning recommendation is to draw a circle with a radius of 0.54 times the CG height around the
forward CG in the top view of the aircraft[28]. That circle should fit within the triangle defined by the gear
struts, see figure 3.5 and equation 3.8. The factor 0.54 is based on dynamic instability considerations and is
statistically supported [28].

1 0.54e
t M I N = tan(sin−1 ( ))(a + b) (3.8)
2 a − c g r ang e

S IDEWAYS TURNOVER CONSTRAINT 2

The second sideways turnover constraint limits the most aft position of the nose gear, based on the aft CG
position and the location of the main gears. Again, a circle of 0.54 times the height of the CG should be drawn
around the location of the CG. As before, this circle should be completely inside the triangle defined by the
landing gear struts. Mathematically, this can be expressed in equation 3.9 [2].

0.54e t
a − c g r ang e = tan(sin−1 ( ) + ∆) − (b + c g r ang e) (3.9)
(b + c g r ang e) sin ∆ 2
In this equation, ∆ is given by at an(2b/t )

3.3. L OAD ON THE NOSE GEAR

If the load on the NLG is too low, the steerability of the aircraft will become problematic. If that load is too
high, there will not be enough load on the main gears to have efficient braking.

M INIMUM NOSE GEAR LOAD CONSTRAINT

During ground operations, the aircraft is usually steered by turning the nose wheels. Therefore, a minimum
load needs to be present on these wheels. This minimum load is defined by the OEM; a conservative value is
6 percent of the aircraft weight [4]. The load on the nose gear is given by:

b
FN = W (3.10)
d

M AXIMUM NOSE GEAR LOAD CONSTRAINT

There is also a maximum to the static load that can be applied on the nose gear. Usually the NLG wheels
are not equipped with brakes. As the load on the nose gear increases, the load on the main gear decreases
and thus braking efficiency is affected. The maximum static load is found with the CG is the most forward
position, see equation 3.11

b + c g r ang e
FN = W (3.11)
d
3.4. O PERATION ON THE AIRFIELD 21

Table 3.1: Aircraft design group classification [30]

Group I Group II Group III Group IV Group V Group VI

Tail height [m] <6 6-9 9 - 13.5 13.5 - 18.5 18.5 - 20 20 - 24.5
Wingspan [m] <15 15 - 24 24 - 36 36 - 52 52 - 65 65 - 80

Table 3.2: Aircraft approach category classification [30]

A B C D E

Approach speed [m/s] <46.8 46.8 - 62.2 62.2 - 72.5 72.5 - 85.4 > 85.4

A conservative value for the maximum NLG load is 20 percent of the aircraft weight.
Some references, such as [4] and [2] also prescribe a maximum load for the MLG. The maximum load that
can be applied to the main gear strut is however half the aircraft weight. More demanding load cases will be
found in chapter 4, so this constraint is omitted.

3.4. O PERATION ON THE AIRFIELD

A set of rules for ground operations is described in AC150/5300-13a [30] for fixed-wing, conventional take-off
and landing aircraft. These rules specify constraints related to the required runway width, manoeuvring on
the taxiway and the castor angle.
Reference [30] subdivides aircraft in Aircraft Design Group (ADG)s based on their physical dimensions and in
Aircraft Approach Category (AAC)s based on their approach speed. The subdivision in these two categories
is given in tables 3.1 and 3.2, where the original values were converted to the metric system.
For the ADG classification, the highest applicable classification should be taken. An aircraft with a tail height
of 14 meter and wing span of 35 meter, would thus be ADG IV.

T URNING ON RUNWAY CONSTRAINT

The minimum required runway width for an aircraft, r 180 deg t ur n , can be found from the AAC and ADG clas-
sification. A 180°turn must be possible on this runway, see equation 3.12 and figure 3.6

t
r 180 deg t ur n = d tan(90 − β) + (3.12)
2
β is the nose steering angle and is typically limited to ± 60°[4].

M ANOUVERING ON THE TAXIWAY

Manoeuvres on taxiways impose another constraint, see figure 3.7. The aircraft must be able to ride on the
taxiway while its wheels stay on the centreline of the track. The radius of this centreline, the inner radius
of the taxiway and the safety margin between the outer wheels and the edge of the taxiway are specified for
every aircraft type in table 3.3.

Figure 3.6: Visualization of the turning on runway requirement [2]

22 3. P OSITIONING CONSTRAINTS TRICYCLE LANDING GEAR

Table 3.3: Operation on taxiway [30]

Group I Group II Group III Group IV Group V Group VI

r cent er l i ne [ft] /[m] 75 / 23 75 / 23 100 / 30 150 / 46 150 / 46 170 / 52

r f i l l et [ft] / [m] 60 / 18 55 / 17 55 / 17 85 / 26 85 / 26 85 / 26
Safety margin S[ft] / [m] 2.5 / 0.76 2.5 / 0.76 13 / 4.0 20 / 6.1 20 / 6.1 25 / 7.6

Figure 3.7: Visualization of the turning on taxiway requirement [1]

The limit to the combination of track width and wheel base is given in equation 3.13:

t
q
r f i l l et = − r 2 cent er l i ne − d 2 + + S (3.13)
2

C ASTOR ANGLE CONSTRAINT

The so-called castor angle is defined as the angle between the nosewheel orientation and the tangent of the
centerline curvature [30]. It should be smaller than the nose wheel steering angle β, leading to equation 3.14:

d
sin−1 ( ) −β ≤ 0 (3.14)
r cent er l i ne
 4
G ROUND REACTION LOADS ON THE WHEELS
The first step of LIT is to determine the loads per gear strut. These loads come from two sources: the aircraft
regulations - described in FAR/CS259 [31] and the OEM. Only the CS-25 load cases are considered in this
thesis.
There are different regulations for the various aircraft categories. The most important ones for the scope of
LIT are CS-23 [32], CS-25 [31], CS-27 [33] and CS-29 [34]. A comparison between these regulating bodies has
been made. It was concluded that CS-25 has the most elaborate rules for the landing gear. Therefore, only
those rules are described in this thesis. The differences between CS-25 and the other regulations can be found
in appendix A. Note that the rules in this chapter are only relevant for tricycle gear layouts.
In section 4.2, it is assumed that the load is evenly divided between both MLG struts. There are more cases to
be considered however; see section 4.3. Before the load cases can be described however,a reference frame is
needed: the global (or aircraft) reference frame. This is introduced in section 4.1.

4.1. R EFERENCE FRAME USED

The reference frame of this thesis is visualized in figure 4.1. The origin of this frame is placed below and in
front of the nose. This results in positive x and z coordinates. The y-axis is positive through the right wing, to
ensure a right-handed axis orientation.

Figure 4.1: Aircraft global reference frame as used throughout this thesis

4.2. S TATIC LOADS

The static load cases can be divided into two scenarios: ground operation loads and touchdown loads. In
total, 41 unique load cases are defined. It is generally unknown which load case will be critical, so all must be
investigated. Only those load cases that turn out to be critical in chapter 8 have been detailed in the current
chapter, however. The other load cases are described in appendix B.
All static load cases are in fact simplifications of a dynamic event. This section is therefore in agreement
with CS-25. In the design process, these dynamic event are all considered. That is however only done in the
detailed design phase, not in the preliminary design phase.
9 For loads on the LG, the CS regulations from America are equivalent to the European FAR regulations

23
24 4. G ROUND REACTION LOADS ON THE WHEELS

4.2.1. B RAKING LOADS

Section 25.493 of the regulations describes 3 different braking scenarios:

1. braking without any load on the nose wheel (2-point braking);

2. braking with load on the nose wheel and zero pitching acceleration (3-point braking);
3. a sudden braking motion with dynamic pitching behaviour as a result (dynamic braking).

Points 1 and 2 must be evaluated at the MRW with a load factor of 1 and at the Maximum Landing Weight
(MLW) with a load factor of 1.2. The third point only needs evaluation at the Maximum Take-Off Weight
(MTOW).
When there is no load on the NG, all weight is carried by the MLG struts, as in equation 4.1

nW
VM = (4.1)
2
Where VM is the vertical force on the main gears, n the load factor and W the aircraft weight. The maximum
drag force per strut must be taken as 0.8 times the vertical load per strut; the side load as 0.
The force on the main gear can be found by taking moments around the attachment point of the nose gear
normal force, see equation 4.2:

nW a
VM = (4.2)
2(d + 0.8e)
With a the distance between the nose gear and the CG, d the distance between the MLG and the NLG, e the
height of the CG and the other symbols as before.
The load on the nose gear strut VN can be calculated by taking moments around the CG and substituting
equation 4.2. The result is equation 4.3:

nW (b + 0.8e)
VN = (4.3)
(d + 0.8e)
where b is the distance between the MLG and the CG. There is an additional drag load on the main gear of 0.8
times the vertical load. The side load is 0.
For the dynamic braking case, an equation is provided by the regulations, see equation 4.4:

W f ·µ·a ·e
VN = [b + ] (4.4)
a +b a +b +µ·e
In this equation, f is the dynamic response factor that is 2.0 unless a lower factor can be proven and µ is the
friction coefficient which is 0.8.
The load that is not carried by the nose gear is divided over the main gear struts. These main gear struts
experience an additional drag load of 0.8 times the vertical load. The side load is 0, for both the MLG and the
NG.

4.2.2. T URNING LOADS

There are 6 turning scenarios defined in CS25.495 and CS25.511:

• Turning left with all tyres inflated

• Turning with one of the nose wheels deflated
• Turning left with the inner tyre of the right strut deflated
• Turning left with the outer tyre of the right strut deflated
• Turning left with the inner tyre of the left strut deflated
• Turning left with the outer tyre of the left strut deflated

Another 6 cases can be thought of, symmetric to the cases described above. Figures 4.2a and 4.2b give a free
body diagram for the case where the inner tyre of the right strut is deflated.
Without any deflated tyres, equations 4.5 to 4.7 can be derived for the vertical load on the gear struts by
evaluating the sum of moments on the nose gear and at the CG:
4.3. S UB LOAD CASES DEPENDENT ON TYRE INFLATION 25

(a) Back view (b) Top view

Figure 4.2: Free body diagram for turning conditions

W e a
VM R = (− + ) (4.5)
2 t d
W e a
VM L = ( + ) (4.6)
2 t d
Wb
VN = (4.7)
d
Where VM R and VM L are the load on the right and left main strut, and t is the main gear track width. There is
a side load of half the vertical load on all three struts; the drag load is 0.

The equations for the other 5 load cases are given in table 4.1. In these equations, u is the distance between
the main gear wheels and v is the distance between the nose gear wheels.

Table 4.1: Loads on left and right main gear strut for different turning load cases.

Scenario Left strut Right strut

¡W ¢³ t b
´ ¡W ¢³ t b
´
Flat nose wheel t 2d a + 0.25e + v d t 2d a − 0.25e − v d
¡ W ¢¡ t ¢ ¡ W ¢ ¡¡ t u
¢ ¢
Flat inner right wheel t +u 2d a + 0.25e t +u 2d + d a − 0.25e
¡ W ¢ ¡¡ t u
¢ ¢ ¡ W ¢¡ t ¢
Flat outer right wheel t −u 2d − d a + 0.25e t −u 2d a − 0.25e
¡ W ¢ ¡¡ t u
¢ ¢ ¡ W ¢¡ t ¢
Flat inner left tyre t +u 2d + d a + 0.25e t +u 2d a − 0.25e
¡ W ¢¡ t ¢ ¡ W ¢ ¡¡ t u
¢ ¢
Flat outer left tyre t −u 2d a + 0.25e t −u 2d − d a − 0.25e

4.2.3. L ANDING WITH SIDE LOAD

The side load condition is described in CS25.485. For this load case, it is assumed that the weight of the
aircraft is divided over both main gear struts. There is an additional side load of 0.8 times the vertical load on
one gear strut, and -0.6 on the other. The drag load is zero.

4.3. S UB LOAD CASES DEPENDENT ON TYRE INFLATION

The load per strut will not always be perfectly divided between both wheels. The regulations describe that in
CS25.511. If a gear strut has two wheels, below five different scenarios must be investigated:

• The load per strut is equally divided; both wheels carrying 50 percent (the 50-50 scenario).
• The left tyre is flat and carries no loads; the right tyre carries 60 percent of the nominal load (the 0-60
scenario).
• The same scenario as above, but now the right tyre is flat (the 60-0 scenario)
26 4. G ROUND REACTION LOADS ON THE WHEELS

• A difference in tyre pressure resulting in an unequally balanced load: 40 percent for the left tyre; 60
percent for the right tyre. This will be referred to as the 40-60 scenario.
• The same as above, but now the left tyre carries 60 percent of the load: the 60-40 scenario.

