Structural Optimization of A Composite Wing
Structural Optimization of A Composite Wing
Structural Optimization of A Composite Wing
A THESIS SUBMITTED TO
THE GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCES
OF
MIDDLE EAST TECHNICAL UNIVERSITY
BY
ZLEM SKMEN
SEPTEMBER 2006
Approval of the Graduate School of Natural and Applied Sciences
I certify that this thesis satisfies all the requirements as a thesis for the degree of Master of
Science.
This is to certify that we have read this thesis and that in our opinion it is fully adequate, in
scope and quality, as a thesis for the degree of Master of Science.
Signature :
iii
ABSTRACT
Skmen, zlem
In this study, the structural optimization of a cruise missile wing is accomplished for the
aerodynamic loads for four different flight conditions. The flight conditions correspond to
the corner points of the V-n diagram. The structural analysis and optimization is performed
using the ANSYS finite element program. In order to construct the flight envelope and to
find the pressure distribution in each flight condition, FASTRAN Computational Fluid
Dynamics program is used.
The structural optimization is performed for two different wing configurations. In the first
wing configuration all the structural members are made up of aluminum material. In the
second wing configuration, the skin panels are all composite material and the other
members are made up of aluminum material. The minimum weight design which satisfies
the strength and buckling constraints are found for both wings after the optimization
analyses.
iv
Z
Skmen, zlem
Bu almada, bir seyir fzesi kanadnn, drt farkl manevra durumundaki aerodinamik
ykler altnda yapsal eniyilemesi gerekletirilmitir. Szkonusu manevra durumlar V-n
grafiinin ke noktalarna karlk gelmektedir. Yapsal analiz ve eniyileme iin ANSYS
sonlu elemanlar program, manevra zarfnn oluturulmas ve her bir uu durumundaki
basn dalmlarnn elde edilebilmesi iin FASTRAN hesaplamal akkanlar dinamii
program kullanlmtr.
v
To My Family
vi
ACKNOWLEDGMENTS
I would like to express my gratitude to Prof. Dr. Mehmet A. AKGN for his unique
guidance, supervision, understanding and encouragement in every phase of the study. I am
deeply grateful to him for additional workload that he spared to me after he retired from
METU. I would also like to thank Prof. Dr. Yusuf ZYRK for his interest and support.
I would like to express my deepest appreciation to Mr. Erdin Nuri YILDIZ for his
invaluable discussions, useful remarks, and support. I cannot thank to him enough.
I would like to thank my coordinator Dr. Mutlu Devrim CMERT and my chiefs Dr.
Serkan GZBYK and Dr. zge EN for their support and understanding.
I would like to thank Mr. H. zgr DEMR, Mr. Bora YAZICI, Mr. Osman BAOLU
and Dr. Ltfi Oktay GN for their help and discussions on the aerodynamic solutions.
Analyses carried out on ANSYS and FASTRAN were performed using the facilities at
TBTAK-SAGE, which is greatly acknowledged.
Special thanks go to my parents Nurhan and Seyit Ahmet SKMEN for their endless
support, patience and encouragement throughout my all education. I would also like to
thank my sister zgl and my brother zkan for their love, support and patience.
Finally, I would like to express my special thanks to Blent SMER for his love,
understanding, patience and support. Without him, none of this would have been possible.
vii
TABLE OF CONTENTS
ABSTRACT ........................................................................................................................ iv
Z ......................................................................................................................................... v
ACKNOWLEDGMENTS................................................................................................. vii
CHAPTERS
1. INTRODUCTION..................................................................................................... 1
viii
3.3 STRUCTURAL OPTIMIZATION WITH ANSYS....................................... 33
5. CONCLUSION........................................................................................................ 90
REFERENCES............................................................................................................ 93
APPENDICES
ix
LIST OF TABLES
Table 2.1 The percent error in the total force perpendicular to chordwise direction........... 24
Table 2.2 Comparison of total forces and moments in FASTRAN and ANSYS (Element
size in ANSYS=40 mm). ............................................................................................ 26
Table 3.1 Design variables, abbreviations and bounds for the aluminum wing. ................. 48
Table 3.2 State variables, abbreviations and bounds for the aluminum wing. .................... 51
Table 3.3 Design variables, abbreviations and bounds for Portion-1 of the composite wing.
.................................................................................................................................... 55
Table 3.4 State variables, abbreviations and bounds for Portion-1 of the composite wing. 57
Table 3.5 Mechanical properties of the unidirectional fiber reinforced epoxy composites
[37].............................................................................................................................. 58
Table 3.6 The geometric properties of the wing used in the comparison analyses. ............ 59
Table 3.7 Results of the analyses for Portion-1 for different epoxy composites. . .............. 60
Table 4.1 Weight change with the number of executions for the aluminum wing. ............. 64
Table 4.2 Best design optimization variables for three analyses with different number of
state variables for the aluminum wing. ....................................................................... 66
Table 4.3 Best design optimization variables for subsequent optimization analyses for the
aluminum wing. .......................................................................................................... 67
Table 4.4 Best design optimization variables for Portion-1 of the composite wing without
buckling constraints. ................................................................................................... 73
Table 4.5 Buckling load factors for different stacking sequences of the upper skin panels
for Portion-1 of the composite wing without buckling constraints. ........................... 75
Table 4.6 Best design optimization variables for Portion-2 of the composite wing without
buckling constraints. ................................................................................................... 77
Table 4.7 Buckling load factors for different stacking sequences of the upper skin panels
for Portion-2 of the composite wing without buckling constraints. ........................... 78
Table 4.8 Best design optimization variables for Portion-1 of the composite wing............ 79
Table 4.9 Best design optimization variables for Portion-2 of the composite wing............ 81
Table 4.10 Best design optimization variables for Portion-3 of the composite wing.......... 82
Table 4.11 Best design optimization variables for Portion-4 of the composite wing (1st
alternative). ................................................................................................................. 84
x
Table 4.12 Best design optimization variables for Portion-4 of the composite wing (3rd
alternative, chosen one). ............................................................................................. 85
Table 4.13 Comparison of the values of the strength constraints found after the
optimization of a portion with the values for the best design for Portions 1-3 of the
composite wing. .......................................................................................................... 86
Table 4.14 The values of the strength and buckling constraints for different stacking
sequences of the upper skin panels for Portion-1 of the composite wing. ................. 89
Table 4.15 The values of the strength and buckling constraints for different stacking
sequences of the lower skin panels for Portion-2 of the composite wing. ................. 89
xi
LIST OF FIGURES
xii
Figure 3.6 Modeling technique for spars and spar caps. ..................................................... 40
Figure 3.7 Offset skin panel and spar cap elements. ........................................................... 41
Figure 3.8 Offset skin panel and rib cap elements............................................................... 42
Figure 3.9 Wing pivot and slot door of Tomahawk cruise missile [25]. ............................. 43
Figure 3.10 Boundary conditions for the wing. ................................................................... 43
Figure 3.11 Graphs of the linear and parabolic thickness functions.................................... 45
Figure 3.12 Percent mass reduction versus thickness ratio graph. ...................................... 45
Figure 3.13 The exaggerated view for parabolically reducing thickness of the upper skin
panels. ......................................................................................................................... 46
Figure 3.14 Distance ratio function for different m values.................................................. 47
Figure 3.15 Mesh of the wing structure............................................................................... 50
Figure 4.1 The von Mises stress (MPa) distribution for top of the shell elements for PLAA
flight condition for the aluminum wing. ..................................................................... 70
Figure 4.2 The von Mises stress (MPa) distribution for bottom of the shell elements for
PLAA flight condition for the aluminum wing........................................................... 70
Figure 4.3 The axial stress (MPa) distribution for beam elements for PLAA flight condition
for the aluminum wing................................................................................................ 71
Figure 4.4 Uy distributions obtained from the buckling analyses for each flight condition for
the aluminum wing. .................................................................................................... 71
Figure 4.5 Uy displacement distributions obtained from the buckling analyses for each
flight condition for composite wing (optimization results for Portion-1 without
buckling constraints)................................................................................................... 75
Figure 4.6 Uy displacement distributions obtained from the buckling analyses for each
flight condition for the composite wing...................................................................... 87
xiii
NOMENCLATURE
xiv
W Weight
Angle of Attack
o Mid-plane Strain
1t Failure Tensile Strain in Longitudinal Direction
1c Failure Compressive Strain in Longitudinal Direction
2t Failure Tensile Strain in Transverse Direction
2c Failure Compressive Strain in Transverse Direction
12 In-plane Failure Strain
Curvature of the Layer
c Buckling Load Factor due to a Combination of Normal and Shear Loads
n Buckling Load Factor due to Normal Loads
s Buckling Load Factor due to Shear Loads
Density
12 In-plane Poissons Ratio
xv
ABBREVIATIONS
C Chordwise Forces
DV Design Variable
GA Genetic Algorithm
N Normal Forces
NS Navier-Stokes
SV State Variable
xvi
CHAPTER 1
INTRODUCTION
The main objective of this thesis is to design and optimize the wing structure of a cruise
missile. The aerodynamic and geometrical properties of the wing are taken from the
Tomahawk Cruise Missile. The aerodynamic loads acting on the wing are found by the
use of a commercial Computational Fluid Dynamics (CFD) program for different flight
conditions. With the aid of predicted aerodynamic properties of the wing, a V-n diagram
is constructed and the design and optimization of the wing is carried out for four
different loading conditions. Deformation of the wing under the applied aerodynamic
loading and the relevant boundary conditions are solved using the ANSYS finite
element program. In addition, weight minimization of the wing structure subjected to
strength and buckling constraints is achieved by use of the optimization routines
embedded into the ANSYS Finite Element program.
The traditional way of designing a wing structure is based on a trial and error procedure,
and in most of the time this procedure involves the participation of different groups in a
company. As soon as the wing shape is determined, the structural sizing of wing
structural components is performed by the stress group of the company based upon
multiple criteria (stress, buckling, aeroelastic considerations, etc.).
Structural design is usually limited to an adequate design that meets the major
requirements of the problem. For a number of load cases, the wing structure is analyzed
until multiple design criteria are satisfied. The design process ends up with a design that
meets the major structural constraints. However, in this design process the final design
may not be the optimum design.
Minimum weight design is an important issue in airframe design process since reduction
in weight results in a high performance and cost effective flight vehicle. More formal
ways of structural design is needed in order to ensure that the resulting design satisfies
1
the minimum weight criterion; in addition to that, all the other structural criteria should
also be satisfied.
Cruise Missiles are tactical missile systems that are used for critical, long range and
precision strike missions against high value targets. After they are launched from a
platform, these missiles are accelerated up to their cruise flight conditions with the aid
of their solid propellant boosters. The thrust needed for the cruise phase is supplied from
a turbojet or turbofan motor depending on the required range. Typical high technology
long range cruise missile design consists of a turbofan engine which gives higher values
of specific impulse when compared to a turbojet system and a wing which increases the
lift-to-drag ratio of the vehicle. The wing of the cruise missile must have sufficient
strength to withstand several loading scenarios. The minimum weight design of both
cruise missile fuselage and wing is important since reduction in the structural weight of
the missile leads to an increase in the useful payload that can be carried. A photograph
of the Tomahawk Cruise Missile is shown in Figure 1.1.
2
Figure 1.1 Tomahawk cruise missile.
A wing is essentially a beam, which gathers and transmits all of the applied airload to
the central attachment to the fuselage. As soon as the basic wing shape has been
decided, a preliminary layout of the wing structure must be indicated to a sufficient
strength, stiffness and lightweight structure with a minimum of manufacturing problems
[1].
During flight, aerodynamic loads on a wing bend it upwards; as a result, the upper side
of the wing is loaded in compression and buckling becomes a problem at these regions.
Stiffeners attached to the skin increase the wing skin resistance to buckling.
3
Figure 1.2 Wing structural components.
In the absence of powerful computing resources and strong numerical methods like the
finite element method (FEM), most of the wing stress analysis depended on the flexural
beam theory which uses simplifying assumptions (see [2] for detailed wing stress
analysis methods).
Today, wing structures consisting of spars, ribs, skin panel and stiffeners with arbitrary
loading conditions, can be easily modeled and solved using the finite element method.
In the finite element method, a wing structure is divided into finite elements and the
problem is solved for each and every one of them. By this way the governing partial
differential equation turn into a set of algebraic equations. The solution of these
equations gives the entire displacement field under the action of applied loads and
boundary conditions.
4
1.4 OPTIMIZATION OF COMPOSITE STRUCTURES
Structural designers seek the best possible design. In the case of aerospace structures the
best possible design would be the one that satisfies the strength and stiffness
requirements with minimum weight. Traditionally structural engineer first looks for a
sufficient design that meets the major requirements of the problem, and makes necessary
changes in the design to make it structurally efficient and light. In the case of a designer
with limited experience such a way to design a structure may require large amount of
iterations that makes the design procedure cumbersome.
Over the past two decades, mathematical optimization, which deals with either the
maximization or minimization of an objective function subject to constraint functions,
has emerged as a powerful tool for structural design [3]. On the other hand the use of
composite materials such as Glass/Epoxy and Carbon/Epoxy has become widespread in
the aerospace industry due to their high strength to weight ratios.
5
Since designers have finite resources, it has been impossible to solve problems with the
most complex models, analysis and optimization simultaneously. Hence much of the
literature can be found on one of the coordinate planes in Figure 1.3. For example
complex optimization is often applied to problem of laminate design with displacement
or stress constraints where the model and analysis are simple. Conversely, simple local
optimization has often been used for complex models, requiring a detailed finite element
analysis [4].
Modeling complexity is related with the choice of modeling the structural problem. For
unstiffened and uniformly stiffened composite panels one can apply simplifying
assumptions like orthotropic material properties and solve the governing equations for
simple loading conditions. For example, [5] gives practices for predicting buckling of
uniform stiffened and unstiffened isotropic/orthotropic shells under various loading
conditions. In the case of a wing structure with composite skin panels and stiffeners
subject to arbitrary loading conditions, the structure has to be modeled with finite
elements.
In terms of optimization complexity the most complex and expensive problems are
reliability based optimization problems, and the easiest problems are the optimization of
a laminate stacking sequence using lamination parameter graphs [4].
The optimization of wing box structures made up of composite laminates has been an
interesting problem. As early as 1973, Khot and Venkayya [7] developed an
optimization method, which is based on constraint gradients, to optimize cantilever
composite wing structure. The spars and panels were idealized by bars and shear panels
and the top and bottom skin were idealized by membrane elements. The two design
variables were the thicknesses of shear panels/membrane elements and the area of the
rod elements. Later they also present a method based on optimality criterion for
6
designing minimum weight fiber reinforced composite structures with stress and
displacements constraints. They solved the structural problem with finite elements and
used 80 composite elements consisting in three layers with fiber orientations 0o, 45o,-45o
directions [8].
Liu and Lin [11] combined the finite element method with a refined optimality criterion
method and optimized an aluminum triple spar wing structure under two independent
loading conditions. Displacement constraints were imposed on the tip in the transverse
direction while size and stress constraints were imposed on all elements. The minimum
weight was found to be 42.28 kg with a 25.4 cm tip displacement constraint and reduced
to a minimum weight of 35.2 kg when the tip constraint was released to a value of 35.2
cm. Buckling constraints were also added as a third case which leaded to a minimum
weight of 34.88 kg.
