;

Semi-Analytical
Composite Oval
Fuselage Mass
Estimation
M.N. Roelofs
Registration number: 082#16#MT#FPP
5 July 2016
ii

Source cover image: Airbus

Semi-Analytical
Composite Oval
Fuselage Mass
Estimation
by

M.N. Roelofs
to obtain the degree of Master of Science
at the Delft University of Technology,
to be defended publicly on Tuesday August 9, 2016 at 09:30 AM.

Student number: 4077407

Project duration: August 1, 2015 – July 1, 2016
Thesis committee: Prof. dr. ir. L. L. M. Veldhuis, TU Delft, chair
Dr. ir. R. Vos, TU Delft, supervisor
Dr. ir. C. Kassapoglou, TU Delft

Registration number: 082#16#MT#FPP

Date: 5 July 2016

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

 Summary
The goal for this thesis was to develop a semi-analytical weight estimation method for pressurized,
composite, oval fuselages. This method was implemented into the FPP Initiator, written in Matlab. In
doing so, the oval fuselage concept can be investigated in conceptual aircraft design of unconventional
configurations, such as the blended wing body. These type of aircraft require a non-circular fuselage
with an aerodynamic shape, for which the oval fuselage is one of few available concepts. With the
help of this thesis work, composite oval fuselage technology may be investigated, following the current
trend in aerospace industry towards composite materials.
The fuselage was modeled as a 1-dimensional beam to compute the internal loads. Application of the
boom method provides the method with internal stresses and shear flow. This method was chosen
not only because it can capture any beam geometry (arbitrary cross-section and taper along the beam
axis), but also because it incorporates the material Young modulus, enabling both isotropic (metal) and
orthotropic (fiber composite) materials.
Using the calculated loads, the structural components are sized: the skin, stringers and sandwich
panels in the trapezoidal structure. An idealized structure was assumed, where the skin carries the
membrane loads due to pressurization and the shear flow, while stringers carry direct stresses resulting
from the bending moments. Finally, sandwich panels carry all loads in the face sheets, while the core
adds bending stiffness.
By assuming initial dimensions for the structural components, consecutively computing loads and then
sizing these components to get more accurate dimensions and repeating this process until it converges
upon the component dimensions the weight estimation method is able to provide a consistently sized
fuselage. This was done such that such a convergence loop does not have to be performed outside of
the FWE.
Three separate optimization methods were developed based on previous work to size the skin panels,
stringers and sandwich panels based on applied loads. This deterministic optimization approach builds
a laminate stacking sequence based on a most critical failure mode. In case of the stringers, the
member widths are determined using an additional optimization loop, similar to how the core thickness
is optimized for sandwich panels. Even though this approach proved to be much faster than a full-blown
optimization, it was found to be too computationally expensive for in-the-loop use.
As a solution to the aforementioned problem, a surrogate modeling approach was adopted. Neural
networks proved to be most capable of representing the complex responses obtained from a design of
experiments and offer an acceptable in-the-loop evaluation time.
One of the main challenges with non-circular fuselages like the oval fuselage is pressurization. The
current method estimates pressurization loads as if each fuselage section were a straight tube, such
that the analytical solution for a circular cylinder may be used in each arc. The mismatch of hoop loads
is carried by the trapezoidal structure. An attempt was made to apply a finite difference method to
compute the membrane loads in an arbitrary shell, but although the method was capable of computing
the hoop load correctly, the longitudinal load was approximated incorrectly and therefore the method
was abandoned. It still remains to be seen whether the 3-dimensional shape of the oval fuselage is
capable of withstanding pressurization loads without deformation.
Validation of the panel failure analysis showed good agreement with finite element analysis or a conser-
vative behavior of the analytical method. The Tsai-Wu failure criterion, however, was underestimated
by around 20 %, but the impact of this on the overall mass estimation is small. Comparison with genetic
algorithms showed the optimization methods implemented are capable of finding optimal designs, even
outperforming the genetic algorithms for stringers and sandwich panels.
Finally, several case studies were performed. Moving from an aluminium design to a composite fuse-
lage resulted in a fuselage mass reduction of on average 15 %. A composite oval fuselage was even
lighter than its aluminium design by 32 %. Additionally, the effect of using stiffened panels instead of
sandwich panels in the trapezoid structure was studied, which showed a significant mass penalty is
associated with this.

iii
 Preface
This report is written as partial fulfillment of the Master’s degree in Aerospace Engineering at Delft Uni-
versity of Technology. It concludes my thesis research and thereby my entire studies. Aircraft design
is one of my great interests and with this thesis I hope to have contributed to the body of knowledge in
novel aircraft design. Through the curriculum and extra-curricular activities like the AIAA Aircraft De-
sign Competition and the DUT Racing Team I feel my time here in Delft has been fruitful and prepared
me well for whatever lies ahead.
First and foremost, I would like to thank my supervisor dr.ir. Roelof Vos for his critical attitude towards
my work and results. Answering his in-depth and to-the-point questions I often reconsidered my solu-
tions and improved them. His ideas and advice have proven invaluable for the outcome of this project.
Furthermore, I would like to thank dr.ir. Christos Kassapoglou for his advice on sizing composite struc-
tures and considerations on composite fuselage design. Additionally, I want to thank ir. Imco van Gent
for helping me out with the program he wrote for his thesis work, which became a pivotal part of my
own research. I would also like to thank my committee: prof.dr.ir. Leo Veldhuis, dr.ir. Roelof Vos and
dr.ir. Christos Kassapoglou.
I would also like to thank my fellow students in ”Kamertje 1”, who made working during this period a
lot more enjoyable. The discussions about the Initiator and on aircraft design in general were very
interesting indeed. My final expression of gratitude is for my family and friends. Especially, I want to
thank my parents, for their continuous support in all of my endeavours.

M.N. Roelofs
Delft, June 2016

v
 Contents
Summary iii
List of Figures ix
List of Tables xi
Nomenclature xi
1 Introduction 1
1.1 Thesis Goal and Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Description of Oval Fuselage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Inside-out Approach for Sizing Oval Fuselage Outer Geometry . . . . . . . . . . . 3
1.3 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Structural Design and Analysis 7
2.1 Initiator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Program Flow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Shear Force and Bending Moment Distribution . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Load Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5 Boom Method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.6 Pressurization Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.7 Structure Sizing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.8 Mass Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Sizing Methods for Composite Structure 17
3.1 Classical Lamination Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.1 Stiffness. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1.2 Strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Skin Panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2.1 Shear Buckling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.2 Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Stringers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3.1 Crippling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3.2 Column Buckling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4 Sandwich Panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4.1 Buckling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4.2 Wrinkling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4.3 Crimping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.5 Sizing Routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.5.1 Monolithic Plate Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.5.2 Stringer Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.5.3 Sandwich Panel Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.6 Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.7 Stiffened Trapezoid. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4 Regression Analysis 33
4.1 Linear Least Squares Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1.1 Skin Panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.2 Stringers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.2 Sandwich panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

vii
viii Contents

5 Pressure Loading 43
5.1 Shell of Arbitrary Shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.2 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.3 Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.4 Effect of Oval Shape in Three Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6 Verification & Validation 47
6.1 Applied Loads Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 Pressure Load Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.3 Boom Method Verification and Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.4 Panel Failure Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.5 Verification of Optimization using GA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.6 Mass Estimation Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.7 Sizing Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
7 Results 59
7.1 Composite A320-200 and Boeing 767-300ER . . . . . . . . . . . . . . . . . . . . . . . . 59
7.2 Oval Composite A320-200 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
7.3 Wide Oval Boeing 767-300ER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
7.4 Stiffened Trapezoid versus Sandwich Trapezoid . . . . . . . . . . . . . . . . . . . . . . . 66
8 Conclusions and Recommendations 69
8.1 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
A Operation Manual 71
A.1 Initiator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
A.2 Design of Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
A.3 Creating Surrogate Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
B Explanation of Messages 75
B.1 Fuselage Weight Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
B.2 Panel Database Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
B.3 Stringer Database Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
B.4 Sandwich Database Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
C Fuselage Weight Estimation Settings 77
Bibliography 79
 List of Figures

1.1 Membrane and shell loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Three concepts for non-circular, pressurized fuselages . . . . . . . . . . . . . . . . . . . 2
1.3 Parametric description of oval cross-section . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1 Initiator process flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Fuselage weight estimation process flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Structural idealization of oval fuselage . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Closed beam section for determining 𝑞 , . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 Pressurization forces FBD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 Weight breakdowns for side and keel panels . . . . . . . . . . . . . . . . . . . . . . . . 16

3.1 Forces acting on a panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3.2 Stringer cross-section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Input and output for sizing procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4 Optimization process for composite panels . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Ply addition for composite panels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.6 Optimization process for composite stringers . . . . . . . . . . . . . . . . . . . . . . . . 28
3.7 Ply addition and core thickness for composite sandwich panels . . . . . . . . . . . . . . 29
3.8 Frame section for elastic center method . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.9 Frame cross-section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.1 Models predicting weight of monolithic panel . . . . . . . . . . . . . . . . . . . . . . . . 37

4.2 Models predicting thickness of monolithic panel . . . . . . . . . . . . . . . . . . . . . . . 37
4.3 Models predicting E-modulus of monolithic panel . . . . . . . . . . . . . . . . . . . . . . 38
4.4 Models predicting weight of C-shape stringer . . . . . . . . . . . . . . . . . . . . . . . . 38
4.5 Models predicting axial stiffness of C-shape stringer . . . . . . . . . . . . . . . . . . . . 39
4.6 Models predicting cross-sectional area of C-shape stringer . . . . . . . . . . . . . . . . 39
4.7 Models predicting weight of sandwich panel . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.8 Models predicting 𝑡 of sandwich panel . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.9 Models predicting 𝑡 of sandwich panel . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.10 Models predicting 𝐸 of sandwich panel . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

5.1 Arbitrary shell element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

5.2 Grid for finite difference method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
5.3 Pressure loads in circular cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.4 Pressure loads in tapered cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

6.1 A320 internal loads from literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

6.2 A320 internal loads from Initiator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.3 Pressurization loads from Abaqus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.4 Direct stress comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.5 Shear stress comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.6 Validation results for panel shear buckling . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.7 Panel pressure deflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.8 Genetic algorithm versus deterministic approach for skin panels . . . . . . . . . . . . . 54
6.9 GA versus ELFO weight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.10 Genetic algorithm versus deterministic approach for sandwich panels . . . . . . . . . . 55
6.11 Generic weight breakdown for large passenger aircraft . . . . . . . . . . . . . . . . . . . 56
6.12 Weight breakdowns from Initiator for aluminium fuselage . . . . . . . . . . . . . . . . . . 57

ix
x List of Figures

6.13 Thickness distributions for metal fuselages from Initiator . . . . . . . . . . . . . . . . . . 58

7.1 Composite Airbus A320-200 thickness distribution . . . . . . . . . . . . . . . . . . . . . 60

7.2 Composite Boeing 767-300ER thickness distribution . . . . . . . . . . . . . . . . . . . . 61
7.3 Weight breakdowns for composite fuselages . . . . . . . . . . . . . . . . . . . . . . . . 61
7.4 Oval, composite Airbus A320-200 thickness distribution . . . . . . . . . . . . . . . . . . 62
7.5 Oval, composite Airbus A320-200 weight breakdown . . . . . . . . . . . . . . . . . . . . 63
7.6 Oval, aluminium Boeing 767-300ER thickness distribution . . . . . . . . . . . . . . . . . 64
7.7 Oval, composite Boeing 767-300ER thickness distribution . . . . . . . . . . . . . . . . . 65
7.8 Weight breakdowns for oval Boeing 767-300ER fuselages . . . . . . . . . . . . . . . . . 65
7.9 Stiffened trapezoid, oval Airbus A320-200 thickness distribution . . . . . . . . . . . . . . 66
7.10 Stiffened trapezoid, oval Boeing 767-300ER thickness distribution . . . . . . . . . . . . 67
7.11 Weight breakdowns for oval composite fuselages with stiffened trapezoid . . . . . . . . 67
 List of Tables

2.1 Design vector for weight estimation method . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Load types per component . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Load cases for fuselage structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4.1 Input values for structural components . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.2 Material properties used for surrogate models . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Regression models for monolithic panel . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4 Regression models for C-shape stringer . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.5 Regression models for sandwich panel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

6.1 Comparison of FEA and analytical hoop pressure stress . . . . . . . . . . . . . . . . . . 48

6.2 Comparison of Tsai-Wu failure criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.3 Estimated Fuselage Mass Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

7.1 Estimated Composite Fuselage Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

xi
 Nomenclature
Abbreviations
Abbreviation Definition
ACT NASA Advanced Composite Technology
ALGO Additive Lay-up Generation Optimizer
ATCAS Boeing’s Advanced Technology Composite Aircraft
Structure
BWB Blended Wing Body
CLT Classical Lamination Theory
ELFO Enumerative Lay-up Family Optimizer
FBD Free Body Diagram
FEA Finite Element Analysis
FF Fiber Failure
FPP Flight Performance and Propulsion
IFF Inter Fiber Failure
MAC Mean Aerodynamic Chord
MTOM Maximum Take-Off Mass
MZFW Maximum Zero Fuel Mass
NEF No Edge Free
OEF One Edge Free
OEI One Engine Inoperative
OEM Operational Empty Mass
PDE Partial Differential Equation
RMSE Root Mean Squared Error
Greek Symbols
Symbol Description Units
𝛽 Angle of wall with vertical rad
𝛿 Deflection m
𝜈 Poisson ratio -
𝜌 Density kg/m
𝜎 Direct stress N/m
𝜏 Shear stress N/m

xiii
xiv Nomenclature

Roman Symbols
Symbol Description Units

𝐴 Area (of cross-section) m

𝑎 Length (of panel) m

1 1
𝑎 Gust decay fts ]ft

𝐴𝑅 Aspect ratio -

𝐵 Boom area m

𝑏 Width (of panel) m

𝑐 Centroid coordinate m

𝑐 Chord length m

𝐶 Lift curve slope rad

𝐸 Young’s modulus N/m

𝐺 Shear modulus N/m

𝑔 Gravitational acceleration m/s

ℎ Height m

𝐼 Second moment of area m

𝐼 Weighted second moment of area Nm

𝐿 Length m

𝑀 Bending moment Nm

𝑁 Line load N/m

𝑛 Load factor -

𝑃 Boom load N

𝑝 Pressure N/m

𝑞 Shear flow N/m

𝑟 Radius m

𝑆 Shear force N

𝑆 Shear strength N/m

𝑆 Surface area m

𝑡 Thickness m

𝑉 Gust velocity ft/s

𝑊 Weight N

𝑤 Width m

𝑋 Longitudinal strength N/m

𝑌 Transverse strength N/m

Nomenclature xv

Subscripts & Superscripts

Symbol Definition
∘
Middle surface of laminate

𝐵 Boom
𝑐 Compression

𝑐 Core
𝑓 Facing

𝑡 Tension

1,2,3 Top arc (1), side arc (2) and bottom arc (3)
buck Buckling

crimp Crimping
crit Critical

crp Crippling
c Ceiling (of trapezoid)

eq Equivalent
fr Frame

f Floor (of trapezoid)

g Gust

hoop Hoop direction

long Longitudinal direction

mat Material
pan Panel

req Required
s Symmetric

ult Ultimate
wr Wrinkling
 1
Introduction
The Blended Wing Body (BWB), aims to reduce fuel consumption and improve aerodynamic perfor-
mance [1]. The concept is one of many novel aircraft configurations devised to compete with and
replace the current conventional tube-and-wing configuration. Decades of improving the latter concept
have exhausted the possibilities for further major improvements and thus the aircraft industry is looking
for new options.
One of the challenges in designing a BWB aircraft is the fuselage, which becomes very wide and
often has an airfoil-like shape in the symmetry plane cross-section. The fuselage has to resist internal
pressure and spheres are the most optimal shape to resist the pressurization loads. Cylinders, as used
in conventional aircraft, are also very efficient pressure vessels. The reason for this is that a circular
shape is able to carry the pressure load in tension entirely; it acts as a membrane. A non-circular shape
will, in addition to the tensile internal loads, develop internal bending moments due to the pressurization
and as such will need more material to cope with these loads. Obviously, such a pressure vessel will
be much heavier. All loads acting on a shell element are shown in Figure 1.1. Membrane loads are
shown in red: 𝑁 , 𝑁 and 𝑁 . For a non-membrane element, the out of plane loads 𝑉 and 𝑉 develop,
shown in blue. Additionally, bending moments develop to resist out-of-plane loads, shown in green.
𝑉
𝑀
𝑁 𝑁
𝑉
𝑀
𝑀
𝑁 𝑀 𝑀 𝑁

𝑀 𝑀

𝑉 𝑁
𝑁
𝑉 𝑀

Figure 1.1: Membrane and shell loads

Several solutions have been proposed to deal with the problem of a non-cylindrical pressurized fuse-
lage. Most commonly, the multi-bubble (see Figure 1.2a) and stiffened-shell (see Figure 1.2b) concepts
are proposed [1–5]. However, a new concept was proposed at the TU Delft: the oval fuselage (see
Figure 1.2c). This concept, unlike the others, does not rely on members that have to be placed in the
cabin to carry the out-of-plane loads and thus results in an unobstructed cabin space. The idea of this
concept is that a cross-section is composed of four circular arcs, which are tangent at the connections.
To carry the difference in membrane forces in these arcs, a trapezoidal structure connects the four
intersections.

1
2 1. Introduction

(a) Multi-bubble concept (b) Integrated skin and shell con-

cept (c) Oval fuselage concept

Figure 1.2: Three concepts for non-circular, pressurized fuselages

Multi-bubble The multi-bubble concept, see Figure 1.2a, uses multiple cylindrical elements which
are linked together in either an open- or closed-cell configuration. Therefore, pressurization loads are
carried in hoop-tension and no bending loads due to pressurization occur [6]. According to Mukhopad-
hyay [3], the multi-bubble configuration appears to be twice as inefficient compared to a cylindrical
configuration, but it could reduce overall weight by about 20% to 30% compared to using flat surfaces.
Extensive research has been conducted by NASA on non-cylindrical fuselages, including composite
ones, and the multi-bubble appears in many of these studies, reinforcing its feasibility [3–5].

Skin and Shell The integrated skin and shell concept, see Figure 1.2b, uses either a deep honeycomb
or stiffened frame construction, following the outer aerodynamic contour of the aircraft, and was chosen
for the BWB study by Liebeck [1], but NASA has also given it attention in several studies into structural
concepts for non-circular fuselages [2, 7, 8]. Walls are present to carry part of the pressurization loads.
Flat shells as used in this concept are not suitable for carrying pressurization, but the simplicity of the
concept makes it attractive nonetheless. One could also replace the honeycomb with stiffening ribs,
which could decrease weight. The lightest configuration would be a double-skin vaulted ribbed-shell
concept [2].

Oval fuselage The oval fuselage, see Figure 1.2c, features an outer skin made from four circular
arcs, which are connected tangentially: a top arc, a bottom arc and two identical side arcs. The cabin
is defined by a trapezoidal structure of straight panels, and the corner points coincide with the arc joints.
As such, the outer skin carries the pressure loads in membrane action, while the trapezoidal structure
carries the remaining loads resulting from pressurization.
The oval fuselage concept was first presented by Hoogreef and Vos [9]. A weight estimation methodol-
ogy was developed that determines the structural members’ thicknesses, depending on pressurization
and wing-bending loads. The longitudinal stresses from pressurization, axial acceleration and fuselage
bending are checked to be below tensile and compressive fatigue limits. The method is combined with
the Torenbeek class II.5 weight estimation for other structural components and wings. It is concluded
that the oval fuselage has the advantage over the previously mentioned two concepts that the cabin
space is unobstructed. A 400 passenger oval-fuselage BWB was shown to have 13% lower OEM and
6% better fuel consumption in comparison to conventional airliners.
This research was carried on by Schmidt [10], where the parameterization was extended to allow for a
non-symmetrical airfoil shape along the fuselage centerline. The weight estimation developed uses the
plane stress assumption and sizes the fuselage based on several load conditions. The outer shell was
modeled as a stiffened skin, while the trapezoidal structure consisted of sandwich panels. Dimpling,
crimping, wrinkling and global buckling of these sandwich beams were taken into account. Inertial and
aerodynamic loads are obtained at each cross-section by representing the fuselage as a 1-dimensional
beam.
In order to validate the above weight estimation method, a finite element analysis was developed using
a knowledge-based engineering approach by De Smedt [11]. It was shown there was good correspon-
dence in lateral and hoop stresses, while longitudinal and shear loads showed significant differences.

Composite Fuselage In addition to the trend towards unconventional configurations, the aircraft in-
dustry is also moving towards new materials. The first, mostly composite, aircraft have already taken
1.1. Thesis Goal and Research Question 3

flight; the Airbus A350 and Boeing 787. Composites may increase lifetime and safety, while reducing
cost and weight, but companies are hesitant to apply them, since the advantages can not be guaran-
teed [12] and production facilities have to be completely modified [13]. Nonetheless, aircraft industry is
applying more and more composite parts in aircraft and it is not unlikely that novel aircraft will feature
major components made from fiber composites. It is therefore the main focus of the present work.

