Professional Documents
Culture Documents
Stegmann PhDThesis
Stegmann PhDThesis
net/publication/235219180
CITATIONS READS
39 2,088
1 author:
Jan Stegmann
Acorn Armor ApS
19 PUBLICATIONS 1,014 CITATIONS
SEE PROFILE
All content following this page was uploaded by Jan Stegmann on 21 May 2014.
Ph.D. Thesis
by
Jan Stegmann
Institute of Mechanical Engineering, Aalborg University
Pontoppidanstræde 101, DK-9220 Aalborg East, Denmark
e-mail: js@ime.aau.dk
ISBN 87-89206-94-0
iii
Preface
This thesis has been submitted to the Faculty of Technology and Science at
Aalborg University in partial fulfillment of the requirements for the Ph.D. degree in
Mechanical Engineering. The underlying work has been carried out at the Institute
for Mechanical Engineering, Aalborg University during the period from July 2001
to August 2004. Based on a preprint version of this thesis a public defence
took place on November 19, 2004, with Professor Kai-Uwe Bletzinger (Technical
University of Munich), Professor Martin P. Bendsøe (Technical University of
Denmark) and Professor Niels Olhoff (Aalborg University) acting as opponents.
Incorporated into this final version of the thesis are minor corrections suggested
by the three opponents.
The project has been supervised by Associate Professor, Ph.D. Erik Lund to
whom I express my sincere gratitude for his competent guidance, endless patience,
support and friendship. I also wish to thank my colleague and friend, Assistant
Professor, Ph.D. Henrik Møller, for many invaluable discussions over a hot cop
of coffee and, not the least, for thorough proof reading of this manuscript.
Furthermore, I wish to thank my friends and fellow Ph.D. students Jens Chr.
Rauhe and Lars R. Jensen for working with me on our joint Master’s thesis in
2001, which helped me get a running start for this work. I would also like to thank
my colleagues at the Institute of Mechanical Engineering for creating a pleasant
and inspiring atmosphere.
Finally I want to thank my family – my wife Ditte for encouraging me and making
the family work around me in times of much work, and my children Maja and
Emilie for bringing me joy and always lifting my spirits after a long days work.
Their love and support is invaluable.
Jan Stegmann
Aalborg, May 2005.
iv
v
Abstract
The objective of the present work is to develop finite element based optimization
techniques for laminated composite shell structures. The platform of implementa-
tion is the finite element based analysis and design tool MUST (MUltidisciplinary
Synthesis Tool) and a number of features have been added and updated. This
includes an updated implementation of finite elements for shell analysis, tools for
investigation of nonlinear effects in multilayered topology optimization and a novel
framework called Discrete Material Optimization (DMO) for solving the material
layout and orientation problem.
Existing methods for solving for optimal material orientation and maximum
stiffness inherently suffer from problems with local optima, which inspired the
development of Discrete Material Optimization (DMO), which is a novel approach
for simultaneous solution for material distribution and orientation. The DMO
method uses an element level parametrization in a weighted sum formulation
that allows the optimizer to choose a single material from a set of pre-defined
materials by pushing the weights to 0 and 1. The success of the method is therefore
dependent on the optimizers ability to push the weights to 0 and 1 and several
weighting schemes are implemented. Numerical examples indicate that the method
is indeed able to solve the combined material distribution and orientation problem.
Furthermore, an industry related design problem of a wind turbine blade main spar
is solved and the obtained results are very encouraging. The DMO method thus
shows promising potential for application to problems of industrial relevance and
no problems with local optima could be identified in the tested examples.
vi
vii
Abstrakt
Formålet med nærværende arbejde er at udvikle finite element baserede op-
timeringsteknikker til laminerede kompositte skalkonstruktioner. Implementer-
ingsplatformen er det finite element baserede analyse- og optimeringsværktøj
MUST (MUltidisciplinary Synthesis Tool), og en række funktioner er blevet
tilføjet og opdateret. Dette inkluderer en opdateret implementering af elementer
til analyse af skaller, værktøjer til undersøgelse af ikke-lineære effekter i multi-
lags topologioptimering samt en ny metode kaldet Diskret Materiale Optimering
(DMO), der kan løse med hensyn til optimal fordeling og orientering af materialer.
Publications
Publications in proceedings
• Lund E, Stegmann J (2004): On Structural Optimization of Composite Shell
Structures Using a Discrete Constitutive Parameterization. In: The Science of
making Torque from Wind, (ed. G.A.M. van Kuik), 19-21 April 2004, DUWind,
Delft University of Technology, pp. 556–567.