For the flat load cases, the side and drag loads must be halved with respect to the 50-50 case.
Combining the above rules with the fact that each load case must be considered at the most forward and the
most aft CG position, there are 10 sub load cases per main load cases. This increases the total number of load
cases to 410.
 5
I NITIAL GEAR SET- UP
Chapter 2 introduced a four-step approach for the determination of the best landing gear concept. The first
step contains 3 routines: to find the loads on the gear strut, to initiate a stick diagram and to do a stress
optimization of the main parts. The derivation of the loads per gear strut was explained in chapter 4. The
next step is to initiate stick diagrams for all possible gear topologies. The POC tool takes a different approach:
a gear design from an existing concept study is used. The aim is to optimize that gear further, where only the
weight will be considered.
This chapter describes that gear in section 5.1. Section 5.2 explains how the loads per component are found.

5.1. I NITIAL STARTING POINT FOR THE OPTIMIZATION

The POC gear was designed as the main gear of a longe-range business aircraft. It is of the Trailing Arm (TA)
type and has a tricycle layout. In the concept study, only the MLG struts were sized. Figures 5.1 and 5.2
give an isometric view of the gear in extended position, and stick diagrams of the aft and side view. A short
description of the gear follows in section 5.1.1, while some limitations of this concept study are introduced in
section 5.1.2.

Figure 5.1: Isometric view of original concept

5.1.1. F EATURES OF THE ORIGINAL CONCEPT STUDY

This gear has 5 main components. The Wheel Axle (WA), A’AA’ in figure 5.2, connects both wheels. The
Trailing Arm (TA), AB in figure 5.2, connects the WA, SA, and Main Fitting (MF). The TA can rotate around the
global y direction, based on the compression of the SA. This shock absorber, DE in figure 5.2, is connected to
the trailing arm and the main fitting. The function of the shock absorber, together with the tyres, is to absorb

27
28 5. I NITIAL GEAR SET- UP

(a) Aft view (b) Side view

Figure 5.2: Stick diagrams of the initial concept

impact loads during taxi and landing. The absorbed energy is then gradually released, resulting in a smooth
ride for the passengers.
The folding side brace consists of two parts: a lower FSB, member CJ, and an upper FSB, member JI. When
the gear retracts, point J moves up and point F is pulled inboard. The rectract actuator that connects point F
to the aircraft, is not shown in figure 5.2.
The main fitting, finally, is member BGH. It connects the trailing arm and the folding side brace to the aircraft
interface points. It has a fixed lug in point E, where the shock absorber is mounted. The main fitting is
subdivided in two parts: the trunnion GFH and the main fitting BF.
Interface points with the airframe are points I, G and H. The points where the tyres touch the ground is given
the point O. The points A to J are called hardpoints. The loads at these hardpoints are calculated with a MBS
package, as a function of the ground loads in the points O. That is explained in section 5.2

5.1.2. L IMITATIONS OF THE ORIGINAL CONCEPT STUDY

There are three sources that describe the original concept study, all internal FLG documents. These sources
are contradicting, however. There is a document that describes the different features of the gear, on a more
general level10 . From that document, the interface points with the aircraft are found. Then, there is a specific
load derivation report11 . That report introduces the hardpoints of the previous section and also specifies the
orientation of the different parts with respect to each other. Finally, there is a CATIA product that visualizes
the gear. The three sources unfortunately disagree on the physical dimensions of the gear.
The first problem is with the interface points to the aircraft: points G, H and I. Point I is not mentioned in the
loads report, and its location is different in the CATIA product and the general description document. Points
G and H are mentioned in all three sources, but now the CATIA product and the general description docu-
ment agree, while the loads description report uses other coordinates. It can thus not be verified which set of
points is correct. For points G and H, the locations from the loads report are used; while the coordinates of
point I as mentioned in the general description document are taken.
The second problem is with the computation of the loads on the folding side brace CJI. In the loads descrip-
tion document, these loads are a function of the ground reaction forces only. In reality, the orientation of the
shock absorber will also have an influence on these loads. This has been corrected for, resulting in differences
in chapter 8.
The third problem is also related to the folding side brace. Besides the incorrect load calculation, the orien-
tation of this member is not properly documented in the loads calculation report. The location of point C is
clearly stated with respect to point B, but the orientation of the member CJI is undocumented. As seen before,
the location of point I can also not be verified. So, the orientation of the folding side brace in the global refer-
ence frame is unknown for the original concept. For the analytical calculation explained in section 5.2.1, this
is not a problem as the method from the load description report is used. For the general solution of section
5.2.2 this is a problem, as will be seen in chapter 8.

10 report number 2010/00060 Section 1 Appendix A

11 report number 2009/00144JTAP003
5.2. D ETERMINING THE LOADS IN THE HARDPOINTS 29

Figure 5.3: The virtual VLM model: isometric and side view

5.2. D ETERMINING THE LOADS IN THE HARDPOINTS

From the ground reaction loads on the wheel axle, the loads in the hardpoints can be found. This is done in
two ways: via an analytical derivation for the original geometry, and via dedicated software for a parametric
gear model.

5.2.1. A NALYTIC SOLUTION

As a first step, the load in every hardpoint is derived analytically. This is done for a single gear geometry only:
the one visualised in Fig. 5.2. To find the loads for a parametric gear model is of more interest, so this section
is kept short.
As all load cases in this thesis are static, each member must be in equilibrium. Using that knowledge, the
loads at every hardpoint can be derived from the load at the wheels. By first evaluating the loads on the
wheel axle, then the trailing arm, the shock absorber, the main fitting, the folding side brace and finally the
trunnion, the load in every hardpoint is found 12 . The analytic model is an exact copy of the original concept.
It is used to study the optimal position of this gear under the aircraft, but not to study the optimal position of
the different members. For that purpose, a multi-body model is created with dedicated software.

5.2.2. U SING DEDICATED SOFTWARE

A fully parametric virtual model was created in the Siemens program LMS Virtual.Lab Motion (VLM). The
required inputs for the model are the location of each hardpoint, the position of the CG and the position of the
nose gear. The lengths of the members and their orientation are thus not directly settable, but indirectly via
the definition of the hardpoints. This model evaluates the static equilibrium when the location of a hardpoint
is moved. The initial set-up of this VLM model is shown in Fig. 5.3. It is equal to the analytic model, with the
exception that the location of point I must be specified to set up the VLM model. This was not needed for the
analytical model.
Where the analytical model describes the gear in terms of part-lengths and orientation angles, the VLM model
is set up with the coordinates of the hardpoints. For the initial point, these definitions are exactly the same. A
global axis system is located at every hardpoint. This is also visible in Fig. 5.3. It is assumed that the left gear
is symmetric to the right gear. It is possible to evaluate the extension and retraction of the shock absorber
with the VLM model.
The type of joint at each point is given in table 5.1. The revolute (REV) allows one rotational degree of freedom
(DOF); the spherical (SPH) allows three rotational DOFs; the translational joint (TRANS) allows a member to
extend or retract; and the bracket joint (BRA) is a fixed connection, like a weld.
To perform an analysis, loads are introduced at the nose gear and at the center of the main gear wheel axles.
By fixing the location of the CG, the forces at the other hardpoints of the main gear are found, in a global axis
system. This can be done in sequence for any number of load cases, directly from MATLAB.
As a post-processing step, the forces in the global axis system are rotated such that each member has a purely
axial force and two shear forces. This rotated axis system will be called the part-local axis system. The orien-
12 for more information, see Fokker report 2009/00144JTAP003
30 5. I NITIAL GEAR SET- UP

Table 5.1: Types of joint at hardpoints.

Hardpoint type of joint

A BRA
B REV
C SPH
D SPH
E SPH
F BRA
G SPH
H SPH
I SPH
J BRA
K TRANS

tation of the part-local axis system with respect to the global axis system is found from the set of hardpoints
used to set up the gear model. Spherical coordinates are used. With these local forces per member, the needed
cross-sectional parameters can be determined.
 6
C OMPONENT WEIGHT ESTIMATION
This chapter explains how each component is sized, based on the acting forces. How to obtain these was de-
scribed in the previous chapter. Every LG component is modelled as an H-beam, a cylinder, or a socket. The
forces found in chapter 5 should be multiplied with a factor of 1.5 before they are used to the size the com-
ponents, to obtain the ultimate loads. Thin-walled theory is not valid in this chapter. A generally accepted
assumption for thin-walled structures is that the diameter of the member must be at least 20 times larger
than its thickness. This is not true for landing gear components. Although this chapter is entitled component
weight estimation, the focus of the chapter is on determining the cross-section of the LG components. Once
a cross-section is defined, the weight is easily found.

6.1. S TRESSES IN AN H- BEAM

The FSB is modelled as an H-beam. It has a purely axial load, as it is mounted between two spherical joints.
The maximum load F max that a beam can support before it buckles is thus given by equation 6.1 [35, p.257]:

π2 E I
F max = (6.1)
L2

where E is the Young’s modulus of the material, L is the beam length and I is the moment of inertia. An
H-beam is modelled as an I-beam with cut-outs, see figure 6.1. That cut-out is used to place sensors, wires,
locks, or actuators. This is done to limit the space taken up by the component when the gear is folded.

Figure 6.1: Model of an H-beam

The relevant moment of inertia, in terms of the design variables introduced in figure 6.1, is given by equation
6.2.

3 3
h 3 · (w o − w i ) w i · t web 2 · w cut out · h cut out
I= + − (6.2)
12 12 12

The maximum force that can be supported by the I-beam should be higher than the compressive force acting
on the member.

31
32 6. C OMPONENT WEIGHT ESTIMATION

Figure 6.2: Forces in x, y, and z direction acting on a cylinder

6.2. S TRESSES IN CYLINDERS 1: EXTERNAL FORCES AND MOMENTS

This section uses the local axis system per component; see figure 6.2.
The method explained in Bruhn [36] is used to find the optimum beam cross section. That method can be
summarized as follows:

1. Find the tensile, bending, shear and torsional stress, based on standard equations

2. For every stress type, compare it to the allowable stress (a material property) and express it as a fraction

3. Find the Margin of Safety (M O S), based on these fractions.

To obtain the tensile stress σt , equation 6.3 is used [36]:

Fx
σt = (6.3)
A

where F x is the force along the local x axis and A is the cross-sectional area of the cylinder.
The bending stress is given by equation 6.4 [36]:

Mz · ro M y · ro
σb = − (6.4)
Iyy I zz

with the moment of inertia around the y and z axis I y y and I zz and r o the distance from the center to the outer
section of the beam.
The shear stress due to an axial torque T is given by equation 6.5 and the (average) shear stress due to shear
forces F y and F z by equation 6.6.

T · ro Mx
σst = = π 3 3
(6.5)
J 2 (r o − r i )
q
F y2 + F z2
σs = (6.6)
A
Knowing the stress in the member, it can be compared to the allowable stress F to find a stress ratio. Equation
6.7 gives the equation for tensile stress; the other ratios are found in the same way:
σt
Rt = (6.7)
Ft u
For stress due to shear forces and stress due to torque, the allowable shear stress of the material must be
taken. For bending stress, the allowable bending stress of the beam is selected. This allowable bending stress
is depending on the ratio between the diameter of the cylinder and its thickness, as described in the metallic
materials properties development and standardization handbook [37]. For 300M steel, figure 6.3 is presented.
This non-linear line is approximated by 3 different linear lines as indicated in the same figure.
Equations 6.3 to 6.6 have all assumed that there is not other load acting than those specified. In reality, all
6 forces act at the same time and their combined effect needs to be considered. For that reason, the M O S
6.3. S TRESSES IN CIRCULAR TUBES 2: INTERNAL PRESSURES 33

Figure 6.3: Allowable bending stress of cylinders made from 300M steel [37]

should be found from the stress ratios. This MoS can be determined with equation 6.8. It should always be
higher than 0, but it can be set to a higher value by the user of LIT if this is required.

1
MoS = q −1 (6.8)
(R b + R a )2 + (R s + R st )2

6.3. S TRESSES IN CIRCULAR TUBES 2: INTERNAL PRESSURES

The SA and Retract Actuator (RA) housing contain compressed hydraulic fluids. Therefore, the radial and
tangential hoop stress in these components must be checked. They are not allowed to exceed the yield stress
of the used material. For thick-walled cylinders, the radial and tangential hoop stress are given by equation
6.9 and 6.10 respectively [36]:

p i r i2 − p o r o2 r i2 r o2 (p 0 − p i )
σh,1 = [ ]+[ ] (6.9)
r o2 − r i2 r 2 (r o2 − r i2 )

p i r i2 − p o r o2 r i2 r o2 (p 0 − p i )
σh,2 = [ ]−[ ] (6.10)
r o2 − r i2 r 2 (r o2 − r i2 )
The variables in the above two equations are best explained by looking at figure 6.4:
The tangential stress is most critical for a cylinder where the outer pressure is negligible. Setting the out-
side pressure to 0 and changing the inner pressure to the overpressure, equation 6.10 can be simplified into
equation 6.11:
34 6. C OMPONENT WEIGHT ESTIMATION

Figure 6.4: Cylinder with hoop stress

r i2 + r o2
σh = p( ) (6.11)
r o2 − r i2
This hoop stress in the material should be lower that the ultimate tensile stress of the material.