1.1. Thesis Goal and Research Question

Former research was conducted focusing on the weight estimation of the oval fuselage concept, made
with metals (usually aluminium). This semi-analytical method uses simple beam theory to compute
loads and analytical expressions for failure criteria of structural components in an effort to size these.
Building upon that research, this research will focus on a similar weight estimation method, but using
fiber composite material for the entire fuselage structure. To analyze the oval fuselage concept in un-
conventional aircraft, the method is implemented in the Initiator, a preliminary aircraft design tool in
development at the TU Delft. Concluding, the research goal can be defined as:

Develop a semi-analytical weight estimation method for a pressurized, composite, oval fuselage to be
implemented in the FPP Initiator by writing a physics based program in Matlab

The research itself focuses on how this goal can be achieved, i.e. what such a method requires and
which methods it needs to estimate relevant parameters. Moreover, the structural design of an oval
fuselage has a large influence on both the methods to be used and the eventual feasibility of the con-
cept. Hence, the research question is:

How can the weight of an oval, composite, pressurized fuselage be estimated in the conceptual design
phase of unconventional aircraft?

This research question is quite broad and has to be split into several sub-questions, listed below.
Obviously, more questions could have been made, because many different aspects come into play
for the weight estimation of a fuselage (load cases, crashworthiness and splicing, to name a few).
However, the three questions below are most important and are the focus of the present research.

• Which failure modes and corresponding equations should be used to size a composite stiffened
skin structure and sandwich structure depending on applied loads?

• How can the skin thickness, stacking sequence and number of layers be determined for a com-
posite laminate, within a time suitable for conceptual design?

• What is the effect of the oval fuselage shape in 3 dimensions on the pressurization loads?

1.2. Description of Oval Fuselage

One of the advantages of an oval fuselage over the multi-bubble or integrated skin and shell is an
unobstructed cabin space. This allows for more freedom in designing the cabin interior, which increases
the flexibility of such a fuselage.

1.2.1. Inside-out Approach for Sizing Oval Fuselage Outer Geometry

An oval cross-section as depicted in Figure 1.3 can be defined by four parameters: ℎ , ℎ , ℎ and
𝑤 . All other parameters can be derived from these four. From a specification of cabin layout (seating
configuration, class distribution and number of passengers), the floor contour can be defined. Secondly,
from cargo requirements an amount of cargo volume (and dimensions if containers are used) can be
deduced. Assuming the cargo hold is below the passenger floor, ℎ is defined in this way. The cabin
height ℎ is a designer’s choice, with a minimum specified by regulations. What remains to be defined
is ℎ . Using an optimization method, ℎ may be found that minimizes the cross-sectional area, while
having the cross-section passing through or around points that define the minimal cabin (floor width,
cabin height, seat height, etc.).
4 1. Introduction

ℎ
𝑤 𝑟

𝛼 2
𝑙
ℎ

𝛼 𝑤

ℎ 𝑟

Figure 1.3: Parametric description of oval cross-section

Defining:

𝑎=𝑤
𝑏 =ℎ ⋅ℎ +ℎ ⋅ℎ −𝑤
𝑐 = −ℎ ⋅ 𝑤 ⋅ (ℎ + ℎ )

the ceiling width 𝑤 is obtained using the quadratic formula:

−𝑏 ± √𝑏 − 4𝑎𝑐
𝑤 = (1.1)
2𝑎

One of the two solutions is either negative or close to zero, leaving only one possible value for 𝑤 .
By defining:

𝑤
𝜃 = tan
ℎ
𝑤
𝜃 = tan
ℎ
𝑤 −𝑤
𝛽 = tan
ℎ
𝛾 = 𝜋 − 2𝜃 − 𝛽
1.3. Outline 5

the radii of the circular arcs (and the length of the wall) may be determined as follows:
𝑤
𝑟 = (1.2)
2 sin 𝜃 cos 𝜃
ℎ
𝑙 = (1.3)
cos 𝛽
𝑙
𝑟 = (1.4)
2 cos 𝛾
𝑤
𝑟 = (1.5)
2 sin 𝜃 cos 𝜃

Now the angles that each of the circular arcs span can be computed as follows:
𝑤
𝛼 = sin (1.6)
𝑟
𝑤
𝛼 = sin (1.7)
𝑟
𝛼 =𝜋−𝛼 −𝛼 (1.8)

Finally, the locations of the centers of the circular arcs need to be determined. Logically, the lateral
positions of the centers of the top and bottom arc lie on the symmetry axis.

𝑐 , =0
𝑐 , =ℎ +ℎ −𝑟
𝑐 , = 𝑤 − 𝑟 sin(𝛾 − 𝛽)
𝑐 , = ℎ − 𝑟 cos(𝛾 − 𝛽)
𝑐 , =0
𝑐 , =𝑟 −ℎ

1.3. Outline
The structure of this report is as follows: chapter 2 discusses the structural analysis and gives an
overview of the entire weight estimation method. The sizing of composite structural components is
elaborated upon in chapter 3. Because the methods outlined in chapter 3 require too long an in-the-
loop computation time, surrogate modeling was employed, which is explained in chapter 4. Regarding
the third sub-question, chapter 5 focusses on the problem of pressurization in an oval fuselage. chap-
ter 6 presents the verification and validation procedures and results. Since both composite and oval
fuselage designs are hard to validate (no data is available), results from the presented method are
not included in the verification and validation chapter, but rather in the separate chapter 7 as case
studies. Finally, in chapter 8 the research question is answered and conclusions are drawn with re-
spect to the applicability and feasibility of the presented approach. Moreover, recommendations for
further research and improvements on the current work are given. Several appendices are included in
this report, being the operational manuals and other information required to run the developed weight
estimation method.
 2
Structural Design and Analysis
Mass estimation is in the strict sense only an analysis of a certain structure or geometry in general.
However, in the current work, only an outline of the fuselage shape is available, but the structural
dimensions have yet to be determined. Therefore, the mass estimation method also needs to size the
fuselage structure.
In this chapter an outline is provided for the entire mass estimation method developed in this thesis work.
First, its role in the Initiator aircraft design tool is described. Secondly, the program flow of the mass
estimation method itself is discussed. The components of this process are subject of later sections and
chapters. Thirdly, the methods to determine applied and internal loads are elaborated upon. Finally,
the mass estimation of secondary structure and non-structural components is described.

2.1. Initiator
As was specified in the thesis goal, the weight estimation method developed during this thesis should
be implemented in the FPP Initiator. The Initiator is a conceptual and preliminary aircraft design tool,
which aims to size and analyze conventional and unconventional aircraft based on a user specified
set of top level requirements and settings. The process flow of a design convergence loop is shown
in Figure 2.1, where it can be seen that the Initiator’s design process can be divided into three major
blocks: Class I, Class II and Class II.5. The former two are based on Roskam’s design process [14],
but the third aims to analyze components of the aircraft using physics and geometry based methods,
rather than empirical relations. As such, the present fuselage weight estimation method fits in the Class
II.5 block of the Initiator.

No

Perform Class I Perform Class II Class II Perform Perform Class Class II.5
Start Weight Weight Converged Yes Mission II.5 Weight Converged Yes End
Estimation Estimation on MTOM? Analysis Estimation on MTOM?

No

Compute Return new

Compute Wing
Fuselage Component
Weight
Weight Weights

Figure 2.1: Initiator process flow

7
8 2. Structural Design and Analysis

2.2. Program Flow

Figure 2.2 shows the flowchart of the weight estimation method. It should be noted that this method
itself is part of a larger convergence loop built into the Initiator, which ensures all aircraft weights become
consistent between all methods implemented, as was shown in Figure 2.1.
The fuselage weight estimation method starts by retrieving the aerodynamic and weight data from the
Class II methods implemented in the Initiator. Moreover, it checks whether a design vector exists
from a previous fuselage weight estimation run. If that is the case, this design will be used as the
initial guess for the convergence loop of the current run. Otherwise, a new design vector is built using
predefined initial values. This design vector describes all structural dimensions and other properties
required to perform the structural analysis and compute the mass. The parameters that are estimated
and converged upon, as well as the initial values, are given in Table 2.1.

Table 2.1: Design vector for weight estimation method

Parameter Unit Type Initial Value

Skin Thickness m 𝑁 × 3 Matrix (Δ𝑝)cruise ⋅ 𝑟arc /𝑋
Stringer Smeared Thickness m 𝑁sections × 3 Matrix 𝜂stringer ⋅ 𝑡skin
Frame Smeared Thickness m 𝑁sections × 3 Matrix 𝜂frame ⋅ 𝑡skin
Ceiling Face Thickness m 𝑁sections Column Vector 8e-3
Wall Face Thickness m 𝑁sections Column Vector 1e-3
Floor Face Thickness m 𝑁sections Column Vector 8e-3
Ceiling, Wall and Floor Core Thickness m 𝑁sections Column Vectors 8e-3
Stringer Pitch 𝑝 m 𝑁sections × 3 Matrix 0.15
Stringer Width 𝑤 m 𝑁sections × 3 Matrix 0.02
Skin Young’s Modulus Pa 𝑁sections × 3 Matrix Material Setting
Stringer Young’s Modulus Pa 𝑁sections × 3 Matrix Material Setting
Sandwich Face Young’s Modulus Pa 𝑁sections × 3 Matrix Material Setting

It should be noted that in Table 2.1 the stringer width parameter is not used by the composite weight es-
timation, whereas the metal weight estimation does not use the last three parameters, i.e. the Young’s
modulus parameters. The first three parameters have computed initial values, where 𝑟arc is the radius
of an arc (𝑖 = 1, 2, 3), 𝑋 is the material strength in tension and 𝜂stringer and 𝜂frame are the stringer-to-
skin and frame-to-skin ratio (in terms of smeared thickness), respectively. These ratios are assumed
to be 0.2, but are only used here as a first guess, so do not have an influence on the final result.
Using the weight and aerodynamic forces from Class II, beam theory is employed to compute the
shear and moment distributions, as is explained in section 2.3. Consecutively, also using Class II
data, the lateral and transverse loads acting on the trapezoidal structure are computed. Finally, the
pressurization loads are computed. With these loads and the load cases from Table 2.3, the loads
corresponding to each load case are computed.
All computations done so far by the fuselage weight estimation will not change during the convergence
loop. In reality, with the changes in thickness each iteration, the structural weight would change and
as such the aerodynamic and other properties of the aircraft would change. However, to recompute
all of these each iteration would be cumbersome and computationally expensive. Moreover, the fuse-
lage weight estimation itself is part of a convergence loop which makes sure that effects as the ones
described here are taken into account.
Each iteration of the fuselage weight estimation, the boom method described in section 2.5 is employed
to determine the shear flow and boom stresses in the fuselage skin. Additionally, the stresses in the
trapezoidal structure are computed. With these stresses, the skin panels, stringers and trapezoidal
members (floor, wall and ceiling) are sized for each section the fuselage is divided into. At the end of
each iteration, the loop checks whether the changes made with respect to the previous design are within
a certain tolerance, in which case the procedure stops. A mathematical description of the convergence
criterion is:
𝑋̄ ( ) /𝑋̄ ( ) − 1 ≤ 𝜖 (2.1)
where 𝜖 is the tolerance set to 1 %. Finally, the mass of the obtained design is computed, complemented
by adding empirical weights (see section 2.8) for those components that were not sized by the analytical
model.
2.2. Program Flow 9

Start
Aerodynamic &
Class II Weight
Data

Previous
Yes No
Results exist?

Design Vector Initial Design

from prior run Vector

Create Shear and

Moment
Distributions

Compute Lateral
and Transverse
Loads

Compute Pressure
Loads

Generate Load
Cases

Error < Tol? Yes

No

Compute Shear
Flow and Boom Compute Mass
Stresses

Size Structural End

Members

Figure 2.2: Fuselage weight estimation process flow

10 2. Structural Design and Analysis

Table 2.2: Load types per component

Inertial
Component Attached to Load type
Wing Point Force + Moment
Engine
Fuselage Point Force
Wing Point Force + Moment
Landing Gear
Fuselage Point Force
Wing Fuselage Point Force + Moment
Wing Point Force + Moment
Fuel Tank
Fuselage Point Force
Furnishing Fuselage Non-linearly Distributed Force
Fuselage - Non-linearly Distributed Force
Systems Fuselage Non-linearly Distributed Force
Operational Items Fuselage Non-linearly Distributed Force
Bulk Cargo Fuselage Non-linearly Distributed Force
ULD Fuselage Point Force
Passenger Fuselage Point Force
Aerodynamic
Component Load type
Fuselage Non-linearly Distributed Force + Moment
Wing Point Force + Moment
Engine Point Force + Moment

2.3. Shear Force and Bending Moment Distribution

Detailed calculation of the internal loads in a fuselage is usually only possible using finite element
analysis or similar techniques. For a conceptual weight estimation method it suffices to model the
fuselage as a cantilever beam. All weights and aerodynamic forces can be projected onto this beam
and statics is used to determine the resulting shear force and moment distributions.
The loads that are projected onto the cantilever beam can be divided into point loads, linearly distributed
loads and nonlinear distributed loads. Passengers are point loads and systems are nonlinear distributed
loads, for example. For a complete overview of the components and types of loads, refer to Table 2.2.

2.4. Load Cases

Several load cases are considered to obtain the critical loads for the structure. The current weight
estimation method makes use of a tentative set of load cases, which are listed in Table 2.3. Each
structural component is sized for all load cases, and the thickest/heaviest design is taken as the final
design for that component.

Table 2.3: Load cases for fuselage structure

Load Case Load Factor Mass Wing Loads Pressure OEI

1 2.5 MTOM 𝑊 ,max , Aero Cruise No
2 2.5 MZFW 𝑊 ,min , Aero Cruise No
3 -1.5 MTOM 𝑊 ,max , Aero Cruise No
4 -1.5 MZFW 𝑊 ,min , Aero Cruise No
5 2.5 MTOM 𝑊 ,max , Aero 0 No
6 2.5 MZFW 𝑊 ,min , Aero 0 No
7 -1.5 MTOM 𝑊 ,max , Aero 0 No
8 -1.5 MZFW 𝑊 ,min , Aero 0 No
9 2.5 MLM 𝑊 ,min Landing No
10 2.5 MLM 𝑊 ,min 0 No
11 Gust up MZFW 𝑊 ,min , Aero Cruise No
12 Gust down MZFW 𝑊 ,min , Aero Cruise No
13 1.0 MTOM 𝑊 ,max , Aero Cruise Yes
2.5. Boom Method 11

The Mass column in Table 2.3 defines the total weight the aircraft flies with in a specific load case.
The Wing Loads column defines the wing bending, which influences the lateral loads imposed on the
trapezoidal structure. 𝑊 ,max is maximum fuel weight in the wings, while 𝑊 ,min is only the minimum
fuel fraction remaining (trapped fuel). Moreover, Aero indicates that aerodynamic bending loads are
present.
In the Pressure column the cabin pressure is defined. Cruise means the differential pressure when the
cabin is at its cruise cabin altitude and the outside pressure at the cruise altitude. Landing is defined
by a setting for the minimum cabin differential pressure. Both landing and cruise cabin pressures are
multiplied by a safety factor of 1.5 and a factor of 1.15 stipulated by regulations [15].
The last column indicates whether a one engine inoperative (OEI) condition is present in the load case,
i.e. whether the vertical tail induces a sideways shear force and the operative engines a moment around
the yaw axis.
The load factor due to gusts is explained in the following. Both gusts induce a certain load factor on
the aircraft, and since these can be more sizing than the other load factors present, they are included
separately. To compute the gust load factors, the so called gust decay in altitude is computed as
follows:
25 − 50
𝑎 = fts 1 ft 1 (2.2)
30000
Then using this decay and the cruise altitude of the aircraft in feet, the gust velocity is:

𝑉 = 𝑎 (ℎ − 20000) + 50 ft/s (2.3)

Now the gust factor is computed with the following two equations:
2𝑊
𝜇= (2.4)
𝑆𝜌𝑔𝐶 𝑐

0.88𝜇
𝑘 = (2.5)
5.3 + 𝜇
where 𝑊 is the aircraft weight, 𝑆 the main wing surface area, 𝐶 the main wing lift curve slope and 𝑐
it’s mean aerodynamic chord (MAC). Finally, the change in load factor due to a gust is:

𝑘 ⋅𝐶 ⋅ 0.5 ⋅ 𝜌 ⋅ 𝑉 ⋅ 𝑉
Δ𝑛 = (2.6)
𝑊/𝑆

and the two gusts impose a load factor being either 1 + Δ𝑛 (gust up) or 1 − Δ𝑛 (gust down).

2.5. Boom Method

To compute the internal direct stresses due to the bending moment and shearflow due to shear forces,
a boom method is applied. An important reason for this decision is that the boom method allows for
a different stiffness of each skin panel and boom, which makes the method very useful for analyzing
anisotropic composite structures. Moreover, using Megson [16] as a guideline, the boom method was
adapted to account for tapering of the fuselage, such that the complex geometry of a BWB fuselage
can be analyzed. It is assumed that skin panels only carry shear and stringers (booms) only take direct
stress. Furthermore, the oval section is analyzed as a single-cell section as can be seen in Figure 2.3,
with the wall thickness added to the side arc thickness. Through this assumption, the ceiling and floor
members carry no shear flow, while the wall member shares the shear flow with the side arc.
The second moments of area including the Young’s modulus of each panel and stringer are defined
as:

𝐼 = ∫ 𝐸 𝑦 d𝐴 (2.7)

𝐼 = ∫ 𝐸 𝑥 d𝐴 (2.8)

𝐼 = ∫ 𝐸 𝑥𝑦d𝐴 (2.9)
12 2. Structural Design and Analysis

Figure 2.3: Structural idealization of oval fuselage

where 𝑥 is the section horizontal coordinate and 𝑦 the section vertical coordinate. Additionally, 𝐸 is the
Young’s modulus, with the subscript 𝑧 indicating it is the modulus in the fuselage longitudinal direction.
Finally, 𝐴 is the cross-section area and the subscript 𝑖 refers to each member over which the integral
is taken.
In each member, the direct stress due to bending moment can then be computed using:

𝑀 𝐼 −𝑀 𝐼 𝑀 𝐼 −𝑀 𝐼
𝜎 = 𝐸 [( )𝑥 + ( ) 𝑦] (2.10)
𝐼 𝐼 −𝐼 𝐼 𝐼 −𝐼

Here the subscript 𝑖 again refers to the member under consideration. 𝑀 is the vertical bending moment
(around the section horizontal axis) and 𝑀 is the horizontal bending moment (around the section
vertical axis).
Using the stresses computed with Equation 2.10, the boom areas can be computed using:

𝑡 ⋅𝑏 𝜎 𝑡 ⋅𝑏 𝜎
𝐵=𝐴 + (2 + )+ (2 + ) (2.11)
6 𝜎 6 𝜎

where the numeric subscripts refer to the skins attached to the boom under consideration, denoted by
subscript 𝐵.
Knowing the boom stresses and areas, the three components of the internal boom loads can be com-
puted as follows:

𝑃, = 𝜎 ,⋅𝐵 (2.12)
d𝑦
𝑃, = 𝑃, (2.13)
d𝑧
d𝑥
𝑃, = 𝑃, (2.14)
d𝑧

In these equations, d𝑥, d𝑦 and d𝑧 are the change in the respective coordinates along the boom (end-
point minus start point).
2.5. Boom Method 13

Using these loads, the boom shear forces are computed using:

𝑆 , = 𝑆 − ∑𝑃 , (2.15)

𝑆 , = 𝑆 − ∑𝑃 , (2.16)

where 𝑆 and 𝑆 are the applied shear forces.

The shear flow is computed as follows:
𝑞 =𝑞 +𝑞 , (2.17)

The first component of the shear flow is given by:

𝑆 , 𝐼 −𝑆 , 𝐼 𝑆 , 𝐼 −𝑆 , 𝐼
𝑞 = −𝐸 , ( )∑𝐵 𝑥 −𝐸 , ( )∑𝐵 𝑦 (2.18)
𝐼 𝐼 −𝐼 𝐼 𝐼 −𝐼

and represents the open section shear flow. For the closed section shear flow, a constant shear flow
𝑞 , is added, which is obtained by equating:

𝑆 𝜂 − 𝑆 𝜉 = ∮ 𝑞 𝑝d𝑠 + 2𝐴𝑞 , − ∑𝑃 , 𝜂 + ∑𝑃 , 𝜉 (2.19)

The variable 𝑝 is the distance from the shear center to a point on the cross-section where the integral is
evaluated. Moreover, as can be seen in Figure 2.4, d𝑠 is a differential distance along the beam section,
𝜂 and 𝜉 are the moment arms of the applied shear forces to the shear center and 𝑞 is the shear flow.