• Stegmann J, Lund E (2003): Optimizing General Shell Structures Using a
Discrete Constitutive Parameterization. In: American Society for Composites 18th
Technical Conference, ASC 18, Gainesville, FL, US, 19-22 October 2003, pp. 1–10.
• Stegmann J, Lund E (2003): Discrete Fiber Angle Optimization of General Shell
Structures using a Multi-Phase Material Analogy. In: Fifth World Congress on
Structural and Multidisciplinary Optimization, WCSMO 5 (ed. C. Cinquini et al),
Venice, Italy, 19-23 May 2003, pp. 1–6.
• Stegmann J, Lund E (2002): Nonlinear topology optimization of laminated shells.
In: 15th Nordic Seminar on Computational Mechanics, NSCM 15 (ed. E. Lund et
al), Aalborg, Denmark, 18-19 October 2002, pp. 215–218.
• Stegmann J, Lund E (2002): Topology Optimization of Multi-Layered Shell
Structures Undergoing Large Displacements, In: Fifth World Congress on Compu-
tational Mechanics, WCCM V (ed. H.A. Mang et al), Vienna, Austria, 7-12 July
2002, pp. 1–10.
1 Introduction 1
1.1 Structural design optimization . . . . . . . . . . . . . . . . . . . . 1
1.2 Background of work . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Finite element analysis of laminated composites . . . . . . . 4
1.2.2 Topology optimization . . . . . . . . . . . . . . . . . . . . . 5
1.2.3 Optimization with orthotropic materials . . . . . . . . . . . 6
1.3 Objectives of work . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.1 The MUltidisciplinary Synthesis Tool – MUST . . . . . . . 7
1.3.2 Outline of thesis . . . . . . . . . . . . . . . . . . . . . . . . 8
6 Conclusions 103
Bibliography 107
Index 114
xiv Contents
1
Introduction
he term optimal is widely used for “the best” but in the present context
T it is necessary to define it more rigourously. Here, optimal denotes the
best design available given the performance criteria and restrictions defined by
the engineer. Obtaining such a solution is called design optimization, meaning
systematic improvement of an initial design by selection of better and better
design parameters. This implies the iterative nature of design optimization where
the design is continuously analyzed, evaluated and improved until no further
improvement can be made and the design is optimal. In its basic form this
procedure is far from new and, indeed, it is probably one of the oldest disciplines
in engineering and, to be philosophical, in human existence. Optimization can
be said to have occurred throughout human history although the process has
been characterized by small and often painstaking steps towards the optimum. In
modern history this task has been left largely to engineers who through knowledge
and skill have managed to find still better solutions to known problems on a
heuristic “trial and error” basis. It is therefore not surprising that solutions
obtained with modern day optimization methods often resemble well known and
well tried solutions from engineering history. One example is frame structures,
which were widely used in the 19th century and the first part of the 20th
century for steel bridges. Two typical examples of such structures are shown in
Fig. 1.1 together with optimal solutions obtained using modern structural design
optimization techniques for similar loading and support conditions.
Figure 1.1: Railroad suspension bridge design (left) and railroad frame bridge design
(right). The black/white structures have been obtained using topology optimization with
similar boundary conditions (from Bendsøe and Sigmund (2003), courtesy Ole Sigmund).
engines and the weight cannot be reduced to a point where the fuselage might
loose stiffness and break. So, finding the best solution available is a far from
simple task and becomes virtually impossible to do “by hand” when the number of
parameters and restrictions is high. Consequently, the development of computers
play an important role in design optimization but at the same time poses a number
of challenges for the engineer who must a priori decide by what measure a design
is good and also, by which bounds the design is limited. In design optimization
the measure of “goodness” is called the objective function since the objective of the
optimization is to “increase goodness” (or “reduce badness”, which amounts to the
same thing) and the limiting factors are called constraints. Common to both is that
they must be quantities that can be computed and evaluated as a number. The
conceptual difference is that the constraints pose nonnegotiable boundaries called
constraint bounds on the design while the objective must simply be improved to a
point where further improvement cannot be made within the constraint bounds.
A constraint is said to be active when it is imposing its bound on the design, i.e.
when the value of the constraint is equal to one of the constraint bounds.
In order to evaluate the objective and constraints the design must be described in
terms of a set of well defined parameters that govern geometry, material properties,
densities, etc. This is called design parametrization and constitutes choosing a
number of characteristic parameters called the design variables, which are the
parameters we wish to change. The goal of the optimization is then to find the
Chapter 1. Introduction 3
combination of design variables that yields the best design while observing the
constraints. Performing the optimization is characterized by three distinct steps.