6.4. S TRESSES ON A SOCKET

The trunnion is modelled as a socket. A socket is a structural member that is designed to hold a pin; in this
case the pins that connect the airframe structure and the landing gear. The main load for a socket is in the
transverse direction; see also figure 6.5.

Figure 6.5: Socket parameters

In figure 6.5, the pin diameter is given by d and the socket thickness by t. The moment is highest in point B
and is given by equation 6.12:

M = W (L 1 + L 2 ) (6.12)
The load in point A can then be derived as equation 6.13 13 :

48M W
F= + (6.13)
11L 22 L2
Based on this force, the burst stress σb and the bearing stress σbr can be calculated with equations 6.14 and
6.15:

F
σb = (6.14)
2t
F
σbr = (6.15)
d
the stress ratio’s are then found with the same method as in equation 6.7, after which the MoS is obtained
with equation 6.16 [38]:

1
MoS = q −1 (6.16)
R b2 + R br
2

13 Internal FLG method provided by Mike Smeets. No scientific source available

 7
D ESCRIPTION OF THE PROOF OF CONCEPT
With all theory described, this chapter explains the set-up of the POC tool, in terms of software. The chosen
optimization architecture will first be explained. The set-up of the tool is the topic of section 7.2. The needed
input to start the tool is explained in section 7.3. Section 7.4 discusses some specific constraints that were
omitted in chapter 3. After that, the output from the tool can be found in section 7.5. The results will be given
in a new chapter: 9.

7.1. MDO ARCHITECTURE CHOSEN

In section 1.2.1, three different MDO architectures were introduced. One of the most important choices that
had to be made is the selection of an architecture.
Studying the works by Chai and Mason [1] and Heerens [2], the first of these uses a methodology closely re-
sembling the IDF architecture and the latter uses a brute-force approach to find a result. Cumnuantip [27]
uses an MDF-like approach in his study. Based on these studies, it is thus inconclusive which MDO architec-
ture is best for the landing gear optimization problem.
Therefore, more research was performed. A usefull article in that respect is the one by Alexandrov and
Kodiyalam [39]. This study took 10 different optimization problems from different engineering disciplines
that have been published. Each of these problems used either MDF, IDF or CO. Alexandrov and Kodiyalam
used all three methods to solve these problems again, to study if one architecture is superior for certain prob-
lems. The results were inconclusive, but they did show that not every architecture can solve every problem.
Another conclusion was that certain architectures are much more efficient than others for the same problem.
So, the choice of MDO architecture is indeed not trivial.
It is concluded that all three discussed MDO architectures will result in a working POC tool. With that in mind,
the choice of architecture is mainly based on code efficiency. It is explained in the next section that object-
oriented programming principles are used during the development of the POC tool. The MDF scheme aligns
best with the principles of OOP, so that optimization architecture is chose. The resulting scheme is visualized
in figure 7.1
The optimization occurs on two levels. Only the top-level optimization is shown in Fig. 7.1, along with the
system-level design variables and the internal variables. The system-level variables are those controlled by
the optimizer (location of the hardpoints); the internal variables are those computed by the different disci-
plines (the other variables). The sizing of each structural part is however an optimization of its own, with
component-level design variables. This optimization module determines the needed cross section for each
part. For the sizing of the trailing arm, for example, there is an optimization routine that uses the force on the
trailing arm and the location of points A and B to determine the inner and outer radius of the trailing arm.
These two radii are the component-level design variables.
This breakdown into a component-level and a system-level optimization was a deliberate choice, for which
two arguments are given. First, it reduces the dimensionality of the problem. Including the component-
level design variables there are over 30, weakly coupled, design variables. It they had all been declared as
system-level design variables, it would have resulted in a 30-D problem. Such a large problem in general has
convergence problems and needs a long computational time [11]. As the design variables are weakly coupled,
the problem is split into several lower-dimension problems that converge much faster.

35
36 7. D ESCRIPTION OF THE PROOF OF CONCEPT

Figure 7.1: Optimization structure of the POC framework.

The second reason for including a component-level optimization is the foreseen expansion of this frame-
work. As described in section 2.4, the parts that have now been modelled as rigid, should be replaced by
modal parts in a later stage. These modal parts are analysed with specialised software. By sizing the rigid
parts in a component-level optimization, they can be replaced by modal parts without making any changes
to the overall framework.
It should be noted that there are other optimization architectures that use a multi-level approach. Examples
are Concurrent Sub-Space Optimization (CSSO) [40] and Bi-Level Integrated System Synthesis (BLISS) [41].
A literature study concluded that these are not the best fit for the LG design problem.

7.2. S OFTWARE IMPLEMENTATION

With the expansion and maintainability of the POC tool in mind, the software was written with OOP princi-
ples in mind. For those unfamiliar with OOP, appendix C gives an introduction into the concept. To summa-
rize that introduction here, OOP gives structure to software by using classes. A class can be seen as a blueprint
for some physical part (e.g. a cylinder). It defines the variables (properties) needed to create (initiate) an ob-
ject, and some specialized functions (methods). These methods only work on that specific class. For instance,
the WA, TA and MF are all cylinders. They are three different instances of the same class.
The inheritance principle is important in OOP. Ideally, no single line of code should be copied. If blocks of
code are needed twice, a parent class can be constructed together with some child classes. The child inher-
its all properties and methods of its parent. Other, new, properties and methods can be defined per child;
properties and methods that its ’brother’ doesn’t have. See for example the classes Cylinder and HBeam in
figure 7.2. These two classes share some properties that they inherited from their parent: lengthMember,
geomInitial, etc. They also have their own properties and methods, for example weightPresHydraulic
for the class Cylinder and HBeam for the class HBeam.
Two top-level parent classes are used for the POC tool: StructuralMember and LGSystem, see also figures
7.2 and 7.3.The construction of a trailing arm gear in figure 7.3 does not benefit from the OOP principles.
However, with LIT as described in chapter 2 in mind, this set-up is useful. It places some of the blocks already
in place.
The benefit of using OOP can be explained with figure 7.2, which shows all classes derived from StructuralMember.
The parent class StructuralMember defines the properties that all derived classes have: material prop-
erties, a length, initial design parameters, and boundaries for these parameters. It is not important that
these design parameters are different for the classes Hbeam and Cylinder. As long as they are a 1xN row,
StructuralMember will accept them. It is checked in the child classes whether the given input is correct. For
example: an object of class Cylinder needs two design variables: the inner radius and the outer radius. If an
object of class Cylinder is initiated with three design variables, the program will throw an error. This is just
one example of the structure given to the program.
The link between the two top-level classes StruturalMember and LGSystem is visible in the class TAGearFSB.
Its properties are specified to be of a certain class. For example, the main fitting must always be of type
Cylinder. This is another example of a structured software set-up.
7.2. S OFTWARE IMPLEMENTATION 37

Figure 7.2: Classes derived from parent StructuralMember.

38 7. D ESCRIPTION OF THE PROOF OF CONCEPT

Figure 7.3: Classes derived from parent LGSystem.

7.3. I NPUT HANDLING 39

Table 7.1: Inputs for POC tool: aircraft parameters.

Parameter Unit Parameter Unit

Aircraft weight(s) N Wing span m
CG location fwd (x,y,z) m Initial position NLG (x,y,z) m
CG location aft (x,y,z) m Initial position MLG (x,y,z) m
Minimum MLG position (x,y,z) m Maximum MLG position (x,y,z) m
Minimum NLG position (x,y,z) m Maximum NLG position (x,y,z) m
Turnover angle deg max pitch angle deg
Max approach speed m/s Tail height m
Max roll angle @ TD deg wing sweep angle deg
distance between secundary tyres m radius sec tyres m
distance between main gear tyres m radius MG tyres m
Minimum load on NLG strut %W Maximum load on NLG strut %W
steering angle nosewheel deg

Table 7.2: Inputs for POC tool: shock absorber parameters.

Parameter unit Parameter Unit

Energy absorption efficiency SA % Energy absorption efficiency tyre %
Load factor at TD - vertical TD velocity m/s
Static SA pressure Pa ratio static over fully extended pressure -
ratio fully compressed over static pressure - Tyre stroke m

The component-level optimizers introduced in section 7.1 are contained in the classes derived from StructuralMember.
The fmincon routine is used, see appendix C for more information. The fmincon function is an optimization
routine from the MATLAB optimization toolbox. It is a gradient-based method, that finds local optima only.
To find a local optimum is fine for the POC tool, as its main objective is to identify improvements to the orig-
inal concept. For a more generic problem however, the use of global optimizers must be considered.
The system-level optimizer is also the fmincon routine. The objective function, the constraints and the de-
sign variables are however different.
This system-level optimizer is defined in the run-file. This file initiates an instance of the class TAgearFSB,
defines the objective, finds its minimum given some constraints, and gives the results. How to start this pro-
cess, which constraints act, and what kind of results to expectis explained in the next sections.

7.3. I NPUT HANDLING

The needed input for the POC tool can be divided into two categories: aircraft data and parameters for the
sizing of the shock absorber. The aircraft data is given in table 7.1, while the SA parameters are given in table
7.2. For this aircraft, type of tyre is fixed by the OEM.
Other inputs are the initial and maximum/minimum allowable cross-sectional parameters of all parts. As
initial parameters, those of the initial concept are used. How the maximum is determined, can be found
in section 7.4. The minimum allowed dimension is in general zero, except for cut-outs. The lengths of the
members are obtained from the MBS model.

7.4. C ONSTRAINTS
Besides the system-level positioning constraints described in chapter 3, there are also component-level con-
straints. Additionally, there are some constraints specific to the POC gear. The component-level constraints
are usually straightforward: the MoS from chapter 6 must be above 0; the inner radius of a part must be
smaller than its outer radius, all dimensions must be larger than zero, the cut-out of the H-beam cannot be
larger/deeper than its height/thickness and every thickness must be larger than 2 mm to make manufac-
turing possible. Although straightforward, these constraints must be specified: the optimization is purely
mathematical and does not take any physics into account.
Then, there is maximum on the cross-sectional parameters. As explained in chapter 2, these parameters are
normally constrained by the gear kinematics: the gear still must fit inside the gear bay. An introduction into
40 7. D ESCRIPTION OF THE PROOF OF CONCEPT

this field of kinematics is provided in chapter 8 of Currey [4]. There was however insufficient time to imple-
ment it in the POC tool. Thus, it was decided to restrict the outer dimensions of the parts to the values used in
the original concept. With these dimensions, the gear fits in the bay. With a larger cross-section, that cannot
be guaranteed. It is allowed that components become thinner or thicker, but it is not allowed to increase the
outer dimensions.

7.5. O UTPUT HANDLING

The resulting gear weight, the position of the gear struts, the final stick diagram from the MBS model, and the
cross sections of each component are obvious output parameters. Additionally, the critical load case for each
part is given as an output, as well as the weight of every individual part.
The most important output parameter is however a message from the component-level optimizers. One
problem encountered with the multi-level optimization is namely that the component-level optimizers can
be erroneous. It can happen that the component-level optimizer does not converge to a correct solution.
This doesn’t have an effect on the system-level optimizer. During development, results were obtained where
parts had an MoS below zero (and thus violated constraints in the component-level optimizer), without the
system-level optimizer giving a warning. The results were presented without any warnings. So, an additional
output is the maximum value of the constraints from the component-level optimizers.
It is the designers task to study the generated output, as there can be several errors in a component-level op-
timizer. Two common errors are described here. The first option is a negative margin of safety. This error can
come from the incorrect calculation of stresses, an incorrect computation of the MoS, or too tight boundaries
on the design variables14 . The first two option occurred during development, but should now be fixed. So
when this error occurs, a relaxation of the boundaries on the design variables should be the first option.
The second possibility is that the optimizer algorithm has convergence problems. This can be tested by
changing the algorithm and running the code again.

14 the optimizer will never use invalid values for the design variables to satisfy the constraints. The boundaries on design variables have
a higher priority than the constraint value
 8
V ERIFICATION OF THE PROOF OF CONCEPT
This chapter compares the results of the analytic and the multi-body model to the results of the original
concept study. This is done for the initial configuration only: the one described in section 5.1. To ensure the
same set of requirements and constraints, the following factors are equal for all three models:

• The maximum allowed outer dimensions of each part

• The global position of the gear struts

• The margin of safety of each gear component

• The evaluated load cases

As an extra verification step, also the critical load case per gear component will be computed. It was described
before that there are 41 different load cases, that the critical case differs per part, and that it is generally
unknown in advance. Therefore, it is relevant output.
The analytic model is modelled in the exact same way as the original concept study. It uses a (fixed) length to
define parts and angles to describe the orientation of the parts with respect to each other. The exact interface
points with the aircraft are not known, as was described in section 5.1. They are not needed to define the
model.
The multi-body model is defined by a coordinate set for all the hardpoints. These coordinates follow from
the description of the original study. The interface points with the aircraft are needed for this model. As these
points are not properly defined, they have been assumed to correspond to the values of the gear description
document.
Table 8.1 breaks down the gear weight in the different component weight and compares them for the original
concept study, the analytical model and the multi-body model. No reliable data is available for the wheel axle,
so that part is not included. The critical load case is also given for each model, as an additional verification
step. As the used data is confidential, the numbers in this chapter are all normalized.
The results are discussed separately for the analytical model and the multi-body model.