The composite version of the three spar wing configuration was optimized by the same
method [12]. The wing structure was idealized by membrane quadrilaterals, shear panels
and bars. The top and the bottom skin were graphite/epoxy layered elements with 0o, 90o
and 45o fibers. The spars and ribs were idealized by aluminum shear panels. Three
7
different constraint conditions were investigated in which stress, local buckling and
twist constraints were imposed on the wing design.
Yurkovich [14] applied Taguchi technique coupled with ASTROS code in order to
optimize wing structures and determined wing external and internal geometry for
minimum wing weight. He used seven design parameters but eventually realized that the
number of spars and ribs were not significant parameters in determining the weight of
the wing. In this study buckling was not taken into account.
Rhl, Mavris and Schrage [15] also used ASTROS for structural optimization of a High
Speed Civil Transport (HSCT) wing in a multilevel decomposition approach. With the
help of ASTROS structural optimization tool, the wing structure was sized subject to
strength, buckling and aeroelastic constraints.
Liu, Haftka and Akgn [16, 17] proposed a two-level wing design optimization
subjected to strength and buckling constraints. They considered an unswept, untapered
wing box with four spars and three ribs made of graphite-epoxy. The wing box was
clamped at the root and subjected to a distributed tip load. In the wing-level
optimization, the objective function was the structure weight and the continuous design
variables were the ply thicknesses with orientations of 0o, 90o and 45o for each panel.
In the panel-level, symmetric and balanced stacking sequence optimization of the wing
panels was accomplished to maximize the buckling load by the use of a permutation
genetic algorithm (GA). A response surface was then fitted to the optimum buckling
load as a function of number of 0o, 90o and 45o stacks and In-plane loads. The resulting
response surface was used for the wing-level optimization. Later, Liu and Haftka [18]
imposed continuity constraints and demonstrated the tradeoffs between weight and
continuity using the same composite wing example.
8
In the case of composite materials a designer can change either the ply thicknesses or
the ply orientation angles to achieve the desired material strength. The optimization ply
angles, which are usually restricted to a small set of discrete angles (0o, 90o and 45o),
can be achieved by the use of GAs. Liu, Haftka, Akgn and Todoroki [19] developed a
gene-ranked GA and used this method in the panel level optimization of a composite
wing box [16, 17].
Kapania and Chun [20] used a simple beam-type structural model and a conjugate
gradient/steepest descent type optimization method to determine the structural weight of
a wing box, subjected to a twist constraint, to carry aerodynamics loads for a wing in
transonic flow field.
Engelstad et al. [21] compared three different optimization strategies used in the
minimum weight design of an F/A 22 horizontal stabilizer structure. The horizontal
stabilizer was a rib stiffened all composite design with honeycomb edges.
9
CHAPTER 2
2.1 INTRODUCTION
Before the structural design of a wing can be made, the external forces acting on the
wing must be known. The main design flight conditions for the cruise missile can be
prescribed by stating the limiting values of acceleration and speed. In graphical form,
design requirements can be represented by plotting load factor (n) versus flight velocity
to obtain a diagram as shown in Figure 2.1. This diagram is generally referred to as V-n
diagram.
10
In level unaccelerated flight, the wing produces sufficient lift to balance the weight of a
cruise missile. When a cruise missile has to make a maneuver, additional loads are
created. The amount of additional load depends on the severity of the maneuvers and is
measured in terms of load factor [1]. Load factor is defined as follows where q=V2/2 is
the dynamic pressure:
L qSC L
n= = (2.1)
W W
The V-n diagram, once constructed for a particular flight vehicle, involves valuable
information about aerodynamic and structural limits of the flight vehicle.
At low speeds maximum load factor (nmax) is a function of the maximum lift coefficient
(CL,max). The equation of the curve AB in Figure 2.1 can be found by plugging CL,max in
Equation (2.1).
At speed V1, flight vehicle can have three possible angle of attack values. At point 1, the
flight vehicle has a low angle of attack value and the resulting load factor is smaller than
the maximum load factor at this speed. At point 2, the flight vehicle has an angle of
attack value sufficient to obtain CL,max and the load factor is maximum. It is not possible
for the flight vehicle to fly at a higher angle of attack at this speed (point 3) since a
further increase in angle of attack results in stalling of the wing.
Maximum load factor increases with velocity up to a certain speed (V*) and beyond that
speed structural damage may occur in the structure. At velocities higher than V* the
flight vehicle has to fly at a CL value less than CL,max which is dictated by the limit load
factor of the flight vehicle. Again the velocity can increase up to a certain limit beyond
which structural damage occurs due to very large dynamic pressures. This limiting
speed is called the design dive speed (VD) and it is the maximum speed for the structural
design. It is a statistically determined speed sufficiently greater than the cruising speed
to provide for safe recovery from inadvertent upsets [1]. The design dive speed is
generally taken as 1.2-1.5 times the cruise speed of the flight vehicle [1, 2, 22, 23, and
24].
11
In Figure 2.1, the points B and C correspond to Positive High Angle of Attack (PHAA)
and Positive Low Angle of Attack (PLAA) flight conditions respectively, and points E
and D correspond to Negative High Angle of Attack (NHAA) and Negative Low Angle
of Attack (NLAA) flight conditions respectively. These conditions represent
symmetrical flight maneuvers; i.e., there is no motion normal to the plane of symmetry
of the flight vehicle [22].
Figure 2.2 Forces acting on the wing for each flight condition.
Bending moments created by the normal forces (N) shown in Figure 2.2 produce
compressive stresses on the upper side of the wing in the PHAA and PLAA conditions,
on the bottom side of the wing in the NHAA and NLAA conditions. On the other hand,
bending moments created by the chordwise forces (C) produce compressive stresses on
the leading edge of the wing in the PHAA and NHAA conditions, on the trailing edge of
the wing in the PLAA and NLAA conditions. Each of these four flight conditions
produces the highest load somewhere on the wing as shown in Figure 2.3 [22].
12
Figure 2.3 Critical locations on the wing for each flight condition.
Generally speaking, if the flight vehicle is designed for the air loads produced by the
velocity and acceleration conditions at points, B, C, D and E, it should be safe from a
structural strength viewpoint if flown within the specified limits regarding velocity and
acceleration [2].
The cruise missile wing has a basic trapezoid shape, an aspect ratio of 6.0, a taper ratio
of 0.5, leading edge sweep angle of 6.34o and NACA 64A208.2 airfoil profile [25]. The
wing planform considered in this study is shown in Figure 2.4.
Figure 2.4 Cruise missile wing planform (All dimensions are in mms).
13
As a starting point for constructing the V-n diagram for the cruise missile wing, the
velocity is chosen as 238.5 m/s which corresponds to Mach 0.7 at sea level conditions
for the PHAA and NHAA flight conditions. At this Mach number and sea level
conditions, Computational Fluid Dynamic (CFD) analyses of the wing are performed for
different angles of attack using the CFD-FASTRAN program. Then, the negative and
positive stall angle of attack values are predicted by solving the Reynolds Averaged
Navier Stokes Equations with Spalart-Allmaras turbulence model.
The CFD solution domain is a rectangular prism with dimensions 30x30x16 m. The
solution domain consists of 1,276,697 hexahedral cells and 5,000 surface elements. The
mesh spacing in the normal direction to the surface is 2x10-6 m for turbulent flow
calculations. The CFD solution domain and the surface mesh are given in Figure 2.5.
After performing several analyses for different angles of attack, the lift coefficient (CL)
versus angle of attack () curve is constructed. CL- curve is given in Figure 2.6, where
lift coefficient is calculated using the following formula:
L
CL = (2.2)
q.S
14
The CL- graph obtained from Euler solutions is also given in Figure 2.6 for
comparison. As expected, the stall phenomenon can not be predicted by solving Euler
equations. But it can be seen that at small angles of attack, the CL- graph is nearly same
for the viscous and inviscid solutions.
Comparison of u velocity contours at 25%, 50% and 75% spanwise locations for =7o
and =8o is given in Figure 2.7 and for =-4o and =-5o is given in Figure 2.8. u
velocities are the velocity components in chordwise direction and the regions with
negative velocities (u0) in these figures are the regions of separated flow.
1.5
CL
1.0
0.5
AOA (degree)
0.0
-15 -10 -5 0 5 10 15 20
-0.5
3D NS TURBULENT
3D EULER
-1.0
Figure 2.6 Lift coefficient versus angle of attack graph for M=0.7.
The stall angle of attack is predicted when one of the two conditions is provided: the
decrease of CL and flow separation. At M=0.7, the positive stall angle of attack is found
as 8o and the negative stall angle of attack is found as -5o. After =7o CL decreases, but
after =-4o CL continues to increase (Figure 2.6). However, flow separation occurs
beyond both of these angles of attack and it is more severe in board of 75% spanwise
location (Figure 2.7 and 2.8). Downwash effects caused by the vorticies occurring at the
tip of the wing prevents the flow separation at these regions.
15
25% Span, M=0.7, =7o 25% Span, M=0.7, =8o
Figure 2.7. Comparison of u velocity contours at different spanwise locations for M=0.7
=7o and M=0.7 =8o.
16
25% Span, M=0.7, =-4o 25% Span, M=0.7, =-5o
Figure 2.8 Comparison of u velocity contours at different spanwise locations for M=0.7
=-4o and M=0.7 =-5o.
17
From Figure 2.9 and 2.10, it is seen that the pressure coefficient distribution differs for
=7o and =8o especially for upper surface and for =-4o and =-5o especially for
bottom surface where the flow separates. Furthermore occurrence of shock, which is a
sudden reduction of Mach number and increase of pressure, vanishes for the regions
where flow separates. Pressure coefficient is calculated using the following formula
where P is the atmospheric pressure:
P P
CP = (2.3)
q
Figure 2.9 Comparison of pressure coefficient distribution for M=0.7 =7o and M=0.7
=8o.
18
M=0.7, = -4o M=0.7, =-5o
Figure 2.10 Comparison of pressure coefficient distribution for M=0.7 =-4o and M=0.7
=-5o.
Angles of attack just before stall are taken as the PHAA and NHAA which are 7o and
-4o, respectively. Lift coefficients at M=0.7 corresponding to -4o and 7o angles of attack
are found as -0.24 and 0.85 for the cruise missile wing. The load factors, using these lift
coefficients and a weight (W) of 1200 kg in Equation (2.1), are calculated as -0.8 and
2.8.
Rooney and Craig [25] found the aerodynamic coefficients by analytical predictions,
sub-scale and full scale wind tunnel tests and flight tests over the Mach number range
0.45 to 0.83 for the same cruise missile geometry considered in this study. Lift
coefficients at M=0.7 corresponding to -4o and 7o angles of attack were found as -0.4
and 0.85 from the flight test results.
The lift coefficient for =7o at M=0.7 found in this study is the same as the lift
coefficient for =8o found in [25]. The lift coefficient for =8o is found as 0.67 in this
study. The lift coefficient for =-4o at M=0.7 is smaller in magnitude than the lift
coefficient for the same condition found from the flight test results in [25]. But it is
satisfactory to see that the results are comparable since only the wing is modeled and
some geometrical properties given in [25] are disregarded in this study. The disregarded
19
geometrical properties for the wing are the 1.716o incidence at wing station (W.S.) 10
(10 inch from the root), -4o twist about trailing edge from W.S. 10 to W.S. 50.91 and
2.747o dihedral at the trailing edge.
Besides M=0.7, CFD analyses are performed for M=0.6 also to compare the maximum
positive load factors. It is found that after =6o, lift starts to decrease and flow separates
for M=0.6. At this speed the lift coefficient for =6o is calculated as 0.68 which
corresponds to a load factor of 1.7. This result is also consistent with [25], in which
CL,max versus M graph is given (without exact numerical values). From [25], it is known
that the cruise missile can experience a load of 2.8g, hence it is verified that taking
M=0.7 for the PHAA and NHAA condition is an appropriate decision.
After determining the positive and negative load factors to determine the PLAA and
NLAA points (point C and D in Figure 2.1), dive speed is chosen as Mach 0.95 which is
1.3 times the cruise speed of the cruise missile (M=0.72). At this Mach number and sea
level conditions, CFD analyses of the wing are performed for different angles of attack
until the load factors obtained are the same as the load factors for PHAA and NHAA
conditions. The angles of attack, which create -0.8 and 2.8 load factors at M=0.95 are
found as -1.3o and 4.2o, respectively.
In Figure 2.11, Mach number distribution is given for PLAA and NLAA flight
conditions. At these flight conditions, shock occurs both at the top and bottom of the
wing. Furthermore, there is a small region of flow separation at the trailing edge. But
this region is not so widespread to affect the flight performance of the cruise missile.
Pressure coefficient distributions for these flight conditions are given in Figure 2.12.
20
25% Span, M=0.95, =-1.3o 25% Span, M=0.95, =4.2o
Figure 2.11 Mach Contours at different spanwise locations for M=0.95 =-1.3o and
M=0.95 =4.2o.
21
M=0.95, = -1.3o M=0.95, =4.2o
Figure 2.12 Pressure coefficient distribution for M=0.95 =-1.3o and M=0.95 =4.2o.
After determining the four critical flight conditions, V-n diagram is constructed as
shown in Figure 2.13. The parabolic lines are constructed for illustration purposes only
by using the value of CL,max and CL,min for M=0.7 in Equation (2.1). In reality, values of
CL,max and CL,min change with velocity.
In general, some modifications to V-n diagram are made to account for gust loads which
can increase angles of attack. But since cruise missiles are one-shot flight vehicles and
large gust speeds occur rarely, gust effect is ignored in this study.
22
cruise condition
A code written in MATLAB is used to arrange the pressure data in a tabular fashion.
This code creates a rectangle which covers the wing surface, creates nodes by dividing
the rectangle into equal cells and evaluates the pressure values with linear interpolation
at the new nodes. If a node is outside the boundary of the surface, the pressure value of
the nearest node at the wing surface is given. This procedure is repeated for upper skin,
bottom skin and tip of the wing for each of the four flight conditions. Then, the location
and pressure data created by MATLAB is read into ANSYS and a pressure table is
created for each surface of the wing.
23
To verify the sufficiency of the load transfer method, the total force perpendicular to
chordwise direction (Fy) obtained from FASTRAN is compared with the total force
obtained from ANSYS. The mesh in MATLAB is chosen to minimize the percent error
in Fy. For the upper and bottom skins, a 65x65 mesh for the PHAA, NHAA and NLAA
flight conditions, and a 66x66 mesh for the PLAA condition are created in MATLAB.
For the tip rib, 49x49 mesh is created for all of flight conditions. The reason for
choosing Fy for comparison is the dominance of it over the forces in other directions.
For the four critical flight conditions, the percent error in Fy for different element sizes
in ANSYS is given in Table 2.1. The computational times for stress and buckling
analyses for different element sizes are also given in Table 2.1. These analyses are
performed for the aluminum wing for one flight condition using a personal
computer with 3 GHz Intel-Pentium processor and 2 GB of RAM.