First, the performance of the design is evaluated by analyzing it with the current
values of the design variables. In structural optimization this is most commonly
achieved using the finite element method. Second, the sensitivity of the design to
changes in the design variables is evaluated for all design variables - this is called
design sensitivity analysis and the sensitivities are the gradients of the objective
and constraints. Third, the sensitivity information is used to update the design
variables in a way that improves the objective. This is most commonly done using
mathematical programming techniques.
In the following the present study will be mapped out in more detail with reference
to related studies.
1 In the present study we employ the Nested ANalysis and Design (NAND) approach in which
the equilibrium equations will be assumed to be satisfied prior to solving the optimization problem
itself.
4 1.2. Background of work
Figure 1.2: Side and edge view of some 2D structure subjected to topology optimization
(left), shape optimization (middle) and size optimization (right).
This brings forward the need for efficient and reliable numerical design tools
– in particular when dealing with large scale structures involving complicated
geometries, multiple layers, multiple materials and multiple load cases. Efforts for
developing such tools are already well under way and this work naturally draws
on results from previous studies in the fields of both finite element analysis and
gradient based optimization. With the vast amount of literature available on these
topics it is well beyond the scope of this brief review to give an exhaustive account
of the works preceding this. In the following particular emphasis will therefore be
placed on topics directly related to the present work.
such as locking and hour-glass modes usually associated with shell finite elements.
The finite element method has, of course, also benefitted tremendously from the
exponential growth in computer power, which has made possible the analysis of
realistic and industry relevant models (Bathe et al., 1997; Noor, 1999).
Several classes of finite elements have been developed and the choice depends
on the level of detail required for design purposes. The simplest and most
popular class combines an Equivalent Single Layer (ESL) laminate description
with finite elements based on First order Shear Deformation Theory (FSDT)
to obtain highly efficient and reliable elements for global response analysis. In
this group the most popular choice are linear elements, rather than higher-order
elements, as it compensates for the increased computational cost of computing the
laminate stiffness. To avoid locking problems when using lower-order elements the
Assumed Natural Strain (ANS) technique developed by e.g. Hughes and Tezduyar
(1981), MacNeal (1982), Dvorkin and Bathe (1984) and Bletzinger et al. (2000) is
employed, resulting in very robust elements. This is also the strategy chosen in this
work (as described in Chapter 3) and other examples from the literature include
Barut et al. (2000), Alfano et al. (2001) and Wagner and Gruttmann (2002). The
ANS elements have also appeared in major commercial codes such as ANSYS,
ADINA and MSC.NASTRAN/MARC while more advanced methods such as layer-
wise elements e.g. Reddy (1993), Brank and Carrera (2000) and To and Liu (2001)
or solid shell elements e.g. Klinkel et al. (1999) and Sze et al. (2002), accounting
for varying degrees of local behavior, are still reserved for research codes. This is
mainly due to their higher complexity and computational cost. For an extensive
review of layer-wise methods in finite element applications see Ochoa and Reddy
(1992), Carrera (2003) and Reddy (2004).
The extension of topology optimization to shells (e.g. Maute and Ramm, 1997) has
2 Solid Isotropic Material with Penalization
6 1.2. Background of work
not received the same amount of attention as other fields in topology optimization.
This is partly due to the fact that the essence of “classical” topology optimization
is the ability to introduce holes in a structure, thereby reducing material usage and
in turn reducing weight. However, the ability to introduce through-the-thickness
holes in shell structures has little practical relevance since holes deteriorate the
membrane load carrying ability of the shell. Furthermore, for the majority of
engineering shell structures (fuselages, wings, ship hulls, turbine blades, pressure
vessels etc.) it is not viable to introduce holes since the boundary is usually
prescribed by other factors (e.g. aerodynamical considerations). Thus, topology
optimization of shell structures only has real engineering applications with the use
of multilayered shell finite elements. In that context topology optimization can
be used to “add or remove” material in specific layers rather than through-the-
thickness. This opens the door to stiffener and core layout design, which has been
treated for plates by e.g. Diaz et al. (1995) and Krog and Olhoff (1999) and for
shells by e.g. Lee et al. (2000), Belblidia et al. (2001) and Belblidia and Bulman
(2002).