8.1. A NALYTICAL MODEL

The analytical model is defined in the exact same way as the original concept study. The following conclu-
sions can be made from table 8.1

• The system weight is predicted within a 2 percent accuracy comparted to the original study

• Except for the folding side brace, all critical load cases are correctly computed by the framework

• The results for the cylindrical components are very similar: within 4 percent

• The results for the H-beams have differences of up to 26 percent

41
Table 8.1: Verification of analytical and multi-body model with respect to original model.
42

Original results Framework Analytical Framework multibody

Component Weight [kg] Critical load case Weight [kg] Critical load case difference [%] Weight [kg] Critical load case difference [%]
Trailing arm 1.337 Turning1 1.335 Turning1 -0.15 1.335 Turning1 -0.15
Main fitting 3.258 2-point braked roll 3.271 2-point braked roll +0.39 1.976 2-point braked roll -39.35 15
Folding Side Brace up 0.522 Side load 0.386 Turning1 -26.14 16 0.472 Turning1 -9.58
Folding side brace low 0.236 Side load 0.191 Turning1 -18.64 17 0.232 Turning1 -1.69
Shock absorber 1 Burst pressure 0.998 Burst pressure -0.20 0.998 Burst pressure -0.20
Retract actuator 0.166 Burst pressure 0.159 Burst pressure -4.00 0.159 Burst pressure -4.00
Trunnion 3.175 2-point braked roll 3.168 2-point braked roll -0.23 4.105 2-point braked roll +29.29 18
Total System 9.693 - 9.507 - -1.91 9.228 - -4.18

15 Due to unknown position of interface point with aircraft. See section 5.1.2
16 Due to different computation of force on member, see section 5.1.2
17 Due to different computation of force on member, see section 5.1.2
8. V ERIFICATION OF THE PROOF OF CONCEPT

18 Due to unknown position of interface point with aircraft. See section 5.1.2
8.2. M ULTI - BODY MODEL 43

The difference in the results for the folding side braces traces back to an error in the original concept study.
The loads on the side brace are incorrectly computed, leading to approximately 20 % higher loads. In the
original concept study, the loads on the folding side brace are independent from the position of and the loads
on the shock absorber. The effect of the shock absorber should however be taken into account.
With that knowledge, it can be concluded that the results from the analytical approach are sufficiently close
to the results from the concept study and the value of the framework is shown. The computational time for
one iteration is reduced from a day to five minutes.

8.2. M ULTI - BODY MODEL

The results for the VLM model are not as expected, although the total gear weight is within 4.2%. The main
problem is that it cannot be verified that the used geometry for the VLM model is equal to the used geometry
for the original model. The location of the interface points with the aircraft is not documented, so they had
to be taken from a different source than the source that describes the gear layout. These two sources do not
correspond to each other, as was seen in section 5.1.2.
For the two components of which it is known that they are correctly modelled, the weight correspond almost
exactly to the original model. These two components are the trailing arm and the shock absorber.
The orientation of the folding side brace is uncertain. Indeed, this component shows a weight difference of
up to 10 percent compared to the original model. This difference on the folding side brace also has an impact
on the main fitting and the trunnion, as these parts still need to be in equilibrium. These two components
also show large differences between the original concepts and the VLM model.
Additional checks were performed to verify that the uncertainty in the model geometry is indeed the cause for
these differences. They were all ruled out, leaving only the geometry uncertainty left as an error possibility.
These checks include:

• Input forces on the wheel axle are the same for the concept study and the VLM model

• Orientation of the gear parts in 3-D space is correctly computed

• The rotation matrices used to change forces from the global to the part-local axis system are correct

• The global forces computed with the VLM software package are correct: all parts are in equilibrium.

As the main objective of this thesis was to develop a working optimization framework and not to optimize
an old concept study, no more effort is spend to further align the results of the concept study and the VLM
model. The differences are accepted as they are. Even more so, the value of the developed framework is again
shown by this verification step. Integration errors such as those made in the original concept are no longer
possible.
 9
R ESULTS
It can be argued that the developed framework, as described in the chapter 7, is the main result of this thesis.
It was decided to keep it out of this chapter however, such that the focus of this chapter can lie on the trade
studies that are possible with the framework.
This chapter then shows some of these results. Section 9.1 shows two case-studies, while section 9.2 has
a more commercial perspective and lists some of the other studies that could be done with the developed
framework.

9.1. R ESULTS OF CASE STUDIES

This section gives the result of two performed trade studies. For the first trade study, the large business jet
trailing arm gear of section 5.1 is analysed with the analytical model. The gear thus has a fixed geometry
and is placed at different locations under the aircraft, to find the optimum location. As this computation is
relatively easy and only a few system-level design parameters are involved, the complete design space can be
visualized.
For the second case study, the gear location is fixed and the relative orientation of the different landing gear
components is varied. This involves more variables and thus visualization becomes harder. As there are now
7 system-level design variables, the complete design space cannot be visualized. The second case uses the
multi-body gear model.

9.1.1. C HANGING THE GEAR LOCATION

For this trade study, the positions of the hardpoints were fixed, using the analytical model. The coordinates
of the NLG and MLG struts are the system-level design variables and their optimal location is designed for.
Normally, only one answer is relevant: which gear position gives the lowest weight? This section takes a
different approach and computes the gear weight at over 700 different, unique positions. This results in figure
9.1, which shows the weight of the MLG as a function of the x position of the NLG, the x position of the MLG
and the y position of the MLG. The gear weight is represented as a fourth dimension: the color. Only a few
slices of the full design space are visualised.
As figure 9.1 may be difficult to interpret, figures 9.2 to 9.4 visualize the same data set in another way. In each
figure, one of the design variables (x position MLG, x position NLG or y position MLG) is fixed at a certain
value, giving the gear weight as a function of the other two design variables. A contour plot of the system
weight is given.
Figure 9.2 gives the gear weight as a function of the y position of the MLG and the x position of the NLG. In
each sub-figure, the x position of the MLG is frozen. It is at its most forward position in the top left figure and
moves aft in clock-wise direction.
Similarly, the x position of the nose gear is frozen in figure 9.3; giving a contour plot of the gear weight as a
function of the MLG position. Again, the NLG is at its most forward location in the top left figure and moves
aft in the clock-wise direction. Figure 9.4 finally gives a contour plot of the gear weight as a function of the x
position of both NLG and MLG. The y position of the MLG is frozen. It is most inboard in the top left figure
and moves outboard in the clockwise direction.
In line with the previous chapter, the values in these figures are again normalised.

45
46 9. R ESULTS

Figure 9.1: 4-D representation of the design space. System weight represented by color as a fourth dimension.

Figure 9.2: X position MLG fixed in each sub-figure. Moving aft from top left and in clockwise direction

The following conclusions are made:

• If the main gear is moved aft, the system weight decreases. This is as expected. If the MLG moves aft, it
moves further away from the CG and it has a lower static load. The load on the NLG becomes higher.
Since only the MLG is sized in this study, it is as expected that a lower load on it leads to a lower system
weight19 . This is clear in figure 9.2, where the system weight decreases in a clockwise direction if the
value for the x position of the MLG increases. It can also be seen in all sub-figures of figures 9.3 and 9.4,
where the system weight decreases as the value for the x position of the MLG increases.

• If the nose gear is moved aft, the system weight decreases. This is also as expected. Again, if the NLG
moves aft, it moves closer to the CG and takes a higher load. This decreases the load on the MLG,
leading to a lower system weight. This can also be seen in all sub-figures of 9.2 to 9.4.

• The influence of the y position of the MLG is marginal. Figure 9.4 shows a contour plot of the system
weight for 4 different track widths. There are slight changes between the different sub-figures, but they
19 the NLG, in contrast, will experience a higher load and become heavier, but that effect is not taken into account yet
9.1. R ESULTS OF CASE STUDIES 47

Figure 9.3: X position NLG fixed in each sub-figure. Moving aft from top left and in clockwise direction

Figure 9.4: Y position MLG fixed in each sub-figure. Moving outboard from top left and in clockwise direction
48 9. R ESULTS

are much weaker than the effect of the other two parameters. A similar behaviour is seen in figure 9.2,
where the lines are almost vertical and the MLG y position has thus little influence. In figure 9.3 the
lines are nearly horizontal, again supporting the conclusion that the MLG y position has little influence
on the system weight.

It should however be noted that figures 9.1 to 9.4 do not look like the results of a typical optimization problem,
where one would expect a single optimum. For the results of this study, there is an area (valley) of optimal
gear locations that all result in the same unique gear weight: 0.893.
This can be explained by studying the results file and by looking back at the load cases defined in section 4.2.
For this particular case study, it is observed that the critical load case for every component is either the side
load or the braked load case at the optimal point. For these two load cases, the gear position does not have
an influence on the strut load. So, as long as these two load cases are critical, the gear weight is always the
same. Moving away from that valley, other load cases become critical, for example the turning load case. For
these points, the gear however is heavier. This is graphically indicated with figure 9.5.

Figure 9.5: Weight of the trailing arm as a function of the main gear location

This figure shows the critical load case and the component weight as a function of the gear position. As long
as the side load case is critical, the gear position has no influence on the component weight. A similar figure
can be drawn for all other components.
A typical evaluation with this level op detail takes around five minutes. To generate figures 9.1 to 9.4 over 700
evaluations were however needed, increasing the runtime to several hours.

9.1.2. C HANGING THE GEAR GEOMETRY

In the figure 9.7, the location of hardpoints B, C, D and E is changed while the location of the gear under the
aircraft is fixed. Figure 5.2 is repeated in this chapter as figure 9.6 to define the hardpoints once more.
Only points B to E can be moved, as I, G, and H are the interface points to the aircraft, and points J and F
are dependent on the location of B and C. Point A was varied in the previous case study and is fixed for this
computation. The top figure of 9.7 shows how much these seven design variables are changed with respect
to their initial point. The second figure shows the gear weight at each iteration. After a relatively large first
step, the solution slowly converges to an optimum in 11 additional iterations. The figure is a perfect example
of what one would expect from an optimization [11]. A reduction in gear weight of almost seven percent is
achieved, from the initial configuration of the multi-body model to the computed optimum. The runtime
is becoming much longer in this evaluation, as the VLM model must be evaluated at each iteration and the
number of system-level design variables is increased from three to seven. The model can still be run overnight
however; it has a typical runtime of seven to eight hours.

9.2. OTHER POSSIBILITIES OFFERED BY THE FRAMEWORK

The possibilities offered by the current framework are more elaborate than the two trade studies described in
the previous section. Some example application are mentioned below:
9.2. OTHER POSSIBILITIES OFFERED BY THE FRAMEWORK 49

(a) Aft view (b) Side view

Figure 9.6: Stick diagrams of the initial concept

Figure 9.7: Optimizing the gear by changing the locations of the hardpoints.

• With the current optimization framework, a landing gear geometry can be found that limits the forces
on the airframe to a pre-defined value. One common problem in landing gear design is namely that the
airframe needs local strengthening at the attachment point of the landing gear, because the introduced
loads are too high. With the presented framework, it is possible to predict in an early stage of the design
process how large the introduced forces are and thus how much reinforcement will be needed.

• If an aircraft manufacturer has not fixed the location of the landing gear yet, a very quick study to find
the optimal location can be performed. This is especially useful if the aircraft configuration itself is not
fixed yet. One example could be that the wing shape and placement is undefined. For all possible wing
configurations, the optimal gear and its weight can be computed. This analysis can be done within 15
minutes.

• It can be investigated what the effect of an increase in aircraft MTOW will be on the weight of the landing
gear system, by simply changing one input parameter.