𝑆𝑦

𝜂 𝑝
𝑞

𝑑𝑠

𝑆𝑥
Figure 2.4: Closed beam section for determining ,
14 2. Structural Design and Analysis

2.6. Pressurization Loads

In the 2D case of the oval fuselage, the pressure stresses in hoop direction in the circular arcs can be
determined by:

𝑝𝑟
𝜎hoop = (2.20)
𝑡

and in the longitudinal direction by:

𝑝𝑟
𝜎long = (2.21)
2𝑡

Because the arcs have different radii, a mismatch in loads exists at the junctions, which have to be
carried by the trapezoidal structure (otherwise these result in bending moments in the skin, requiring
large thicknesses to cope with that). Suppose the line loads (N/m) in the circular arcs are called 𝑁 ,
𝑁 and 𝑁 and the loads in the trapezoid 𝑁 , 𝑁 and 𝑁 , which denote the ceiling, floor and wall,
respectively. In Figure 2.5 a free-body-diagram (FBD) of these forces is shown. These loads are
positive for a member in tension and negative for compression. First the following angles are defined:

𝑁 𝑁

𝑁
𝑁

Figure 2.5: Pressurization forces FBD

𝑤
𝜓 = cos
𝑟
𝜙 = 𝜋/2 − 𝜓
𝑤
𝛾 = cos
𝑟
𝜒 = 𝜋/2 − 𝛾
2.7. Structure Sizing 15

Then the horizontal and vertical components of the member line loads are (note that 𝑁 and 𝑁 only
have a horizontal component):

𝑁 , = 𝑁 cos 𝜙
𝑁 , = 𝑁 sin 𝜙
𝑁 , ,upper = 𝑁 cos 𝜙
𝑁 , ,upper = 𝑁 sin 𝜙
𝑁 , ,lower = 𝑁 cos 𝜒
𝑁 , ,lower = 𝑁 sin 𝜒
𝑁 , = 𝑁 cos 𝜒
𝑁 , = 𝑁 sin 𝜒
𝑁 , ,upper = 𝑁 sin 𝛽
𝑁 , ,upper = 𝑁 cos 𝛽
𝑁 , ,lower = −𝑁 , ,upper
𝑁 , ,lower = −𝑁 , ,upper

The components of the side arc load have different magnitudes at the upper and lower junctions. The
wall components do not, because the direction of the wall is the same at both junctions. However,
the signs of the wall load components at the upper junction are logically opposite of those at the lower
junction.
Then at the junction of arc 1 and 2 and the ceiling and wall, the following force equilibrium equations
hold:

𝑁 , ,upper +𝑁 , ,upper −𝑁 −𝑁 , =0 (2.22)

𝑁 , −𝑁 , ,upper −𝑁 , ,upper =0 (2.23)

The second equilibrium equation can be solved directly for 𝑁 , , from which 𝑁 follows by geometry.
At the lower junction, the following equilibrium equations hold:

𝑁 , ,lower −𝑁 , ,lower −𝑁 −𝑁 , =0 (2.24)

𝑁 , ,lower +𝑁 , ,lower −𝑁 , =0 (2.25)

The second of these equations is not needed, since all those terms are already known. From the first
equation then follows 𝑁 .

2.7. Structure Sizing

Based on the material defined by the user to be used for the different components of the fuselage,
either a metal sizing routine or a composite sizing routine is used. The metal sizing routines were
developed in Ref. [10] and will not be discussed here. The sizing routines for composite members (flat
skin panels, stringers and sandwich panels) are discussed in chapter 3.
The main purpose of the sizing procedure is to determine the dimensions of all structural members, such
that the weight may be computed. However, for composite structures, the laminate stacking sequences
also need to be determined. These result in the axial and bending stiffness of each member, which are
used by the boom method.

2.8. Mass Estimation

For all metal secondary structure and non-structural masses, empirical relations from Howe [17] are
used [9].
For the composite structural masses, trends from the NASA Advanced Composite Technology (ACT)
program [18] and Boeing’s Advanced Technology Composite Aircraft Structure (ATCAS) [19] are used.
The skin, stringer and frame masses are predicted by the presented method and the masses for splic-
ing, assembly, cargo floor and reinforcements around cutouts are estimated as a fraction of these.
16 2. Structural Design and Analysis

From the side-panel study [18], the weight breakdown in Figure 2.6a is obtained. For the keel panels
[19] the weight breakdown in Figure 2.6b is obtained.

Splicing Assembly: 7% Splicing Assembly: 8%

Passenger Floor: 18%

Cargo Floor: 19%

Skin: 42%

Skin: 51%

Cutouts: 11% Frames: 7%

Frames11% Stringers: 15%

Stringers: 12%

(a) Side panels (b) Keel panels

Figure 2.6: Weight breakdowns for side and keel panels

From these breakdowns it can be estimated that skin/panel mass should lie around 42-52 % of the
structural mass, stiffeners form around 12-15 % and frames between 7 and 11 %. For the side panels,
these three items combined form 65 % of the mass. With that in mind, the window belt, skin panel
assembly and door reinforcements may be sized as a fraction of the side panel primary structure mass.
This fraction becomes 0.17 for the three items combined, from Figure 2.6a. The cargo floor is sized
similarly, but now the keel panel primary structure (skin + stringers + frames) mass is used. Thus,
from Figure 2.6b, the cargo floor fraction is 0.26. Finally, the splicing and assembly mass is taken as a
fraction of the total primary structure mass with a value of 0.11.
Finally, the nose and aft shell (structure in front of and aft of the passenger cabin) masses are estimated
by taking the average total smeared thickness in the side panels and multiplying this by the surface area
of these shells, which are sized elsewhere in the Initiator. Multiplication with the composite material
density consecutively gives the mass.
 3
Sizing Methods for Composite Structure
Using the internal loads computed from the boom method and load cases, each structural element can
be sized to carry the applied loads. In order to do this, just like for metal, the fuselage is divided into
sections in the longitudinal direction, where each section is in between two frames. A frame spacing of
0.5 m is assumed (or another value specified by the user), then it is computed how many frames that
would give (rounding to the nearest lower integer value) and finally, the frame spacing is recomputed
using the integer amount of frames.
Each section comprises of a top arc, side arc and bottom arc, as well as the ceiling, wall and floor
members of the trapezoid structure. Each of these will be sized separately using the methods outlined
in this chapter.
First, a review of Classical Lamination Theory (CLT) is given, which is used to represent the material
properties of a fiber composite laminate. Consecutively, the sizing criteria used for the (flat) skin panels,
stringers and sandwich plates are given in that order. Following that, the routines for programmatically
determining the stacking sequence and dimensions of each of these three members is discussed.
Finally, a frame sizing routine is discussed.

3.1. Classical Lamination Theory

Prediction of the structural behavior of composites can be divided into two areas: stiffness response
and strength response. Unlike metals, fiber composites exhibit orthotropic or even anisotropic behavior.
Therefore, prediction of the strength is rather complicated, while the stiffness can be fairly accurately
be predicted using Classical Lamination Theory (CLT).

3.1.1. Stiffness
A laminate is constructed by bonding several laminae on top of each other, with a certain stacking
sequence, i.e. different fiber orientation for each ply (lamina) in a certain order. Classical Lamination
Theory can be used to determine the stress-strain relation for a lamina and laminate, as described by
Jones [20].
For an orthotropic material under plane stress, the stress-strain relations in principal material coordi-
nates is given by:

𝜎 𝑄 𝑄 0 𝜖
[ 𝜎 ] = [𝑄 𝑄 0 ][𝜖 ] (3.1)
𝜏 0 0 𝑄 𝛾

17
18 3. Sizing Methods for Composite Structure

where the constants are given by:

𝐸
𝑄 =
1−𝜈 𝜈
𝐸
𝑄 =
1−𝜈 𝜈 (3.2)
𝜈12𝐸 𝜈 𝐸
𝑄 = =
1−𝜈 𝜈 1−𝜈 𝜈
𝑄 =𝐺

When describing the stresses in any other axis system, rotated with an angle 𝜃 from the principal axes,
Equation 3.1 changes into:
𝜎 𝑄̄ 𝑄̄ 𝑄̄ 𝜖
[ 𝜎 ] = [𝑄 ̄ 𝑄 ̄ 𝑄̄ ] [ 𝜖 ] (3.3)
𝜏 𝑄̄ 𝑄̄ 𝑄̄ 𝛾
which can be rewritten in a short notation for ply 𝑘:
{𝜎} = [𝑄]̄ {𝜖} (3.4)

Before continuing, several statements have to be made regarding the following derivations:
• The laminae are presumed to be perfectly bonded, which is not necessarily an assumption but
can be achieved in practice. This implies that the laminae can not slip relative to each other
• The normal to the middle surface remains straight and perpendicular to the middle surface, such
that 𝛾 = 𝛾 = 0
• The aforementioned normal is of constant length, such that 𝜖 = 0
The forces and moments on a laminate can be described by Equation 3.5 and Equation 3.6, respec-
tively:
𝑁 𝜎
[ 𝑁 ] = ∑ ∫ [ 𝜎 ] 𝑑𝑧 (3.5)
𝑁 𝜏

𝑀 𝜎
[𝑀 ] = ∑∫ [ 𝜎 ] 𝑧𝑑𝑧 (3.6)
𝑀 𝜏
Finally, these equations can be combined with Equation 3.3 to obtain the final stiffness equations for
laminates:
𝑁 𝜖∘ 𝜅
[ 𝑁 ] = [𝐴] [ 𝜖 ∘ ] + [𝐵] [ 𝜅 ] (3.7)
𝑁 𝛾∘ 𝜅
𝑀 𝜖∘ 𝜅
[ 𝑀 ] = [𝐵] [ 𝜖 ∘ ] + [𝐷] [ 𝜅 ] (3.8)
𝑀 𝛾∘ 𝜅
where

𝐴 = ∑ (𝑄̄ ) (𝑧 − 𝑧 )

1
𝐵 = ∑ (𝑄̄ ) (𝑧 − 𝑧 ) (3.9)
2

1
𝐷 = ∑ (𝑄̄ ) (𝑧 − 𝑧 )
3
3.2. Skin Panels 19

In Equation 3.9, 𝑖, 𝑗 = 1, 2, 6 and 𝐴 are the extensional stiffnesses, 𝐵 are bending-extension coupling
stiffnesses and 𝐷 are bending stiffnesses. When [𝐵] is nonzero, a laminate under tension would also
bend and/or twist. A symmetric laminate, where both geometry and material properties are symmetric
about the middle surface, [𝐵] is zero and hence no such coupling exists.

3.1.2. Strength
First-ply failure is analyzed using the Tsai-Wu failure criterion:

𝜎 𝜎 1 1 1 1 1 𝜏
+ −√ 𝜎 𝜎 +( − )𝜎 + ( − )𝜎 + =1 (3.10)
𝑋 𝑋 𝑌 𝑌 𝑋 𝑋 𝑌 𝑌 𝑋 𝑋 𝑌 𝑌 𝑆

Here, 𝑋 , 𝑋 , 𝑌 and 𝑌 denote the longitudinal and transverse ply strengths in tension and compression
and 𝑆 is the in-plane shear strength (all strengths should be provided in magnitude only, i.e. positive
values). All of these strengths are decreased by three knockdown factors: elevated temperature wet,
barely visible impact damage and material scatter [21]. The values for these knockdown factors are
0.8, 0.65 and 0.8, respectively [21].
The stresses are obtained by first using Equation 3.7 and Equation 3.8, rewritten in terms of the strains
and curvature, with the applied loads as inputs. Then using the computed strains in Equation 3.3, the
stresses in each ply can be computed.
As will become apparent later, laminates in the current study are subjected to 𝑁 , 𝑁 and 𝑁 . To com-
pute the material failure loads when all three are applied simultaneously, a Newton-Raphson iterative
method is applied. The three loads, 𝑁 , 𝑁 and 𝑁 are divided by the maximum of these three, giving
ratios from 0 to 1. Thus, with 𝐹 being the maximum (magnitude) of 𝑁 , 𝑁 and 𝑁 :

1
[𝑁 𝑁 𝑁 𝑀 𝑀 𝑀 ] ⋅ = [𝑅 ] (3.11)
𝐹

where [𝑅 ] denotes the ratios of the loads to 𝐹 . Note that 𝑀 , 𝑀 and 𝑀 are zero.
The Newton-Raphson method starts with an initial guess 𝑁 , which is multiplied by [𝑅 ] to get a new
load vector. Then the Tsai-Wu failure criterion is used to compute first-ply failure as described earlier.
Now a residue is computed:
𝑟 =𝑅 −1 (3.12)

where 𝑅 is the Tsai-Wu failure criterion. A second load vector is computed using 𝑁 + 𝛿 , which is
a little offset from the original load. Now a residue 𝑟 is computed analogous to 𝑟 . Now the derivative
between these two residues can be computed:
𝑟 −𝑟
𝜕𝑟 = (3.13)
𝛿

which provides a new guess for the next iteration:

𝑟
𝑁 =𝑁 − (3.14)
𝜕𝑟

This process repeats until 𝑟 is (close to) zero, which means the Tsai-Wu failure criterion was satisfied.
The value for 𝑁 obtained this way is now multiplied with [𝑅 ] to give a vector describing the critical
loads for material failure.

3.2. Skin Panels

Skin panels are assumed to carry shear forces and a transverse pressure load. Under shear, no failure
or buckling may occur. Under pressure, a maximum deflection is constraining. Moreover, a transverse
pressure in a pressure vessel such as a fuselage results in hoop and longitudinal tensile stresses.
Failure is checked with first-ply failure using the Tsai-Wu criterion as explained above. A panel with all
possible loads acting on it is shown in Figure 3.1.
20 3. Sizing Methods for Composite Structure

𝑝
𝑁
𝑁 𝑏

𝑎
𝑁

Figure 3.1: Forces acting on a panel

3.2.1. Shear Buckling

An approximate method to obtain the critical shear buckling load is presented in Ref. [21] and repeated
here:
𝑁 ,crit = (3.15)
. . .
√ + +

This equation holds when 0.5 ≤ 𝑎/𝑏 < 1, where 𝑎 is the panel length and 𝑏 the panel width. Moreover:

𝑎 𝑎
𝐷1 = 𝐷 +𝐷( ) + 2 (𝐷 + 2𝐷 ) ( ) (3.16)
𝑏 𝑏
𝑎 𝑎
𝐷2 = 𝐷 + 81𝐷 ( ) + 18 (𝐷 + 2𝐷 ) ( ) (3.17)
𝑏 𝑏
𝑎 𝑎
𝐷3 = 81𝐷 + 𝐷 ( ) + 18 (𝐷 + 2𝐷 ) ( ) (3.18)
𝑏 𝑏
When 𝑎/𝑏 = 0 a different approach should be used, which will be explained directly hereafter. When
0 ≤ 𝑎/𝑏 < 0.5, the result should be linearly interpolated between the values for 𝑎/𝑏 = 0 and 𝑎/𝑏 = 0.5.
Note that 𝑎/𝑏 = 0 is never achieved in practice and no threshold was set after which panels were
deemed long, so the linear interpolation is executed for all panels with 𝑎/𝑏 < 0.5.
For long plates under shear (𝑎/𝑏 = 0), a different approach is used. Now, 𝐴𝑅 = 𝑎/𝐿 where 𝐿 is, instead
of 𝑏 the length over which the buckling pattern is confined. Then the critical buckling load is given by:

𝜋
𝑁 ,crit = [𝐷 (1 + 6 tan 𝛼𝐴𝑅 + tan 𝛼𝐴𝑅 )
2𝐴𝑅 𝑎 tan 𝛼
+2 (𝐷 + 2𝐷 ) (𝐴𝑅 + 𝐴𝑅 tan 𝛼) + 𝐷 𝐴𝑅 ] (3.19)

This equation cannot be solved directly, because the parameters tan 𝛼 and 𝐿 are not known. Taking
the partial derivative with respect to the two unknowns and setting these equal to zero allows to solve
for the critical buckling load:
/
𝜕𝑁 ,crit 𝐷
= 0 ⇒ 𝐴𝑅 = [ ] (3.20)
𝜕𝐴𝑅 𝐷 tan 𝛼 + 2 (𝐷 + 2𝐷 ) tan 𝛼 + 𝐷

𝜕𝑁 ,crit
= 0 ⇒ 3𝐷 𝐴𝑅 tan 𝛼 + (6𝐷 𝐴𝑅 + 2 (𝐷 + 2𝐷 ) 𝐴𝑅 ) tan 𝛼
𝜕 tan 𝛼
− (𝐷 + 2 (𝐷 + 2𝐷 ) 𝐴𝑅 + 𝐷 𝐴𝑅 ) = 0 (3.21)

3.2.2. Pressure
Under internal pressure, a hoop load and longitudinal load develop in the fuselage skin. Conventional
fuselages are approximately circular cylinders, and analytical solutions for the developed stresses can
3.3. Stringers 21

be obtained. Equation 2.20 gives the hoop stress, and the longitudinal stress is half that, as in Equa-
tion 2.21. In the present study, it is assumed that when looking at the individual skin panels, they are
approximately elements in a circular cylinder and as such the same tensile stresses develop due to
pressurization. Therefore, a tensile load is applied in both hoop and longitudinal direction and first-ply
failure is computed. It should be remarked that this assumption leads to an underestimation of pressur-
ization loads and as such a lighter structure is obtained. In fact, through this assumption, panels would
not line up with each other entirely and the gaps would have to be modeled as flat pressure bulkheads
to compensate for the mass reduction obtained otherwise.
As an additional failure mode, a maximum deflection of the skin panel is imposed when under pressure
loading. Again, the methodology from Ref. [21] is used and summarized here. It is important to note
that this theory is applicable to flat, rectangular panels, whereas the fuselage in reality consists of
curved panels. However, each time a panel between two stringers and two frames is considered, such
that the curvature may be small. Additionally, it is expected that flat panels under transverse pressure
exhibit a larger deflection than curved panels and as such, the current approach is conservative.
Writing the out-of-plane deflection of a plate and the applied pressure load as a Fourier series, Equa-
tion 3.22 is obtained, which describes the center deflection of a monolithic panel under pressure 𝑝 .
Using an iterative method, 𝑚 and 𝑛 are increased until the computed deflection in the current iteration
is within a certain margin of the previously computed deflection. It is assumed that 𝑚 = 𝑛.

𝑚𝜋 𝑛𝜋
𝛿 = ∑∑ sin sin (3.22)
2 2
𝐷 ( ) + 2 (𝐷 + 2𝐷 ) ( ) +𝐷 ( )

3.3. Stringers
Stringers are analyzed using the exact same procedure as described in Ref. [22]. Only C-stringers
have been modeled in the present study. However, any other shape of stringer is also possible with
slight modifications to the code.
Figure 3.2 shows that all three flanges can have different widths, thicknesses and layups. However,
properties of the entire cross-section are required to perform the structural analysis. The Young’s
modulus 𝐸 of flange 𝑖 can be computed using:

1
𝐸 = (3.23)
(𝑎 ) 𝑡

The bending stiffness of the entire stringer then can be computed as follows:

(width) (height)
𝐸𝐼eq = ∑(𝐸𝐼) = ∑ 𝐸 [ +𝐴 𝑑 ] (3.24)
12

The equivalent axial stiffness can simply be computed using:

𝐸𝐴eq = ∑(𝐸𝐴) = ∑ 𝐸 𝐴 (3.25)

As will become clear later, the current weight estimation method only takes into account bonded
stringers. Therefore, the inter-rivet buckling failure is not discussed here. It should be noted, though,
that inter-rivet buckling is taken into account in the computer program and therefore fastened stringers
are also a possibility if the designer/researcher wants to analyze these.

3.3.1. Crippling
Flanges of a stringer feature a local stability failure mode, called crippling, where the flange buckles
locally under compression. A semi-empirical approach is used, based on Ref. [21]. As can be seen
in Figure 3.2, the web is analyzed as a no-edge-free (NEF) member, while the top and bottom flanges
are analyzed as one-edge-free (OEF) members.
22 3. Sizing Methods for Composite Structure

Laminate 3 OEF
𝑡

Laminate 2

𝑏 NEF

Laminate 1 OEF

𝑏
Figure 3.2: Stringer cross-section

For OEF members, the following equation can be used to compute the critical crippling load.

crit,crp 2.151𝑁crit,mat
,
𝑁 , = .
(3.26)
crp
( )
crp
Failure only occurs when 𝑏 ≤ 2.91𝑡 , otherwise failure is dictated by material failure.
Similarly, the following equation gives the critical crippling load for NEF members, where a flange is
crp
only crippling sensitive when 𝑏 ≤ 11.07𝑡 .

crit,crp 14.92𝑁crit,mat
,
𝑁 , = .
(3.27)
crp
( )

3.3.2. Column Buckling

Column buckling load under compression is given by:
𝑐𝜋 𝐸𝐼
𝑃crit = (3.28)
𝐿
where the buckling constant 𝑐 can attain different values depending on the boundary conditions of the
beam. For the present study, a value of 1.0 was used, which corresponds to pinned boundary condition
on both beam ends.