As the element size decreases, percent error in Fy decreases, but the increase in
computational time for the analyses are much higher especially when the element size is
smaller than 20 mm. Since too many analyses have to be performed for optimization and
small element size will increase optimization time, element size is chosen as 40 mm for
the skin panels and 30 mm for the spar and rib webs.
Table 2.1 The percent error in the total force perpendicular to chordwise direction.
Wing surface mesh used in ANSYS with element size of 40 mm is given in Figure 2.14.
It must be noted that the sizes of the elements are not actually 40 mm in the structural
model. When an element size is defined in ANSYS, the number of divisions is
24
automatically calculated from the line length. Quotients of the division of line length to
element size is rounded upward to next integer and taken as the number of divisions for
that line. For example, if 40 mm element size is defined for an 81x120 mm rectangle, a
3x3 mesh is created on that area.
Figure 2.14 Wing surface mesh used in ANSYS with element size of 40 mm.
The occurrence of shock has a negative effect on the percent error in Fy, since it causes
high gradients in pressure distribution. When the element size in ANSYS is increased,
the possibility of catching the maximum pressure values decreases.
In Table 2.2, comparison of total forces and moments in all directions obtained from
FASTRAN with the values obtained from ANSYS for the critical flight conditions is
given. The coordinate system is the same for both models in FASTRAN and ANSYS as
given in Figure 2.14. The structural model used has an element size of 40 mm. It can be
seen that other forces and moments are also close to each other. The reason for higher
percent errors in Fy in NHAA and NLAA conditions is because of relatively small
values of Fy. Although the percent error in NHAA condition is the highest, the total
force difference can be ignored.
25
Table 2.2 Comparison of total forces and moments in FASTRAN and ANSYS (Element
size in ANSYS=40 mm)
Fx Fy Fz Mx My Mz
(N) (N) (N) (N.mm) (N.mm) (N.mm)
PHAA
FASTRAN -874 16618 -1418 -9421800 -133970 3204500
ANSYS -874 15860 -1420 -8918555 -147549 3123262
NHAA
FASTRAN -36 -4641 -1603 2615700 368300 -306700
ANSYS -45 -4059 -1609 2275154 350635 -277611
PLAA
FASTRAN 1905 16675 -1377 -9351600 1381700 5090300
ANSYS 1915 16085 -1389 -8991225 1375126 4867998
NLAA
FASTRAN 1965 -4718 -1438 2468400 1439400 -569210
ANSYS 1939 -4377 -1450 2261601 1406833 -580058
The upper surface pressure distribution in FASTRAN, the distribution after transferring
pressure data into MATLAB, and then to ANSYS for the NLAA flight condition can be
seen in Figure 2.15. Although the contour settings are not exactly the same, the pressure
distributions in the three programs seem quite agreeable. But the main justification for
the sufficiency of the load transfer method comes from the comparison of the total
forces and moments obtained from ANSYS and FASTRAN given in Table 2.1 and
Table 2.2.
26
FASTRAN MATLAB ANSYS
27
CHAPTER 3
3.1 INTRODUCTION
The finite element method is a numerical procedure that can be used to obtain solutions
to a large class of engineering problems. ANSYS has many finite element analysis
capabilities, ranging from a simple, linear, static analysis to a complex, nonlinear,
transient dynamic analysis. The growing popularity of structural optimization as a tool
for industrial applications is generating demand for the introduction of optimization
capabilities into general-purpose analysis packages like ANSYS, NASTRAN and I-
DEAS.
There are many advantages of using ANSYS as a basic Finite Element Analysis package
for structural optimization. It integrates preprocessing, solution, postprocessing and
optimization processors in one package and is a multi-physics analysis package for
which coupled analysis is possible. It provides a very powerful macro language (APDL:
ANSYS Parametric Design Language) and also provides interface to call external
programs in macro language, which makes the interaction between the optimizer and the
analysis easier.
In this study, weight minimization of a cruise missile wing is accomplished for both
metal and composite wing configurations using the ANSYS finite element solver and
optimization algorithms. In the first wing configuration all the structural members are
made up of aluminum. In the second wing configuration the skin panels are all
composite and other members (spars, ribs, spar caps and rib caps) are made up of
aluminum.
28
In order to perform optimization analyses in ANSYS, a parametric model for the cruise
missile wing must be built. In the parametric model, the quantities to be used as
optimization variables (design variables, state variables and objective function) must be
available as parameters. By the use of APDL, ANSYS command input files for the two
wing configurations are written using a text editor.
In this chapter, after brief information about finite element method and optimization
procedures in ANSYS, modeling details, boundary conditions, loading types and
optimization variables for wing configurations are presented.
In ANSYS, as the number of design variables increases it gets more difficult to obtain a
global optimum. For this reason, it is important to keep the number of design variables
as low as possible. So techniques used for decreasing the number of design variables are
also presented.
The finite element method is a numerical procedure for analyzing structures and
continua. Usually the problem is too complicated to be solved satisfactorily by classical
analytical methods. The finite element procedure produces many simultaneous algebraic
equations, which are generated and solved on a digital computer [27].
ANSYS is one of the commercial packages which is used to solve structural mechanics
problems. The easiest way to communicate with ANSYS is by using the ANSYS menu
system, called the Graphical User Interface (GUI). The GUI consists of windows,
menus, dialog boxes, and other components that let you enter input data and execute
ANSYS functions simply by picking buttons with a mouse or typing in responses to
prompts. Another way to communicate with ANSYS is to use the ANSYS Parametric
Design Language (APDL) which is a scripting language that one can use to automate
common task or build the finite element model. APDL also provides features like
29
repeating a command, macros, if-then-else branching, do loops, scalar, vector and
matrix operations. APDL is also the foundation for design optimization using ANSYS.
Parameters
Parameters are APDL variables, which are similar to FORTRAN variables. The
parameter does not to be declared as integer or real, and all numeric values are stored as
double-precision values. Scalar and array parameters can be defined in ANSYS.
CYL4, 0, 0, rad
CYL4, 0, 0, 10
which creates a solid circle with radius 10 and center at the origin.
Array parameters can also be defined in ANSYS. This option is useful when distributed
pressure or temperature data is read into ANSYS; ANSYS reads the distributed pressure
data and stores it as an array inside the program with the array parameter name.
By the use of APDL one can input model dimensions, material properties, etc. in terms
of parameters rather than numbers and retrieve information from the ANSYS database,
such as maximum stress and displacement on a node.
30
3.2.2 ANSYS Finite Elements Used in Structural Analysis
SHELL93 Element
SHELL93 is an eight node shell element which is particularly well suited to model
curved shells. The element has six degrees of freedom at each node; translations in the
nodal x, y, and z directions and rotations about the nodal x, y, and z-axes. The
deformation shapes are quadratic in both in-plane directions. The geometry, node
locations, and coordinate systems for this element are shown in Figure 3.1. The element
is defined by eight nodes, four thicknesses, and material properties.
SHELL99 Element
SHELL99 can be used for layered applications of a structural shell model. The element
is defined by eight nodes, average or corner layer thicknesses, layer material direction
angles, and orthotropic material properties. Shell 99 can allow up to 250 layers. The
element has six degrees of freedom at each node; translations in the nodal x, y, and z
directions and rotations about the nodal x, y, and z-axes. The geometry, node locations,
and coordinate systems for this element are shown in Figure 3.2.
31
Figure 3.2 ANSYS SHELL99 element [28].
BEAM189 Element
32
3.3 STRUCTURAL OPTIMIZATION WITH ANSYS
Optimization of an aircraft structure seeks most of the time for a minimum weight
design. The weight of the structural element is the objective function. In general the
choice of the structural design has some limits on it, such as displacement constraints,
allowable stress and natural frequency constraints. In addition, the optimization problem
has some design variables so that by changing the design variables the optimum design
is found. Design variables can be either continuous or discrete.
r r
Minimize f (x) xXn
r
Subjected to h i ( x ) = 0, i = 1,......, n e
r
g j ( x ) 0, j = 1,......, n g
r r r
xL < x < xU
r
In this mathematical formulation x is a vector of n components to describe design
variables and Xn is the domain of design variables. The objective function is designated
r r
with f ( x ) and the equality and inequality constraints are designated with h ( x )
r r r
and g ( x ) , respectively. The elements of the vectors x L and x U are the lower and upper
bounds on the values of design variables.
A typical design optimization problem in ANSYS consists of three parts [28, 29]:
33
Design Variables: Design characteristics that are varied to achieve the objective
(e.g, thickness of the wing skin panel, etc).
State Variables: Conditions that the design must meet. (e.g. allowable stress on
the wing, allowable tip deflection of the wing, etc).
A feasible design is within all the Design Variable (DV) and State Variable ranges.
Generally, the DVs are always inside their permitted range. An infeasible design is one
that violates at least one constraint. ANSYS can achieve an optimum design even if the
initial design is infeasible.
The best design is one that has the lowest objective function value and most closely
meets all the constraints. If no feasible designs are available, the best design is one that
most closely meets all the constraints, not the one with the lowest objective function
value. The design domain (or design space) is the region defined by all possible feasible
designs.
34
Figure 3.5 ANSYS optimization procedure [29].
In ANSYS, optimization begins with building a parametric model of the initial design
and creating an analysis file. The basic requirement is that quantities to be used as DVs,
SVs, and objective function must be parameters. For example, if the objective function
is the weight of the wing then the volume of the wing must be available as a parameter
as well as all the design constraints.
The procedure to build and analyze a parametric model is the same as for a normal
ANSYS analysis except that parameters are used wherever appropriate. The following
steps are recommended in order to build and analyze an initial design [28, 29].
In this step the standard Name=Value format is used to define parameters. For
parameters to be used as DVs, the values specified are used for the initial design only.
Appropriate parameters are used in this step to construct the geometrical model, mesh
and apply the loads and boundary conditions.
35
Step 4 - Parameterize the Results
The results from the initial analysis, such as the maximum stress/deflection, volume of
the geometry are retrieved at this step and stored as parameters. These parameters are
then used as state variables and objective function.
In this step the analysis file is created. The analysis file contains ANSYS commands for
the complete parametric analysis including geometric modeling, meshing, loading,
solution and post processing. The optimizer uses it to loop through multiple designs.
Upon the creation of the analysis file the next step is to enter the optimization module,
specify analysis file and declare the optimization variables. ANSYS allows the user to
specify a single objective function. Maximum 60 design variables and 100 state
variables can be used in the model.
After these steps, the design is optimized using one of the optimization methods
available in ANSYS. In the next section, ANSYS Optimization methods are presented.
36
Approximations for Objective Function and State Variables
State variables are handled in the same manner. An approximation is generated for each
state variable and updated at the end of each loop.
A linear, quadratic or quadratic plus cross terms fit can be used for the curve fitting
approximation.
Convergence Checking
At the end of each loop, a check for convergence is made. The problem is said to have
converged if the current, previous, or best design is feasible and any of the following
conditions are satisfied
The change in objective function from the best feasible design to the current
design is less than the objective function tolerance.
The change in objective function between the last two designs is less than the
objective function tolerance.
37
The changes in all design variables from the current design to the best feasible
design are less then their respective tolerances.
The changes in all design variables between the last two designs are less than
their respective tolerances.
In the First Order Method the constrained optimization problem is turned into an
unconstrained one by adding penalty functions like in the Subproblem Approximation
Method. However in this method no approximation is made and the actual finite element
representation is minimized.
The First Order Method uses gradients of the dependent variables with respect to the
design variables. Each iteration may involve several analyses (loops through the analysis
file) to determine the proper search direction. Various steepest descent and conjugate
direction searches are employed during each iteration. For each iteration, gradient
calculations are performed in order to determine a search direction, and a line search
strategy is adopted to minimize the unconstrained problem.
Convergence Checking
First order iterations continue until convergence is achieved. The problem is said to have
converged if, when comparing the current iteration design set to the previous and best
sets, one of the following conditions is satisfied:
The change in objective function from the best design to the current design is
less than the objective function tolerance.
The change in objective function from the previous design to the current design
is less than the objective function tolerance.
It is also a requirement that the final iteration uses a steepest descent search, otherwise
additional iterations are performed.
38
3.3.3 Exploring the Design Domain
Using the optimization tools of ANSYS, different designs can be tested in order to
measure and understand the design domain of the problem. The following tools are
available [28, 29]:
This tool performs one loop and produces one FEA solution at a time. Single Loop Run
provides a quick look at the design with a set of design variables.
Multiple loops are performed, with random design variable values at each loop. This
tool is useful for studying the overall design space, and for establishing feasible design
sets for subsequent optimization analysis.
Sweep Generation
Starting from a reference design set, this tool generates several sequences of design sets.
Specifically, it varies one design variable at a time over its full range using uniform
design variable increments. This tool makes global variational evaluations of the
objective function and of the state variables possible.
Factorial Evaluation
This is a statistical tool that is used to generate design sets at all extreme combinations
of design variable values. The primary aim is to compute main and interaction effects
for the objective function and the state variables.
Gradient Evaluation
At a user-specified reference design set, this tool calculates the gradients of the objective
function and the state variables with respect to the design variables. Local design
sensitivities can be investigated using this tool.
39
3.4 MODELING DETAILS FOR THE WING
A two-sided L-shaped cross section (which looks like a T-shaped section all together) is
used for spar caps and a one-sided L-shaped cross-section for rib caps. In general, caps
are connected to skin panels, spars and ribs with rivets. Cap flanges and webs for such
connections should have enough lengths to insert rivets. Since the distance between the
upper and lower skins for the wing geometry is too narrow, upper spar/rib cap webs
almost touch lower spar/rib cap webs when adequate distance to use rivets for
connections is considered. So, in the finite element model, thickness of the spar cap web
is included into thickness of the spar web as seen in Figure 3.6.
The flanges of the spar caps are modeled with rectangular cross section beam elements.
Likewise, thickness of the rib cap web is included into thickness of the rib web and
flanges are modeled with beam elements of a rectangular cross section.
40
By default, nodes of a shell element are located at the midsurface of the element and
nodes of a beam element are located at the centroid of the element. Offset elements are
used in the model of skin panels so that the upper side of the element coincides with the
wing geometry. Since SHELL93 does not have node offset option, layered shell-type
elements (SHELL99) are used for both aluminum and composite skins. Beam elements
that are used to model spar caps and rib caps are also offset so that they obey the wing
geometry. Offset skin panel, spar cap and rib cap elements are shown in Figure 3.7 and
Figure 3.8. The rectangular cross-sections for the beams with proper offset distances are
created using common sections provided by ANSYS.
Using non-offset elements results in higher wing moment of inertia values which leads
to unconservative results. Besides, it causes wrong moment calculations. For very thin
shell elements, the difference between the results obtained from offset and non-offset
elements is negligible. But, in preliminary studies it has been seen that the difference
cannot be neglected for the wing considered in present study.
41
Figure 3.8 Offset skin panel and rib cap elements.
After the Tomahawk cruise missile is launched from a platform, its wings, tail fins and
engine air intake is deployed, allowing the missile to cruise towards its target. Doors in
the missiles flanks open to let the wings swing out, then close again to streamline the
slots through which the wings protrude [30].