• Application of the framework is not limited to the main landing gear of a long-range business aircraft.
The framework can also be used to study the MLG or NLG of a helicopter, or the NLG of a short- to
medium-range passenger aircraft.
 10
C ONCLUSION
Four weaknesses were identified for the traditional preliminary landing gear design process. First of all, it
is relying too much on existing topologies. Secondly, it is not very efficient. Many man-hours are wasted
by performing repetitive design tasks instead of performing creative work. Communication between disci-
plines is also a problem. In preliminary design, many concepts are considered which all have their own set of
parameters. Informal communication between designers leads to mistakes and inconsistent results. Lastly,
integration between the design of the landing gear and the design of the aircraft is often sub-optimal.
This thesis had two main objectives to improve this preliminary landing gear design process. First, a frame-
work was described that can perform the concept design of a landing gear automatically: LIT. The analogy
to an onion was made to explain its different layers: step by step, more design variables are added to the
problem until the optimal landing gear solution is found. As a first step, the locations of the gear struts are
determined based on static load cases and some aircraft parameters. Then, the gear geometry is studied in
more detail. As a third step, dynamic load cases are added and the problem is restarted. As a final step, the
rigid parts are studied in more detail by interchanging them with modal parts.
A proof-of-concept (POC) tool is developed to show the working principles of LIT. This POC studies one par-
ticular fictitious gear: the main landing gear of a long-range business class aircraft. A trade-study to find the
optimal location of this gear is shown in the results chapter. The runtime for this trade study is shorter than
five minutes. Also, the position of the hardpoints was varied to analyse if this yields a lower weight. With a
runtime of eight hours this analysis takes longer, but it can still be done overnight. Compared to the initial
configuration, a weight decrease of almost seven percent was realized.
These two trade studies are only some of the options offered by the framework. Another possibility would be
to tailor the loads introduced at the airframe, in the earliest stages of the design process. It is also possible to
find a trend in the weight of the landing gear versus the MTOW of the aircraft. Furthermore, this framework
is not limited in application to the main gears of a long-range business aircraft. It is equally suited to size the
NLG or MLG of a helicopter or the NLG of a short- to medium-range passenger aircraft.
The analytic model used for the first trade study does not exactly correspond to the multi-body model used
for the second trade study, as there is an insecurity in the position of the interface points with the aircraft.
The difference in system weight is however below 4 percent. As the main objective of this thesis was to de-
velop a working framework for landing gear design, and not to study one particular gear in full detail, it can
be concluded that the research objective is met. The further development still needs a lot of work, but the
proof-of-concept version is there.

51
 11
R ECOMMENDATIONS
This thesis explained the final lay-out for LIT and the development of a POC tool. This chapter describes the
road towards completion of LIT. The tool must be expanded both in width (adding more capabilities) as in
depth (adding more detail). That expansion is treated in sections 11.1 and 11.2.

11.1. E XPANDING THE TOOL : ADDING MORE CAPABILITIES

At this point, the tool is only capable to analyse aircraft with a tricycle gear layout. Another limitation is that
only the TA gear design is implemented. To add more functionality, the following expansions are recom-
mended:

• Derive the strut loads for a taildragger gear layout, such that a comparison between both layouts can
be made. This is mainly relevant for helicopters.

• Also military aircraft requirements should be implemented. This allows LIT to be used for the LG of
fighter jets. The articles by Kempf [42] and Thorby et all [43] might serve as a starting point.

• A module should be added that estimates the LG cost. This has several implications, as the problem
now transforms into a multi-objective optimization.

• Throughout this thesis, it was assumed that the loads are perfectly alligned with the gear struts. This is
not true in real life. Significantly higher loads of up to 40 percent more may be the result, according to
NACA technical note 2596 [44]. Although the aircraft regulations do not mention such eccentric loads,
this deserves further attention.

• Flotation characteristics20 are not considered in the current method. This becomes important when a
landing on unpaved runways is considered. A good starting point in this field would be chapter 7 of [1]
and the references there.

• The result from LIT should be visualized in CATIA to obtain graphic results instead of numerical tables.
This requires the definition of parametric landing gear parts in CATIA, as well as a method to initiate
them. Sending data from MATLAB to CATIA can be done via iSight or data sheets.

• At this moment, only the trailing arm gear is modelled. The other gear topologies from chapter 2 should
be added.

• In this thesis, an existing concept gear was further optimized. Sizing rules should be defined to initiate
a first concept for any type of aircraft.
20 where the load per tyre is compared to the runway strength

53
54 11. R ECOMMENDATIONS

11.2. E XPANDING THE TOOL : ADDING MORE DETAIL

Besides adding more capabilities to the tool, a more detailed analysis of the landing gear should be possible.
The following topics are proposed:

• Retraction characteristics of the gear are not considered. It should be checked if the gear can fold in the
LG bay.

• Energy absorption of the tyres is not properly modelled in this thesis. Tyre models are already created
at FLG, and these should be incorporated with the work of this thesis.

• During the creation of this tool, the fmincon function from the MATLAB optimization toolbox is used.
This function has many benefits, but also some weaknesses. In general, it only finds the local minimum
of an objective function and not the global minimum. When there is no initial design to start from, the
use of global optimizers might be needed. These are available in the global optimization toolbox of
MATLAB.

• It was also assumed that the design space is continuous. That might not always be true. If the design
space is discontinuous, a gradient-based optimizer such as fmincon is not the best solution and a direct
search method such as genetic algorithms or particle swarm optimization should be considered. For
some problems, the use of gradient based methods is still possible for a discontinuous design space,
however.

• The current toolset minimizes the landing gear weight. A related problem is to minimize the weight of
the landing gear + aircraft structure. This might become a feasible research direction. A good starting
point in that direction is the research at the German Aerospace Center (DLR) [3], [27], [45], [46].

• The shock absorber is sized using simplified hand calculations. With VLM it is possible to add 1D SA
models. This improves the solution, and allows the implementation of dynamic load cases.

• Sizing of the components is done with simple hand calculations. Modal parts should be added to in-
clude the effect of displacements and to study critical points better.

• This thesis only considers the main landing gear components such as the wheel axle, the shock ab-
sorber and the main fitting. Smaller components such as lugs and pins should be added.
 A
L OAD CASES FROM REGULATIONS

55
 56

Subject CS25 [31] CS23 [32] CS27 [33] CS29 [34] STANAG4761 [47]
Ground loads are external forces, and
471: Ground loads -
the complete CG range No mention made of CG range Same as CS-25 Same as CS-25 Same as CS-23
General
should be considered
The vertical velocity should be 10 fps as The vertical velocity must be
The vertical velocity is 10 fps maximum and 7 fps as minimum, determined from a free-fall
473: Landing load Same as CS-27, but drop height
at MLW and 6 fps at MTOW, and MTOW should always be taken, from a height of 0.33m, Same as CS-23
conditions and assumptions should be 0.20m
the aircraft lift equals its weight Wing lift equals 2/3 of the weight, and Always use MTOW,
a load factor of 2.0 must be used Rotor lift is 2/3 of the weight
475: tyres and Tyres in static position, and
- - Same as CS-27 -
shock absorbers shock absorber in critical position
479: Level landing see page 61 No drift landing No drift landing No drift landing No drift landing
481: Tail-down landing See page 62 Same as CS-25 Same as CS-25 Same as CS-25 Same as CS-25
483: One-wheel landing See page 62 Same as CS-25 Same as CS-25 Same as CS-25 Same as CS-25
Same as CS-25, but extra condition
Vertical force is 1.33 times Vst at i c , and
3-point landing must be considered
485: Side load See page 25 side load is 0.5 and 0.33 times vertical same as CS-27 Same as CS-23
as well: side load on nose is 0.8
force
times vertical load
491: Taxi, take-off
See page 58 Not mentioned Not mentioned Not mentioned Not mentioned
and landing roll
Load factor is 1.33 for 3-point landing,
Load factor is always 1.33 and
493: Braked roll See page 24 Load factor is 1.0 for 2-point landing, Same as CS-27 Same as CS-23
dynamic braking not mentioned
no dynamic braking is mentioned
495: Turning See page 24 Not mentioned Not mentioned Not mentioned Not mentioned
497: supplementary conditions for tail wheels. Not considered during this thesis
The nose wheel should be able to withstand:
V = 2.25*Vst at i c and D = 0.8*V, and
499: supplementary
See page 60 V = 2.25*Vst at i c and D = -0.4*V, or Not mentioned Not mentioned Same as CS-23
conditions for nose wheels
V = 2.25*Vst at i c and S = 0.7*V
but no combination of the above
501: Ground loading conditions: landing gear with skids. Not relevant for LIT
503: pivoting See page 60 Not mentioned Not mentioned Not mentioned Not mentioned
505: Supplementary conditions for ski-planes. Not relevant for LIT
Load factor vertical: 1.35,
507/519: Jacking See page 61 Any combination of side and drag load Not mentioned Not mentioned Same as CS-23
of 0.4 times static vertical load
507: Reversed braking See page 60 Not mentioned Not mentioned Not mentioned Not mentioned
Same conditions as CS-25, but
509: Towing loads See page 60 Not mentioned Not mentioned Same as CS-23
different table
The 50-50, 40-60, 60-40,
511: unsymmetric loads
0-60 and 60-0 scenarios Essentially the same as CS-25 Not mentioned Essentially the same as CS-25 Essentially the same as CS-25
on multiple-wheel units
of section 4.3
A. L OAD CASES FROM REGULATIONS
 B
L OAD CASES DESCRIPTION
This appendix has been written as a stand-alone chapter and details all 41 static load cases from CS-25. They
can be divided into 2 scenarios: ground load cases and touchdown load cases.

B.1. G ROUND LOADS

This section treats the derivation of the ground loads, as specified in paragraphs 491 until 519 of CS-25. These
include:

• taxi loads
• braking loads
• turning loads
• jawing loads
• pivoting loads
• towing loads
• jacking loads

S TATIC LOADS
The static load condition is used as a basis for several load cases. Therefore, it is included in this chapter.
Figure B.1 shows an Free Body Diagram (FBD) of this condition, where n equals 1. The load carried by the
main gear struts, VM , is shared by the left and right strut and must thus be divided by 2 to obtain the load per
strut.

Figure B.1: Free body diagram for static load condition. Side view

Taking moments about the point where the load on the main gear struts acts, equation B.1 can be derived to
calculate the static load on the nose gear:

Wb
VN = (B.1)
d

57
58 B. L OAD CASES DESCRIPTION

Where VN is the load on the nose gear, W the aircraft weight, b the distance between the main gear and the
centre of gravity and d the distance between the nose and main landing gear.
Taking moments about the point where the normal force of the NLG acts, the static load per MLG strut can
be calculated according to equation B.2

Wa
VM = (B.2)
d
With VM the load on the main gear and a the distance between the nose gear and the CG. For the static load
condition, the drag and side loads on the struts are 0.

TAXI LOADS
AC25-491 states that for taxi load, a load factor of 1.7 times the static loads can be assumed to give the max-
imum loading during taxi21 [48]. A more elaborate dynamic analysis should be made later in the design
process, for example as described by Freund et all[49].
For now, the factor of 1.7 will be used. When doing so, figure B.1 and equations B.1 and B.2 can be applied,
where the load factor should be added to the numerator.
A second requirement is the combined taxi case. In this case, the vertical force must be taken as 90 percent of
the above vertical load, and the drag and side loads must be 20 percent of the vertical load.

B RAKING LOADS
Section 25.493 of the regulations describes 3 different braking scenarios:

1. braking without any load on the nose wheel and with the pitching moment of the aircraft resisted by its
angular acceleration (2-point braking);
2. braking with load on the nose wheel and zero pitching acceleration (3-point braking);
3. a sudden braking motion with dynamic pitching behaviour as a result (dynamic braking).

Points 1 and 2 must be evaluated at the MRW with a load factor of 1 and at the MLW with a load factor of 1.2.
The third point only needs evaluation at the MTOW.
When there is no load on the NG, all weight is carried by the MLG struts, as in equation B.3

nW
VM = (B.3)
2
Where n is the load factor. The maximum drag force per strut must be taken as 0.8 times the vertical load per
strut; the side load as 0.
The force on the main gear can be found by taking moments around the attachment point of the nose gear
normal force, see equation B.4:

nW a
VM = (B.4)
2(d + 0.8e)
With e the height of the CG and the other symbols as before.
The load on the nose gear strut VN can be calculated by taking moments around the CG and substituting
equation B.4. The result is equation B.5:

nW (b + 0.8e)
VN = (B.5)
(d + 0.8e)
There is an additional drag load on the main gear of 0.8 times the vertical load. The side load is 0.
For the dynamic braking case, an equation is provided by the regulations, see equation B.6:

W f ·µ·a ·e
VN = [b + ] (B.6)
a +b a +b +µ·e
In this equation, f is the dynamic response factor that is 2.0 unless a lower factor can be proven and µ is the
friction coefficient which is 0.8.
The load that is not carried by the nose gear is divided over the main gear struts. These main gear struts
experience an additional drag load of 0.8 times the vertical load. The side load is 0, for both the MLG and the
NG.
21 This is true for multi axle gears. For single axle gear, a load factor of 2.0 should be applied according to Advisory Circular (AC) 25-491
B.1. G ROUND LOADS 59

T URNING LOADS
There are 6 turning scenarios defined in CS25.495 and CS25.511:

• Turning left with all tyres inflated

• Turning with one of the nose wheels deflated
• Turning left with the inner tyre of the right strut deflated
• Turning left with the outer tyre of the right strut deflated
• Turning left with the inner tyre of the left strut deflated
• Turning left with the outer tyre of the left strut deflated

Another 6 cases can be thought of, symmetric t the cases described above. Figures B.2a and B.2b give a free
body diagram for the case where the inner tyre of the right strut is deflated.

(a) Top view (b) Top view

Figure B.2: Free body diagram for turning conditions

Without any deflated tyres, equations B.7 to B.9 can be derived for the vertical load on the gear struts by
evaluating the sum of moments on the nose gear and at the CG:

W e a
VM R = (− + ) (B.7)
2 t d
W e a
VM L = ( + ) (B.8)
2 t d
Wb
VN = (B.9)
d
Where VM R and VM L are the load on the right and left main strut, and t is the main gear track width. There is
a side load of half the vertical load on all three struts; the drag load is 0.