3.4. Sandwich Panels

A sandwich panel consists of two facesheets and a core of, for example, honeycomb material. The ABD
matrices of the facesheets can be determined using CLT, since these are just normal laminates. How-
ever, the sandwich in its entirety has a different D matrix, which can be computed using the following
equation:
𝑡 +𝑡
𝐷 = 2𝐷 , + 2𝐴 , ( ) (3.29)
2
With this new D matrix, the failure modes of a sandwich panel can be computed. First of all, there
is global buckling under compression. Secondly, shear buckling and buckling under the combined
3.4. Sandwich Panels 23

load case. Thirdly, sandwich wrinkling under compression and shear is taken into account. Sandwich
crimping is the final failure mode. The loads applied to sandwich panels are the same as for monolithic
panels, so refer to Figure 3.1 for visualization of these loads. Do note that only half of the applied 𝑁 ,
𝑁 and 𝑁 need to be carried by the separate face sheets.
Since sandwich panels are most likely to be used for the trapezoidal structure, a transverse load due
to passenger weight needs to be accounted for. The same analysis is used as for monolithic panels,
i.e. Equation 3.22 is used to compute the panel deflection, which is constrained by a user-defined
maximum.

3.4.1. Buckling
Buckling of the entire sandwich panel is computed using:
𝑁 ,crit
𝑁 ,crit = (3.30)
,crit
1+

where 𝑁 ,crit is given by:

𝜋 𝐴𝑅
𝑁 ,crit = [𝐷 𝑚 + 2 (𝐷 + 2𝐷 ) 𝐴𝑅 + 𝐷 ] (3.31)
𝑎 𝑚

Here, 𝑚 is an integer value that minimizes the right hand side of the equation. Equation 3.31 is therefore
computed for a range of values for 𝑚, and the minimum resulting 𝑁 ,crit is used. For shear buckling,
the following equation is used:
(𝐺 + 𝐺 )𝑡
𝑁 ,crit = ( )
(3.32)
2+
,crit,pan

where 𝑁 ,crit,pan is computed as was done for monolithic panels in Equation 3.15 and Equation 3.19.
When both a compressive load and a shear load are applied, the following interaction curve is used:

𝑁 𝑁
𝑅buck = +( ) (3.33)
𝑁 ,crit 𝑁 ,crit

which indicates buckling failure when the right hand side becomes larger than 1.

3.4.2. Wrinkling
Three wrinkling modes are possible: symmetric, asymmetric and mixed mode wrinkling. In the present
analysis, only the first two are taken into account. One can either take into account waviness (which
leads to wrinkling at a lower load) or not. Logically, the first approach is more conservative. Therefore,
this approach was used in the present work. However, this approach may also be too conservative, so
the method that does not take into account waviness is also discussed here.

Without Waviness For symmetric wrinkling, when the following holds:

/
𝐸 𝐸
𝑡 ≥ 1.817𝑡 ( ) (3.34)
𝐺

the critical wrinkling load is:

/
𝑁 ,wr = 0.91𝑡 (𝐸 𝐸 𝐺 ) (3.35)
and otherwise the critical wrinkling load is given by:

𝐸 𝐸 𝑡 𝑡
𝑁 ,wr = 0.816√ +𝐺 (3.36)
𝑡 6

For asymmetric wrinkling, when:

/
𝐸 𝐸
𝑡 ≥ 3𝑡 ( ) (3.37)
𝐺
24 3. Sizing Methods for Composite Structure

the critical wrinkling load is:

/
𝐺 𝑡
𝑁 ,wr = 0.51𝑡 (𝐸 𝐸 𝐺 ) + (3.38)
3
and otherwise:

/ 𝐸 𝐸
𝑁 ,wr = 0.59𝑡 √ + 0.378𝐺 𝑡 (3.39)
𝑡

With Waviness For symmetric wrinkling the critical load is given by:

/
𝑁 ,wr = 0.43𝑡 (𝐸 𝐸 𝐺 ) (3.40)

and for asymmetric wrinkling by:

/
𝐺 𝑡
𝑁 ,wr = 0.33𝑡 (𝐸 𝐸 𝐺 ) + (3.41)
3

Although only for thin cores asymmetric wrinkling occurs [21], both modes are used in the current
analysis, since very thin cores may lead to an optimal design for certain panels in the fuselage structure.
Therefore, the lowest critical load (either symmetric or asymmetric) is used as critical wrinkling load.
Wrinkling under shear load can be analyzed analogous to the presented wrinkling analysis, because a
shear load can be decomposed into a compressive and tensile load. Only the core shear stiffnesses
and face Young’s modulus should be computed as if the panel were rotated by 45∘ . The core shear
stiffnesses can be computed as:

𝐺 +𝐺
𝐺 ̄ = (3.42)
2
𝐺 +𝐺
𝐺 ̄ = (3.43)
2

For wrinkling under combined compression and shear, again an interaction curve similar to Equa-
tion 3.33 is used.

3.4.3. Crimping
Crimping is similar to asymmetric wrinkling and under compression can be shown to give a critical load
as given by:
𝑁 ,crimp = 𝑡 𝐺 (3.44)

Under shear load, the critical crimping load is given by:

𝑁 ,crimp = 𝑡 √𝐺 𝐺 (3.45)

3.5. Sizing Routines

The equations presented in the previous sections need to be implemented into routines that determine
the laminate(s) of the structural components, depending on applied loads. Figure 3.3 shows what inputs
should lead to what outcome for the three structural components considered in the present work.
Evidently from Figure 3.3, monolithic (skin) panels take shear, line and pressurization loads, stringers
take longitudinal loads and sandwich panels have to resist all of these. Sizing processes for the mono-
lithic panels and stringers were developed in Ref. [22], but several modifications were made to these
processes. Therefore, an explanation of these processes is presented here. The first and most impor-
tant modification is the addition of the sandwich panel. Moreover, sizing of the monolithic panels was
modified to include a transverse pressure load.
3.5. Sizing Routines 25

Width a
Length b
Monolithic Panel
Shear Load Nxy Laminate
Sizing
Line Loads Nx & Ny
Pressure p

Length L
Flange Laminates
Compression Load Stringer Sizing
Flange Widths
Tensile Load

Width a
Length b
Sandwich Panel Face Laminate
Shear Load Nxy
Sizing Core Thickness
Line Loads Nx & Ny
Pressure p

Figure 3.3: Input and output for sizing procedures

3.5.1. Monolithic Plate Optimization

Optimization of monolithic plates is performed almost identically to the Additive Lay-up Generation
Optimizer (ALGO) routine developed in Ref. [22]. However, the pressurization loads as discussed
earlier were added to the process. Figure 3.4 shows the optimization process. Reference A points to
the flowchart in Figure 3.5, which in turn references back to B in Figure 3.4.
As a start, a [±45/0/90]s laminate is used. Then this laminate is analyzed, which means all failure
modes are checked. If no failure is found, the laminate is valid and the optimization ends. If not,
Figure 3.5 shows how a ply is selected which should be added to the laminate. Additionally, in the
second and later iterations, four alternatives for the main laminate are created: one with a midply (if
the main laminate has none), or one without a midply if the main laminate has one, the main laminate
with an additional ply (in the symmetric part) and as a fourth alternative the main laminate with an
additional ply in the symmetric part and a midply. These alternatives may result in a more efficient
laminate than the main laminate, simply because they add some thickness. All five (including the main
laminate) designs are analysed and the main laminate is replaced with the thinnest option capable
of withstanding the loads. If no designs can handle the loads, the main laminate is modified again as
depicted in Figure 3.5. When the plate is found to be shear buckling critical, it is not directly evident what
ply should be added. Therefore, each D-term is perturbed by 1% and the shear buckling calculation is
performed for each perturbation separately. The term that gives the largest increase in critical shear
buckling load determines the ply that should be added in the next iteration.
Finally, as can be seen in Figure 3.4, when a suitable laminate has been found, an iterative technique
starts that tries to remove as many plies from the laminate without making it fail. Since the optimization
may jump from one critical failure mode to another, some redundancy may develop in the laminate.
This process aims to minimize the weight penalty associated with that.

3.5.2. Stringer Optimization

The Enumerative Lay-up Family Optimizer (ELFO) from Ref. [22] was used to optimize a stringer for
a maximum tensile and maximum compression load simultaneously, without any modifications to this
method. ELFO aims to include as many 0 plies as possible, since only axial loads are considered for
stringers. Furthermore, basic laminate design rules are used, including but not limited to: balanced
and symmetric laminates, 10% rule and maximum ply stack rule.
26 3. Sizing Methods for Composite Structure

Start

Remove Next Ply

Yes (start at inner ply)
Initialize Fixed
Parameters

Start Loop with More plies can Design still

Initial Laminate Yes No Restore removed ply
be removed? meets criteria?

Create Alternative
Laminates and Yes
Analyze These
No End

Some Design Replace Main

A
Better than Yes Design with
Main Design? Superior One

Does Design
No Meet all
Criteria?

Figure 3.4: Optimization process for composite panels

3.5. Sizing Routines 27

Analyze Current
Design

(Shear)
Buckling Yes
Critical?

No

Add Ply for

Shear Strength
Yes Maximum Nxy
Critical?
Strength
Add Ply for Most
Critical D Term
No

Pressure
Add Ply for
(Deflection) Yes
Maximum D22
Critical?

No Return New Design

Nx Strength Add Ply for

Yes
Critical? Maximum D11

B
No

Ny Strength Add Ply for

Yes
Critical? Maximum D22

Figure 3.5: Ply addition for composite panels

28 3. Sizing Methods for Composite Structure

ELFO generates a family of laminates, which are subsequently applied to each member (flange or
web) of the stringer under consideration. Each member can have a different laminate, and a maximum
difference in amount of plies is set, such that excessive ply-drops from one member to another are pre-
vented. With the laminates fixed, ELFO starts a gradient based optimization routine to determine the
optimal flange widths (i.e. minimum weight, while satisfying failure constraints). When all different com-
binations of layups have been analyzed, the lightest design meeting the criteria is chosen, concluding
the procedure. A flowchart of the described process can be seen in Figure 3.6.

Compute
Initilize Optimization Generate Valid
Compressive and Compute total
Start Settings, Material Layups for each
Tensile Strength of amount of designs N
and Parameters Member
each Laminate

Return
Perform Member
Set Layups for
Select Design i Gradient Based Widths that i=N?
each Member
Optimization Minimize
Weight

No: i + 1

Return Lightest
Stringer Design
End Yes
Meeting Load
Requirements

Figure 3.6: Optimization process for composite stringers

3.5.3. Sandwich Panel Optimization

The optimization process for sandwich panels (see Figure 3.4 again) is identical to the Additive Lay-up
Generation Optimizer (ALGO) routine as presented for panels. However, Reference A now points to
Figure 3.7. The process for determining the failure criteria is shown in Figure 3.7 and is different from
Figure 3.5. Logically, different failure modes are present. Moreover, analysis of a sandwich panel is not
simply computing the failure criteria as was done for monolithic panels. For sandwich panels, a core
thickness optimization loop is performed each time the face laminate is analyzed. This loop increases
the core thickness from a set minimum (0.1 mm) to a set maximum (200 mm), or until all failure criteria
are met.

3.6. Frames
Frames transfer loads from attached components to the fuselage skin and prevent global instability
of the fuselage. Furthermore, they prevent expansion of the fuselage skin under pressurization loads.
Frames are statically indeterminate structures and the multitude of loads make their analysis and sizing
difficult. Finally, the unconventional shape of the oval frames rule out use of simple empirical solutions.
The elastic center method as presented by Bruhn [23] was used for sizing of the fuselage frames. As
such, the shear forces from the boom method can be taken into account to compute the internal bending
moment in a frame. This moment then is used to size a cross-section that minimizes the weight while
preventing material failure and crippling. Finally, a minimum cross-sectional area moment of inertia
results from the Shanley criterion, in order to prevent global buckling of the fuselage structure. This
section presents this shortly described procedure in more detail.
The elastic center method is a method to analyze a statically indeterminate structure. For a closed
section like a frame, a virtual cut is made, and one edge is assumed to be attached to a rigid arm
connecting it to the so-called elastic center. The other end (of the section, where the cut was made) is
3.6. Frames 29

Compute Strength
and Failure Ratio

Analyze Current
Design

Start With
Minimum
Buckling Compression Core
Yes No Shear Buckling? Thickness
Critical? Buckling?

Yes Yes

Compute Buckling,
Increase Core
Add Ply for Add Ply for Most Wrinkling, Crimping
No Thickness
Maximum D66 Critical D Term and Deflection Failure

No

Add Ply for

Shear Strength
Yes Maximum Nxy
Critical?
Strength
Design Meets Maximum Core
No
Criteria? Thickness?
No Return New Design

Pressure
Add Ply for Yes
(Deflection) Yes
Maximum D22
Critical? Return Failure
Yes
Ratios
B
No

Wrinkling Add Ply for

Yes
Critical? Maximum D11

No

Nx Strength Add Ply for

Yes
Critical? Maximum D11

No

Ny Strength Add Ply for

Yes
Critical? Maximum D22

Figure 3.7: Ply addition and core thickness for composite sandwich panels
30 3. Sizing Methods for Composite Structure

assumed to be clamped.

𝐹
1
𝑄
6
7
Inelastic bracket
8

𝑦′

Elastic center
𝑋
𝑀 𝑦

𝑦̄

Reference axis

Figure 3.8: Frame section for elastic center method

Since the frames considered here are symmetrical around the vertical axis, only half of the section is
considered, which is analyzed as a curved beam. The cut in the cross-section is made exactly at the
top of the cross-section. Considering a section of this curved beam, as shown in Figure 3.8, 𝑄 and 𝐹
are the applied forces and 𝑉, 𝐻 and 𝑀 are the internal forces and bending moment. Starting from the
free end and working towards the clamped end, Equation 3.46, Equation 3.47 and Equation 3.48 relate
the internal forces to the external forces.

𝑉 =𝑉 +𝐹 (3.46)
𝐻 =𝐻 +𝑄 (3.47)
𝑀 = 𝑀 + 𝑉 𝑑 − 𝐻 ℎ + 𝐹𝑎 − 𝑄𝑏 (3.48)

The horizontal and vertical forces applied are the components of the shear force acting on the frame,
which is computed by taking the difference in shear flow of the two adjacent sections of skin, analogous
to Megson [16].
Now the assumption of the rigid arm connected to the free end of the beam comes into play, which
introduces an additional horizontal and vertical force to the entire beam, as well as a bending moment.
However, first the position of the elastic center has to be found. Since the cross-section is symmetrical
around the vertical axis, the elastic center lies on the vertical axis and only the vertical location of the
elastic center should be found. This is done by assuming a reference axis (𝑦 = 0 in this case) and
computing the product of elastic weight 𝑤 with the arm of a section to the reference axis 𝑦′, where
𝑤 = d𝑠/𝐼 and d𝑠 the length of a section. Then the location of the elastic center 𝑦̄ is given by:

∑ 𝑤𝑦′
𝑦̄ = (3.49)
∑𝑤

Then the elastic moment of inertia (called 𝐼 for convenience) is given by:

𝑦 d𝑠
𝐼 =∑ (3.50)
𝐸𝐼
3.6. Frames 31

where 𝑦 is the arm from a section on the cross-section to the elastic center location 𝑦,̄ see Figure 3.8.
Calling the internal bending moments computed from Equation 3.48 𝑀 , the concept of elastic moments
Φ are introduced, which are computed as follows:
𝑀 d𝑠
Φ = (3.51)
𝐸𝐼
Now the moment and horizontal force in the elastic center can be computed. Note that a vertical force
is absent there due to symmetry.
−∑Φ
𝑀 = (3.52)
∑ d𝑠/𝐸𝐼
∑Φ ⋅ 𝑦
𝑋 = (3.53)
𝐼
Finally, the internal bending moments in the frame are given as:
𝑀 =𝑀 +𝑀 −𝑋 ⋅𝑦 (3.54)

𝑡
ℎ

Figure 3.9: Frame cross-section

A gradient-based optimization routine is used to optimize the cross-section for each section of each
frame. As a constraint on the minimum cross-sectional area moment of inertia, the Shanley criterion is
used:
𝐶 𝑅𝑀
𝐸𝐼req = (3.55)
8ℎ𝐿fr

The frame is checked for both material failure and crippling considerations. For metal frames, the Von
Mises yield criterion is used to predict material failure:

√𝜎 − 𝜎 𝜎 + 𝜎 + 3𝜎
<1 (3.56)
𝜎ult
For composites the Tsai-Wu failure criterion is used as was presented in subsection 3.1.2.
For crippling in a metallic cross-section, the top flange (attached to the fuselage skin) is assumed to
be a no-edge-free (NEF) flange with both sides clamped. The web is also assumed to be NEF, but
with one side clamped and the other simply supported. Finally, the lower flange is approximated as a
one-edge-free member with one edge simply supported. The general crippling formula for a metallic
flange is given by:
𝐸 𝑡
𝜎crp = 𝑘𝜋 ( ) (3.57)
12(1 − 𝜈 ) 𝑏
[23] where the constant 𝑘 depends on the constraints of the flange. For the top flange 𝑘 = 6.98, for the
web 𝑘 = 5.42 and for the lower flange 𝑘 = 0.43.
Crippling failure occurs when:
𝜎
>1 (3.58)
𝜎crp
where 𝜎 is the compressive stress acting on the flange.
Crippling failure for composite frames is determined exactly as was shown in subsection 3.3.1, with
both the top flange and web being NEF members and the lower flange an OEF member.
32 3. Sizing Methods for Composite Structure

3.7. Stiffened Trapezoid

Not always are sandwich panels desirable to build the trapezoidal structure. When cutouts need to be
made in the trapezoid, for windows, doors and possibly overhead bins (on top of the ceiling member),
stiffened panels may be more beneficial than sandwich panels. The same sizing methods can be used
for a stiffened panel trapezoid as for the stiffened skin panels. However, a lateral compression force
may act on certain members of the trapezoid, and since these are not aligned with the stringers, an
additional beam is required that carries these loads. Such beams are connected to the frames and
prevent buckling of the stiffened trapezoid panels. Sizing of these beams is done analogous to the
stringer sizing outlined in this chapter.
 4
Regression Analysis
Use of the procedure outlined in the previous chapter directly in the weight estimation method is pro-
hibitive due to its computational time. Even though the method is much faster than a full-blown opti-
mization, such as a genetic algorithm, a single panel or stringer may take from 0.1 s to little under a
minute. Obviously, when several hundreds or even thousands of panels and stringers are sized this
way, total computation time amounts to hours.
In order to use the results from the composite analysis, a surrogate modeling approach was adopted.
This allows the relevant results (weight, axial and bending stiffness, thickness and/or cross-sectional
area) to be modeled as function of the relevant parameters (panel dimensions, axial and/or transverse
load and shear load, for example). Since such a model only maps input to output, the expensive
detailed calculation does not have to be executed and the in-the-loop calculation time of the weight
estimation method remains acceptable.
Neural networks are used in the current mass estimation method as surrogate models. Neural networks
proved to be very capable of capturing the behavior of this specific problem. However, additionally, lin-
ear regression models were created as well, since these can be easily distributed using a mathematical
description. This chapter discusses the results of both these approaches. Table 4.1 shows the ranges
of inputs used to create the models as shown in Figure 3.3. It should be noted in advance that in order
to use the following models, only values within these ranges should be used if an accurate result is to
be expected. Even neural networks have difficulty in extrapolating their data.
The bounds for 𝑎 and 𝑏 were deduced from what seem to be logical sizes for panels in between frames
and stringers. According to Kassapoglou [21], a minimum stringer pitch of 0.12 m provides a good
weight/cost trade-off. The upper bound is somewhat arbitrary. The panel length depends on the frame
spacing, which is assumed not to exceed 0.8 m. Since composite panels are used in the trapezoidal
structure, and there the x-direction is along the pitch axis of the aircraft, the panel length is defined by
the floor width and as such a length of 10 m was chosen.
The 𝑁 and 𝑁 bounds were selected based on loads applied by Ambur and Rouse [24], who dictate
an ultimate load of 430 kN/m. In order to have some margin, a maximum load of 500 kN/m was
chosen for these two loads. Even though this is already higher than the ultimate found in literature, this
load is multiplied by the safety factor in the composite sizing routines, providing even more tolerance.