Rotation of the wing over a pivot can be achieved by a mechanism which is connected
to spars. So, this unfolding behavior causes the root of the spars to be the most critical
part of the wing structure. Wing pivot and slot door is shown in Figure 3.9.
42
Figure 3.9 Wing pivot and slot door of Tomahawk cruise missile [25].
In the finite element model, displacement degrees of freedom in all directions (ux, uy, uz)
are constrained at the root of the spars and displacement degree of freedom in y
direction (uy) is constrained at the root of skin panels as shown in Figure 3.10.
43
Four load cases are created to account for all critical flight conditions (PHAA, PLAA,
NHAA and NLAA) mentioned in Chapter 2. For each flight condition, pressure tables
for upper surface, lower surface and tip rib are created separately. Pressures are applied
on the wing surfaces using these tables. The pressure values at the nodes of the wing
surfaces are automatically calculated by linear interpolation using the pressure tables.
Besides the surface pressures obtained from FASTRAN, atmospheric pressure (1 atm) is
applied on the inner surfaces of the wing.
To decrease the number of design variables, thicknesses of the skins, spars, ribs and
spar/rib cap flanges and widths of the spar/rib cap flanges are defined as a function of
the spanwise location. The functions provide the tapering of the thickness and width
values from root to tip. For the thicknesses and width of the spar/rib cap flanges, linear
functions are used. For the thicknesses of skins, spars and ribs, parabolic functions are
used. A parabolic function is defined with three constants. The three constants can be
the thickness/width value at the root and tip sections and the slope at the tip section. In
defining the parabolic functions the slope of the thickness function is taken zero at the
tip. Thus, a parabolic function with minimum cross-section area is used, and in addition
number of design variables is decreased. For a root thickness of 5 mm and a thickness
ratio (root thickness/tip thickness) of 10, graphs of the linear and parabolic thickness
functions are given in Figure 3.11.
44
Figure 3.11 Graphs of the linear and parabolic thickness functions.
45
To be able to define different thickness/width values at each node as a function of
spanwise location, a subprogram is written in the ANSYS input files. This subprogram
determines the spanwise locations of the nodes for each element from the root, then
calculates the thickness/width values using the location values. Each shell element has a
thickness which decreases linearly in the spanwise direction while each beam element
has a constant thickness and width values which are given the average of their nodal
values calculated by the subprogram. An exaggerated view of the parabolically reduced
thickness of the upper skin panels is shown in Figure 3.13.
Figure 3.13 The exaggerated view for parabolically reducing thickness of the upper skin
panels.
To aid reduction of the number of variables, ribs locations are defined with only one
variable. Toward that end, the distance between two adjacent ribs is increased linearly
from root to tip since the loading at the root side is more severe. A subprogram is
written in ANSYS to find the locations of the ribs by using the number of ribs (n) and
the slope of the distance ratio function (m). The distance ratio function is defined as
x=m(i-1)+1, i = 1, 2, ..., n
46
where, i is the rib number from root to tip. This function determines the ratio of the
distance between the ith rib and the (i-1)th rib to the distance between the first rib and the
root. In this definition, 0th rib corresponds to the root. As a result the distance ratios
between two adjacent ribs are taken as 1: m+1: 2m+1m(n-1)+1 from root to tip. For
example, if the slope is zero, then the distances between the ribs are equal; if the slope is
one, then the distance ratios are taken as 1, 2, 3n from root to tip. The graph of the
distance ratio function for different slope values is given in Figure 3.14.
The design variables defined in the ANSYS input file, abbreviations and the upper and
lower bounds are listed in Table 3.1. As mentioned before spar/rib thicknesses include
also spar/cap flange thicknesses. The locations of the spars are the distances from the
leading edge given as percent of the chord length.
47
Table 3.1 Design variables, abbreviations and bounds for the aluminum wing.
The reason for taking the ratios (root value/tip value) as design variables instead of tip
values for thicknesses and widths is to provide the tapering of thicknesses and widths
from root to tip for all conditions.
48
process) at the optimization module. With this command optimization loop starts from
preprocessor module and design variables are processed during looping.
The analysis file is created with 20 design variables. But after some trial analyses it has
been seen that spar/rib cap flange widths can be taken as constant since the minimum
and maximum values are very close to each other. The spar cap flange width is taken as
linearly decreasing from 12.5 mm at the root to 9 mm at tip. The rib cap flange width is
taken as linearly decreasing from 12 mm at root to 9 mm at the tip.
Optimization analyses are continued with 16 design variables. Among them, four design
variables determine the number of ribs and location of ribs and spars and changing the
values of these design variables lead to mesh change. After examining the results of the
optimization analyses, it is observed that design variables for the best design have the
values which create the largest element sizes and the results are highly mesh-dependent.
So, it is decided to take the design variables which lead to a mesh change as constant.
The values taken for these design variables are:
The value of 0.5 for m means that the distance ratios are taken as 1, 1.5, 2, 2.5 from root
to tip as seen in Figure 3.14. For subsequent analyses, the inner and outer meshes of the
wing structure with 1636 elements and 4138 nodes are given in Figure 3.15.
49
Figure 3.15 Mesh of the wing structure.
From the preliminary studies it has been seen that high strength aluminum is necessary
for the cruise missile wing. So, 7075 series aluminum is used which has a yield strength
of nearly 400 MPa. The material properties for the aluminum are taken as
E=70000 MPa
=0.3
=2700 kg/m3
After the static analysis, a buckling (eigenvalue) analysis is performed for each of the
four load cases mentioned above. After these analyses, maximum von Mises stresses for
shell elements and axial stresses for beam elements in each portion of the wing,
buckling load factors (eigenvalues) and the volume of the wing are retrieved and
parameterized. Axial stresses for beam elements are referred to as von Mises stresses,
since other stress components are negligible compared to axial stresses in a beam
element.
Stresses are retrieved using element tables which use the average results for top, middle
or bottom of shell elements. So, top and bottom stresses for shell elements are retrieved
separately, and then the maximum of the results for each portion is used. All the
elements from root to first rib constitutes the first portion, including the first rib; all the
elements from first rib to second rib constitutes the second portion, excluding the first
rib and including the second rib and so on until the fourth portion.
50
The state variables (strength, buckling and size constraints) defined in the ANSYS input
file, abbreviations and the upper and lower bounds are listed in Table 3.2. Objective
function for the optimization analyses is the mass of the wing.
Table 3.2 State variables, abbreviations and bounds for the aluminum wing.
51
3.7 OPTIMIZATION VARIABLES FOR THE COMPOSITE
WING
For the composite wing, four ribs and two spars are used as in the aluminum wing with
the same locations. The spar/rib cap flange widths are also taken constant having the
same values with the aluminum wing for the same reason as explained in the previous
section.
Optimization of composite structures requires too many design variables to be used. But
as mentioned earlier, ANSYS optimization module does not have the capability of
dealing with too many design variables. So for the composite wing, four portions are
separately optimized. The elements constituting these four portions are as in the
aluminum wing.
In each portion, spar web and spar flange thicknesses are defined as a function of the
spanwise location. The functions provide the linear reduction of the thicknesses from the
beginning to the end of each portion.
Composites differ from metals having coupling behavior between in-plane and bending
deformations. In other words, in-plane loads can create out-of-plane deformations in
addition to in-plane deformations and vice versa. This behavior is described with the
following formulae [3]:
N = B+Ao (3.1)
o
M= B +D (3.2)
where,
52
Detailed formulations for the matrices given above can be found in almost every book
about composites, [3] being one of them.
In this study symmetric balanced stacking sequence is used for the composite skin
panels having layers with 0o, 90o and 45o orientations. For each portion, numbers of
each orientation for upper and lower skin panels are taken as design variables.
As the B matrix vanishes with the use of symmetric laminates, coupling between in-
plane and out-of-plane responses of a laminate is eliminated. Balanced laminates are
obtained by placing a -45o ply orientation for every occurrence of a 45o one. Vanishing
of A16 and A26 terms in A matrix with the use of balanced laminates, shear-extension
coupling is eliminated. Furthermore, balanced laminates with adjacent layers of 45o and
-45o orientation have smaller bending-twisting coupling terms (D16 and D26).
For a wing skin panel, in-plane loads are usually dominant. So, considering only number
of layers for each orientation is a good approximation for the strength calculations of
wing skin panels. But on the contrary, buckling is an out-of-plane phenomenon which
depends on flexural stiffness matrix and is highly affected by the stacking sequence.
Initially, number of layers for each ply orientation subject to strength constraints alone
was optimized for the first two portions of the wing. Buckling analyses were then
performed for different stacking sequences. It was seen that buckling was not a problem
for the first portion, since the skin panels were thick enough to resist buckling. The
second portion, on the other hand, buckled whatever the stacking sequence was, since
the thin skin panels were sufficient for strength constraints only. So, it is understood that
strength and buckling optimization must be performed simultaneously.
One of the remedies to handle such a complex problem is to create a response surface
for the optimum buckling load as a function of number of 0o, 90o and 45o stacks and in-
53
plane loads and use genetic algorithms for stacking sequence optimization [16, 17]. In
these studies analytical buckling formulations are used.
Genetic algorithms are effective in producing global optimum solutions when only a few
variables are involved, and analyses are not so expensive [33]. So, use of genetic
algorithm may not be effective for the stacking sequence optimization for the wing
considered in this study.
In the current study, stacking sequence is not dealt with and a stacking sequence of
[(90o)n90/(0o)n0/(45o)n45]s is used for every skin panels, where n90, n0 and n45 are half
the number of 90o, 0o, 45o layers, respectively. In 0o layers, fibers are in the chordwise
direction. In 90o layers, fibers are in the spanwise direction. Layer thickness is taken as
0.125 mm. With the assumption of perfectly bonded layers, it is possible to take (90o)n90
layers as one layer with a thickness of 0.125x(n90) and (0o)n0 layers as one layer with a
thickness of 0.125x(n0).
The design variables defined in the ANSYS input file, abbreviations and the upper and
lower bounds for the first portion are listed in Table 3.3. As in the aluminum wing,
spar/rib web thicknesses include also spar/rib cap web thicknesses. For the other
portions, thicknesses of spar web and spar cap flanges at the beginning of each portion
are taken the same as the values of the thicknesses at the end of the previous portions.
So, these portions have two design variables less than the first portion.
54
Number of layers for each orientation found in an optimization iteration are used at the
following iteration by rounding it to an integer number as explained in the previous
section.
In order to use 0 as a lower bound for number of layers for each orientation, design
variables are used with adding 1 to them since design variables cannot take the value
0.
Table 3.3 Design variables, abbreviations and bounds for Portion-1 of the composite
wing.
After the static analysis, a buckling (eigenvalue) analysis is performed for each of the
four load cases as in the aluminum wing. After these analyses, the volume of the portion
optimized, maximum von Mises stresses for shell elements and axial stresses for beam
elements for the aluminum parts and buckling load factors (eigenvalues) are retrieved
and parameterized.
55
Stresses of the portion optimized are retrieved using element tables which use the
average results for top, middle or bottom of shell elements. So, top and bottom stresses
are retrieved separately, and then the maximum of the results for each portion is used.
There are various types of failure criteria for composite layers: The most frequently used
ones are the maximum stress, maximum strain, Tsai-Hill, Hoffman and Tsai-Wu
criteria. While the application of maximum stress and maximum strain criteria is
straightforward, these criteria fail to represent interactions of different stress
components in failure mechanisms. Quadratic failure criteria similar to the von Mises
criterion such as Tsai-Hill, Hoffman and Tsai-Wu can account for this interaction [3].
Tsai-Wu criterion is one-of the most commonly used among others.
The failure criteria mentioned above are all referred to as first-ply failure criteria since
only the most critical ply is considered. Failure of only one layer does not always mean
failure of the whole structure. So, in some applications degraded laminate failure (or
referred to as progressive failure) analysis is performed in which each layer of a
laminate is assumed to be degraded and, thus, to carry loads with a reduced
performance.
In ANSYS, three different failure criteria for composites are available: maximum stress,
maximum strain and Tsai-Wu. For Tsai-Wu failure criteria there are two different
formulations used by different researchers. ANSYS offers both of them and
distinguishes them with the names Tsai-Wu and inverse Tsai-Wu. The criterion referred
to as Tsai-Wu in ANSYS is the 3-D version of the failure criterion reported in Tsai and
Hahn [34] and Grdal et al. [3]. The criterion referred to as inverse Tsai-Wu in ANSYS
is the 3-D version of the failure criterion reported in of Tsai [35] and Barbero [36].
In this study, maximum stress, maximum strain and inverse Tsai-Wu failure criteria are
used. For the inverse Tsai-Wu failure criterion, -1 is used for the coupling coefficient
(Cxy) which is the default value in ANSYS and the recommended value in Barbero [36].
Cxy is twice the value of F*xy used by Tsai and Hahn and the value of f12 given in
Barbero. The reason for selecting the inverse Tsai-Wu instead of Tsai-Wu failure
criterion is explained later.
56
Failure criteria values are calculated in integration points or nodes of elements in
ANSYS. So, a subprogram is written in the ANSYS input file in order to use element
stresses and strains for the failure criteria value calculations for the portion optimized.
The state variables (strength, buckling and size constraints) defined in the ANSYS input
file, abbreviations and the upper and lower bounds for the first portion are listed in
Table 3.4. The state variables for the other portions are the same except for the
abbreviations for the size constraints. Objective function for the optimization analyses is
the mass of the wing.
Table 3.4 State variables, abbreviations and bounds for Portion-1 of the composite wing.
smax_alu1
Max. von Mises stress for aluminum
smax_alu2
elements for PHAA, NHAA, PLAA,
smax_alu3 - 400
NLAA (MPa)
smax_alu4
57
For the aluminum parts, 7075 series aluminum is used with the same material properties
given in the previous section. In order to choose the composite material, four different
composite types are compared first. Mechanical properties used for these composites
with a fiber volume fraction of 0.6 are given in Table 3.5.
Table 3.5 Mechanical properties of the unidirectional fiber reinforced epoxy composites
[37].
S: strength
: failure strain
1: longitudinal (fiber) direction
2: transverse (perpendicular to fiber) direction
t: tension
c: compression
58
The geometry and stacking properties used in comparing the four composite materials
are given in Table 3.6. The thicknesses of the portions are 5.5, 5, 4, 3 mm from root to
tip for both upper and lower skin panels.
Table 3.6 The geometric properties of the wing used in the comparison analyses.
The strength and buckling results for the four flight conditions are given in Table 3.7.
fcs1, fce1, fctw1, fctwi1 are the abbreviations for maximum stress, maximum
strain, Tsai-Wu and inverse Tsai-Wu failure criteria values respectively for the first load
case (PHAA flight condition).
59
Table 3.7 Results of the analyses for Portion-1 for different epoxy composites.
60
With the use of e-glass/epoxy, skin panels become very critical (high failure criteria
values) for PHAA and PLAA conditions when strength is considered, and also von
Mises stresses for the aluminum parts are greater than the yield stress of 7075. It is
understood from the results that root thickness must be very thick if e-glass/epoxy is
used for the skin panels.