The equations for the other 5 load cases are given in table B.1. In these equations, u is the distance between
the main gear wheels and v is the distance between the nose gear wheels.

Table B.1: Loads on left and right main gear strut for different turning load cases.

Scenario Left strut Right strut

¡W ¢³ t b
´ ¡W ¢³ t b
´
Flat nose wheel t 2d a + 0.25e + v d t 2d a − 0.25e − v d
¡ W ¢¡ t ¢ ¡ W ¢ ¡¡ t u
¢ ¢
Flat inner right wheel t +u 2d a + 0.25e t +u 2d + d a − 0.25e
¡ W ¢ ¡¡ t u
¢ ¢ ¡ W ¢¡ t ¢
Flat outer right wheel t −u 2d − d a + 0.25e t −u 2d a − 0.25e
¡ W ¢ ¡¡ t u
¢ ¢ ¡ W ¢¡ t ¢
Flat inner left tyre t +u 2d + d a + 0.25e t +u 2d a − 0.25e
¡ W ¢¡ t ¢ ¡ W ¢ ¡¡ t u
¢ ¢
Flat outer left tyre t −u 2d a + 0.25e t −u 2d − d a − 0.25e
60 B. L OAD CASES DESCRIPTION

N OSE WHEEL JAW LOADS

For nose wheel jaw, there are 2 cases described in CS25.499. In the first, the vertical forces on the struts are
equal to the static case. There is an additional side load of 0.8 times the vertical load on the nose gear. This
load should be balanced by the side loads on the main gears; this is shared evenly.
In the second case, it is assumed that brakes are used to lock 1 of the MLG struts. The airworthiness specifi-
cations for this case can be translated into figure B.3, where a top view and a side view are given:

Figure B.3: Free body diagram for the second nose wheel jaw condition. Top view on the left and side view on the right.

The load on the main gear can be derived by taking moments around the nose gear, resulting in equation
B.10.

aW
VM = (B.10)
2(d + 0.4e)
The load on the nose gear is given by equation B.11. It can be found by summing the moments around the
CG in the side view of figure B.3, and substituting equation B.10.

b + 0.4e
VN =W (B.11)
d + 0.4e
The side load on the nose gear is obtained by evaluating the moment around point C in the top view of figure
B.3 and is given below in equation B.12. The side load on the main gear strut balances the side load on the
nose gear.

0.2t aW
SN = (B.12)
d (d + 0.4e)
The drag load on the nose gear can be taken as 0, while the drag load on the main gear should be taken as 0.8
times the vertical load.

P IVOTING LOADS
The vertical, drag and side loads during pivoting are equal to the static loading conditions, following CS25.503.
There is however an additional moment that must be taken into account for the NLG, as computed by equa-
tion B.13.

M Z = 0.4Vm v (B.13)

R EVERSED BRAKING LOADS

The reversed braking load case is described in CS paragraph 25.507. In this case, the vertical load is equal to
the static case. There is an additional drag load of 0.55 times that vertical load and no side load.

T OWING LOADS
For towing, the regulations provide a table of load cases that need to be checked in paragraph CS25.509; see
table B.2. The first step in computing the towing loads is to calculate the required towing force. That towing
force must be taken as 0.3 times the Maximum Ramp Weight (MRW) if the MRW is below 30,000 pounds; 0.15
times the MRW if the MRW is over 100,000 pounds and calculated according to equation B.14 if the MRW is
in between these weights.
B.2. T OUCHDOWN LOADS 61

6 · M RW + 450, 000
F t ow = [l bs] (B.14)
70
Then, the side and drag loads on the struts need to be calculated according to table B.2. On top of that, the
gears should also support the static vertical loads.

J ACKING LOADS
For jacking, described in CS25.519, the vertical force is 1.33 times the static value; combined with a horizontal
load of 0.33 times the static vertical load. Worst-case scenarios to check here are any horizontal force vectors
pointing 45°inboard or outboard, and any multiple of 90°.

B.2. T OUCHDOWN LOADS

The different touchdown load cases are described in paragraphs 479 till 487 of CS-25. This section again
treats every load case in a separate section. For all touchdown load cases in CS-25, it can be assumed that
the lift provided by the wings equals the weight of the aircraft, unless some system drastically alters the lift
generating capability of the aircraft.

L EVEL LANDING
The level landing case is illustrated in figure B.4 and described in CS25.479. For this condition, 4 load cases
need to be checked:

1. Normal landing on 3 struts

2. Drift landing on 3 struts
3. Normal landing on main gears only
4. Drift landing on main gears only

Figure B.4: Level landing condition [31].

For a normal landing, the vertical forces on the struts are the same as those in the static load case. The drag
loads must be taken as 25 percent of these vertical loads and the side loads can be assumed to be zero. This

Table B.2: Load cases for aircraft towing [31].

Load
Tow Point Position
Magnitude No. Direction
1 Forward, parallel to drag axis
0.75*F t ow per 2 Forward, at 30°to drag axis
Main Gear
main gear unit 3 Aft, parallel to drag axis
4 Aft, at 30°to drag axis
5 Forward
Swivelled forward
6 Aft
1.0*F t ow
7 Forward
Swivelled aft
8 Aft
Nose gear
Swivelled 45° 9 Forward, in plane of wheel
from forward 10 Aft, in plane of wheel
0.5*F t ow
Swivelled 45° 11 Forward, in plane of wheel
from aft 12 Aft, in plane of wheel
62 B. L OAD CASES DESCRIPTION

holds for both the NLG and the MLG.

For the drift landing, the vertical load per strut is 75 percent of the load in the normal landing case. The drag
load is then 40 percent of the vertical load and the side load 25 percent of the vertical load.
When there is a normal landing on the main gears, the weight of the aircraft is divided over both gear struts;
both carrying half the weight. The drag load is again 25 percent of the vertical load and the side load is zero.
The nose gear takes no loads.
When the drift landing on the main gears is considered, the vertical load is 75 percent of the vertical load in
the normal 2-point load case. The factors for the drag load and side load are again 0.4 and 0.25. The nose gear
takes no loads.

TAIL DOWN LANDING

In the tail down landing scenario, CS25.481, the load case is similar to the third load case described above:
the main gears each carry half the aircraft weight and the drag load is equal to 25 percent of the vertical load.
In this case, however, the aircraft must be assumed to be at its maximum allowable pitch angle θ.

Figure B.5: Tail down landing condition [31].

As the aircraft has a significant angle with respect to the ground, the loads experienced by the struts have to
be transformed according to equations B.15 and B.16. The side load in this case is zero. The nose gear is zero.

Vst r ut = sin θ · D M + cos θ · VM (B.15)

D st r ut = − sin θ · VM + cos θ · D M (B.16)

O NE GEAR LANDING
The loads for the one gear landing should be taken equal to the normal landing on two wheels, where it is
assumed that the unbalanced external load is reacted by the aircraft inertia. This load case is thus irrelevant.

L ANDING WITH SIDE LOAD

The side load condition is described in CS25.485. For this load case, it is assumed that the weight of the
aircraft is divided over both main gear struts. There is an additional side load of 0.8 times the vertical load on
one gear strut, and -0.6 on the other. The drag load is zero; see Fig. B.6:

Figure B.6: Side load landing condition [31].

 C
D ETAILED DESCRIPTION OF LIT SOFTWARE
This appendix gives a more detailed description of the code structure. Before describing that in detail, some
important design choices are explained. Why is object oriented programming (OOP) used, and what are its
benefits; how does the fmincon routine in MATLAB work and why is it chosen?

C.1. N OTES ON O BJECT-O RIENTED P ROGRAMMING

This section starts with an often made mistake about OOP. No: it will not save time during development of
a program. It will save time when expanding a program, making changes and having several other people
involved, however. So, why would you want to use OOP?
In normal procedural programming, everything is about data, variables, and functions. As the program grows
in complexity, more lines of code are added, more functions are written and more variables are introduced.
Often, pieces of code that are needed in more than 1 place are copied. As a program grows and grows (for this
thesis, over 3000 lines of code was the moment my eyes opened), things become very messy and unstruc-
tured. Input and output handling to functions is a nightmare, data is sent from function to function via Excel
sheets, a typo adds new fields to a structure and the overall structure of your program is just impossible to
explain to someone else. You start to look for another way of writing your code and the answer is found: OOP.
So, what is OOP?
Object-Oriented Programming is all about classes. A class can best be explained as an intelligent MATLAB
structure. A class has predefined properties (similar to MATLAB fields). If one tries to change the name of
these properties, it is not allowed and an error is thrown. Also, if an incorrect number of inputs is given when
initiating an instance of the class, this is not allowed. So in that respect, classes are much more robust than
structures. They are ordered more rigidly and typos result in errors instead of unexpected outcomes. The
properties of a class can be classes itself (for example, referring back to chapter 7, the trailing arm is of type
Cylinder: a self-defined class), but also structures or any other type. The properties of a class are always
available once the class is initiated, much like global variables. The difference is that the use of global vari-
ables is generally strongly discouraged, and that classes function properly.
The class is intelligent because it has methods. A method can be explained as a function that only works for
that particular class. For example, there are different sizing functions for the different structural members
used in this thesis (I-beams, Cylinders, Sockets, etc.). All have the same name: findWeight. But if a part is
defined to be of the type Cylinder, it will never be possible to use another sizing function on it than the one
belonging to Cylinder.
Helper functions, functions that are only needed for a particular class, can be defined below the class defini-
tion in the same file. This keeps the number of functions in the folder structure lower. Because the set-up of
a class always follows the same structure, it is easy to understand for another programmer (assuming he/she
knows a bit about OOP).

One major advantage of using OOP is that data is always available. For example, when sizing a member,
the acting loads are needed. In procedural programming there are 2 options to obtain that data, when there
is a separate function that calculates these loads. As a first option, the function calculating these loads could
be called, with the relevant in- and output. That input should be gathered somewhere and many outputs of

63
64 C. D ETAILED DESCRIPTION OF LIT SOFTWARE

these functions are probably not needed, but some are. It quickly becomes an administrative nightmare to
keep track of the correct calling routines, especially if these functions are called at several places. Addition-
ally, if this function takes a bit of time and needs to be evaluated for every member, this option is not feasible.
The second option is to write the output to some file from which it can be read. This is not very efficient
though, and these writing files quickly becomes a mess. At some point in time, another value is needed, shift-
ing the locations of the other values. These locations are hard-coded and spread over different files. One
forgets to edit one of these hard-coded values, and a nasty error is the result.
Using OOP, the load data is a property of the gear. Thus, when sizing the trailing arm, the loads data can be
obtained by just typing, for example, obj.loads.pointB. No function calls or temporary writing files, but
simple and clean coding.
Another advantage of OOP is that some properties can be fixed or hidden, by making the constants or by pro-
tecting their access. For example the gravitational constant g is stored as the constant property gear.g. So,
the field gear.g cannot be overwritten by accidence at a later point. This in contrast to ordinary structures,
where every field can be overwritten.
The third major advantage of OOP is the DRY principle. Don’t Repeat Yourselves. This is where the inheri-
tance principle of OOP should be mentioned. The basic principle is that no piece of code should be occurring
twice. If it is needed twice, a parent and child function should be created. This has been done several times,
but will be explained based on the class ShockAbsorber.
Both the retract actuator and the shock absorber are modelled as a cylinder. The main failure mechanism
for both members is the burst pressure case, as a result of the compressed hydraulic fluid inside these mem-
bers. Where the pressure inside the retract actuator is more-or-less constant22 , the pressure inside the shock
absorber varies with the load on the wheels. There is an additional calculation needed to determine this pres-
sure, but the principle strength calculations are similar for both members. The retract actuator is modelled as
the class Cylinder.The additional calculations for the shock absorber call for a child class ShockAbsorber,
which is constructed from the class Cylinder. It inherits all sizing methods available to Cylinder, but has
its own method to calculate the needed pressure. The methods of the class Cylinder are available to the
shock absorber and the retract actuator, while they have been specified just once. The methods of the class
ShockAbsorber are only available to the shock absorber, and not to the retract actuator.
When using procedural programming, it is also possible to not repeat yourselves. This requires to make func-
tions for every computation that should be done more than once. This typically results in many functions
and does not make the understandability of the code any easier.