Table 4.1: Input values for structural components

Monolithic panel Sandwich panel

Input Start End Input Start End
Stringer
𝑎 [m] 0.25 0.8 𝑎 [m] 0.1 5.0
Input Start End
𝑏 [m] 0.1 0.5 𝑏 [m] 0.1 0.8
𝐿 [m] 0.3 0.6
𝑁 [kN/m] 0 1500 𝑁 [kN/m] 0 500
𝑁 , [kN] 0 2000
𝑝 [kPa] 0 100 𝑁 [kN/m] 0 500
𝑁 , [kN] 0 2000
𝑁 [kN/m] 0 500 𝑁 [kN/m] 0 1500
𝑁 [kN/m] = 𝑁 /2 𝑝 [kPa] 0 100

33
34 4. Regression Analysis

Additionally, shear load bounds were defined based on values encountered in the load calculations for
several aircraft.
For an A320, the minimum cabin pressure (maximum cabin altitude) is at 2,400 m, corresponding to
75.2 kPa atmospheric pressure according to the ISA. The service ceiling is 12.5 km, where the outside
pressure is 17.9 kPa. Therefore a differential pressure of 57.3 kPa acts on the fuselage skin. In terms
of ultimate load, this differential pressure reaches 86 kPa and therefore a slightly higher upper limit
was selected at 100 kPa. It should be noted that this is valid for most if not all aircraft, because the
maximum cabin altitude is defined by regulations and the service ceiling will likely not be much higher
than the 12.5 km used here.
Dividing each parameter range into a certain number of values, and optimizing for each possible combi-
nation of these values results in a huge set of designs that need to be analyzed. Reducing this amount
of designs by limiting the range or increasing the step size reduces the fidelity of the modeling space
and may exclude certain important effects. Therefore, Latin Hypercube Sampling is used to obtain an
evenly spaced set of input values for which the structural sizing is performed.
Additionally, before the regression techniques presented in the next two sections were performed, the
data was scaled. Since the values of data vary widely (thicknesses in the order of 10 and Young’s
modulus in the order of 10 ), this ensures that the relative importance of each predictor is appropriately
taken into account and is not dominated by the absolute value a predictor or response has. The scaling
technique used is called feature scaling and as Equation 4.1 shows, it scales each value to a [0,1]
domain, referring to the ratio of the value to the range of the parameter (𝑥 is the normalized value).
𝑥 − min(𝑥)
𝑥 = (4.1)
max(𝑥) − min(𝑥)
For the neural networks, this scaling is not necessary, but it may make it easier to train. Conversely,
for the linear regression and especially interpretation of the coefficients, scaling is of paramount impor-
tance.
The material properties used for the surrogate models are listed in Table 4.2. If different materials are
to be used in the weight estimation method, one has to modify these values and create new surrogate
models with these materials. How that works is explained in detail in Appendix A.
Table 4.2: Material properties used for surrogate models

UD material
Property Value Unit
𝐸 137.9 GPa
𝐸 11.7 GPa
Core material
𝜈 0.29 -
Property Value Unit
𝐺 4.82 GPa
𝐸 131 MPa
𝑡ply 0.1524 mm
𝐺 41.4 MPa
𝑋 2068 MPa
𝐺 20.7 MPa
𝑋 1723 MPa
𝜌 48.2 kg/m
𝑌 96.5 MPa
𝑌 338 MPa
𝑆 124 MPa
𝜌 1609 kg/m

4.1. Linear Least Squares Regression

Linear regression uses least-squares to fit the model to data. It assumes a linear relationship between
the dependent variable and the coefficients of the function describing that variable (usually a polyno-
mial). A linear model can be written as:
𝑦̄ = 𝑋̄ 𝛽̄ + 𝜖̄ (4.2)
where:
𝑦
𝑦̄ = ( ⋮ ) (4.3)
𝑦
4.1. Linear Least Squares Regression 35

𝑥 ⋯ 𝑥
𝑋̄ = ( ⋮ ⋱ ⋮ ) (4.4)
𝑥 ⋯ 𝑥

𝛽
𝛽̄ = ( ⋮ ) (4.5)
𝛽

𝜖1
𝜖̄ = ( ⋮ ) (4.6)
𝜖
In the above equations, 𝑛 is the number of data points and 𝑝 is the amount of independent variables or
predictors.
The least-squares solution to this problem is a vector 𝑏,̄ which is an estimation of the vector 𝛽.̄ It can
be computed using:
𝑏̄ = (𝑋̄ 𝑋)
̄ 𝑋̄ 𝑦̄ (4.7)
Predicting values of 𝑦̄ then simply becomes:

𝑦̂ = 𝑋̄ 𝑏̄ (4.8)

where 𝑦̂ denotes that it is a prediction from the model.

Important to note is that linear regression does not necessarily mean that the model is linear in the
predictor variables. Any (polynomial) function can be used with this type of regression, since the term
linear refers to that the model is linear with respect to the coefficients 𝛽̄ of the model.
For each response, an initial model was assumed based on engineering sense and the failure criteria
as outlined in chapter 3. Using that model, linear regression is performed using a Matlab routine called
fitlm. A stepwise linear regression procedure was executed for each of these models, using the
Matlab routine step, to improve the prediction. Based on the p-value of each predictor, which indicates
the significance of a certain predictor, terms are added or removed from the model to improve the fit.
These modified linear regression models generally produce a better prediction of the response variable
than the original model.
In order to understand the results presented in the following three sections, the Wilkinson notation
is shortly discussed. Wilkinson notation describes the factors in models, with Aˆk (with k a positive
integer) meaning a factor to the power k, A*B meaning the product of A and B, but also the terms A and
B separately and, finally, A:B meaning the product of A and B only. With a - sign, the following term is
excluded from the model (for example, A*B*C - A:B:C would include the terms A, B, C, A:B, A:C and
B:C).

4.1.1. Skin Panels

For the monolithic skin panels, three responses have to be predicted: weight, thickness and the Young’s
modulus (in the length direction, which is aligned with the fuselage longitudinal axis). The independent
variables are the length and width of the panel, the applied shear load, transverse pressure and the
hoop line load. The axial line load is not included, since this is exactly half the hoop line load, based
on the analytical formula for membrane loads in a circular cylinder. The resulting linear regression
coefficients are shown in Table 4.3.
Inspecting Table 4.3, it can be concluded that all loads have a more or less equal effect on weight,
thickness and Young’s modulus. However, for the weight, it is evident that the panel area (𝐿 ∶ 𝑊) and
pressure times the panel area (𝐿 ∶ 𝑊 ∶ 𝑝) are the most sizing parameters. In terms of thickness, the
shear load (𝑁 and related terms) has a significant influence and again the term 𝐿 ∶ 𝑊 ∶ 𝑝 appears
sizing. The Young modulus is mainly determined by its intercept value, which indicates there is little
deviation in the Young’s modulus response. Additionally, the model performs rather poorly, with a high
root-mean-squared error and low R-squared. This is attributed to the discrete response of the Young’s
modulus.
The performance of the linear regression models in Table 4.3 can be compared to the neural networks
modeling the same data in Figure 4.1 and Figure 4.2. The residuals are the predicted response minus
the actual response value for that design. Frequency is the amount of times a certain residual is
36 4. Regression Analysis

Table 4.3: Regression models for monolithic panel

Thickness
Weight
Predictor Coefficient E-modulus
Predictor Coefficient
(Intercept) -0.080721 Predictor Coefficient
(Intercept) -0.005651
𝐿 0.14307 (Intercept) 0.47881
𝐿 0.012173
𝑊 0.041909 𝐿 -0.094144
𝑊 -0.19099
𝑁 0.50351 𝑊 -0.14212
𝑁 0.05416
𝑝 0.2528 𝑁 -0.13604
𝑝 0.12869
𝑁 0.10449 𝑝 -0.44394
𝑁 0.015108
𝐿∶𝑊 0.14214 𝑁 -0.061836
𝐿∶𝑊 0.28119
𝐿∶𝑝 -0.10001 𝐿∶𝑊 -0.012931
𝐿∶𝑁 0.072602
𝑊∶𝑁 -0.3556 𝐿∶𝑝 0.031517
𝐿∶𝑝 -0.11328
𝑊∶𝑝 0.36016 𝐿∶𝑁 -0.077556
𝐿∶𝑁 0.024681
𝑊∶𝑁 -0.074105 𝑊∶𝑁 0.33761
𝑊∶𝑝 0.065117
𝑁 ∶𝑝 -0.26575 𝑊∶𝑝 -0.24639
𝑁 ∶𝑝 -0.10847
𝑝∶𝑁 -0.065339 𝑁 ∶𝑝 0.307
𝑝∶𝑁 -0.040007
𝐿 -0.1476 𝑁 ∶𝑁 0.096899
𝐿 -0.033441
𝑊 0.216 𝑝∶𝑁 0.10574
𝑊 0.27371
𝑁 -0.15517 𝐿 0.14127
𝑁 0.02327
𝑝 -0.096116 𝑁 -0.11578
𝑝 -0.061339
𝐿∶𝑊∶𝑁 -0.072308 𝑝 0.25078
𝐿∶𝑊∶𝑁 -0.031432
𝐿∶𝑊∶𝑝 0.46325 𝐿∶𝑊∶𝑝 -0.18054
𝐿∶𝑊∶𝑝 0.59799
𝑁 0.19727 RMSE 0.101
RMSE 0.0247
RMSE 0.0397 R-squared 0.439
R-squared 0.977
R-squared 0.951

observed in the entire data set. Moreover, the residuals are taken on the non-dimensional responses,
such that they can be interpreted as a percentage (when multiplied by 100). The neural networks have
smaller residuals and can therefore predict the data better, but the linear regression models also have
satisfactory performance.

4.1.2. Stringers
The main failure criteria for the stringers are the crippling criteria, which are linear in the applied com-
pression load. However, material failure produces a quadratic term. Moreover, the stringer length ap-
pears as a quadratic term in the column buckling equation. Therefore, the initial model for the stringers
is taken as a quadratic function in all three predictor variables. The resulting models are presented in
Table 4.4.

Table 4.4: Regression models for C-shape stringer

Weight
Predictor Coefficient
(Intercept) 0.012449 Axial stiffness 𝐸𝐴 Cross-sectional area 𝐴
𝐿 0.0237 Predictor Coefficient Predictor Coefficient
𝑁 , 0.22002 (Intercept) 0.041409 (Intercept) 0.043999
𝑁 , 0.22122 𝑁 , 0.37273 𝑁 , 0.36518
𝐿∶𝑁 , 0.11853 𝑁 , 0.54427 𝑁 , 0.5292
𝐿∶𝑁 , 0.35877 𝑁 , ∶𝑁 , -0.73508 𝑁 , ∶𝑁 , -0.72362
𝑁 , ∶𝑁 , -0.54752 𝑁 , 0.18761 𝑁 , 0.1865
𝐿 0.022379 𝑁 , 0.5652 𝑁 , 0.57527
𝑁 , 0.1366 RMSE 0.0349 RMSE 0.034
𝑁 , 0.43064 R-squared 0.979 R-squared 0.98
RMSE 0.0282
R-squared 0.98
4.1. Linear Least Squares Regression 37

Neural Network for Monolithic Panel Weight Modified Linear Regression Model for Monolithic Panel Weight
450 1200

400

1000
350

300 800
Frequency

Frequency
250
600
200

150 400

100
200

50

0 0
-0.12 -0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.08 -0.2 -0.1 0 0.1 0.2 0.3 0.4
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.1: Models predicting weight of monolithic panel

Neural Network for Monolithic Panel Thickness Modified Linear Regression Model for Monolithic Panel Thickness
400 900

800
350

700
300

600
250
Frequency
Frequency

500
200
400

150
300

100
200

50
100

0 0
-0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.2: Models predicting thickness of monolithic panel

What becomes immediately apparent from Table 4.4 is that the stringer length does not play a role for
the axial stiffness and cross-sectional area of the stringer. This would mean that indeed crippling is the
constraining failure mode, which does not depend on the stringer length, but only on its cross-sectional
properties.
Another interesting observation is that the product 𝑁 , :𝑁 , for all three models has a negative coef-
ficient. From a physical point of view, it does not make much sense for this term to exist, since the
tensile and compression load never occur simultaneously in the same stringer. However, the stringer
is sized for both these loads. One would expect that sizing for both tension and compression would
actually lead to a less efficient design and, therefore, a heavier design. As such, the product term may
be interpreted as a compensation for the compression and tension loads existing simultaneously in the
model.
Finally, the magnitude of coefficients including 𝑁 , are generally higher than those including 𝑁 , , which
indicates a stronger correlation between failure and compression load than failure due to tension. This
makes perfect sense, since buckling failure occurs earlier than material failure, generally.
From Figure 4.4, Figure 4.5 and Figure 4.6 it can be seen that the linear regression models perform
much worse than the neural networks modeling the same data. Not only are the residuals for the latter
at least 10 times smaller, but also the shape of the histogram resembles a normal distribution (the red
lines) better, which indicates that the residuals are evenly distributed around zero. However, the linear
38 4. Regression Analysis

Neural Network for Monolithic Panel E-mod Modified Linear Regression Model for Monolithic Panel E-mod
300 600

250 500

200 400
Frequency

Frequency
150 300

100 200

50 100

0 0
-0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.3: Models predicting E-modulus of monolithic panel

Neural Network for Stiffener Weight Modified Linear Regression Model for Stiffener Weight
60 200

180

50
160

140
40

120
Frequency

Frequency

30 100

80

20
60

40
10

20

0 0
-0.025 -0.02 -0.015 -0.01 -0.005 0 0.005 0.01 0.015 -0.1 -0.05 0 0.05 0.1 0.15 0.2
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.4: Models predicting weight of C-shape stringer

regression models still perform well from an absolute point of view, so the results presented may be
used with good accuracy.
4.2. Sandwich panels 39

Neural Network for Stiffener EA Modified Linear Regression Model for Stiffener EA
30 140

120
25

100
20

80
Frequency

Frequency
15

60

10
40

5
20

0 0
-0.015 -0.01 -0.005 0 0.005 0.01 0.015 0.02 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.5: Models predicting axial stiffness of C-shape stringer

Neural Network for Stiffener A Modified Linear Regression Model for Stiffener A
70 140

60 120

50 100

40 80
Frequency

Frequency

30 60

20 40

10 20

0 0
-0.02 -0.015 -0.01 -0.005 0 0.005 0.01 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.6: Models predicting cross-sectional area of C-shape stringer

4.2. Sandwich panels

Surrogate models for sandwich panels were made predicting weight, face thickness, core thickness and
face Young modulus in the panel length direction. As with the previous models, the neural networks in
Figure 4.7a, Figure 4.8a, Figure 4.9a and Figure 4.10a perform well, with a small residual and evenly
distributed around zero.
The least-squares regression model in Figure 4.7b for the sandwich panel weight performs slightly
worse than the neural network, with a larger spread, but still evenly around zero. The model predicting
face thickness performs almost identical to its neural network counterpart, see Figure 4.8b. The core
thickness, on the other hand, is best approximated by the neural network. Finally, Figure 4.10b shows
an almost identical performance as Figure 4.10a for the facing Young’s modulus.
The linear regression models obtained for sandwich panels are shown in Table 4.5. Several interesting
observations can be made from the coefficients. It is again noted that although the predictors and
responses are non-dimensional, the original ranges of the predictor variables have an influence on the
magnitude of the corresponding coefficients. With that in mind, 𝑁 can be seen as the most sizing load.
For the weight, its products with length and width form the largest coefficients. In the thickness model
its first and second order terms are the most influential coefficients. For the core thickness and Young’s
modulus models these effects are observed again. Secondly, the area 𝐿 ∶ 𝑊 plays an important role
40 4. Regression Analysis

Neural Network for Sandwich Panel Weight Modified Linear Regression Model for Sandwich Panel Weight
80 200

180
70

160
60
140

50
120
Frequency

Frequency
40 100

80
30

60
20
40

10
20

0 0
-0.06 -0.04 -0.02 0 0.02 0.04 0.06 -0.1 -0.05 0 0.05 0.1 0.15
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.7: Models predicting weight of sandwich panel

Neural Network for Sandwich t f Modified Linear Regression Model for Sandwich tf
45 150

40

35

30 100
Frequency

Frequency

25

20

15 50

10

0 0
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 -0.35 -0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.8: Models predicting of sandwich panel

in the panel weight and core thickness. Thirdly, especially for core thickness, width seems more sizing
than length. This is surprising when considering the actual values for length are much larger than the
width (see Table 4.1). Apparently, changing the length of a long panel does not have a large effect
on the weight and dimensions, but widening such a panel leads to early failure, driving the weight and
thickness.
4.2. Sandwich panels 41

Table 4.5: Regression models for sandwich panel

Weight
Predictor Coefficient
Core Thickness
(Intercept) 0.098285
Predictor Coefficient
𝐿 -0.14894
(Intercept) -0.032037
𝑊 -0.21669
𝐿 0.19221
𝑁 -0.049661
𝑊 -0.30578
𝑁 -0.15907
Face Thickness 𝑁 0.13566
𝑁 -0.026317
Predictor Coefficient 𝑁 0.27082
𝑝 -0.021847
(Intercept) -0.032381 𝑁 0.073194
𝐿∶𝑊 0.49268
𝑊 -0.021734 𝑝 0.071702
𝐿∶𝑁 0.065658
𝑁 0.015829 𝐿∶𝑊 0.18115
𝐿∶𝑁 0.30703
𝑁 0.45328 𝐿∶𝑁 -0.052638
𝐿∶𝑁 0.035387
𝑁 -0.017122 𝐿∶𝑁 -0.074424
𝐿∶𝑝 0.043546
𝑊∶𝑁 0.036574 𝐿∶𝑝 0.087353
𝑊∶𝑁 0.047226
𝑁 ∶𝑁 -0.12726 𝑊∶𝑁 -0.39912
𝑊∶𝑁 0.22087
𝑁 ∶𝑁 0.078797 𝑊∶𝑝 0.54994
𝑊∶𝑁 0.035151
𝑁 0.17519 𝑁 ∶𝑁 -0.060692
𝑊∶𝑝 0.054924
𝑁 0.24541 𝑁 ∶𝑝 -0.22727
𝑁 ∶𝑁 0.01512
𝑁 0.1198 𝐿 -0.2216
𝑁 ∶𝑁 -0.024174
RMSE 0.0462 𝑊 0.46685
𝑁 ∶𝑁 0.024123
R-squared 0.962 𝑁 -0.053974
𝑁 ∶𝑝 -0.019371
𝑁 0.087076
𝑊 0.059175
𝑁 -0.091354
𝑁 0.027342
𝑝 -0.10567
𝑁 0.045483
RMSE 0.0622
𝑁 0.016449
R-squared 0.832
RMSE 0.0206
R-squared 0.98
E-modulus
Predictor Coefficient
(Intercept) 0.70905
𝐿 -0.037006
𝑊 0.062763
𝑁 0.021223
𝑁 -1.2989
𝑁 -0.30791
𝐿∶𝑁 0.070033
𝑁 ∶𝑁 -0.062684
𝑁 ∶𝑁 0.4013
𝑊 -0.058873
𝑁 0.74196
RMSE 0.0876
R-squared 0.713
42 4. Regression Analysis

Neural Network for Sandwich t c Modified Linear Regression Model for Sandwich tc
160 300

140
250

120

200
100
Frequency

Frequency
80 150

60
100

40

50
20

0 0
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 -0.4 -0.2 0 0.2 0.4 0.6 0.8
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.9: Models predicting of sandwich panel

Neural Network for Sandwich Face Ef Modified Linear Regression Model for Sandwich Face Ef
100 300

90

250
80

70
200

60
Frequency

Frequency

50 150

40

100
30

20
50

10

0 0
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
Residual [-] Residual [-]

(a) Neural network (b) Linear regression model

Figure 4.10: Models predicting of sandwich panel

 5
Pressure Loading
One of the reasons for the oval fuselage concept is the challenge of dealing with internal pressure in a
non-cylindrical fuselage. For the 2D cross-section, the assumption that each arc carries the pressure
load in membrane forces seems valid. The trapezoid structure can take on any difference in membrane
forces between arcs and retain the section’s shape. However, when considering a 3D geometry with
the oval cross-section, it gets complicated, because the radii of the arcs can change in the longitudinal
direction, which makes the shape in that direction non-circular nor straight. The aim of this chapter is
to investigate the validity of the 2D assumption in the 3D case.

5.1. Shell of Arbitrary Shape

Flügge discusses in the book ”Stresses in Shells” [25] how loads are distributed in shells. Not only
shells of revolution are discussed, which generally have an analytical solution for the stresses due to
internal pressure, but also a method is derived for determining the stresses in a shell of arbitrary shape.