With the use of graphite/epoxy, von Mises stresses for the aluminum parts become
lower than the yield stress of 7075. Furthermore, as the modulus of the graphite/epoxy
increases, von Mises stresses for the aluminum parts decrease and buckling load factors
increase. But, mass and failure criteria values for composite skin panels increase. So,
high strength graphite is used for skin panels, since the strength for skin panels are more
critical than the aluminum parts.
Tsai-Wu failure criteria values are much greater than the other criteria values for PHAA
and NHAA conditions as seen in Table 3.7. So, different failure criteria values are also
compared for a split disk test conducted in accordance with ASTM D 2290-00 [38]
which is presented in Appendix C. It is observed that, Tsai-Wu failure criterion gives
conservative results compared to other three failure criteria, namely maximum stress,
maximum strain and inverse Tsai-Wu.
61
CHAPTER 4
4.1 INTRODUCTION
For the subproblem method, initially N+2 design sets must exist to form the
approximations (N=number of DVs). Otherwise, random design sets are generated until
the required number is obtained. By supplying known good designs the quality of the
approximations can be improved.
For the optimization of two wing configurations, initial designs are created with sweep
and random tool, and then several restarts are performed using subproblem method and
N+2 best designs of the previous run. After some restarts, only one best design is
selected for the next optimization run. By this way, the optimizer is forced to follow a
different path by creating some random design sets.
After performing some optimization executions, certain design variables are eliminated
for subsequent optimizations. Then analyses are performed using the first order method.
Finally local sensitivity analyses using the gradient tool are performed to obtain a better
design.
In this chapter, the procedures followed when performing optimization analyses and the
results are presented for both the aluminum and composite wing.
62
4.2 RESULTS FOR THE ALUMINUM WING
Although the ANSYS input file is created with 20 design variables given in Table 3.1, 8
variables are decided to be taken constant as explained in the previous chapter.
Furthermore, from the global sensitivity analyses performed with the sweep tool in
ANSYS, it is seen that the effect of the rib web and rib flange thicknesses on the
strength and buckling results are negligible. So, for the subsequent optimization
analyses, 4 design variables representing rib web/cap flange thicknesses and ratios are
eliminated and taken constant as:
After every static analysis, von Mises stresses for each element; and after every buckling
(eigenvalue) analyses, buckling load factors are retrieved. To see the effect of the
number of state variables for stresses and buckling load factors, three input files are
created with different number of state variables. In addition to the four size constraints
for the tip thicknesses for upper skin, bottom skin, spar web and spar cap flange, the
following constraints are added:
In the first input file (OPT1), two state variables are used: maximum von Mises stress
and minimum buckling load factor of the four load cases.
In the second input file (OPT2), eight state variables are used: maximum von Mises
stress and minimum buckling load factor for each of the four load cases.
In the third input file (OPT3), twenty state variables are used: maximum von Mises
stress for each of the four portions and four load cases and minimum buckling load
factor for each of the four load cases.
A design with 8.55 kg weight found from the previous trial optimization analyses is
used as the initial design. Optimization analyses with 16 executions are performed using
subproblem method for the three input files.
63
Weight change with the number of executions for three optimization analyses with
different number of state variables is given in Table 4.1. ANSYS refers to each iteration
as a set. Number of sets (iterations) in each execution is counted from the very
beginning.
Table 4.1 Weight change with the number of executions for the aluminum wing.
For the optimization analysis with six state variables, the best design is obtained in the
13th execution with 814 sets; for the optimization analysis with twelve state variables,
the best design is obtained in the 9th execution with 255 sets; for the optimization
analysis with twenty-four state variables, the best design is obtained in the 9th execution
with 291 sets.
As the number of state variables is increased from six to twelve, a great enhancement is
achieved in optimization time and weight minimization, and a further enhancement is
obtained with the use of twenty-four state variables.
The location or load case for the maximum stress and the load case for the maximum
buckling load factor may change in each loop. So, choosing only one maximum for the
stress and one minimum for the buckling load factor may result in poor quality
64
approximations for the state variables. On the other hand, choosing so many state
variables may result in a local minimum. Using twenty-four state variables seems to
work well for subsequent optimization analyses.
It is also observed that only one execution is not enough for the wing optimization. By
using restarts with the best designs of previous executions, much better designs are
obtained.
The initial design and best design optimization variables for three analyses are given in
Table 4.2. Percent mass reduction with respect to initial design is also given in Table
4.2. The meaning for each abbreviation is given in the previous chapter in Table 3.1 and
Table 3.2.
To obtain a better design, optimization analyses are continued by taking the best design
given in Table 4.2 as the initial design using the subproblem method with smaller
objective function tolerance. Then the first order method and again the subproblem
method are used with the best designs of previous analyses. After all, local sensitivity
analyses are performed with the gradient tool.
The best design optimization variables and percent mass reduction for subsequent
optimization analyses with different methods are given in Table 4.3. Percent mass
reduction is calculated with respect to initial design given in Table 4.2.
65
Table 4.2 Best design optimization variables for three analyses with different number of
state variables for the aluminum wing.
66
Table 4.3 Best design optimization variables for subsequent optimization analyses for
the aluminum wing.
67
After nine executions with subproblem method (OPT3 results given in Table 4.1 and
Table 4.2), little changes are obtained with the subsequent executions with the
subproblem and first order methods as seen from Table 4.3. However, by only slightly
changing the design variables with the gradient tool, more reduction in weight is
obtained compared to other methods after performing several restarts with subproblem
method.
First order method is so slow and found as inefficient for the optimization of the wing.
One iteration takes approximately 35 minutes for the first order method and 2 minutes
for the subproblem method using a personal computer with 3 GHz Intel-Pentium
processor and 2 GB of RAM.
After gradient analyses, slight changes are obtained for the values of four design
variables representing the root thickness and ratio of the spar web (tr_sw, t_ratio_sw),
and the root thickness and ratio of the spar cap flange (tr_sc, t_ratio_sc) leading to the
lower bound values for the tip thicknesses of spar web and spar cap flanges (tt_sw,
tt_sc).
The PHAA (smax1_i) and PLAA (smax3_i) flight conditions are more critical than the
NHAA (smax2_i) and NLAA (smax4_i) flight conditions for strength. It is an expected
result since the load factors are higher for the PHAA and PLAA conditions as explained
in Chapter 2.
It can be seen from Table 4.3 that buckling is more critical rather than strength for the
aluminum wing. Because, buckling load factors for PHAA, PLAA and NLAA flight
conditions (buckling1, buckling3, buckling4) take the lower bound value 1, whereas
maximum von Mises stresses (smax3_1) are below the upper bound value 400 MPa
except the first portion for the PLAA flight condition.
The upper skin thickness (4.05 mm) is thicker than the lower skin thickness (2.82 mm)
as it is expected. The aerodynamic loads on the wing during PHAA and PLAA flight
conditions bend it upwards, as a result the upper side of the wing is loaded in
compression and buckling becomes a problem at these regions. In addition, the
aerodynamic loads on the wing during NHAA and NLAA flight conditions bend it
68
downwards, as a result the lower side of the wing is loaded in compression and buckling
becomes a problem at these regions. But since positive load factor is greater than the
negative load factor in magnitude, upper skin panels are more critical than the lower
skin panels.
Root thickness for the spar web is found to be very close to the upper bound value. This
is an expected result because of the boundary conditions which are selected considering
the mechanism for unfolding wings. This choice of boundary condition makes the root
of the spars to be the most critical part of the wing structure as explained in Section 5 of
Chapter 3.
The von Mises stress distribution for top of the shell elements are given in Figure 4.1
and for bottom of the shell elements are given in Figure 4.2. The axial stress distribution
for beam elements is given in Figure 4.3. Stress distributions given are the results of the
PLAA flight condition. The skin panels carry also bending load especially at the root as
understood from the difference in results between top and bottom of shell elements.
Uy distributions for the most critical panels obtained from the buckling analyses for each
flight condition are given in Figure 4.4. Upper skin panels at the second and third
portions are the most critical parts for PHAA and PLAA flight conditions, lower skin
panels at the third portion are the most critical parts for NHAA flight condition and
lower skin panels at the fourth portion are the most critical parts for NLAA flight
condition when considering buckling.
Hence, the best design weight for the wing is found as 15.652 kg (7.826x2). Weight for
the wing is 1.3% of the total weight of the cruise missile. For different aircrafts this ratio
changes between 7.5-12% [1]. But using the weight estimation formulae for light utility
aircraft given in [1], the estimated weight is found as 19.241 kg for the cruise missile
wing which is 1.6% of the total weight of the cruise missile. The estimated weight is
greater than the best design weight found by ANSYS as expected. This is because,
weight found for the cruise missile will increase when connection elements are included
and fatigue and thermal loads are taken into consideration. Another factor that reduces
the weight of the wing is the absence of control surfaces.
69
Figure 4.1 The von Mises stress (MPa) distribution for top of the shell elements for
PLAA flight condition for the aluminum wing.
Figure 4.2 The von Mises stress (MPa) distribution for bottom of the shell elements for
PLAA flight condition for the aluminum wing.
70
Figure 4.3 The axial stress (MPa) distribution for beam elements for PLAA flight
condition for the aluminum wing.
Figure 4.4 Uy distributions obtained from the buckling analyses for each flight condition
for the aluminum wing.
71
4.3 RESULTS FOR THE COMPOSITE WING
For the composite wing, four portions are separately optimized. The optimization
analyses are started for the first portion which includes all the elements from root to first
rib, including the first rib. When this portion is optimized the values of the design
variables for the other portions are taken as constant. These values are chosen after some
preliminary analyses in order to provide the other portions not to buckle and also not
have an unrealistic geometry. Because, buckling load factor is given only for the most
critical part of the wing structure after a buckling analysis. So, it is important to provide
the buckling of the portion to be optimized instead of others in order to have a good
approximation for the buckling. The selected design variable values for other portions
and the values of initial design for the first portion are given in Table 3.6.
After the first portion is optimized, optimization analyses are started for the second
portion which includes all the elements from first rib to second rib, excluding the first
rib and including the second rib. When this portion is optimized, the values of the design
variables for the first portion are taken as constant which has the values of best design of
the first portion. The values of the design variables for the third and fourth portions and
the values of initial design for the second portion are the same with the values used in
the first portion optimization as given in Table 3.6. Optimization analyses have the same
procedure for the third and fourth portions.
From the global sensitivity analyses performed with the sweep tool in ANSYS, it is seen
that the effect of the rib web and rib flange thicknesses on the strength and buckling
results are negligible as in the aluminum wing. So, for the subsequent optimization
analyses, two design variables representing rib web and rib cap flange thicknesses (t_rw,
t_rc) are eliminated for each portion and taken constant as given in Table 3.6.
It must be also noted that the upper bound for the root thickness of the spar web (tr_sw)
is taken as 10 mm in the first analyses but it is seen that infeasible designs are created by
the optimizer. So, root thickness of the spar web is increased to 15 mm for the
subsequent analyses.
72
4.3.1 Optimization of the First Portion without Buckling
Constraints
Table 4.4 Best design optimization variables for Portion-1 of the composite wing
without buckling constraints.
73
Without buckling constraints, upper skin thickness (2.75 mm) is found as nearly half of
the bottom skin thickness (5.75 mm). In PHAA and PLAA flight conditions which is
more critical than the NHAA and NLAA flight conditions, upper skin panels are loaded
in compression and lower skin panels are loaded in tension. This is the reason of thicker
lower skin panels since the tensile strength is much smaller than the compressive
strength in transverse direction for high strength carbon/epoxy material (Table 3.5).
It is also observed that numbers of 0o ply orientations for both upper and lower skin
panels are found as zero after optimization analysis for the first portion. The stacking
sequences of the best design are (907/(45)2)s for the upper skin panels and (905/(45)9)s
for the lower skin panels. This is an indication of much greater stiffness requirement in
spanwise direction than the requirement in chordwise direction as expected. Because, 0o
ply orientations increase the stiffness in chordwise direction.
Root thickness for the spar web is found to be very close to the upper bound as in the
aluminum wing. This is an expected result because of the boundary conditions as
explained in the previous section.
After performing the optimization analyses for the first portion, buckling analyses are
performed and minimum buckling load factor is obtained as 0.92 for the PLAA flight
condition at the upper skin panels of the first portion. Uy displacement distributions
obtained from the buckling analyses for each flight condition is given in Figure 4.5.
Lower skin panels at the second portion is more critical than the first portion for NHAA
and NLAA flight conditions as seen in Figure 4.5.
74
PHAA NHAA PLAA NLAA
(upper skin) (lower skin) (upper skin) (lower skin)
Figure 4.5 Uy displacement distributions obtained from the buckling analyses for each
flight condition for composite wing (optimization results for Portion-1 without buckling
constraints).
In order to have a buckling load factor greater than one, analyses are repeated for
different stacking sequences of the upper skin panels for the first portion and at the first
trial a feasible design is obtained. The buckling load factors with different stacking
sequences are given in Table 4.5. The maximum stress values and failure criteria values
for the stacking sequences given in Table 4.5 are very close to each other.
Table 4.5 Buckling load factors for different stacking sequences of the upper skin panels
for Portion-1 of the composite wing without buckling constraints.
75
Buckling load factors for the NHAA and NLAA flight conditions (buckling2,
buckling4) are nearly the same for different stacking sequences given in Table 4.5, since
buckling occurs at the lower skin panels for these flight conditions and the stacking
sequence of the lower skin panels is not changed. Furthermore, buckling load factors are
very large for NHAA and NLAA flight conditions. Because, lower skin panels are very
thick to have sufficient strength for tensile loading in PHAA and PLAA flight
conditions.
Using a stacking sequence of (902/45/904/45/90)s for the upper skin panels of the first
portion instead of (907/(45)2)s, 9% increase in the buckling load factor for the PHAA
flight condition (buckling1) and 15% increase in the buckling load factor for the PLAA
flight condition (buckling3) is obtained. Differences in buckling load factors are caused
by the differences in flexural stiffness matrix (D) for different stacking sequences.
Using the best design optimization variables for the first portion given in Table 4.4,
optimization analyses are then performed for the second portion without buckling
constraints. The best design optimization variables found for the second portion are
given in Table 4.6.
End thicknesses for spar web and spar cap flanges take the lower bound values after
optimization analyses and the thicknesses for both upper and lower skin panels are
found as 1.5 mm as seen in Table 4.6. The stacking sequence of the best design is
(90/03/45)s for the upper skin panels and (902/(45)2)s for the lower skin panels of the
second portion.
76
Table 4.6 Best design optimization variables for Portion-2 of the composite wing
without buckling constraints.
After performing the optimization analyses for the second portion, buckling analyses are
performed and buckling load factors are obtained as 0.252, 0.620, 0.248, 0.482 for the
PHAA, NHAA, PLAA and NLAA flight conditions respectively. The minimum
buckling load factor is obtained for the PLAA flight condition at the upper skin panels
of the second portion. So, analyses are repeated for different stacking sequences of the
upper skin panels of the second portion for the PLAA flight condition. The buckling
load factors with different stacking sequences are given in Table 4.7.