C.2. N OTES ON FMINCON

Throughout this thesis, the solver has been a black box. It was there to find the minimum; it was needed as
the minimum of the objective function could not be found without it; but which solver was used and why was
not properly addressed. This section goes into some more detail, but without using lots of mathematics.
Within the scope of this thesis, where finding a local minimum is the goal and the global minimum is not
considered yet, the MATLAB optimization toolbox is needed. This toolbox offers several different pre-defined
optimization routines. Depending on the problem at hand, multiple solvers may be chosen. A nice article is
written by MATLAB to explain the different solvers and how to choose one23 . It turns out that for a smooth
but non-linear objective function, with smooth but non-linear constraints, fmincon is the recommended al-
gorithm.
To run the optimization, fmincon needs an objective function with a single output, an initial vector contain-
ing all design variables, and constraints. These constraints are placed in a dedicated constraints function.
Boundaries for the design variables are specified: physical dimensions should be larger than zero and cannot
be infinite. This is a trivial example, but as the optimization is pure mathematics, the constraints need some
attention. Another example of a trivial but critical constraint is that the outer radius should be larger than the
inner radius 24 . If that is not told to MATLAB, very strange results can be the result.
The function fmincon can take 4 different types of constraints: linear equalities, non-linear equalities, linear
inequalities and non-linear inequalities. All constraints have been specified as being non-linear, even if they
are in fact linear. This is less efficient in terms of computation time, but it allows all constraints to be grouped
in the same function. This makes the set-up of the program easier to be understood.

22 it is dependent on the weight of the gear, which doesn’t change too much in a single iteration
23 google for "choosing a solver Matlab"
24 Alternatively, one could specify the outer radius as the inner radius plus some thickness. In that case, the thickness must be positive
C.3. S TRUCTURE OF THE SOFTWARE FOLDER 65

Table C.1: Runtime needed for the different algorithms of fmincon

Algorithm Runtime [sec] iterations [-] function calls [-]

active-set 85.34 2 8
sqp 134.13 3 12
interior-point 653.45 6 57

The optimizer, the objective function and the constraint function are placed in a nested function. All cal-
culations are performed in 1 function: computeall. This function is called once per iteration and sends all
correct values to objective function, constraint function, and top-level optimization function. This reduces
the runtime with a factor of two.
There are many options to be set while performing optimizations with fmincon. Most importantly, probably,
is the choice of algorithm. Four different ones have been defined within MATLAB for fmincon: interior-point,
trust-region-reflective, sqp and active-set. Not all algorithms work on all problems. For the same problem set-
up, some algorithms of fmincon are able to find an answer, and others are not. It can not be concluded in
general which algorithm is best for a certain objective function. For the first problem presented in the results
chapter, where a fixed gear configuration is used to find the optimal gear location, the four different algo-
rithms were tried in the system-level optimizer. The results of that investigation are presented in table C.1.
Where all algorithms found the same optimum, the time needed for it varied significantly. No explanation
can be given for the fact that the active-set algorithm performs best.
The trust-region-reflective algorithm could not be used, as the gradient of the objective function needs to be
specified for this algorithm. The gradient is not available for this problem.

C.3. S TRUCTURE OF THE SOFTWARE FOLDER

The folder Software contains 7 sub-folders: CLI_tools, Data, MATLAB, Parts_Catia, Parts_VLM, READ_NOTES_IN_THIS_FOLDER,
Results and SubMechs. When operating the software, this folder must be placed in a location that has no
spaces in its pathname, or error will occur. The standard local folder My Documents, should thus be avoided.
This section describes the files contained in each folder, while sections C.4 and C.5 describe the written MAT-
LAB files in some detail.
The sub-folder CLI_tools contains a windows batch file that allows the VLM model to be ran fully automatic.
The batch file opens VLM, initiates the model, runs the load cases that are specified in the solution manager
of VLM, and closes VLM again. This batch file can be opened directly from MATLAB, once the correct path is
specified.
The folder Data contains two Excel sheets that are needed to set up the VLM model. The file BASE_geometryV4_Original
contains the initial coordinates of all hardpoints. That file is never edited during an optimization run, to
keep track of the initial design. The file BASE_geometry is changed during each iteration. The values from
BASE_geometryV4_Original are obtained, the global optimizer of MATLAB modifies them, and writes them
in the file BASE_geometry. From that file, the VLM model is initiated.
The folder MATLAB has many files. These are discussed separately in the next sections.
The folder Parts_VLM contains three CATIA parts: two related to the definition of the global coordinate sys-
tem and one that defines the interface points between the airframe structure and the landing gear.
The folder READ_NOTES_IN_THIS_FOLDER contains two m-files that should be placed in the installation
folder of MATLAB. The files are self-defined classes which are needed to read the VLM results and convert it
to a MATLAB structure. If they are placed in the MATLAB folder however, they are somehow not found; an
error is thrown when running the code. Placing them in the installation folder of MATLAB solves this prob-
lem. The folder contains the results from the VLM model. It has a separate sub-folder for each loadcase. All
sub-folders are investigated by the function InvestigateVLMResultsfile, which selects the wanted results
and creates a table from it. This table is available for further analysis, and is also written in an Excel file: Stat-
icLoadCase.xlsx. The results table has around 110 columns.
The first few contain the load case number (1 column), the length of the shock absorber (1 column), the x,y
and z coordinates of points A and D for the left and the right gear (3x2x2=12 columns) and the input forces on
the nose gear, the left gear and the right gear (6x3=18 columns). The VLM model allows the contraction and
extension of the shock absorber. With the output in the second column it can be checked if that is done cor-
rectly; the next 12 columns allow the calculation of the orientation of the trailing arm in the global reference
frame. For other components this is done from the input data, but the trailing arm moves when the shock
66 C. D ETAILED DESCRIPTION OF LIT SOFTWARE

absorber extends or retracts. Therefore, its coordinates are output data.

The other columns define the output loads. In the first 6, the forces that are needed to fix the CG are given.
After that, the loads in point A to J follow; first for the left gear and then for the right gear.
The folder SubMechs contains the parts definition of the VLM model. The right gear is mirrored from the left
gear, by making all y coordinates postive. Each part is simply defined as a line between two hardpoints. The
definition of these hardpoints is contained in the Data folder; they are linked to each other.

C.4. I NPUT FILES

The MATLAB folder contains two more input files: inputGlobal7000.xslx and inputSizes.xslx. The first con-
tains general aircraft data such as the maximum allowable roll and pitch angle, tyre geometry, tail height
and wingspan. It also defines the different aircraft weights and their corresponding forward and aft CG po-
sitions. As a third set, this file defines the initial, minimum and maximum global strut positions. The file
inputSizes.xslx contains initial sizing parameters for the different LG components and specifies upper and
lower bounds for these design variables. This file has three sheets. The first contains input data for cylin-
drical components, the second contains data for H-beams, and the third specifies some parameters for the
shock absorber.

C.5. MATLAB FILES

Figures 7.2 and 7.3 list eight MATLAB classes. Each of these classes is defined in one MATLAB file, that has
the same name as its class. With eight class files and 11 MATLAB scripts, this leaves three scripts. The first of
these is runThisFile.m, which should be ran to execute the tool and get an answer. It is explained in more
detail in section C.5.3. The function computeBU estimates the ultimate bending stress of a cylindrical mem-
ber according to figure 6.3. It is needed in several files and therefore placed apart. The function initialGeom
is placed apart for the same reason. This function contains the hardpoint coordinates. For the static shock
absorber position this data is contained in the Data folder; for the extended shock absorber position, it is
found in the Results folder.
The other files can be divided into two groups: one with StructuralMember as its parent and one with
LGSystem as its parent. The connection between the two is that all components of the POC gear are modelled
as some StructuralMember.

C.5.1. MATLAB FILES FOR THE LANDING GEAR

This section describes all files that inherit from the class LGSystem. A short description of what these files
do was already given in section 7.2, while the broader set-up of these files was described in section 2.1 This
section is more elaborate than the main body of this thesis.
A sub-folder BlockA is present in the Matlab folder. Its files are of similar set-up as the main files, only the
analytical model is used to find the forces in the hardpoints and not the VLM model. These files will therefore
not be described in any more detail.

LGS YSTEM . M
The parent MATLAB file is LGSystem.m. It has a method to calculate the ground reaction forces acting on a
LG strut: deriveStrutLoads. The position of the gears, the gear layout, and the certification type are needed
as input. Besides these parameters, the aircraft parameters are gathered and some constants are defined.
Regarding certification type, only CS-25 is fully tested at the moment but the file is set up to allow evaluation
of CS-23, CS-27 and/or CS-29 as well, however.
The helper function strutLoadsTricycle calculates the loads on the gear struts, depending on the position
of these struts. As described in chapter 4, all load cases should be considered, as it is not known beforehand
which load cases are critical. The outcomes of this function are 4 MATLAB structures: mleft, mright, n and
cg for the loads acting at the left main gear, the right main gear, the nose gear strut and the CG, respectively.
Each structure contains 41 fields: the different load cases described in chapter 4 and Appendix B. The field
m.Taildown for example contains the load cases for the tail-down load case of section B.2. Each field is a
structure itself, containing the 10 sub load cases that were explained in section 4.3. How these 10 sub load
cases are obtained, is explained below.
Each field has again 2 sub-fields: aft and fwd. Thus, taking the field m.Taildown again as an example, it
C.5. MATLAB FILES 67

contains 2 subfields: m.Taildown.Aft and m.Taildown.Fwd. These fields contain the loads for the most
forward and the most aft CG position. This is initiated with the first loop in the code, in the lines

cg = {'Aft' 'Fwd'};
for pos=1:2 % This loop is present such that every equation is evaluated at the
% most forward CG and the most aft CG
if pos==1
A = aAft; B = bAft; E = constantValues(8);
else
A = aFwd; B = bFwd; E = constantValues(9);
end

Finally, for every load case and for every CG position, 5 tyre conditions should be evaluated: the 50-50, 40-60,
60-40, 0-60 and 60-0 sub load cases as explained in section 4.3. As a field name in MATLAB cannot start with
a number, these have been stored in the field Sym, aSym1, aSym2, Flat1 and Flat2, respectively. For the
ground handling cases, the evaluation of these 5 tyre condition is done in two steps. Specific requirements
for the flat tyre load case are mentioned in CS25.511. These requirements are implemented in the function
strutLoadsTricycle, in the loop that is initiated with the following piece of code:

for cond=1:2 % This loop is present such that every equation is evaluated for flat tyres
% and for normal tyres.
if cond==1
mu=factors.mu(1);
tow=factors.tow(1);
reversed=factors.reversed(1);
%disp('Below results are for all inflated tyres');
else
mu=factors.mu(2);
tow=factors.tow(2);
reversed=factors.reversed(2);
%disp('Below results are for 1 flat tyre');
end

This results in two sub load cases: the 50-50 and the 0-60 load case. The remaining three load cases are com-
puted in the helper function constructLoadCasesGR. An example input of this function is m.Static.Aft,
containing both the 50-50 and the 0-60 sub load case. Per main load case, there are now ten different fields
with loads. For the example of m.Static, these are:
• m.Static.Fwd.Sym • m.Static.Aft.Sym

• m.Static.Fwd.aSym1 • m.Static.Aft.aSym1

• m.Static.Fwd.aSym2 • m.Static.Aft.aSym2

• m.Static.Fwd.Flat1 • m.Static.Aft.Flat1

• m.Static.Fwd.Flat2 • m.Static.Aft.Flat2
A similar thing happens for the touchdown load cases, but these load cases do not have a separate computa-
tion for the flat tyre conditions in the helper function strutLoadsTricycle. This is all done in the function
constructLoadCasesTD. The sub-fields of the field m.Taildown are the same as those of m.Static. In total,
410 load cases are now considered.
It is assumed in the function strutLoadsTricycle that the loads act in the middle of the wheel axle. For the
asymmetric and flat load cases, this results in a moment in that point. This is accounted for in the functions
constructLoadCasesTD and constructLoadCasesGR.

LGS TRUT. M
LGStrut is a child of LGSystem and thus inherits its properties and functions. In LGStrut, the loads at dif-
ferent hardpoints of the landing gear are computed, based on the loads per strut. This is done via a call to the
VLM program.
The VLM program is set up via the file BASE_geometry, so this file is overwritten at every iteration. The initial
configuration is contained in another excel file, which is opened and read in the function LGStrut.m. The
values of the design vector are then added to the coordinates and the resulting coordinates are written in the
file BASE_geometry.
68 C. D ETAILED DESCRIPTION OF LIT SOFTWARE

After running the VLM model (this requires browsing through several folders; this is all hardcoded so do not
change the folder structure!), the forces need to be rotated to their part-local axis system. That is done with
standard transformation matrices in the helper functions transformY and transformZ. The loads in point F
are also obtained.
These loads in the part-local axis system are then used to do the stress calculations in TAGear.

TAG EAR . M
In TAGear, all components of the POC gear are modelled. Since several members are of the same type, a class
has been created for all main structural components; see section C.5.2. It is specified for every component
which class they should be. This is done in the following MATLAB lines. The trailing arm, for example, can
only be of type Cylinder.

properties
trailingArm@Cylinder;
wheelAxle@Cylinder;
mainFitting@Cylinder;
retractActuator@Cylinder;
shockAbsorber@Cylinder;
foldingSideBraceUp@IBeam;
foldingSideBraceLow@IBeam;
trunnion@Socket;
pistonSA@ShockAbsorber;
orientation
lenghtmembers
end

The component sizing is done based on the loads in the hardpoints that can be calculated in the parent class
TAGear. As every component is a class of itself, it has an optimization routine. The only needed input for
this routine are the loads, the material, the initial geometry, and the boundaries for the geometry. This sizing
function will be described when the different classes are explained.