𝑁̄
𝑁̄
𝑁̄
𝑥 𝑁̄

𝑁̄ 𝑦
𝑁̄
𝑁̄ 𝑁̄

𝑁 𝑁
𝑁

𝑁 𝑁
𝑁 𝑁

Figure 5.1: Arbitrary shell element

Consider an element of any shell, as depicted in Figure 5.1. The element is obtained by cutting the shell
in the longitudinal and transverse directions. Hence, when projecting the element on the x-y plane, a
rectangular element is obtained. On this transformed domain, a stress function Φ is defined which is
described as follows:

𝜕 Φ𝜕 𝑧 𝜕 Φ 𝜕 𝑧 𝜕 Φ𝜕 𝑧 𝜕𝑧 𝜕𝑧 𝜕 𝑧 𝜕 𝑧
−2 + = −𝑝̄ + 𝑝̄ + 𝑝̄ + ∫ 𝑝̄ 𝑑𝑥 + ∫ 𝑝̄ 𝑑𝑦 (5.1)
𝜕𝑥 𝜕𝑦 𝜕𝑥𝜕𝑦 𝜕𝑥𝜕𝑦 𝜕𝑦 𝜕𝑥 𝜕𝑥 𝜕𝑦 𝜕𝑥 𝜕𝑦

43
44 5. Pressure Loading

Furthermore:

𝜕 Φ
𝑁̄ = − ∫ 𝑝̄ d𝑥 (5.2)
𝜕𝑦
𝜕 Φ
𝑁̄ = − ∫ 𝑝̄ d𝑦 (5.3)
𝜕𝑥
𝜕 Φ
𝑁̄ = − (5.4)
𝜕𝑥𝜕𝑦

Transforming the projected values (with a bar) to the real values in the shell, the following transforma-
tions are used:

cos 𝜃
𝑁 = 𝑁̄ (5.5)
cos 𝜒
cos 𝜒
𝑁 = 𝑁̄ (5.6)
cos 𝜃
𝑁 = 𝑁̄ (5.7)

Finally, for the transformed applied loads the following holds:

𝑝̄ 𝑝̄ 𝑝̄ d𝐴 √1 − sin 𝜒 sin 𝜃
= = = = (5.8)
𝑝 𝑝 𝑝 d𝑥d𝑦 cos 𝜒 cos 𝜃

5.2. Discretization
Obviously, Equation 5.1 is a partial differential equation (PDE) in two variables, and therefore obtaining
an exact solution is complicated if not impossible. However, numerical techniques can be used to solve
such an equation.
The PDE in Equation 5.1 is differentiable in two variables: x and y. Consider an 𝑚×𝑛 grid of coordinates
as shown in Figure 5.2, formed by x-y pairs, and on each coordinate both a z position is defined (the
geometry) and the 𝑁 and 𝑝 loads. Then the derivatives of all relevant parameters can be determined
using finite differences and Equation 5.1 becomes:

1 𝜕 𝑧
(Φ , − 2Φ , +Φ , )( ) −
𝑑 𝜕𝑦 ,
1 𝜕 𝑧
(Φ , −Φ , −Φ , +Φ , )( ) +
2𝑑𝑒 𝜕𝑥𝜕𝑦 ,
1 𝜕 𝑧
(Φ , − 2Φ , +Φ , )( ) = (5.9)
𝑒 𝜕𝑥 ,
𝜕𝑧 𝜕𝑧
−(𝑝̄ ) , + (𝑝̄ ) , ( ) + (𝑝̄ ) , ( ) +
𝜕𝑥 ,
𝜕𝑦 ,
𝜕 𝑧 𝜕 𝑧
( ) ⋅ (𝑝̄ ) , ⋅𝑑+( ) ⋅ (𝑝̄ ) , ⋅𝑒
𝜕𝑥 ,
𝜕𝑦 ,

Equation 5.9 was derived using central differencing. On the edges and corners of the grid, however,
such a finite differencing scheme can not be used, because it would refer to non-existent grid points.
Figure 5.2 also shows the boundary conditions used on the grid boundaries. The stress function is put
to zero at all points outside the modeled fomain. Such a boundary condition is indicated by Flügge as
being valid [25], since the stress function should be constant along the boundary when the shape is
symmetric in at least one direction and may attain an arbitrary value.
5.3. Solution 45

0 0 0

0 1,1 d 1,2 1,m 0

0 2,1 2,2 0

0 n,1 n,m 0

0 0 0

Figure 5.2: Grid for finite difference method

5.3. Solution
As was presented in the previous section, for each grid point the stress function can be expressed as
a function of itself and neighboring points. By putting all function values at the grid points in a column
vector, a 𝑚 × 𝑛,𝑚 × 𝑛 matrix A can be constructed which contains the coefficients with which each
stress value Φ is multiplied, i.e. each row in this matrix represents Equation 5.9 for one grid point.
Since Equation 5.9 also contains constant terms, independent of any stress function value, these can
be put in another column vector C. Performing these operations allows Equation 5.9 to be written for
all grid points as a system of linear equations:
A⋅Φ=C (5.10)
This system is solved using Gaussian Elimination with pivoting.
In order to verify the implemented approach, a circular cylinder was analyzed, for which analytical
solutions exist. The hoop stress in a circular cylinder is given by Equation 2.20 and the longitudinal
stress by Equation 2.21. Multiplying these equations by the thickness results in a line load in N/m,
which is the load that can be obtained for a membrane solution as presented above.
Modeling a cylinder with 𝑟 = 3 m and 𝑝 = 45 kPa, the hoop line load should be 135 kN/m, and the
longitudinal line load 77.5 kN/m. As can be seen in Figure 5.3b, the hoop load indeed approaches
the analytical value, but the longitudinal load (Figure 5.3a) is not constant and does not approach the
analytical value. Performing the analysis with a longer cylinder (but same grid size) does not improve
the situation: the longitudinal load increases, while retaining a nonlinear, symmetrical distribution along
the cylinder.
With a coarse grid, the hoop load is close to the analytical value, but with a finer grid the value is better
approximated. Similarly, the longitudinal load does not change greatly when refining the grid. On the
contrary, computation time increases drastically with a finer grid.

5.4. Effect of Oval Shape in Three Dimensions

Even though the longitudinal load is incorrectly determined by the finite difference method when com-
pared to the analytical result for a circular cylinder, it might be interesting to see the effects of curvature
in the longitudinal direction as well. Figure 5.4a and Figure 5.4b show the longitudinal and hoop loads,
respectively. At the pinched location much larger loads are observed than in the rest of the section,
and larger than the circular cylinder in Figure 5.3a and Figure 5.3b. However, these results are of
little significance because the longitudinal loads are possibly incorrect, but nonetheless point out that
the effect of curvature in all directions may affect pressurization loads significantly. It is therefore an
important subject for further study.
46 5. Pressure Loading

(a) Longitudinal load (b) Circumferential load

Figure 5.3: Pressure loads in circular cylinder

(a) Longitudinal load

(b) Circumferential load

Figure 5.4: Pressure loads in tapered cylinder

 6
Verification & Validation
Verification and validation of a fuselage weight estimation is not straightforward: many intermediate
calculations are performed and all their outcomes should be checked, because it may occur that the
final predicted weight is within an acceptable margin from the real weight, while the skin is sized too
thin, the stringers too large and the empirical estimation for bulkheads compensates for the two, for
example. However, detailed information of the actual fuselage structure dimensions are hard to obtain,
which also holds true for the loads acting on the structure. As such, those aspects of the presented
method that could be verified and/or validated are discussed in this chapter, but it is noted beforehand
that this excludes some parts of the method.
In the order of procedures performed by the weight estimation method, first the shear and moment dis-
tribution calculations are verified for an Airbus A320-200. Consecutively, the boom method is verified.
Sizing of the monolithic panels and stringers was verified in Ref. [22] to produce the same result as
a computationally more expensive genetic optimization algorithm. However, since a pressure load
was added for the present study, and sandwich panels were implemented, FEA analyses have been
performed on these to validate the correct failure point. Finally, the weight estimation itself is verified.

6.1. Applied Loads Verification

For an Airbus A320-200, the shear forces and bending moments acting on the fuselage are shown in
Figure 6.1a and Figure 6.1b from Van Tooren [13], respectively. Running the Initiator with an input file
resembling the A320 gives a shear force and moment distribution similar to those presented by Van
Tooren, as can be seen in Figure 6.2a and Figure 6.2b. Please note that these latter two distributions
only include the portion of the fuselage acting as passenger cabin, and therefore the nose and tail
sections are not shown. The shear force magnitude is similar for both plots, but the shape differs quite
significantly. Conversely, the bending moment diagram shape is similar between both distributions
(except the Initiator features a larger jump at the wing root), but the magnitude is several orders of
magnitude off. It is unknown what the original source of Figure 6.1a and Figure 6.1b is, and therefore
care should be taken to draw conclusions from these results. Nonetheless, the load cases implemented
in the Initiator may need revision.

6.2. Pressure Load Validation

FEA of an oval section with internal pressurization was performed using Abaqus. In order to model the
pressure loads, symmetry boundary conditions were applied to the front and rear edges of the section,
as well as the edges in the symmetry plane. A pressure of 45 kPa was applied and the Poisson’s
ratio of the material was set to 0.5, modeling a perfectly incompressible material. This way no load
alleviation takes place due to contraction of the section in the longitudinal direction.
The pressure loads in longitudinal and hoop direction are shown in Figure 6.3a and Figure 6.3b, respec-
tively. The exact values from the analytical method and an approximate average of the FEA results
are shown in Table 6.1. It can be seen that both the hoop and longitudinal stresses are approximated
accurately. On a critical note, though, the presented section is straight, i.e. the cross-section does not

47
48 6. Verification & Validation

(a) Shear force envelope in A320-200 [13]

(b) Bending moment envelope in A320-200 [13]

Figure 6.1: A320 internal loads from literature

change in longitudinal direction. Therefore the analytical solution for pressurization stress holds. Fur-
ther research has to point out whether such good agreement is maintained when the fuselage shape
is more complex (i.e. changes in longitudinal direction).

Table 6.1: Comparison of FEA and analytical hoop pressure stress

Member FEA [MPa] Analytical [MPa] Error

Arc 1 35.4 35.3 -0.28%
Arc 2 10.5 10.5 0%
Arc 3 26.8 26.7 -0.37%
Ceiling -10.9 -11.0 -0.92%
Wall 8.00 8.05 0.63%
Floor -7.2 -7.17 0.42%

6.3. Boom Method Verification and Validation

The implementation of the boom method is verified by examining the direct stress resulting from the
applied bending moment and the shear flow resulting from the applied shear forces. Because a sym-
metrical cross-section was assumed and all loads are applied in the axis of symmetry, it is expected
that both the stress and shearflow distributions along the cross-section are symmetrical (for stress ex-
actly equal, for shear flow equal magnitude, opposite sign). Moreover, the shear flow becomes zero at
exactly the top and bottom of the cross-section and has a maximum in between. The direct stress is
6.3. Boom Method Verification and Validation 49

(a) Shear force envelope in A320-200 from Initiator

(b) Bending moment envelope in A320-200 from Initiator

Figure 6.2: A320 internal loads from Initiator

maximum at the top or bottom of the fuselage.

To test above statements, an arbitrary oval cross-section was created and a fuselage section of ap-
proximately 5 m was created. A constant shear force of 998,800 N and constant bending moment of
9,537,600 Nm were applied, where the bending moment is oriented such that the upper side of the fuse-
lage is in compression (which is not necessarily an actual load case). The resulting boom stresses and
shear stress computed by the boom method are shown in Figure 6.4a and Figure 6.5a, respectively.
As can be seen from Figure 6.4a, the stress is indeed negative in the upper portion of the fuselage
section and its positive maximum is exactly at the bottom. However, the minimum is located at the
junction of the top and side arc (and ceiling and wall). The stress shown in this figure was scaled to
the smeared stringer thickness (for comparison with FEA, presented hereafter), which is not equal to
the boom stress. Because the boom areas include the skin and those in the side arc also the wall, the
boom stress is different from the actual stress in the (smeared) stringers. This also explains the jump
in direct stress observed in Figure 6.4a and why the minimum stress is not observed in the top of the
section.
Figure 6.4b and Figure 6.5b show the same fuselage section as discussed earlier, but now modeled
using Abaqus. No stringers were modeled; instead the smeared stringer thickness was used to model
50 6. Verification & Validation

(a) Longitudinal stress in oval section (b) Hoop stress in oval section

Figure 6.3: Pressurization loads from Abaqus

(a) Initiator (b) Abaqus

Figure 6.4: Direct stress comparison

the stringers as a skin. Even with this approximation and the difference that in the FEA model a trape-
zoid structure is present, which is omitted in the boom method, good agreement is reached between
the boom method and FEA. As expected, the signs of both direct and shear stress are identical for both
methods. More interestingly, the magnitude of both stresses is also similar for both methods, giving
confidence in the implemented boom method.

6.4. Panel Failure Validation

Of major importance are the sizing procedures implemented for stringers, panels and sandwich panels.
Two aspects of the procedure needed to be verified and validated: firstly, the analytical failure criteria
should be compared with FEA and secondly, the deterministic optimization should be verified using a
more robust (but more computationally expensive) optimizer.
The shear buckling analysis implemented was checked with FEA results. Three different plate dimen-
sions, representing three different aspect ratios, were used and for each multiple shear loads were
applied. First the analytical method was run to optimize a certain laminate for that specific panel and
load combination. Secondly, that panel with the optimized laminate was modeled in Abaqus and a
buckling analysis was performed to see whether the failure modes match. The boundary conditions in
Abaqus resemble a simply supported plate, which was also assumed in the analytical methods.
In Figure 6.6b and Figure 6.6c it can be seen that fairly good agreement is achieved between the
analytical method and FEA. The error ranges from 5% to 20% and from 13% to 66%, respectively.
6.4. Panel Failure Validation 51

(b) Abaqus
(a) Initiator

Figure 6.5: Shear stress comparison

Table 6.2: Comparison of Tsai-Wu failure criterion

𝑎 𝑏 𝑁 𝑁 Layup Analytical FEA Difference

[m] [m] [kN/m] [kN/m] Tsai-Wu [-] Tsai-Wu [-] [%]
0.5 0.3 250 500 [±45/90 /0/90 /±45/0]̄ s 0.37 0.44 -17
0.5 0.5 500 500 [±45/90 /0/90 /±45/90 /0/90/±45/0]̄ s 0.50 0.56 -10
0.5 0.3 100 100 [±45/90/0]s 0.29 0.36 -20
0.3 0.3 200 200 [±45/90/0]s 0.66 0.72 -8.8
0.3 0.3 200 0 [0]s 0.55 0.76 -29
0.3 0.3 0 200 [90]s 0.55 0.76 -29
0.3 0.3 200 200 [(±45) ]s 0.66 0.72 -8.8

The square panel shear buckling failure as obtained from Figure 6.6a shows a large error which is
approximately constant around 55 %. Note that the analytical results in Figure 6.6a and Figure 6.6b
were obtained using an empirical relation, while Figure 6.6c was computed using a semi-analytical
relation, as was explained in subsection 3.2.1.
The difference in Figure 6.6a is large, but a conservative result is obtained since a lower critical load is
predicted by the analytical approach. Additionally, panels of this aspect ratio are rarely found in fuselage
structures. However, the results from Figure 6.6c are applicable to panels in fuselage structures and
these results are non-conservative. In order to account for this, the estimated critical shear buckling
load may be scaled to more closely represent the FEA data. A scaling factor obtained from Figure 6.6c
is provided by:
𝑅 = 0.88846 ⋅ 𝑁 . (6.1)

This factor should be multiplied by the analytical result to obtain the FEA result. Note that the unit for
𝑁 is kN/m. Obviously, to obtain a more reliable scaling factor additional cases have to be run.
Additionally, the maximum deflection of panels under transverse pressure was compared with FEA.
A pressure of 60 kPa was applied, with panels of length 0.8 m and varying widths. The results are
shown in Figure 6.7, where good agreement is observed between the analytical routine and FEA. The
maximum error is around 4%.
A final test is required to conclude this validation, namely checking whether the Tsai-Wu failure crite-
rion as computed by the deterministic approach matches the one computed from FEA. By doing this,
computation of the internal stresses is compared, albeit not directly.
Several panel size and load cases were run and the results are shown in Table 6.2. Each panel from
the ALGO program was modeled using Abaqus. The first three cases were optimized by ALGO based
on dimensions and applied loads, while all other laminates were pre-defined as additional test cases.
The Tsai-Wu failure values were checked at each ply and the highest values present in the laminate
are shown here. On a final note, 0̄ means it is a midply, i.e. is not repeated in the symmetry pattern.
52 6. Verification & Validation

Analytical versus FEA Shear Buckling

Panel 0.75 x 0.75 m
7000

Critical Shear Load, Nxy,crit, [kN/m]

6000

5000

4000

3000 Analytical
FEA Mode 1
2000

1000

0
150 300 450 600 750 900 1050 1200 1350 1500
Applied Shear Load, Nxy, [kN/m]

(a) Panel shear buckling failure for /

Analytical versus FEA Shear Buckling
Panel 0.6 x 0.8 m
4000
Critical Shear Load, Nxy,crit, [kN/m]

3500

3000

2500

2000
Analytical
1500 FEA Mode 1

1000

500

0
150 300 450 600 750 900 1050 1200 1350 1500
Applied Shear Load, Nxy, [kN/m]

(b) Panel shear buckling failure for / .

Analytical versus FEA Shear Buckling
Panel 0.3 x 0.8 m
14000
Critical Shear Load, Nxy,crit, [kN/m]

12000

10000

8000

6000 Analytical
FEA Mode 1
4000

2000

0
150 300 450 600 750 900 1050 1200 1350 1500
Applied Shear Load, Nxy, [kN/m]

(c) Panel shear buckling failure for / .

Figure 6.6: Validation results for panel shear buckling

6.5. Verification of Optimization using GA 53

Figure 6.7: Panel pressure deflection

From Table 6.2 it is clear that the analytical calculation under-predicts material failure. This means that
a non-conservative design is obtained. However, to put this error into perspective, take the third panel
from Table 6.2. When a 0 ply is added (in symmetry, so 2 plies in total), the FEA shows a Tsai-Wu
failure value of 0.3054, which is slightly above the value obtained with the deterministic approach for
the original panel. Considering an A320 fuselage, with 335 m skin area, this results in a mass penalty
of 164 kg, or 1.84 % of the total fuselage mass.

6.5. Verification of Optimization using GA

In order to verify that the deterministic optimization routine arrives at the lightest panel possible for
the required dimensions and applied loads, 150 random combinations of dimensions and loads were
generated and both the analytical procedure and a genetic algorithm were run to find the most suitable
laminate.
The genetic algorithm was set up with a design vector where the first entry denotes the ply angle of
the midply, or the absence of a midply. The next 𝑛 entries denote an amount of plies, where 𝑛 can
range from 0 to 5 (the maximum amount of identical, consecutive plies). The last 𝑛 entries denote the
ply angles. As a final note, the stacking sequence defined this way is made symmetrical around the
midply. To illustrate, suppose 𝑛 = 30 and there is a midply. Then the maximum number of plies in the
laminate is 301 (= 2 × (30 × 5) + 1). The same failure analysis is used for both GA and deterministic
optimization and also a laminate validity check is performed in the GA constraint function, ensuring,
among other things, the 10% rule and maximum ply stack.
For the skin panels, the weight difference between GA and ALGO is shown in Figure 6.8. The exact
same histogram is obtained for the panel thickness, obviously. The difference was computed by dividing
the GA weight by the ALGO result, such that a positive difference indicates that ALGO outperforms the
GA and a negative difference indicates that the GA outperforms ALGO.
Most designs ( 90%) optimized by the GA are significantly lighter than the ALGO routine. A possible ex-
planation is that the deterministic approach might jump from one critical failure mode to the next, adding
plies in different directions every time. This might lead to a less efficient laminate than the one where all
ply angles are optimized simultaneously. It should be noted, however, that the deterministic approach
takes slightly more than a minute to complete 100 designs, while the GA requires approximately 10
hours.
For the stringers, just like with the deterministic approach, a family of laminates is generated beforehand
and then a combination of laminates is taken and the flange widths are optimized. Therefore, the GA
design vector now only contains six variables: three index variables pointing to a laminate in the family
for each member and the three member widths. The results are shown in Figure 6.9, where an excellent
similarity is seen between GA and ELFO. All designs have a positive difference between GA and ELFO,
meaning that the GA always finds a heavier design than ELFO. This is likely a result of the fact that
ELFO adopts a brute force approach (it analyzes all possible designs and chooses the lightest option).
Finally, the sandwich panel optimization was also compared to its GA counterpart. The results are
54 6. Verification & Validation

GA Weight versus ALGO Weight

30

25

20

Number of designs
15

10

0
-50 -40 -30 -20 -10 0 10 20 30
Difference [%]

Figure 6.8: Genetic algorithm versus deterministic approach for skin panels

GA Weight versus ELFO Weight

120

100
Number of designs

80

60

40

20

0
-10 0 10 20 30 40 50 60 70
Difference [%]

Figure 6.9: GA versus ELFO weight

shown in Figure 6.10a for the weight, Figure 6.10b for the facing thickness and Figure 6.10c for the core
thickness. It becomes immediately clear that the difference is much larger between the deterministic
approach and the genetic algorithm than for the monolithic panels and stringers. The exact cause for
this is unclear, but the large difference happens on the positive side of the scale, meaning that there
the genetic algorithm performs worse than the deterministic approach. Then, the error may arise from
the amount of plies used in the design vector for the GA. With too few design variables it may be
impossible to find a suitable laminate, while too many design variables make the GA optimize mainly
the face laminate, while the core thickness is found to have less effect. One can see this in Figure 6.10c,
where the GA has a thinner core for nearly all designs. This may partially explain the lighter designs
from the deterministic approach.