77
Table 4.7 Buckling load factors for different stacking sequences of the upper skin panels
for Portion-2 of the composite wing without buckling constraints.
buckling3
(03/45/90)s 0.143
(0/45/0/90/0)s 0.201
(0/90/45/02)s 0.242
(90/03/45)s 0.248
(45/90/03)s 0.260
(90/0/45/02)s 0.264
(90/45/03)s 0.276
The buckling load factor changes between 0.143 and 0.276 as seen in Table 4.7. The
upper skin panels of the second portion buckle whatever the stacking sequence is, since
very thin skin panels are sufficient for strength constraints. It is also observed that if the
90o layers are on the outside and the 0o layers are in the interior, maximum buckling
load factor is obtained; minimum buckling load factor is obtained in the opposite case.
The optimization analyses without buckling constraints are stopped at this point and, it
is understood that strength and buckling optimization must be performed
simultaneously. Then the optimization analyses are performed with the addition of
buckling constraints for each portion.
Initially, optimization analyses are performed for the first portion. After creating initial
designs with sweep and random tools, several restarts are performed using subproblem
method. After a while, it is observed that the stresses and failure criteria values are much
smaller than the allowable values and the buckling load factors are much greater than
one for the NHAA and NLAA flight conditions. It is also observed that, root thickness
of the spar web takes the upper bound value 15 mm, numbers of 0o ply orientations for
both upper and lower skin panels take the lower bound value 0 and the thickness of
the spar cap flanges at the end of the first portion takes the lower bound value 0.5 mm.
So for the subsequent analyses, two load cases are eliminated (NHAA and NLAA flight
78
conditions) and four design variables (n0_us1, n0_bs1, t_ratio_sc1, tr_sw1) are taken as
constant. By this way, weight minimization is obtained in a shorter time. Then
optimization analyses are finished with the gradient tool. The best design variables for
the first portion are given in Table 4.8.
Table 4.8 Best design optimization variables for Portion-1 of the composite wing.
The stacking sequences of the best design for the first portion are (908/(45)2)s for the
upper skin panels and (906/(45)9)s for the lower skin panels. When compared with the
results without buckling constraints (Table 4.4), an increase of 0.25 mm in both the
upper and lower skin thicknesses is obtained with the addition of buckling constraints.
79
In addition, half the number of 45o and 0o ply orientations are the same for both of the
cases, and numbers of 90o ply orientations obtained with buckling constraints are one
more than the numbers of 90o ply orientations obtained without buckling constraints.
Mass of the best design for the first portion is found as 2.115 kg which is very close to
the mass found without buckling constraints (2.058 kg) as expected. Because,
thicknesses are very thick to have sufficient strength for the loading in PHAA and
PLAA flight conditions.
After the optimization of the first portion is completed, optimization analyses are started
for the second portion. Since the thickness of the spar cap flanges is found to be the
lower bound value 0.5 mm at the end of the first portion, thickness of the spar cap
flanges at the second portion is taken as constant (t_ratio_sc2=1). In addition, after a
while it is observed that, the thickness of the spar web at the end of the second portion
takes the lower bound value 1.5 mm. So, the subsequent optimization analyses are
continued with six design variables (half the number of 0o, 90o and (45/-45)o ply
orientations).
Optimization analyses with two different initial designs for the second portion lead to
two different best designs with a mass of 1.64 kg. Then subsequent analyses are
performed using these two designs as initial designs. After performing several restarts
using subproblem method, an optimum design with a mass of 1.57 kg is obtained which
is 50% higher than the mass obtained without buckling constraints (1.044 kg). Local
sensitivity analyses are also performed for the best design by decreasing the each
number of ply orientations by 1, but for six of the analyses infeasible designs are
obtained. The best design variables for the second portion are given in Table 4.9.
The stacking sequences of the best design for the second portion are (90/0/(45)6)s for
the upper skin panels and (904/(45)2)s for the lower skin panels. When compared with
the results without buckling constraints (Table 4.6), thicknesses are increased from 1.5
mm to 3.5 mm for the upper skin panels and from 1.5 mm to 2 mm for the lower skin
panels.
80
The upper skin thickness for the second portion (3.5 mm) is found as greater than the
upper skin thickness for the first portion (3 mm). A reduction in skin panels from root to
tip can be obtained by adding a rib in between the first and second ribs.
It is seen from Table 4.9, that strength constraints are smaller than the upper bound
values and buckling load factors are very close to lower bound value of one. So,
buckling constraints have primary importance on the thicknesses of the skin panels for
the second portion.
Table 4.9 Best design optimization variables for Portion-2 of the composite wing.
81
4.3.5 Optimization of the Third and Fourth Portions
With the completion of optimization analyses for the second portion, optimization
analyses for the third portion then for the forth portion are performed. Since the
thickness of the spar cap flanges is found to be the lower bound value 0.5 at the end of
the first portion and the thickness of the spar web at the end of the second portion is
found to be the lower bound value 1.5, thicknesses of the spar cap flanges and spar
webs at the third and fourth portions are taken as constant. The best design variables for
the third portion are given in Table 4.10.
Table 4.10 Best design optimization variables for Portion-3 of the composite wing.
82
The stacking sequences of the best design for the third portion are (90/02/(45)4)s for the
upper skin panels and (902/02/45)s for the lower skin panels. The thickness is found as
2.75 mm for the upper skin panels and as 1.5 mm for the lower skin panels.
Using different methods, objective function tolerances and limits for the design
variables, three optimum designs are obtained for the fourth portion with the same mass:
- The stacking sequence of the first alternative design is (90/03/45)s for the upper
skin panels and (90/0/45)s for the lower skin panels.
- The stacking sequence of the second alternative design is (90/0/(45)2)s for the
upper skin panels and (02/45)s for the lower skin panels.
- The stacking sequence of the third alternative design is (90/45/03)s for the upper
skin panels and ((45)2)s for the lower skin panels.
The buckling coefficients are nearly the same, but strength values differ for these
optimum designs. Maximum von Mises stresses are nearly the same for the second and
third alternative designs, but failure criteria values are smaller for the third one. First
alternative design has the smallest values for the maximum von Mises stresses, but
failure criteria values are larger than the third one. Although maximum von Mises
stresses are so small compared to yield stress of aluminum, failure criteria values are
close to one for the NLAA flight condition in the first alternative design. So, the third
alternative design is chosen for the fourth portion of the composite wing.
The best design variables for the fourth portion are given in Table 4.11 for the first
alternative and in Table 4.12 for the second alternative.
Subsequent analyses are performed using these three designs as initial designs. But a
better design cannot be obtained by this way. It is also observed that when there is more
than one best designs with the same objective function value, ANSYS lists the best
design with the smallest set number. However, in order to use the best design for a
subsequent analysis, ANSYS selects the best design with the largest set number.
83
It must be also noted that, no better results can be obtained with the first-order method
for any portions of the composite wing and this method is found so slow and inefficient
as in the aluminum wing.
Table 4.11 Best design optimization variables for Portion-4 of the composite wing
(1st alternative).
84
Table 4.12 Best design optimization variables for Portion-4 of the composite wing
(3rd alternative, chosen one).
It is seen from Table 4.10 and Table 4.12, that strength constraints are smaller than the
upper bound values and buckling load factors are very close to lower bound value of
one. So, buckling constraints have also primary importance on the thicknesses of the
skin panels for the third and fourth portions as for the second portion.
With the completion of optimization analyses for the fourth portion, optimum design is
obtained for the whole composite wing with a total mass of 11.432 kg (5.716x2). With
the use of carbon/epoxy instead of aluminum for the skin panels, a mass reduction of
27% is obtained.
85
4.3.6 Final Checks and Adjustments for the Best Design of the
Composite Wing
For the composite wing, when a portion is optimized the values of the design variables
for the other portions are taken as constant. Although the values of the design variables
for the portions at the root side have the best design values, the values of the design
variables for the portions at the tip side of the wing are taken different from the best
designs of each portion. So, after the completion of optimization analyses for the four
portions, strength constraints are recalculated for the first, second and third portions.
Comparison of the values of the strength constraints found after the optimization of a
portion with the values for the best design is given in Table 4.13. The constraint values
under the column of before in Table 4.13 is the values obtained after the optimization
analyses for that portion which are given before in Table 4.8, Table 4.9 and Table 4.10.
The constraint values under the column of after is the values obtained for that portion
for the best design of the wing. In addition, P-1, P-2 and P-3 represents Portion-1,
Portion-2 and Portion-3 respectively.
Table 4.13 Comparison of the values of the strength constraints found after the
optimization of a portion with the values for the best design for Portions 1-3 of the
composite wing.
86
As seen from Table 4.13, there are negligible changes for the values of the strength
constraints except the maximum von Mises stresses for the aluminum parts of the
second and third portions for the NLAA flight condition (smax_alu4). When the stresses
are examined, it is seen that the location of the maximum stress is changed from spar
web to rib web. Although these stresses are still much below the yield stress of the
aluminum, for a similar study it is recommended to include the rib at the root side and
exclude the rib at the tip side for the portion optimized.
Uy distributions for the most critical panels obtained from the buckling analyses for each
flight condition are given in Figure 4.6. Upper skin panels at the first and second
portions are the most critical parts for PHAA and PLAA flight conditions, lower skin
panels at the second portion are the most critical parts for NHAA and NLAA flight
conditions when considering buckling.
Figure 4.6 Uy displacement distributions obtained from the buckling analyses for each
flight condition for the composite wing.
87
Laminates with more than four contiguous layers of the same fiber orientation are
generally assumed to be not practical because of thermal stresses created during the
curing process, which can lead to matrix cracking [3]. So, different stacking sequences
for the most critical panels with not more than four contiguous layers, stress and
buckling analyses are repeated for the four load cases.
The values of the strength and buckling constraints for different stacking sequences of
the upper skin panels for the first portion is given in Table 4.14 and for different
stacking sequences of the lower skin panels for the second portion is given in
Table 4.15.
For the upper skin panels of the first portion, fourth stacking sequence
(902/45/904/45/902)s gives the best results compared to other stacking sequences as
seen in Table 4.14. For the lower skin panels for the second portion, first stacking
sequence (45/902/45/902)s gives the best results compared to other stacking
sequences as seen in Table 4.15. But it is also seen that, there is not much difference in
the values of the strength and buckling constraints for different stacking sequences.
88
Table 4.14 The values of the strength and buckling constraints for different stacking
sequences of the upper skin panels for Portion-1 of the composite wing.
Table 4.15 The values of the strength and buckling constraints for different stacking
sequences of the lower skin panels for Portion-2 of the composite wing.
89
CHAPTER 5
CONCLUSION
The load factors for the cruise missile wing were found as +2.8 and -0.8. The negative
load factor was found a little small compared to other types of airplanes. In FAR 23
[24], it is mentioned that negative load factor may not be less than 0.4 times the positive
load factor for the normal and utility airplanes and not less than 0.5 times the positive
load factor for the acrobatic airplanes. But since a cruise missile need not make an
upside down flight, a negative load factor smaller than one in magnitude is acceptable.
Nevertheless using a higher velocity for the NHAA flight condition would increase the
absolute value of the negative load factor.
The pressure distributions for the flight conditions at the corner points of the V-n
diagram were then transferred to ANSYS to create the four load cases of the
optimization analyses. Since the fluid and the structural meshes do not match at the
wing surface, transfer of the pressure data creates some errors. As the element size of
the structural model decreases, errors also decrease. But, since too many analyses have
to be performed for optimization and small element size will increase optimization time,
90
a compromise was made between the transfer error and optimization time for the
element size.
The design variables which determines the number of ribs and location of ribs and spars
were taken as constant since changing the values of these design variables caused the
mesh to change and the results were highly mesh-dependent. However, it is possible to
include these design variables using a finer mesh in the finite element model and
performing the analyses using more powerful computers.
For the composite skin panels, maximum stress, maximum strain and inverse Tsai-Wu
failure criteria were used. For Tsai-Wu failure criteria there are two different
formulations used by different researchers. ANSYS offers both of them and
distinguishes them with the names Tsai-Wu and inverse Tsai-Wu. Different failure
criteria values were compared for a split disk test and it was observed that, Tsai-Wu
failure criterion gives conservative results compared to other three failure criteria.
91
After the optimization analyses, total mass of the wing was found as 15.652 kg for the
aluminum wing and 11.432 kg for the composite wing. So, a mass reduction of 27% was
obtained with the use of carbon/epoxy instead of aluminum for the skin panels.
Optimization analyses were successfully performed for the aerodynamic loads for four
different flight conditions. Since the weight of the cruise missile wing was so small,
gravity loads were not included. For a future study, other types of loadings such as
thermal loads can be taken into consideration.
Structural optimization may easily involve more than 10 design variables, which is
about the number that can be handled by ANSYS efficiently. A code written specifically
for structural optimization purposes, GENESIS being one of them, may be preferable in
a future study.
92
REFERENCES
[1] Niu M.C.Y., Airframe Structural Design, 2nd Edition, Hong Kong Conmilit
Press Ltd., Hong Kong ,2002, pp 247.
[2] Bruhn E.F., Analysis and Design of Flight Vehicle Structures, 2nd Edition,
Jacobs Publishing, Inc., Indianapolis, IN, 1973.
[3] Grdal Z., Haftka R.T., Hajela P., Design and Optimization of Laminated
Composite Materials, John Wiley & Sons, Inc, New York, 1999.
[5] Buckling of Thin Walled Circular Cylinders, NASA SP-8007, NASA Space
Vehicle Design Criteria (Structures), September 1965.
[6] Venkataraman S., Haftka R.T., Structural Optimization: What Has Mooress
Law Done for Us, AIAA-2002-1342, 43rd AIAA/ASME/ASCE /AHS/ASC
Structures, Structural Dynamics and Materials Conference, Denver, CO, Apr.
22-25, 2002.
[7] Khot N.S., Venkayya V.B., Johnson C.D., Tischler V.A., Optimization of
Fiber Reinforced Composite Structures, International Journal of Solids and
Structures, 1973. Vol.9, pp 1225-1236.
[8] Khot N.S., Venkayya V.B., Berke L., Optimum Design of Composite
Structures with Stress and Displacement Constraints, AIAA Journal, Vol.14
No.2, February 1976, pp 131-132.
[10] Haftka R., Starnes J.H., Preliminary Design of Composite Wings for
Buckling, Strength and Displacement Constraints, AIAA Journal, Vol.16
No.8, August 1979, pp 564-570.
93
[11] Liu I.W., Lin C.C., A Refined Optimality Criterion Technique Applied to
Aircraft Wing Structural Design, Computers & Structures, Vol.33 No.2,
1989, pp 427-434.
[12] Liu I.W., Lin C.C., Optimum Design of Wing Structures by a Refined
Optimality Criterion, Computers & Structures, Vol.17, 1991, pp 51-65.
[14] Yurkovich R., The Use of Taguchi Techniques with the ASTROS Code for
Optimum Wing Structural Design, AIAA-94-1484
[15] Rhl P.J., Mavris D.N., Schrage D.P., HSCT Wing Design Through
Multilevel Decomposition AIAA 95-3944, 1st Aircraft Engineering,
Technology and Operations Congress, Los Angeles, CA, Sept. 19-21, 1995.