C.5.2. MATLAB FILES FOR STRUCTURAL MEMBERS

To find the optimal properties of each gear component, different structural member classes are defined. The
properties of these structural members are the subjects of the following section.
The (component-level) optimization function follows a similar set-up for all classes. Every sizing function
has three nested functions: one to obtain the part weight, one to evaluate the constraints, and one to perform
calculations. In the sizing functions, first some properties that are needed in the different nested functions
are initialized. Then the initial values of the design vector are read, as well as the boundaries. This is enough
to set the fmincon routine up.
The main calculations are performed in the nested function computeall 25 . For a particular set of design
variables, this function calculates the margin of safety of the part; the critical load case and the weight. It uses
the equations of chapter 6 to do so. This calculation is performed once per iteration. The weight is needed
in the objective function; the margin of safety is needed in the constraints function. The critical load case is
wanted as output.

S TRUCTURAL M EMBER . M
There are three principle structural members in the landing gear: cylinders, H-beams, and a socket. One
could argue that a socket is a special type of a cylinder, but as it requires completely different calculations
(see chapter 6), it has been modelled as a different part.
All structural members share some properties, and these are stored in the class StructuralMember. All
structural members need material properties such as density, yield strength, etc. Three different materials
are modelled: 300M steel that is often used in landing gear parts, a more standard steel and an often-used
aluminium alloy. Furthermore, all structural members will have a length, an initial geometry, a final ge-
ometry, a margin of safety, and upper and lower bounds for its design variables. This is defined in the file
StructuralMember.m

25 Sometimes this function has a different name, but it always starts with compute
C.5. MATLAB FILES 69

C YLINDER . M
The cylinder is a child of StructuralMember and inherits its properties. Additional properties are the loads
acting on it, the hydraulic pressure inside it and the name of the member.
Based on the loads acting on a cylinder and its geometry, a sizing can be performed. This is done according
to 3 different routines: weightExtForces, weightPresHydraulic and weightVertForce. Depending on
the loads acting on the member, the relevant function is called.
If there are forces acting in the local x, y and z direction, the sizing method described in section 6.2 is used
in the function weightExtForces, following a component-level optimization. Based on the geometry of the
cylinder, the margin of safety is computed for all 410 load cases. The critical load case, the one resulting in the
lowest margin of safety, is remembered. The geometry resulting in the lowest weight (given the constraints
and the upper and lower boundaries) is the output of this function.
If there are hydraulic forces, a similar optimization is performed in the function weightPresHydraulic.
This optimization is based on the physics described in section 6.3. One difference is that there is no critical
load case. The sizing is purely done based on the pressure of the hydraulic fluid inside the component. This
pressure is depending on the aircraft weight (for the shock absorber) and the LG weight (for the RA).
For the shock absorber and the retract actuator, it is assumed that there is a compressive axial force acting
on them. Thus, the buckling criteria (described in section 6.1 for an I-beam, but similarly applicable to a
cylinder) should be checked. That is done in the function weightVertForce.
Depending on the acting forces, the relevant functions are computed. The main fitting, for example, does not
have any hydraulic fluids inside it, so weightPresHydraulic is not evaluated for this part. If the sizing is
done for two different sets of forces, the heaviest result is selected.
Specific constraints are that the computed margin of safety must be larger the specified MoS and that the
outer radius of the member must be larger than the inner radius, with a minimum wall thickness of 2mm.

HB EAM . M
The set-up of HBeam is similar to Cylinder. It has only one input: the axial force on the member. Other
properties are inherited from StructuralMember. As the force on an H-Beam is always axial, there is only
one sizing function: findWeight. It is based on the physics described in section 6.1.
Specific constraints are that the outer width must be larger than the inner width, that the web thickness must
be smaller than the beam height, the that cut-out height cannot be larger than the beam height - with a
margin of 4mm on both sides, and that the cut-out thickness cannot be larger than the web thickness - with
a margin of 2mm. Again, the computed margin of safety must be larger or equal to the specified MoS.

S OCKET. M
The file socket.m again has a similar set-up as the above two classes. Inputs include the acting loads, the
length of the pins inside the socket and the geometry of the socket. The other design variables are inherited
from StructuralMember.
The equations used to size a socket were given in section 6.4 and are implemented in this file. Again, a
component-level constraint is that the computed margin of safety must be larger or equal to the computed
margin of safety. Additionally, the thickness of the socket must be positive.

S HOCK A BSORBER . M
Sizing of the shock absorber is based on the method described in Currey [4]. First, the needed piston diameter
is computed. Design variables for this calculation are the shock absorber efficiency, the tyre efficiency, the
load factor, the maximum vertical touchdown velocity, the tyre stroke, the static pressure, the pressure when
the shock absorber is fully extended and the pressure when the shock absorber is fully compressed.
As the shock absorber is a child of the class Cylinder, the thickness of the shock absorber piston can be
computed with the function findWeight that is specified for this parent class.

C.5.3. RUNFILE
In the runfile, the first step is to find the optimal location of the gears under the aircraft. This is done with
the analytical model from the folder BlockA. The coordinates of the optimal location are saved in the file
BASE_geometry.xlsx.
The next step is to vary the gear layout. To do so, an object of TAGear is initiated and a design vector is defined,
as well as upper and lower boundaries.
The boundaries are specified in such a way, that no global constraints are needed. There is thus no constraint
70 C. D ETAILED DESCRIPTION OF LIT SOFTWARE

function.
An additional function outfun ensures that there is some real-time information while the optimizer runs. A
figure is created that plots the current values of the design vector, the current value of the objective function
and the maximum constraint violation.
 B IBLIOGRAPHY
[1] S. Chai and W. Mason, Landing Gear Integration in Aircraft Conceptual Design, Tech. Rep. MAD 96-09-01
(NASA Ames Research Centre, 1997).

[2] N. Heerens, Landing gear design in an automated design environment, Master’s thesis, Delft University
of Technology (2014).

[3] S. Cumnuantip, M. Spieck, and W. Krueger, eds., An approach for sizing and topology optimization in-
tegrating multibody simulation (46th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics &
Materials Conference, 2005).

[4] N. Curry, Aircraft Landing Gear Design: Principles and Practices, 1st ed. (American Institute of Aeronau-
tics and Astronautics, 1988).

[5] J. Roskam, Airplane Design part IV: Layout design of landing gear and systems, 2nd ed. (DARcorporation,
1989).

[6] Kable Intelligence Limited, http://www.army-technology.com/projects/lynx-mk7/lynx-mk72.

htmL (2015), accessed 21-apr-2015.

[7] Fokker B.V, http://www.fokker.com/Landing_Gear_Programs (2015), accessed 21-apr-2015.

[8] The Aviation History Online Museum, http://www.aviation-history.com/boeing/b47.html

(2015), accessed 21-apr-2015.

[9] J. Heeren, Landing gear design handbook for engineering (Fokker Landing Gear B.V., 1997).

[10] P. Papalambros and D. Wilde, Principles of optimal design: modeling and computation (Cambridge uni-
versity press, 2000).

[11] P. E. Gill, W. Murray, and M. H. Wright, Practical optimization, Vol. 5 (Academic press London, 1981).

[12] C. B. Chapman and M. Pinfold, Design engineering—a need to rethink the solution using knowledge based
engineering, Knowledge-based systems 12, 257 (1999).

[13] T. van den Berg, Harnessing the potential of Knowledge Based Engineering in manufacturing design, Ph.D.
thesis, Delft University of Technology (2013).

[14] S. Cooper, I.-s. Fan, and G. Li, Achieving competitive advantage through knowledge-based engineering:
a best practice guide (Prepared for the Dept. of Trade and Industry by Dept. of Enterprise Integration,
Cranfield University, 1999).

[15] M. Stokes, M. Consortium, et al., Managing engineering knowledge: MOKA: methodology for knowledge
based engineering applications (Professional Engineering Publ., 2001).

[16] G. la Rocca, Knowledge based engineering: Between ai and cad. review of a language based technology to
support engineering design, Advanced Engineering Informatics 26(2), 159–179 (2012).

[17] S. A., J. Nayfeh, and R. Zarda, An integrated design and optimization environment for industrial large
scaled systems, Research in Engineering design 16, 86 (2005).

[18] S. Yi, J. Shin, and G. Park, Comparison of mdo methods with mathematical examples, Structural and
Multidisciplinary Optimization 35, 391 (2008).

[19] N. Tedford and M. J.R.R.A., eds., On the common structure of MDO problems: a comparison of architec-
tures (11th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, 2006).

71
72 B IBLIOGRAPHY

[20] A. Keane and P. Nair, Computational approaches for aerospace design: the pursuit of excellence (John
Wiley & Sons, 2005).

[21] J. Martins and C. Marriage, eds., An object-oriented framework for multidisciplinary design optimization
(3rd AIAA Multidisciplinary Design Optimization Specialist Conferernce, 2007).

[22] I. Kroo, Mdo for large-scale design, in Multidisciplinary design optimization: state-of-the-art, ICASE/-
NASA Lanley Workshop on Multidisciplinary Design Optimization (SIAM, 1997) pp. 22–44.

[23] K. Deb, Multi-objective optimization, in Search methodologies (Springer, 2014) pp. 403–449.

[24] M. v. Tooren, G. L. Rocca, and T. Chiciudean, Advance Design Methodologies Course Book (TU Delft,
2013).

[25] A. Myklebust, S. Jayaram, and P. Gelhausen, Acsynt-a standards-based system for parametric computer
aided conceptual design of aircraft, AIAA Paper , 92.

[26] G. La Rocca, T. Langen, and B. Y.H.A., eds., The design and engineering engine. Towards a modular system
for collaborative aircraft design (28th international congress of the aeronautical sciences, 2012).

[27] S. Cumnuantip, ed., Landing gear conceptual design and structural optimization of a large blended wing
body civil transport aircraft (5th Challenges in European Aerospace Conference, 2015).

[28] E. Torenbeek, Synthesis of subsonic airplane design: an introduction to the preliminary design of subsonic
general aviation and transport aircraft, with emphasis on layout, aerodynamic design, propulsion and
performance (Springer Science & Business Media, 1982).

[29] Leibowitz, Bruce, http://http://flightlineaviationmedia.com/planespotting/

2-engines-tail/ (2015), accessed 13-jul-2015.

[30] F. A. Circular, 150/5300-13, Airport Design (1994).

[31] E. CS25, Certification specifications and acceptable means of compliance for large aeroplanes, Amend-
ment 15 (2014).

[32] E. CS23, Certification specifications for normal, utility, aerobatic, and commuter category aeroplanes,
Amendment 3 (2012).

[33] E. CS27, "certification specifications for small rotorcraft, Amendment 3 (2012).

[34] E. CS29, "certification specifications for large rotorcraft, Amendment 3 (2012).

[35] T. H. G. Megson, Aircraft structures for engineering students (Elsevier, 2012).

[36] E. F. Bruhn, R. Bollard, et al., Analysis and design of flight vehicle structures (Jacobs, 1973).

[37] Federal Aviation Authority, Metallic Materials Properties Development and Standardization (Batelle
Memorial Institute, 2013).

[38] C. Niu, Airframe structural design: practical design information and data on aircraft structures (Conmilit
Press, 1988).

[39] N. M. Alexandrov and S. Kodiyalam, Initial results of an mdo method evaluation study, AIAA paper 4884,
1998 (1998).

[40] J. Sobieszczanski-Sobieski, Optimization by decomposition: a step from hierarchic to non-hierarchic sys-

tems, Recent Advances in Multidisciplinary Analysis and Optimization , 51 (1988).

[41] J. Sobieszczanski-Sobieski, J. S. Agte, and R. R. Sandusky, Bilevel integrated system synthesis, AIAA journal
38, 164 (2000).

[42] G. Kempf, ed., Development of undercarriage design loads (Advisory Group for Aerospace Research and
Development, 1991).
B IBLIOGRAPHY 73

[43] D. Thorby, J. Johnson, A. Auld, H. Newman, and M. Brooker, eds., The Special Requirements of a VSTOL
Aircraft (Advisory Group for Aerospace Research and Development, 1991).

[44] R. Yntema and B. Milwitzky, An impulse-momentum method for calculating landing-gear contact condi-
tions in eccentric landings, (1952).

[45] W. Krueger, Integrated Design Process for the Development of Semi-Active Landing Gear for Transport
Aircraft, Ph.D. thesis, University of Stuttgart (2000).

[46] W. Krueger and M. Spieck, Interdisciplinary landing gear layout for large transport aircraft, (1998).

[47] N. standardization Agency, "standardization agreement 4674 - unmanned aerial vehicles systems airwor-
thiness requirements (usar), Amendment 1 (2009).

[48] Federal Aviation Authority, Taxi, takeoff and landing roll design loads.

[49] D. Freund, D. R. McKissack, L. C. Hanson, and H. Brodman, Dynamic taxi, take-off and landing roll
analyses for large business jet aircraft, AIAA paper 1526 (2000).