6.6. Mass Estimation Verification

Combining all elements from the method presented so far, the weight estimation method should be
verified as a whole. Two major problems arise though: no data is available on the weight of composite
fuselages (the Airbus A350 and Boeing 787 are the only passenger aircraft currently with such a fuse-
lage). Secondly, the oval fuselage is still a concept and has never been used in a real-life application.
Therefore, verification is performed on several conventional aircraft, with metallic fuselages.
A generic weight breakdown for metallic, conventional fuselages is given in Figure 6.11 [26], which is
used as a reference for the breakdowns provided by the implemented method. It should also be noted
that the aircraft presented here are not the actual aircraft, but approximations, because the Initiator
designs these aircraft based on a collection of settings and top-level requirements. Therefore, the
6.6. Mass Estimation Verification 55

GA Weight versus ALGO Weight

20

18

16

14

Number of designs
12

10

0
-50 0 50 100 150 200 250
Difference [%]

(a) GA versus ALGO weight

GA Face Thickness versus ALGO Face Thickness
25

20
Number of designs

15

10

0
-100 0 100 200 300 400 500
Difference [%]

(b) GA versus ALGO face thickness

GA Core Thickness versus ALGO Core Thickness
35

30

25
Number of designs

20

15

10

0
-100 -50 0 50 100 150 200
Difference [%]

(c) GA versus ALGO core thickness

Figure 6.10: Genetic algorithm versus deterministic approach for sandwich panels

masses computed are not compared to the actual aircraft, but rather the fuselage mass fraction and
difference with Torenbeek’s empirical method are compared with.
Table 6.3 shows the fuselage mass as predicted by the Initiator compared to the empirical relation
from Torenbeek [27]. Additionally, the MTOM from the Initiator and literature are provided. Finally,
the fuselage fraction of total mass is given for both the Initiator and reality. Although several aircraft
shown in Table 6.3 have a large error between the Initiator mass prediction and the empirical mass,
56 6. Verification & Validation

Windows: 8%
Shell (skin, stiffeners, frames): 43%

Bulkheads: 10%

Door Assembly: 11%

Floor Assemblies: 12%

Keel, Wheelwell: 16%

Figure 6.11: Generic weight breakdown for large passenger aircraft

Table 6.3: Estimated Fuselage Mass Comparison

fus,Ini TO,Ini fus,Ini fus,Tor

Aircraft 𝑀fus,Ini 𝑀fus,Tor −1 𝑀TO,Ini 𝑀TO,Act −1
fus,Tor TO,Act TO,Ini TO,Act
[kg] [kg] [%] [kg] [kg] [%] [%] [%]
Airbus A320-200 9,655 8,798 9.74 86,435 73,500 17.60 11.17 11.97
Boeing 727-200 8,936 10,994 -18.72 62,133 83,800 -25.86 14.38 13.12
Boeing 767-300ER 28,273 19,791 42.86 261,270 186,880 39.81 10.82 10.59
Boeing 747-100 49,873 39,015 27.83 360,300 333,400 8.07 13.84 11.70
Airbus A340-300 34,989 28,570 22.47 279,520 276,500 1.09 12.52 10.33

the fuselage fraction is very similar for all aircraft considered. This means that the error may arise
from other predictions performed by the Initiator (e.g. dimensions and MTOM), but that the fuselage
mass estimation provides an acceptable result despite of this. The structural weight breakdowns of the
Airbus A320-200 and Boeing 767-300ER are shown in Figure 6.12a and Figure 6.12b, respectively.
Comparing to the generic breakdown in Figure 6.11, several observations are made. The stringer and
frame mass seem to be sensitive to fuselage size. The Boeing 767-300ER has a longer fuselage with
larger radius than an A320-200. The loads around the wing root increase significantly and therefore,
as can be seen in Figure 6.13b, the stringer pitch reduces drastically in this area. The outlier frames
were removed from the mass estimation, because the jump in shear flow in these sections results in
unrealistically sized frames. However, still a large fraction of the fuselage mass consists of frames.
Compared to Figure 6.11, the shell mass (skin, stringers and frames) is over-estimated by the current
method (note that the nose and aft shell masses should be added as well). Another observation is that
the (pressure) bulkhead mass is underestimated quite significantly, if the bulkhead mass in Figure 6.11
indeed only consists of pressure bulkheads. If also frames where the landing gear and/or wing are
attached count as bulkheads, part of this mass is accounted for in the frames mass from the Initiator.
The floor mass appears to be overestimated in the present method, but note that the cargo floor is
included here. When the cargo floor is accounted for in the keel portion in Figure 6.11, this may again
cancel each other out.

6.7. Sizing Verification

Returning to Figure 6.13a and Figure 6.13b, it may be concluded that sizing of the metal structure was
done as expected. In the area where the wing is mounted and largest compression is found in the lower
portion of the fuselage, the stringer pitch is small and many stringers are present. Additionally, the skin
panels are thicker there and the thickest frames are observed there as well. At the same position, but
in the upper portion of the fuselage, a larger stringer smeared thickness is found, which is necessary
6.7. Sizing Verification 57

PressureBulkheads:4% PressureBulkheads:2% NoseShell:2%

NoseShell:5%
Stringers:18%
Floors:15%
Stringers:29%

Floors:18%

Windows:5%
Skin:11%

Doors:7%

AftShell:5%
Windows:11%
Skin:12%

Frames:18% Frames:21%
Doors:10% AftShell:8%

(a) Airbus A320-200 (b) Boeing 767-300ER

Figure 6.12: Weight breakdowns from Initiator for aluminium fuselage

to cope with the high tensile loads. Finally, the smallest stringers are observed in the side arc, where
direct stresses are smallest (closest to the neutral axis).
The location around the wing root is heavily reinforced with thick skins and stringers. Normally, a cutout
in the fuselage is present there, allowing for the wing box to carry through. It is a matter of bookkeeping
whether this carry-through structure is part of the fuselage structure mass or wing structure mass.
Additionally, the skins and stringers sized in this area may be seen as a replacement for a keel beam.
However, further study would have to point out if this yields an acceptable prediction.
58 6. Verification & Validation

Skin thickness [mm]

3

Z position [m]
2 2.5
2
0
1.5
1
-2
5 10 15 20 25 30
X position [m]
Stringer smeared thickness [mm] 5
Z position [m]

4
2
3
0 2

1
-2
5 10 15 20 25 30
X position [m]
Frame smeared thickness [mm]
80
Z position [m]

2 60

40
0
20
-2
5 10 15 20 25 30
X position [m]

(a) Airbus A320-200

(b) Boeing 767-300ER

Figure 6.13: Thickness distributions for metal fuselages from Initiator

 7
Results
This chapter focuses on showing some case studies of composite and oval fuselages, which do not
exist in reality. The purpose is to get an idea of how the weight estimation method performs in these
cases and if reasonable results are obtained. Moreover, it proves whether the method actually works
for its intended goal.
The figures showing thickness distributions, shear flow and direct stress only cover the part of a fuselage
where the passenger cabin is present. The rest of the fuselage is not modeled with the analytical
methods and the mass of the front and aft shell are computed based on their wetted area and the
average smeared thickness of skins, stringers and frames combined in the side arc.

7.1. Composite A320-200 and Boeing 767-300ER

Using the exact same input files as for a standard A320-200 and B767-300ER, but with aluminium
replaced by the carbon fiber composite with the properties from Table 4.2, a lighter fuselage is expected.
Moreover, with composites, one expects to find thicker skins and stringers.
Looking at the thicknesses in Figure 7.1 and Figure 7.2, the expectation of thicker skins and other
structural components is met. The frame thickness follows the same trend as for metal fuselages,
which may be expected considering the method used; only the bending moment and shear flow size
the frames. The skin thickness distribution looks similar for both aircraft and also to the aluminium skin
thickness distribution.
The stringer pitch distribution is also similar for both aircraft and similar to the metal distribution: in the
bottom arc around the wing root the smallest stringer pitch is observed, along with the largest stringer
smeared thickness. Again this is as expected, since the largest compression forces occur there. Along
the side arc, least and smallest stringers are present and some thicker stringers are present in the top
arc to cope with the high tensile loads there.
The weight breakdown for the composite A320 fuselage in Figure 7.3a shows the same trend as was
observed in the NASA ACT papers (see section 2.8): the largest portion of the primary structure is
taken up by skins and stringers and the frames form the other main portion of weight. Figure 7.3b
shows approximately the same breakdown. Apparently, the stringer weight scales exponentially with
the compression load (as was also observed from the linear regression coefficients) since the stringer
mass becomes the main portion of fuselage mass. The large Boeing 767 fuselage has to cope with
high bending loads, due to the large lift and mass of the aircraft. A remark is in order here, for actual
fuselages have a cutout in this region to allow the wingbox to run through. Then it is a matter of
bookkeeping whether the wingbox carry-through structure is part of the fuselage or wing mass and in
case of the former option, whether the stringers, skins and frames sized in this region may account for
the carry-through structure. The NASA results showed that skins form most of the primary structure,
so it is likely that assuming skins take no direct stresses results in this underestimation. However, the
stringer weight makes up for this, such that the weight estimation itself is not thwarted too much by this.
An interesting observation is the large cargo floor mass in Figure 7.3a and Figure 7.3b. This mass
was obtained as a fraction of the bottom arc primary structure mass (skins, stringers and frames). It is
likely that the bottom arc is estimated larger in the Initiator’s geometry sizing than the keel panels are

59
60 7. Results

Skin thickness [mm]

3.6

Z position [m]
3.4
2
3.2

0 3
2.8

-2 2.6
5 10 15 20 25 30
X position [m]
Stringer smeared thickness [mm]
10
Z position [m]

2 8

6
0
4

-2 2
5 10 15 20 25 30
X position [m]
Frame smeared thickness [mm]
80
Z position [m]

2 60

40
0
20
-2
5 10 15 20 25 30
X position [m]

Figure 7.1: Composite Airbus A320-200 thickness distribution

Table 7.1: Estimated Composite Fuselage Mass

Aircraft Mass [kg] Weight Saving [%]

Airbus A320-200 7,790 19
Boeing 767-300ER 25,334 10
Boeing 727-200 7,584 15

in actual aircraft. Logically, the cargo floor mass is then sized inappropriately as well.
Finally, the composite fuselage mass is 7,790 kg and 25,334 kg for the A320 and Boeing 767, respec-
tively, as can be seen in Table 7.1. The Boeing 727-200 fuselage is also included here for a better
comparison of the weight saving obtained with composite material. Comparing the composite fuse-
lage masses with the aluminium masses in Table 6.3, a weight saving of (on average) around 15 % is
observed.
7.1. Composite A320-200 and Boeing 767-300ER 61

Figure 7.2: Composite Boeing 767-300ER thickness distribution

Frames:11% Cargofloor:11% Frames:11% Cargofloor:14%

AftShell:6% Cutouts:3%
AftShell:6% Cutouts:2%

SplicingAssembly:6%
NoseShell:5% SplicingAssembly:7%
NoseShell:9%

Floor:4%

Floor:6%

Stringers:24% Skin:16%

Stringers:35%
Skin:24%

(a) Composite Airbus A320-200 (b) Composite Boeing 767-300ER

Figure 7.3: Weight breakdowns for composite fuselages

62 7. Results

7.2. Oval Composite A320-200

Using the same dimensions as the conventional A320-200, but with a trapezoidal structure to obtain
the oval cross-section and with composite material, the thickness distribution in Figure 7.4 is obtained.
This distribution is almost identical to the one in Figure 7.1, except for that the skin thickness in the side
arc is reduced, because the wall structure takes up some of the shear flow. The weight breakdown in
Figure 7.5 is also almost identical to Figure 7.3a, but obviously a wall and ceiling are present. These
are only slightly lighter than the passenger floor. Again the cargo floor forms a significant portion of the
fuselage mass. The total fuselage mass was increased to 8,068 kg, which is a 3.6 % mass increase
compared to the (conventional) composite fuselage.

Skin thickness [mm]

3.8
Z position [m]

2 3.6
3.4
0
3.2
-2
5 10 15 20 25 30
X position [m]
Stringer smeared thickness [mm]
Z position [m]

10
2 8
6
0
4
2
-2
5 10 15 20 25 30
X position [m]
Frame smeared thickness [mm]
Z position [m]

80
2 60
40
0
20
-2
5 10 15 20 25 30
X position [m]

Figure 7.4: Oval, composite Airbus A320-200 thickness distribution

7.3. Wide Oval Boeing 767-300ER 63

Frames:10% Cargofloor:11%

AftShell:5% Cutouts:3%

SplicingAssembly:6%
NoseShell:9%

Ceiling:3%
Wall:2%

Floor:3%
Stringers:23%

Skin:25%

Figure 7.5: Oval, composite Airbus A320-200 weight breakdown

7.3. Wide Oval Boeing 767-300ER

By fitting in slightly more passengers, in a wider configuration and with the fuselage set to an oval con-
figuration, an extra wide Boeing 767 was created. For this configuration, both aluminium and composite
material were analysed consecutively and the results are presented and compared in this section.
As expected, the oval fuselage shows similar thickness distribution trends as the conventional fuselage.
The composite fuselage in Figure 7.7 shows approximately identical skin thickness in the bottom and
upper arc, while the side arc has a lower thickness. The latter is a result of the wall accounting for some
of the shear flow. The upper and lower arc are probably mainly sized by pressure loads, which reduce
the influence of shear flow. The stringer smeared thickness shows logical results and again the frame
thickness is distributed just like in the conventional configurations.
The weight breakdown for the aluminium fuselage, see Figure 7.8a, shows that frames become a
significant portion of the fuselage mass, being more than the skin and stringers together. Apparently,
the oval shape induces large internal loads and moments in the frames. The weight breakdown for
the composite fuselage in Figure 7.8b shows the same trend, albeit the frames contribute less to the
total fuselage mass there. For both oval fuselages, the wall adds negligible weight and the ceiling
contributes less than the floor.
The aluminium oval fuselage is estimated at 39,020 kg, while the empirical Torenbeek method esti-
mates it at 18,189 kg. The empirical methods are unable to capture the deviating oval geometry and
as such underestimate the fuselage mass.
For the composite oval fuselage a mass of 26,602 kg is estimated, which is less than the aluminium
weight by 32 %.
64 7. Results

Figure 7.6: Oval, aluminium Boeing 767-300ER thickness distribution

7.3. Wide Oval Boeing 767-300ER 65

Figure 7.7: Oval, composite Boeing 767-300ER thickness distribution

PressureBulkheads:2%
NoseShell:3%
Stringers:14% Frames:16% Cargofloor:15%

Floors:19%
Cutouts:2%

Skin:11% AftShell:7%
SplicingAssembly:6%

Windows:3% Wall:< 1%
NoseShell:7%
Ceiling:4%
Doors:6%
Ceiling:2%
Wall:< 1%
Floor:2%
AftShell:11%
Stringers:25%

Skin:16%
Frames:27%

(a) Aluminium (b) Composite

Figure 7.8: Weight breakdowns for oval Boeing 767-300ER fuselages

66 7. Results

7.4. Stiffened Trapezoid versus Sandwich Trapezoid

As a final case study, the composite, oval fuselages for the Airbus A320-200 and Boeing 767-300ER
from the previous sections were taken. However, the ceiling and wall panels are now made as stiffened
panels, instead of sandwich panels.

Skin thickness [mm]

3.8
Z position [m]

2 3.6

3.4
0
3.2
-2
5 10 15 20 25 30
X position [m]
Stringer smeared thickness [mm]
Z position [m]

10
2
8
6
0
4

-2 2
5 10 15 20 25 30
X position [m]
Frame smeared thickness [mm]
80
Z position [m]

2 60

40
0
20
-2
5 10 15 20 25 30
X position [m]

Figure 7.9: Stiffened trapezoid, oval Airbus A320-200 thickness distribution

Both Figure 7.9 and Figure 7.10 show that the stringer smeared thickness and frame smeared thick-
ness distributions are unaffected by the transition to stiffened panels in the trapezoid (as compared to
Figure 7.4 and Figure 7.7. However, the skin thickness is different along the side arc. The stiffened
trapezoid wall carries less shear flow than the sandwich wall did, resulting in thicker side arc skins
for the stiffened trapezoid concept. Additionally, the skins in the top and bottom arc are also thicker.
Likely, the stiffened panels are thinner than the sandwich panels, reducing the section second moment
of area, hence increasing direct stress due to bending.
Inspecting the weight breakdowns in Figure 7.11a and Figure 7.11b and comparing these to Figure 7.5
and Figure 7.8b, respectively, shows that the contributions to the total mass have not shifted signifi-
cantly. On the other hand, the ceiling member does undergo a significant mass increase. The A320-200
fuselage mass is now estimated at 8,849 kg, which is 10 % more than its sandwich counterpart. The
Boeing 767-300ER fuselage mass is now estimated at 28,146 kg, which is a weight increase of 5.8 %
over its sandwich alternative. So having a stiffened trapezoid (excluding floor) definitely results in a
heavier structure, but it is unclear as of yet this would be offset by the larger weight penalty involved
with cutouts in sandwich panels (as opposed to cutouts in stiffened panels).
7.4. Stiffened Trapezoid versus Sandwich Trapezoid 67

Figure 7.10: Stiffened trapezoid, oval Boeing 767-300ER thickness distribution

Frames:9% Cargofloor:10%
Frames:15% Cargofloor:15%

AftShell:5% Cutouts:3%

SplicingAssembly:6% Cutouts:2%
NoseShell:8% AftShell:7%
SplicingAssembly:6%

NoseShell:6%
Ceiling:8%

Stringers:21%
Ceiling:5%
Wall:3%
Wall:1% Stringers:24%
Floor:3% Floor:2%

Skin:16%
Skin:23%

(a) Airbus A320-200 (b) Boeing 767-300ER

Figure 7.11: Weight breakdowns for oval composite fuselages with stiffened trapezoid
 8
Conclusions and Recommendations
The goal for this thesis was to develop a semi-analytical weight estimation method for pressurized,
composite, oval fuselages. This method was implemented into the FPP Initiator, written in Matlab. In
doing so, the oval fuselage concept can be investigated in conceptual aircraft design of unconventional
configurations, such as the blended wing body. These type of aircraft require a non-circular fuselage
with an aerodynamic shape, for which the oval fuselage is one of few available concepts. With the
help of this thesis work, composite oval fuselages may be investigated, following the current trend in
aerospace industry towards composite materials.
The implemented approach was verified and validated on different aspects. It was concluded that the
load cases and shear force and bending moment distributions may need revision. Pressurization loads
in an oval fuselage section were compared to FEA results, and good agreement was obtained with
errors less than one percent. Additionally, the implemented boom method showed good agreement
with stresses predicted by FEA. It is therefore concluded that this method is implemented correctly and
is able to satisfactorily predict internal stresses and shear flows.
Each composite panel is sized according to applied loads and the critical loads. Computation of these
critical loads is therefore important and was compared to critical loads obtained from FEA. The findings
are listed here:

• Shear buckling failure was found to be under-predicted by the implemented approach (i.e. the
critical buckling load was overestimated) for panels with aspect ratio below 0.5. Scaling was used
to compensate for this. For higher aspect ratios (up to 1), the opposite was found: a conservative
shear buckling load was predicted.

• Deflection due to transverse pressure was compared with FEA and showed good agreement.

• Material failure due to tensile loads as predicted by the Tsai-Wu failure criterion seems to under-
predict stresses as compared to FEA. Therefore, a non-conservative result is obtained. Investi-
gation into different failure theories is in order.

Comparing the optimum found by the deterministic optimization approach outlined in section 3.5 with
a genetic algorithm showed that the deterministic approach outperforms the GA for stringers, which is
attributed to the brute-force approach adopted. For monolithic panels the GA performs better, which
is attributed to the inherent inability of the deterministic approach to take into account all failure modes
at once. Finally, sandwich panels were best designed by the deterministic approach, which produced
lighter designs than the GA by a comfortable margin.
For aluminium fuselages, it was found that a similar fuselage mass fraction (i.e. the ratio of fuse-
lage mass to MTOM) is predicted by the implemented approach as found in reality. Additionally, the
weight breakdown is similar to a generic breakdown for passenger aircraft from literature. Inspection of
thicknesses obtained from the implemented method shows satisfactory sizing of the primary structure.
These observations were also made for composite and oval fuselage sizing. A weight saving of 10 to 20
% was obtained as compared to the aluminium counterparts. Oval fuselage sizing shows satisfactory
results as well, but the cargo floor placement in the Initiator may need revision.

69
70 8. Conclusions and Recommendations

Concluding, the research question was answered and with that the thesis goal was attained. Meth-
ods similar to those used for conceptual sizing of conventional fuselages and materials may be used,
but the effect of the oval geometry and varying stiffness of composite laminates has to be taken into
account. Additionally, pressurization loads may be accurately predicted using the analytical solution
for cylindrical cylinders when the oval cross-section does not change in the longitudinal direction. The
sub-questions were answered in chapter 3, chapter 4 and chapter 5.

8.1. Recommendations
Several improvements can be made to the current method, and several recommendations for further
research are presented in this section.

• In chapter 5 an attempt was made to find out the effect of the curvature of the oval fuselage
pressure shell on the loads that form in this structure due to internal pressure. The finite difference
method presented there can correctly approximate the hoop load, but the longitudinal load was
probably incorrect. Due to time restrictions, no solution for this problem was found. However, by
performing several FEA analyses of arbitrarily shaped pressure vessels and comparing these to
the finite difference method might give more insight into what goes wrong. A working FD method
can be implemented into the Initiator to better predict pressurization loads in the oval fuselage
and therefore make the weight estimation more robust and accurate.

• The wing loads are introduced at one frame in the current method. However, for most aircraft,
these loads are divided over two spars, so a better approximation may be obtained when the wing
loads are introduced at two frames.