[16] Liu B., Haftka R.T., Akgn M.A., Composite Wing Structural Optimization
Using Genetic Algorithms and Response Surfaces AIAA 98-4854,
7th AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis
and Optimization, St Louis, MO, Sept. 2-4, 1998.
[17] Liu B., Haftka R.T., Akgn M.A., Two-Level Composite Wing Structural
Optimization Using Response Surfaces Structural and Multidisciplinary
Optimization, Vol.20, 2000, pp 87-96.
[18] Liu B., Haftka R.T., Composite Wing Structural Design Optimization with
Continuity Constraints AIAA 2001-1205, 42nd AIAA/ASME/ASCE/
AHS/ASC Structures, Structural Dynamics and Materials Conference, Seattle,
WA, Apr. 16-19, 2001.
[19] Liu B., Haftka R.T., Akgn M.A., Todoroki A., Permutation Genetic
Algorithm for Stacking Sequence Design of Composite Laminates,
Computational Methods in Applied Mechanical Engineering, Vol. 186, 2000,
pp 357-372
[20] Kapania K.R., Chun S., Preliminary Design of a Structural Wing-Box under a
Twist Constraint, AIAA 2003-2004, 44th AIAA/ASME/ASCE/ AHS/ASC
Structures, Structural Dynamics and Materials Conference, Norfolk, VA, Apr.
7-10, 2003.
[21] Engelstad S.P., Barker D.K., Ellsworth C.S., Proctor L., Optimization
Strategies for the F/A-22 Horizontal Stabilizer, AIAA 2003-1456, 44th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials
Conference, Norfolk, VA, Apr. 7-10, 2003.
94
[22] Peery D.J., Azar J.J., Aircraft Structures, McGraw- Hill, Inc.,1982
[26] Reddy J.N., Mechanics of Laminated Composite Plates and Shells: Theory
and Analysis, Boca Raton, CRC Press, 2nd edition, 2004
[27] Cook R.D., Malkus D.S., Plesha M.E., Concepts and Applications of Finite
Element Analysis, 3rd Edition, John Wiley & Sons, New York, 1989.
[29] Design Optimization Training Manual for Release 5.6, Volume 1; Mar.2000,
ANSYS Inc.
[32] Lekhnitskii S.G., Anisotropic Plates, Translated from the second Russion
edition by Tsai S.W and Cheron T., Gordon and Breach Science Publishers
Inc, 1968
[33] Juan P.L., Ghosh D.K., Rastogi N., A New Approach in Stacking Sequence
Optimization of Composite Laminates Using GENESIS Structural Analysis
and Optimization Software, AIAA 2002-5451, 9th AIAA/ISSMO Symposium
on Multidisciplinary Analysis and Optimization, Atlanta, Georgia, 4-6 Sept.
2002
[34] Tsai S.W. and Hahn H.T., Introduction to Composite Materials, Section 7.2,
Technomic Publishing Company, 1980
95
[35] Tsai S.W., Composites Design, Third Edition, Section 11.6, Think
Composites, Dayton, Ohio, 1987
[38] ASTM D 2290-00, Standard Test Method for Apparent Hoop Tensile
Strength of Plastic or Reinforced Plastic Pipe by Split Disk Method, 2000
96
APPENDIX A
!INPUT VARIABLES
wr_sc=12.5 !root width of the spar cap flange
w_ratio_sc=12.5/9 !root width/tip width for the spar cap flange
wr_rc=12 !root width of the rib cap flange
w_ratio_rc=12/9 !root width/tip width for the rib cap flange
n=4 !number of ribs except root rib
m=0.5 !space between two ribs (if 0: equal space, if 1: 1, 2n)
c1=25 !% c placement of spar1
c2=60 !% c placement of spar2
tr_rw=1 !root thickness of the rib web
t_ratio_rw=1 !root thickness/tip thickness for the rib web
tr_rc=1 !root thickness of the rib cap flange
t_ratio_rc=2 !root thickness/tip thickness for the rib cap flange
!DESIGN VARIABLES
tr_us=4.7898 !root thickness of the upper skin
t_ratio_us=5.0699 !root thickness/tip thickness for the upper skin
tr_ls=4.3175 !root thickness of the lower skin
t_ratio_ls=7.3244 !root thickness/tip thickness for the lower skin
tr_sw=7.9125 !root thickness of the spar web
t_ratio_sw=5.1151 !root thickness/tip thickness for the spar web
tr_sc=3.9442 !root thickness of the spar cap flange
t_ratio_sc=6.6761 !root thickness/tip thickness for the spar cap flange
/PREP7
!resuming the tables of pressure distributions for different flight conditions and array of
!airfoil data
PARRES,CHANGE,'pressure_and_airfoil_data_NS-opt'
97
tt_rc=tr_rc/t_ratio_rc !tip thickness of the rib cap flange
wt_rc=wr_rc/w_ratio_rc !tip width of the rib cap flange
!calculating x(i) (for i:1n) for finding the z locations of the ribs (x(i)=m(i-1)+1)
!It makes the ratios of distances between two ribs as 1:m+1:2m+1m(n-1)+1 from root
!to tip.
x=0
xt=0
*do,i,1,n
xt=xt+m*(i-1)+1
*enddo
!L.E. radius = 0.455 percent c
!slope of mean line at LE = 0.084
98
*enddo
FITEM,3,26+i*6
BSPLIN, ,P51X
!creating L.E. radius with fillet
LFILLT,1+i*8 ,2+i*8 ,0.455*c/100, ,
!adding two lines at the upper and lower parts of the rib
LCOMB,1+i*8 ,4+i*8 ,0
LCOMB,2+i*8 ,5+i*8 ,0
!reversing the lower trailing edge, upper leading&mid edge lines of the ribs
*do,i,0,n
lreverse,2+8*i,0
lreverse,4+8*i,0
lreverse,7+8*i,0
*enddo
99
ln=(n+1)*8 !total line numbers on the ribs
an=(n+1)*3 !max area number on the ribs
!CREATING COMPONENTS
ASEL,S, , ,4,an,1,1
CM,RIBS,AREA
ASEL,S, , ,an-2,an,1,1
CM,TIP_RIB,AREA
ASEL,S, , ,an+1,an+1+n*8,8,
ASEL,A, , ,an+2,an+2+n*8,8,
CM,SPARS,AREA
ASEL,S, , ,an+3,an+5,1,
*do,i,0,n-1
ASEL,A, , ,an+3+i*8,an+5+i*8,1,
*enddo
CM,UPPER_SKINS,AREA
ASEL,S, , ,an+6,an+8,1,
*do,i,0,n-1
ASEL,A, , ,an+6+i*8,an+8+i*8,1,
*enddo
CM,LOWER_SKINS,AREA
ASEL,S, , ,an+3,an+8,1,
*do,i,0,n-1
ASEL,A, , ,an+3+i*8,an+8+i*8,1,
*enddo
CM,SKINS,AREA
100
lsel,s,,,ln+6,ln+6+(n-1)*6,6
*do,i,0,2
lsel,a,,,ln+3+i,ln+3+i+(n-1)*6,6
*enddo
CM,SPAR_CAPS,LINE
lsel,s,,,ln+3,ln+3+(n-1)*6,6
lsel,a,,,ln+5,ln+5+(n-1)*6,6
CM,UPPER_SPAR_CAPS,LINE
lsel,s,,,ln+4,ln+4+(n-1)*6,6
lsel,a,,,ln+6,ln+6+(n-1)*6,6
CM,LOWER_SPAR_CAPS,LINE
LSEL,S, , ,9,15,3
*do,i,1,n
LSEL,A, , ,1+i*8,7+i*8,3
LSEL,A, , ,2+i*8,8+i*8,3
*enddo
CM,RIB_CAPS,LINE
LSEL,S, , ,9,15,3
*do,i,2,n
LSEL,A, , ,1+i*8,7+i*8,3
*enddo
CM,UPPER_RIB_CAPS,LINE
LSEL,S, , ,10,16,3
*do,i,2,n
LSEL,A, , ,2+i*8,8+i*8,3
*enddo
CM,LOWER_RIB_CAPS,LINE
ALLSEL,ALL
ET,2,BEAM189
ET,3,SHELL99
KEYOPT,3,2,1 !tapered layer
KEYOPT,3,8,1 !storage of all layers data
KEYOPT,3,11,2 !nodes @ top face (for the skins)
!MESHING
es1=40 !element size
101
ESIZE,es1,0
MSHKEY,1 !MAPPED MESH
type,3
AMESH, UPPER_SKINS
AMESH, LOWER_SKINS
!defining the beam section orientations and properties for the spar cap flanges with the
!appropriate keypoints
!LATT, MAT, REAL, TYPE, --, KB, KE, SECNUM
lsel,s,,,ln+3,ln+3+(n-1)*6,6
LATT,1,,2, ,4,,1
lsel,s,,,ln+4,ln+4+(n-1)*6,6
LATT,1,,2, ,3,,2
lsel,s,,,ln+5,ln+5+(n-1)*6,6
LATT,1,,2, ,6,,1
lsel,s,,,ln+6,ln+6+(n-1)*6,6
LATT,1,,2, ,5,,2
!defining the beam section orientations for the rib cap flanges with the appropriate
!keypoints
*do,i,1,n
lsel,s,,,1+i*8
LATT,1,,2, , ,6+i*6,3
lsel,s,,,4+i*8
LATT,1,,2, , ,4+i*6,3
lsel,s,,,7+i*8
LATT,1,,2, ,4+i*6,6+i*6,3
lsel,s,,,2+i*8
LATT,1,,2, , ,5+i*6,4
lsel,s,,,5+i*8
LATT,1,,2, , ,3+i*6,4
lsel,s,,,8+i*8
LATT,1,,2, ,3+i*6,5+i*6,4
*enddo
ALLSEL
LMESH,SPAR_CAPS
LMESH,RIB_CAPS
102
CM,UPPER_SKINS_ELEMENTS,ELEM
CMSEL,S,LOWER_SKINS
ESLA,S
CM,LOWER_SKINS_ELEMENTS,ELEM
CMSEL,S,SPARS
ESLA,S
CM,SPARS_ELEMENTS,ELEM
CMSEL,S,SPAR_CAPS
ESLL,S
CM,SPAR_CAPS_ELEMENTS,ELEM
CMSEL,S,UPPER_SPAR_CAPS
ESLL,S
CM,UPPER_SPAR_CAPS_ELEMENTS,ELEM
CMSEL,S,LOWER_SPAR_CAPS
ESLL,S
CM,LOWER_SPAR_CAPS_ELEMENTS,ELEM
CMSEL,S,RIBS
ESLA,S
CM,RIBS_ELEMENTS,ELEM
CMSEL,S,RIB_CAPS
ESLL,S
CM,RIB_CAPS_ELEMENTS,ELEM
CMSEL,S,UPPER_RIB_CAPS
ESLL,S
CM,UPPER_RIB_CAPS_ELEMENTS,ELEM
CMSEL,S,LOWER_RIB_CAPS
ESLL,S
CM,LOWER_RIB_CAPS_ELEMENTS,ELEM
!/ESHAPE,1.0
103
lsel,s,,,ln+1+(i-1)*6,
en=(distkp(1+(i-1)*6,7+(i-1)*6))/es1 !number of elements on a line
*if,en,gt,nint(en),then
en=nint(en)+1
*else
en=nint(en)
*endif
dz=(kz(7+(i-1)*6)-kz(1+(i-1)*6))/en !element length in z direction
*do,j,1,en
z1=kz(1+(i-1)*6)+dz*(j-1)
z2=kz(1+(i-1)*6)+dz*j
!only the following two commands differ from the linearly reducing thickness definition
tz1=tr_us-(tr_us-tt_us)*2/bh*z1+(tr_us-tt_us)/bh**2*z1**2
tz2=tr_us-(tr_us-tt_us)*2/bh*z2+(tr_us-tt_us)/bh**2*z2**2
!for Shell99
r, rn+j,1,0
rmore
rmore,1,0,tz1,tz1,tz2,tz2
Lzs1=wr_sc-(wr_sc-wt_sc)*z1/bh
Lzs2=wr_sc-(wr_sc-wt_sc)*z2/bh
Lzs=(Lzs1+Lzs2)/2
!Defining the T-beam section properties for the upper spar cap flanges
SECTYPE,sn+j,BEAM,RECT,
SECOFFSET,USER,0,-(tz1+tz2+tzs)/2
SECDATA,2*Lzs,tzs
CMSEL,S,UPPER_SKINS_ELEMENTS
NSEL,S,LOC,Z,z2
ESLN,R
EMODIF,ALL,REAL,rn+j,
CMSEL,S,UPPER_SPAR_CAPS_ELEMENTS
NSEL,S,LOC,Z,z2
ESLN,R
EMODIF,ALL,SEC,sn+j,
*enddo
sn=sn+en
rn=rn+en
*enddo
104
!PARABOLICALLY REDUCING THICKNESS FOR LOWER SKINS AND
!LINEARLY REDUCING WEB-FLANGE THICKNESS&WIDTH FOR LOWER
!SPAR CAP FLANGES
*do,i,1,n !number of lines along half span
lsel,s,,,ln+1+(i-1)*6,
en=(distkp(1+(i-1)*6,7+(i-1)*6))/es1 !number of elements on a line
*if,en,gt,nint(en),then
en=nint(en)+1
*else
en=nint(en)
*endif
dz=(kz(7+(i-1)*6)-kz(1+(i-1)*6))/en !element length in z direction
*do,j,1,en
z1=kz(1+(i-1)*6)+dz*(j-1)
z2=kz(1+(i-1)*6)+dz*j
tz1=tr_ls-(tr_ls-tt_ls)*2/bh*z1+(tr_ls-tt_ls)/bh**2*z1**2
tz2=tr_ls-(tr_ls-tt_ls)*2/bh*z2+(tr_ls-tt_ls)/bh**2*z2**2
r, rn+j,1,0
rmore
rmore,1,0,tz1,tz1,tz2,tz2
Lzs1=wr_sc-(wr_sc-wt_sc)*z1/bh
Lzs2=wr_sc-(wr_sc-wt_sc)*z2/bh
Lzs=(Lzs1+Lzs2)/2
!Defining the T-beam section properties for the lower spar cap flanges
SECTYPE,sn+j,BEAM, RECT,
SECOFFSET,USER,0,-(tz1+tz2+tzs)/2
SECDATA,2*Lzs,tzs
CMSEL,S,LOWER_SKINS_ELEMENTS
NSEL,S,LOC,Z,z2
ESLN,R
EMODIF,ALL,REAL,rn+j,
CMSEL,S,LOWER_SPAR_CAPS_ELEMENTS
NSEL,S,LOC,Z,z2
ESLN,R
EMODIF,ALL,SEC,sn+j,
*enddo
sn=sn+en
rn=rn+en
*enddo
105
!PARABOLICALLY REDUCING THICKNESS FOR SPAR WEBS
*do,i,1,n !number of lines along half span
lsel,s,,,ln+1+(i-1)*6,
en=(distkp(1+(i-1)*6,7+(i-1)*6))/es1 !number of elements on a line
*if,en,gt,nint(en),then
en=nint(en)+1
*else
en=nint(en)
*endif
dz=(kz(7+(i-1)*6)-kz(1+(i-1)*6))/en !element length in z direction
*do,j,1,en
z1=kz(1+(i-1)*6)+dz*(j-1)
z2=kz(1+(i-1)*6)+dz*j
tz1=tr_sw-(tr_sw-tt_sw)*2/bh*z1+(tr_sw-tt_sw)/bh**2*z1**2
tz2=tr_sw-(tr_sw-tt_sw)*2/bh*z2+(tr_sw-tt_sw)/bh**2*z2**2
r,rn+j,tz1,tz1,tz2,tz2
CMSEL,S,SPARS_ELEMENTS
NSEL,S,LOC,Z,z2
ESLN,R
EMODIF,ALL,REAL,rn+j,
*enddo
rn=rn+en
*enddo
r,rn+i,tz
!Defining the L-beam section properties for the upper rib cap flanges
SECTYPE,sn+i,BEAM, RECT,
SECOFFSET,USER,-(tz+Lzr)/2,-tzu-tzr/2
SECDATA,Lzr,tzr
106
!Defining the L-beam section properties for the lower rib cap flanges
SECTYPE,sn+n+i,BEAM,RECT,
SECOFFSET,USER,-(tz+Lzr)/2,-tzb-tzr/2
SECDATA,Lzr,tzr
CMSEL,S,RIBS_ELEMENTS
NSEL,S,LOC,Z,z
ESLN,R
EMODIF,ALL,REAL,rn+i,
CMSEL,S,UPPER_RIB_CAPS_ELEMENTS
NSEL,S,LOC,Z,z
ESLN,R
EMODIF,ALL, SEC,sn+i
CMSEL,S,LOWER_RIB_CAPS_ELEMENTS
NSEL,S,LOC,Z,z
ESLN,R
EMODIF,ALL, SEC,sn+n+i
*enddo
rn=rn+n
sn=sn+2*n
LSEL,S,LOC,Z,0,
DL,ALL,,UY
CMSEL,S,SPARS
ALLSEL,BELOW,AREA
LSEL,R,LOC,Z,0
DL,ALL,,UX
DL,ALL,,UZ
107
!LOAD CASE 2 (NHAA)
CMSEL,S,UPPER_SKINS
SFA,ALL,1,PRES, 0.1
SFA,ALL,2,PRES, %XZP_UST_M07_AOA_4%
CMSEL,S,LOWER_SKINS
SFA,ALL,1,PRES, 0.1
SFA,ALL,2,PRES, %XZP_ALT_M07_ AOA_4%
CMSEL,S, TIP_RIB
SFA,ALL,1,PRES, 0.1
SFA,ALL,2,PRES, %XYP_TIP_M07_ AOA_4%
ALLSEL,ALL
LSWRITE,2
/POST1
ETABLE,VOLU,VOLU
SSUM
108
*GET,V,SSUM,,ITEM,VOLU
mass=v*ro*1000 !kg
port=0
ASEL,S, , ,4+port*3,6+port*3,1,1 !(port+1)th portion of the ribs
ASEL,A, , ,an+1+port*8,an+2+port*8 !(port+1)th portion of the spars
ASEL,A, , ,an+3+port*8,an+8+port*8,1, !(port+1)th portion of the skins
(composite areas)
ESLA,S
shell,top
ETABLE,seqvtop,S,EQV
ESORT,ETAB,seqvtop
*GET, seqvtop,SORT,,MAX
shell,bot
ETABLE,seqvbot,S,EQV
ESORT,ETAB,seqvbot
*GET, seqvbot,SORT,,MAX
ASEL,S, , ,4+port*3,6+port*3,1,1 !(port+1)th portion of the ribs
ASEL,A, , ,an+1+port*8,an+2+port*8 !(port+1)th portion of the spars
ALLSEL,BELOW,AREA
ESLL,S !selecting elements associated with the lines
ETABLE,AXS,LS,1
ESORT,ETAB,AXS,0,1, , !sort elements on absolute value descending
order
*GET,axs,SORT,,MAX
smax1_1=max(seqvtop, seqvbot, abs(axs))
port=1
ASEL,S, , ,4+port*3,6+port*3,1,1 !(port+1)th portion of the ribs
ASEL,A, , ,an+1+port*8,an+2+port*8 !(port+1)th portion of the spars
ASEL,A, , ,an+3+port*8,an+8+port*8,1 !(port+1)th portion of the skins
!(composite areas)
ESLA,S
shell,top
ETABLE,seqvtop,S,EQV
ESORT,ETAB,seqvtop
*GET, seqvtop,SORT,,MAX
shell,bot
ETABLE,seqvbot,S,EQV
ESORT,ETAB,seqvbot
*GET, seqvbot,SORT,,MAX
ASEL,S, , ,4+port*3,6+port*3,1,1 !(port+1)th portion of the ribs
ASEL,A, , ,an+1+port*8,an+2+port*8 !(port+1)th portion of the spars
ALLSEL,BELOW,AREA
ESLL,S !selecting elements associated with the lines
ETABLE,AXS,LS,1
ESORT,ETAB,AXS,0,1, , !sort elements on absolute value descending
order
*GET,axs,SORT,,MAX
smax1_2=max(seqvtop, seqvbot, abs(axs))
109
port=2
ASEL,S, , ,4+port*3,6+port*3,1,1 !(port+1)th portion of the ribs
ASEL,A, , ,an+1+port*8,an+2+port*8 !(port+1)th portion of the spars
ASEL,A, , ,an+3+port*8,an+8+port*8,1, !(port+1)th portion of the skins
(composite areas)
ESLA,S
shell,top
ETABLE,seqvtop,S,EQV
ESORT,ETAB,seqvtop
*GET, seqvtop,SORT,,MAX
shell,bot
ETABLE,seqvbot,S,EQV
ESORT,ETAB,seqvbot
*GET, seqvbot,SORT,,MAX
ASEL,S, , ,4+port*3,6+port*3,1,1 !(port+1)th portion of the ribs
ASEL,A, , ,an+1+port*8,an+2+port*8 !(port+1)th portion of the spars
ALLSEL,BELOW,AREA
ESLL,S !selecting elements associated with the lines
ETABLE,AXS,LS,1
ESORT,ETAB,AXS,0,1, , !sort elements on absolute value descending
order
*GET,axs,SORT,,MAX
smax1_3=max(seqvtop, seqvbot, abs(axs))
port=3
ASEL,S, , ,4+port*3,6+port*3,1,1 !(port+1)th portion of the ribs
ASEL,A, , ,an+1+port*8,an+2+port*8 !(port+1)th portion of the spars
ASEL,A, , ,an+3+port*8,an+8+port*8,1, !(port+1)th portion of the skins
(composite areas)
ESLA,S
shell,top
ETABLE,seqvtop,S,EQV
ESORT,ETAB,seqvtop
*GET, seqvtop,SORT,,MAX
shell,bot
ETABLE,seqvbot,S,EQV
ESORT,ETAB,seqvbot
*GET, seqvbot,SORT,,MAX
ASEL,S, , ,4+port*3,6+port*3,1,1 !(port+1)th portion of the ribs
ASEL,A, , ,an+1+port*8,an+2+port*8 !(port+1)th portion of the spars
ALLSEL,BELOW,AREA
ESLL,S !selecting elements associated with the lines
ETABLE,AXS,LS,1
ESORT,ETAB,AXS,0,1, , !sort elements on absolute value descending
order
*GET,axs,SORT,,MAX
smax1_4=max(seqvtop, seqvbot, abs(axs))
ALLSEL
/SOLU
ANTYPE,BUCKLE !Buckling analysis
110
BUCOPT,LANB,1 !Use Block Lanczos solution method, extract 1
mode
MXPAND,1 !Expand 1 mode shape
SOLVE
/POST1
*GET,buckling1,MODE,1,FREQ
opvar, tr_us,dv,2,6
opvar, t_ratio_us,dv,2,12
opvar, tr_ls,dv,2,6
opvar, t_ratio_ls,dv,2,12
opvar, tr_sw,dv,4,15
opvar, t_ratio_sw,dv,2,10
opvar, tr_sc,dv,1,5
opvar, t_ratio_sc,dv,2,10
opvar,smax1_1,sv,,400
opvar,smax1_2,sv,,400
opvar,smax1_3,sv,,400
opvar,smax1_4,sv,,400
opvar,smax2_1,sv,,400
opvar,smax2_2,sv,,400
opvar,smax2_3,sv,,400
opvar,smax2_4,sv,,400
opvar,smax3_1,sv,,400
opvar,smax3_2,sv,,400
opvar,smax3_3,sv,,400
opvar,smax3_4,sv,,400
opvar,smax4_1,sv,,400
opvar,smax4_2,sv,,400
opvar,smax4_3,sv,,400
opvar,smax4_4,sv,,400
opvar, buckling1,sv,1
opvar, buckling2,sv,1
opvar, buckling3,sv,1
opvar, buckling4,sv,1
opvar, tt_us,sv,0.5
111
opvar, tt_ls,sv,0.5
opvar, tt_sw,sv,1.5
opvar, tt_sc,sv,0.5
opvar,mass,obj,,,
optype,sweep
opsweep,,10
opexe
OPSAVE,' KANAT_alu-1'
OPSEL,10
optype,random
oprand,50,50
opexe
OPSAVE,' KANAT_alu-2'
OPSEL,10
optype,subp
OPSUBP,100,100
opexe
OPSAVE,'KANAT_alu-3'
opvar,mass,obj,,,0.001
OPSEL,10
optype,subp
OPSUBP,100,100
opexe
OPSAVE,'KANAT_alu-4'
OPSEL,10
optype,subp
OPSUBP,100,100
opexe
OPSAVE,'KANAT_alu-5'
OPSEL,1
optype,subp
OPSUBP,100,100
opexe
OPSAVE,'KANAT_alu-6'
112
APPENDIX B
Nx, Ny: normal loads per unit length in x and y directions, N/mm
Nxy: shear load per unit length, N/mm
ux,, uy,, uz,: displacement degrees of freedom in x, y, and z directions, mm
113
Test Case 5 (Nx&Ny): E-glass/epoxy plate, (0/90)s
(Specially orthotropic plate, Bij=0, D16, D26=0)
uz=0
ux=0 uz=0
uz=0 Nx=1 N/mm
uy=0, uz=0
ux=0 uz=0
uz=0 Nx=1 N/mm
uy=0, uz=0
114
1: Analytical formula for specially orthotropic and simply-supported plate subjected to
normal loads Nx&Ny [31]:
4 2 2 4
m m n n
D11 + 2(D12 + 2 D66 ) + D22
n ( m ,n ) =2 a a b b
2 2
m n
NX + Ny
a b
where m and n are the number of half waves (bubbles) in the buckled plate in the x and
y directions, respectively. The pair (m,n) that yields the smallest value of n(m,n) varies
with the loading case, total number of plies, material and the plate aspect ratio (a/b).
n(m,n) is the buckling load factor due to normal loads Nx and Ny.
2
2 E bm a t3 1
n ( m) = + 2
12(1 2 ) a bm b NX
115
uz displacement distributions
Test Case 1 Test Case 2
Test Case 5
ux=0 uz=0
uz=0 Nxy=1 N/mm
Nxy=1 N/mm
116
Test Case 6: Isotropic plate (E=10000 MPa, =0.3)
Using analytical formula for elastic shear buckling of flat plates [22]:
t3 1
s = K .E.
2
b N xy
Using analytical formula for specially orthotropic and infinite strip (a>>b) simply-
supported plate subjected to shear loads Nxy [31]:
D11 D22
=
D12 + 2D66
4 1 ( D11 D22
3 1/ 4
)
s = 2
for 1
b N xy
117
Buckling load factors
a/b ANSYS
(element size=10 mm)
1 8.934
1.5 6.478
2 5.754
3 5.174
4 5.018
6 4.893
8 4.854
10 4.838
12 4.827
10
ANSYS
K 8
Figure 11.44 of [22]
4
0 2 4 6 8 10 12
a/b
118
Test Case 7: E-glass/epoxy plate, (0/90)s
(Specially orthotropic plate, Bij=0, D16, D26=0)
Using analytical formula for specially orthotropic and infinite strip (a>>b) simply-
supported plate subjected to shear loads Nxy [31] which is given above:
a/b ANSYS
(element size=10 mm)
1 14.369
2 7.068
3 6.431
4 6.125
5 5.999
6 5.963
8 5.891
10 5.866
12 5.846
As seen from the buckling load factors for different a/b values for Test Cases 6-7,
buckling load factor found with the analytical formula for specially orthotropic and
infinite strip gets incompatible with the values found with numerical solutions as a/b
gets smaller. In addition, the difference between numerical and analytical solutions
increases too much especially for a/b smaller than four.
119
3. SIMPLY-SUPPORTED PLATE SUBJECTED TO COMBINATION OF
NORMAL AND SHEAR LOADS
c=1.011 (analytical)
120
APPENDIX C
Finite element analysis is performed for a split disk test and then different failure criteria
values are compared. A static analysis is performed for the maximum load in the test.
The tests are conducted in BARI Elektrik Endstrisi A.. in accordance with ASTM D
2290-00 [38]. The test specimen is loaded through the split disk test fixture which
applies tensile stress to the test ring.
Material: E-glass/epoxy
Stacking sequence: (45o)3
Material properties: given in Table 3.5
Tensile test machine: Instron 4200
121
Assumptions in the finite element model:
- Only the ruptured region (notch) is modeled in ANSYS as a flat plate with
Shell99 elements.
- Bending moment created during the tensile test is ignored since the text fixture
is designed to minimize the effect of this moment [38].
- In order to prevent the boundary condition effects, loading and boundary
conditions are defined far from the interested region (notch).
- Half of the maximum load is applied as a pressure distribution to the one side of
the model and the nodes of the other side is constrained in all directions. Then a
static analysis is performed.
122
Stress Distributions (MPa) for all layers obtained after the analysis
Shear
123
Maximum failure criteria value distribution obtained after the analysis
Cxy: coupling coeffficient for Tsai-Wu & inverse Tsai-Wu failure criteria
124
From the stress distribution graphs given above, it is seen that the stresses in transverse
direction and shear stresses are greater than their strength values in the ruptured region
although the stresses in fiber direction are smaller than the strength values in fiber
direction. As seen from the figure of ruptured test specimens after the tensile test, there
is not any failure in the fiber direction and the failure occurs in the transverse direction
for each layer. So, the results of the finite element analysis agree qualitatively with the
test results.
Failure of a composite structure is assumed when the failure criterion value is greater
than one. So, after the analysis it is expected to obtain maximum failure criterion values
to be nearly one for the ruptured regions. Figures of the maximum failure criterion value
distribution for different failure criteria and coupling coefficients are given above. From
these figures, it is seen that such a result is obtained for maximum strain, maximum
stress and inverse Tsai-Wu failure criterion. But for Tsai-Wu failure criterion, very high
failure criterion values are obtained. So, it can be said Tsai-Wu failure criterion gives
conservative results compared to other three failure criteria, namely maximum stress,
maximum strain and inverse Tsai-Wu.
125