• So far, the oval fuselage concept has been studied in terms of its mass and in terms of feasibility
concerning internal pressurization. However, crashworthiness is another important topic that dic-
tates the feasibility of the fuselage concept. The crash structure might also induce a significant
weight penalty and the oval shape of fuselage frames (if present at all) requires additional con-
sideration. A possibility could be to employ frames that only span the bottom arc of the fuselage
and support a transverse beam that in turn supports the passenger floor. Struts may be used for
additional energy absorption. The trapezoidal structure should then carry the remaining loads.
• The load cases used in the current method might be incomplete, and a more detailed and com-
plete set of load cases is advisable.
• Instead of the Tsai-Wu failure criterion, which is only accurate for single plies, the Puck criterion
may be used. The Puck criterion is a physically based model, that does not rely on linear stress-
strain relationships and accounts for multiple failure modes, either fiber failure (FF), or inter fiber
failure (IFF) [28–31]. Moreover, the Puck criterion is the most widely accepted in aerospace
industry [29], because disadvantages of both interactive and non-interactive failure theories are
avoided; a distinction is made between fiber and matrix failure modes; non-linear stress and strain
analysis is relied on; and it includes physically based action plane related fracture criteria [32]. It
is therefore recommended to use this criterion instead.
• Modeling of a keel beam and the area around the wing-box carry-through structure may improve
the weight estimation. Additionally, bulkheads should be sized using an analytical approach,
since their weight seems to be underestimated in the current method.

• The presented frame analysis method does, apart from the shape, not take into account the
trapezoidal structure. At the junctions with the trapezoidal members, forces have to be introduced
in the frame that represent the trapezoid members. It is expected that these will reduce the loads
in the frame, because the alleviate the internal bending moment.
• The boom method as implemented is valid for a single-cell cross-section, which is certainly not
the case for the oval fuselage. Therefore, it should be extended by implementing the multi-cell
theory.
 A
Operation Manual
This manual explains how to perform several aspects of the weight estimation method outlined in this
thesis. The first section discusses how to run the Initiator and get the fuselage weight results. Secondly,
a description of how to run a new DOE for each type of composite member is provided, where it is also
explained how to modify the used material. The third section continues by elaborating upon how to
generate new surrogate models (using the newly created DOE results). Finally, the fourth section is a
short explanation of the finite difference method presented in chapter 5.

A.1. Initiator
The Initiator is a Matlab program which can be obtained by using TortoiseSVN (or another SVN tool)
and performing a checkout on the correct trunk. The details of these operations are not discussed here,
since the URLs might change and working with SVN is beyond the scope of this thesis. Obviously, an
installation of Matlab is also required. The version depends on new functionality added to the Initiator,
but at the time of writing version 2015b or higher is recommended.
In Matlab, browse to the main folder of the Initiator. In the CleanInputFiles folder one finds .xml
files containing settings specific to certain aircraft. Copy either an aircraft file and its corresponding
settings file to the main folder, or type restore in the command window and press Enter. Now type
Initiator --interactive in the command window, press Enter, then type the name of the aircraft
for which to run the weight estimation and press Enter again. After the aircraft settings have been
loaded, type run FuselageWeightEstimation and press Enter. The Initiator will now perform
all tasks in order for it to run the fuselage weight estimation and then perform the weight estimation
method. It terminates by displaying the resulting fuselage mass along with the empirical masses from
Torenbeek, Raymer, Howe and Nicolai. Moreover, several plots have been displayed (if not, the Matlab
property DefaultFigureVisible is set to 0). All results have been written to the aircraft file, so to examine
the exact data one should look there.

A.2. Design of Experiments

The CompositeDOE folder contains all files needed to perform the design of experiments for panels,
stringers and sandwich panels. The somewhat cumbersome process of modifying material properties
is explained next. After that, an explanation of how to run the DOE follows.
For monolithic plates, locate the optimizePlate.m file in the @Plates folder. In there, there is a line
p l a t e . mat = [ 1 3 7 . 9 e9 11.7 e9 4.82 e9 0.29 0.1524e−3 1 6 0 9 ] ;
which denotes the material properties 𝐸 , 𝐸 , 𝐺 , 𝜈 , 𝑡 and 𝜌 in the same units as Table 4.2. For
the material strengths and other properties, locate the createOptWorkspace_plate.m in the same
folder. Modify the necessary parameters and run this file from the command window. It will save a file
to the Opt_Settings folder. Don’t modify that filename.
For sandwich panels, modifying the material for the facesheets is identical to the procedure for mono-
lithic panels, but now the code is located in optimizeSandwich.m and createOptWorkspace_sandwich.m,

71
72 A. Operation Manual

respectively. To modify the core material, locate the GenerateSandwichDatabase.m file in the root
folder. There are a few lines defining the core:
core . E = 131*10^6; % Pa
core . Gxz = 4 1 . 4 * 1 0 ^ 6 ; % Pa
core . Gyz = 2 0 . 7 * 1 0 ^ 6 ; % Pa
core . rho = 4 8 . 2 ; % kg /m^3
Finally, for the stringers, locate createOptWorkspace_v2.m in the @Stringers folder and modify
the required properties. Also modify the line
mat = [ 1 3 7 . 9 e9 11.7 e9 4.82 e9 0.29 0.1524e−3 1 6 0 9 ] ;
in the GenerateStringerDatabase.m file, located in the root folder.
In the root folder, the three files GenerateSandwichDatabase.m, GenerateStringerDatabase.m
and GeneratePanelDatabase.m will perform the design of experiments. Simply run the script and
all necessary actions will be performed. Please note that depending on the amount of experiments and
the computer used it may take several hours to days for it to complete! Moreover, the experiments will
be run in a parallel pool, so much of the computer’s resources will be used. It is therefore advised to
either run the scripts on a computer that is not needed for other tasks, or change the parfor loops to
normal for loops.
In each of the three files mentioned earlier, one finds the following two lines:
num = 30000;
x = l h s d e s i g n ( num , 5 ) ;
This generates a matrix with 30,000 rows (the experiments) and 5 columns, with values ranging from
0 to 1. This is the Latin Hypercube Sampling. The amount of columns reflects the parameters used
to specify the component dimensions and loads and does not need to be changed. However, the num
variable may be set to any number preferred.
Since the sampling only returns numbers from 0 to 1, these need to be scaled to meaningful values.
The variable preds does that:
preds = [ 0 . 3 0.12 0 0 0 ; . . .
0 . 8 0 . 8 2000 100 5 0 0 ] ;
The comments indicate which number is for which parameter, but the first row always is the minimum
value and the second and last row the maximum. For dimensions meters are used as unit, for loads
it’s in either kN/m or kPa.
The last thing one may want to change is the filename where the results are written to. The last line in
each Generate file can be modified:
save ( ’ plate_database ’ , ’ plateArray ’ , ’ plate_data ’ , ’x ’ , ’ preds ’ ) ;
Change the first string in the save command to change the filename. The extension .mat does not
have to be included.

A.3. Creating Surrogate Models

The surrogate models can be easily created by running the scripts located in the ModelGeneration
folder. The main three files are:
1. Plate_forecasting.m
2. Sandwich_forecasting.m
3. Stringer_forecasting.m
However, before these files can use the newly created DOE data, copy the files created by the DOE
into the current (ModelGeneration) folder. Then locate the CreateExcel scripts:
1. CreateExcelPlate.m
2. CreateExcelSandwich2.m
A.3. Creating Surrogate Models 73

3. CreateExcelStringer.m

In each of these files the following two lines may be modified:

load ( ’ plate_database_final ’ ) ;
filename = ’ plate_final_nondim . xlsx ’ ;
The first loads the DOE data, so make sure to edit this string to refer to the correct file. The second
line defines the filename where the Excel file will be written to. It does not necessarily have to be
changed; it will overwrite the current file if it wasn’t. Otherwise, make sure to also edit the filename in
the forecasting files in a line similar to:
ds = d a t a s e t ( ’ x l s f i l e ’ , ’ sandwich_data_nondim . x l s x ’ ) ;
With the Excel files created, the forecasting scripts can be executed without further editing. Each
will create a collection of neural networks and linear regression models for the parameters of in-
terest for each type of component. Additionally, new files are created that contain the neural net-
works for use in the Initiator. Copy these files (plate_model.mat, stringer_model.mat and
sandwich_model.mat) to the Opt_Settings folder in the Initiator folder (and replace the exist-
ing ones, or better yet: rename those and copy the new files afterwards). If this operation needs to be
performed multiple times, it may be advantageous to create a function and a setting in the Initiator that
will load the correct neural networks depending on the new material setting. Then each file put in the
Opt_Settings folder needs a unique name.
 B
Explanation of Messages
The weight estimation method and member sizing methods may output some messages to the Matlab
Command Window during runtime. A list of messages for each of these methods is provides below,
accompanied by a description of what they mean and what the user should do.

B.1. Fuselage Weight Estimation

FuselageWeightEstimation assumes first cabin to be main cabin, i.e. this cabin lies on the floor
of the trapezoid structure and its height is the trapezoid height The FWE always starts with this
message and the user does not have to do anything with this, except for checking whether the cabins
have been correctly declared in the aircraft settings file as stated by this message.

Warning: Discrepancy between fuel tank mass and total fuel mass of xxx kg This warning is
common and means that the fuel mass that can geometrically fit in the fuel tanks is not equal to the fuel
mass required for the harmonic mission. No action need be undertaken, since this message is only
meant for informative purposes.

Scaling fuel tank mass with ratio of geometric fuel volume and total fuel mass Follow-up mes-
sage of the previous message, again for information, meaning that the method assumes that all tanks
are only filled up to the amount required for the harmonic mission.

Fuselage weight estimation scaling pax, cargo and fuel masses to match previous weight re-
sults Again a message only for informative purposes. What happens is that the amount of passen-
gers, cargo and/or fuel is not equal to the amounts as specified for the harmonic mission. In this case
the FWE scales these masses such that agreement is obtained.

Iteration x, Max Abs Deviation: xxx

Iteration x, Amount of deviating variables: xxx The iterative process of estimating the structural
thicknesses is busy and provides the user with this information each iteration. The maximum absolute
deviation of the design variables is shown, which gives an idea of how close to converging the method
is. Additionally, the amount of variables that have a deviation larger than the tolerance from their
previous value is provided.

Too large/small input value for stringer! Only for composite sizing this message may appear.
When either a stringer dimension or load is provided to the neural network that is not within the limits of
values used to generate the neural network, this message appears. It may or may not be a cause for
errors in the weight estimation. A new neural network may be made which includes the out-of-bound
values to solve this issue.

75
76 B. Explanation of Messages

Too large/small input value for panel! Only for composite sizing this message may appear. When
either a panel dimension or load is provided to the neural network that is not within the limits of values
used to generate the neural network, this message appears. It may or may not be a cause for errors
in the weight estimation. A new neural network may be made which includes the out-of-bound values
to solve this issue.

Too large/small input value for sandwich! Only for composite sizing this message may appear.
When either a sandwich panel dimension or load is provided to the neural network that is not within
the limits of values used to generate the neural network, this message appears. It may or may not be
a cause for errors in the weight estimation. A new neural network may be made which includes the
out-of-bound values to solve this issue.

B.2. Panel Database Generation

Total time to complete p(i=xx/xx): xxx s Appears after each completed panel, giving the user an
indication of how long it takes to size one panel. Since the loop is executed in parallel, the index/total
indication is not in any particular order, so no conclusion may be drawn on the total completion of the
database generation.

Axial loads can not be compressive (unless buckling calculation is adapted) This error mes-
sage appears when a negative axial load is specified (i.e. 𝑁 or 𝑁 ). Currently, the method only
supports tension in these directions, and as stated, the buckling calculation should be modified when
compression forces are required.

Warning no convergence in strength determination The Newton-Raphson method has not con-
verged within a set amount of iterations, such that the critical loads computed are likely incorrect. When
this happens, there is likely a bug leading to incorrect input for the StrengthDetermination func-
tion.

Execution stopped because initial laminate does not meet requirements Each ply addition func-
tion can produce this message, but it only appears when the initial laminate specified is invalid (accord-
ing to the laminate stacking rules) or when there is a bug in the program.

For some reason a ply could not be added to the laminate Most of the ply addition functions
can produce this message, meaning they failed in adding a ply to the laminate without violating the
stacking sequence constraints. In general this should never happen, but if it does, a good first solution
is to check the process selecting a ply addition function based on the most critical failure load.

B.3. Stringer Database Generation

Optimization routine finished... Warning! No optimum subdesigns could be found Because a
given number of laminates is generated beforehand and all combinations are analyzed, it may happen
that none can cope with the applied loads within the dimension bounds. Either increase nogen to
increase the amount of laminates or increase maxgendiff to allow more difference in number of plies
per member. Note that both these options increase computational time. As a final option, decrease
the maximum loads for which to generate the database.

Warning no convergence in strength determination The Newton-Raphson method has not con-
verged within a set amount of iterations, such that the critical loads computed are likely incorrect. When
this happens, there is likely a bug leading to incorrect input for the StrengthDetermination func-
tion.

B.4. Sandwich Database Generation

The same or similar messages may occur as for the monolithic panels. Therefore, they are not repeated
here.
 C
Fuselage Weight Estimation Settings
FWESkinMaterial Set a material tag for the material to be used for skin sizing and failure analysis.
Generally, al2024t3 is set for conventional, aluminium aircraft. Set to cfrp to use fiber composite
material.

FWEFrameMaterial Set a material tag for the material to be used for skin sizing and failure analysis.
Generally, al7075t6 is set for conventional, aluminium aircraft. Set to cfrp to use fiber composite
material.

FWEFacingMaterial Set a material tag for the material to be used for skin sizing and failure analysis.
Generally, al2024t3 is set for conventional, aluminium aircraft. Set to cfrp to use fiber composite
material.

FWECoreMaterial Set a material tag for the material to be used for skin sizing and failure analy-
sis. For conventional, aluminium aircraft the value honeycomb is used, generally. For a composite
fuselage, this setting is not used.

FWECargoFloorMaterial Set a material tag for the material to be used for skin sizing and failure
analysis. Generally, al2024t3 is set for conventional, aluminium aircraft. Set to cfrp to use fiber
composite material.

FWEMinimumGageThickness Set the minimum thickness for metal components in m.

FWEPaxTransverseLoad An empirical estimate of the load passenger exert on the floor in Pa. Is by
default set to 17 kPa.

FWECargoTransverseLoad An empirical estimate of the load cargo exerts on the cargo floor in Pa.

FWEBurstPressureSafetyFactor Safety factor for burst pressure load case.

FWEUseCabinAltitude Set to true to use a specified cabin altitude. Otherwise, the maximum dif-
ferential pressure is used.

FWEMinimumCabinAltitude The minimum cabin altitude in m.

FWEMaximumCabinDifferentialPressure Maximum cabin differential pressure in psi. Used when

a cabin altitude is not specified.

77
78 C. Fuselage Weight Estimation Settings

FWEMinimumCabinDifferentialPressure Minimum cabin differential pressure in psi. Used in burst

and landing load cases.

FWELandingLoadFactor Load factor applied in landing load cases.

FWELandingGearLoadFraction Fraction of aircraft weight carried by landing gear during landing.

FWEStiffenerType Stiffener type for metal sizing. Available options: z, hat, z-simple and z-
simple-optim.

PanelTypeFloor Optional setting. Only used for composite sizing. Set by default to sandwich,
however, can be set to stiffened. Models the trapezoid floor either as a sandwich beam or as
stiffened panels with cross-beams.

PanelTypeWall Optional setting. Only used for composite sizing. Set by default to sandwich, how-
ever, can be set to stiffened. Models the trapezoid wall either as a sandwich beam or as stiffened
panels with cross-beams.

PanelTypeCeiling Optional setting. Only used for composite sizing. Set by default to sandwich,
however, can be set to stiffened. Models the trapezoid ceiling either as a sandwich beam or as
stiffened panels with cross-beams.
 Bibliography
[1] R. H. Liebeck. Design of the blended wing body subsonic transport. Journal of Aircraft, 41(1):10–
25, 2004.

[2] V. Mukhopadhyay. Structural concepts study of non-circular fuselage configurations. In SAE/AIAA

World Aviation Congress, pages 1–13. AIAA, 1996.

[3] V. Mukhopadhyay, J. Sobieszczanski-Sobieski, I. Kosaka, G. Quinn, and G. N. Vanderpaats. Anal-

ysis, design, and optimization of noncylindrical fuselage for blended-wing-body vehicle. Journal
of Aircraft, 41(4):925–930, 2004.

[4] V. Mukhopadhyay. Blended-wing-body fuselage structural design for weight reduction. In 46th
Structures, Structural Dynamics and Materials Conference, number April, pages 1–8. AIAA, 2005.

[5] V. Mukhopadhyay and M. R. Sorokach. Composite structure modeling and analysis of advanced
aircraft fuselage concepts. In AIAA Modeling and Simulation Technologies Conference, number
June, pages 1–14. AIAA, 2015.

[6] F. J. J. M. M. Geuskens, O. K. Bergsma, S. Koussios, and A. Beukers. Analysis of conformable

pressure vessels: Introducing the multi-bubble. AIAA Journal, 49(8):1683–1692, 2011.

[7] V. Mukhopadhyay. Hybrid-wing-body vehicle composite fuselage analysis and case study. In 14th
AIAA Aviation Technology, Integration and Operations Conference, pages 1–12. AIAA, 2014.

[8] V. Mukhopadhyay. Hybrid wing-body pressurized fuselage modeling, analysis, and design for
weight reduction. In 53rd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Ma-
terials Conference, number April, pages 1–14. AIAA, 2012.

[9] R. Vos and M.F.M. Hoogreef. Semi-analytical weight estimation method for fuselages with oval
cross-section. In 54th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Mate-
rials Conference, pages 1–15. AIAA, 2013.

[10] K. Schmidt and R. Vos. A semi-analytical weight estimation method for oval fuselages in conven-
tional and novel aircraft. In 52nd Aerospace Sciences Meeting, number January, pages 1–20.
AIAA, 2014.

[11] S. de Smedt and R. Vos. Knowledge-based engineering approach to the finite element analysis
of the oval fuselage concept. In 53rd AIAA Aerospace Sciences Meeting, number January, pages
1–15. AIAA, 2015.

[12] R. Marissen. Two main challenges for the future composites technology, costs reduction and
strength prediction. In Tenth European Conference on Composite Materials, pages 1–11, 2002.

[13] M. J. L. van Tooren. Sandwich Fuselage Design. PhD thesis, Delft University of Technology,
1998.

[14] J. Roskam. Airplane Design Part V: Component Weight Estimation. Darcorporation, 1985.

[15] Anon. Certification specifications for large aeroplanes. Technical Report September, EASA, 2007.

[16] T.H.G. Megson. Aircraft Structures. Elsevier, 4th edition, 2007.

[17] D. Howe. Blended wing body airframe mass prediction. Journal of Aerospace Engineering,
215(June):319–331, 2001.

79
80 Bibliography

[18] D.R. Polland, S.R. Finn, K.H. Griess, J.L. Hafenrichter, C.T. Hanson, L.B. Ilcewicz, S.L. Metschan,
D.B. Scholz, and P.J. Smith. Global cost and weight evaluation of fuselage side panel design
concepts. Technical report, NASA, 1997.

[19] B.W. Flynn, M.R. Morris, S.L. Metschan, G.D. Swanson, P.J. Smith, K.H. Griess, M.R. Schramm,
and R.J. Humphrey. Global cost and weight evaluation of fuselage keel design concepts. Technical
report, Boeing Commercial Airplane Group, 1997.
[20] R. Jones. Mechanics of Composite Materials. Taylor and Francis, 2nd edition, 1999.

[21] C. Kassapoglou. Design and Analysis of Composite Structures: With Applications to Aerospace
Structures. Wiley, 2nd edition, 2013.

[22] I. van Gent and C. Kassapoglou. Cost-weight trades for modular composite structures. Structural
and Multidisciplinary Optimization, 2013.

[23] E.F. Bruhn. Analysis and Design of Flight Vehicle Structures. Tri-State Offset Company, 1973.

[24] D. Ambur and M. Rouse. Design and evaluation of composite fuselage panels subjected to com-
bined loading conditions. Journal of Aircraft, 42(4):1037–1045, 2005.

[25] W. Flügge. Stresses in Shells. Springer-Verlag, Berlin, 3rd edition, 1966.

[26] R. W. Johnson, L. W. Thomson, and R. D. Wilson. Study on utilization of advanced composites
in fuselage structures of large transports. Technical report, NASA, 1985.
[27] E. Torenbeek. Synthesis of Subsonic Airplane Design. Delft University Press, Delft, 1982.

[28] A. Puck and H. M. Deuschle. Progress in the puck failure theory for fibre reinforced composites :
Analytical solutions for 3d-stress. Composites Science and Technology, 62(3):371–378, 2002.

[29] G. Lutz. The puck theory of failure in laminates in the context of the new guideline vdi 2014 part
3. Technical report, VDI, 2014.

[30] M Knops. Analysis of Failure in Fiber Polymer Laminates, volume 0. Springer, 2008.
[31] H. M. Deuschle and A. Puck. Application of the puck failure theory for fibre-reinforced composites
under three-dimensional stress: Comparison with experimental results. Journal of Composite
Materials, 47(6-7):827–846, 2013.

[32] N. Kosmas. Predicting Onset of Damage in Special Class of Laminates under Tension. Master,
Delft University of Technology, 2015.