The above mentioned works are all concerned with linear problems. However,
nonlinearities play an important role in the failure of large composite structures,
such as wind turbine blades, and should therefore be considered as well. Some
of the earliest work involving geometrical nonlinearities in stiffness design of
continuum structures was that of Jog (1996) and Yuge et al. (1999). Recent
developments of importance to the present work include Buhl et al. (2000),
Gea and Luo (2001) and Bruns et al. (2002) who used topology optimization on
2D structures with geometrical nonlinearities. To the best of our knowledge no
work prior to this addresses the influence of nonlinearities on the stiffening topology
of laminated composite shell structures as treated in Chapter 4.
have contributed as well, e.g. Luo and Gea (1998) who used a very similar
approach to that of Pedersen for plates, and Thomsen and Olhoff (1990) who used
optimality criteria combined with mathematical programming to solve for fiber
orientation and density in discs and plates. Other approaches have been taken by
e.g. Miki and Sugiyama (1993) and Foldager et al. (1998) who used lamination
parameters to overcome the inherent difficulties with local minima in this type
of problems. The prevailing method seems to be that of Bruyneel and Fleury
(2002) and Moita et al. (2000) who use customized mathematical programming
techniques and such methods have been implemented in the commercial software
packages BOSS QUATTRO from SAMTECH and OPTISTRUCT from ALTAIR.
In the present work the methods above are not extended further but used as
reference in Chapter 5.
The last point springs from our collaboration with the Danish wind turbine
industry who wants to develop still longer and lighter wind turbine blades, which
are subject to very large displacements under running conditions. In the context of
such large laminated composite structures it has been chosen to focus on stiffness
maximization (compliance minimization) as the optimization objective.
Postprocessor
(FEPlot)
Input module
(ANSYS, COSMOS, ODESSY)
Database module
(Result generation)
Analysis modules
(Solid, fluid, thermal, FSI) Optimization modules
MUST (Shape, size, topology, fiber)
Figure 1.3: Major components of the MUltidisciplinary Synthesis Tool (MUST). The
gray components are those changed in the present work.
The major reason for choosing MUST as platform of implementation is that the
need for developing a finite element framework from scratch is circumvented.
Furthermore, the use of MUST allows the implemented methods to be used by
colleagues and students – an opportunity that has already been exploited in
several graduate studies. This supports the philosophy behind MUST which is
to support both research and education by providing understandable program
code and allowing easy implementation of new features.
and used largely by Associate Professors Erik Lund and John Rasmussen.
Chapter 1. Introduction 9
Chapter 5 is devoted to material layout and fiber angle optimization and introduces
the concept of discrete material optimization (DMO), which is tested for various
2D and 3D examples.
his chapter lays out the analysis and optimization tools necessary
T for solving the generic optimization problem stated in (1.1). Solving the
optimization problem is a process involving several steps as illustrated in Fig. 2.1.
Starting from an initial design (with defined design variables) a design analysis is
performed (solution of the physical problem) and subsequently, analysis and design
improvement is performed consecutively to gradually obtain the final, optimal
design.
The optimization process is thus a chain of events, each leading (hopefully) towards
the optimal design. As usual, the chain is only as strong as its weakest link and so,
a considerable amount of time has in this work been invested in implementation
of reliable finite element technology. The methodology for doing so is introduced
in Section 2.1 and later in Chapter 3, the particular implemented elements will be
described in greater detail. Another important aspect of solving the optimization
problem is the design improvement step, Fig. 2.1. In this work a shortcut has
Yes
Sensitivity
analysis
Improve design
(optimizer)
New design
variables
Figure 2.1: Flow chart of the solution process for the generic optimization problem in
(1.1). The gray boxes indicate topics treated in this chapter.
12 2.1. Analyzing the design
been taken in this step since an “off the shelf” optimizer has been implemented in
MUST, only requiring an appropriate interface to be programmed. A thorough
treatment of the theory behind optimizers will be left the established literature
but the fundamentals will be presented in Section 2.2 with special emphasis on
the applicability of such methods in this work.
In this work both linear and nonlinear problems are considered and as the
formulation of the latter encompasses the former, emphasis will be on deriving
the nonlinear expressions. The starting point is the governing equations for the
static structural problem, which are stated as an axiom:
Z Z Z
sT δ dV − pb δu dV + ps δu dA = 0 (2.1)
V V A
| {z } | {z }
Internal work External work
In finite element analysis the governing equations (2.1) are recast in vector form
by introducing the strain-displacement matrix, B, which is defined from = Bu
where u indicates nodal values. Alternatively, the strain-displacement matrix can
∂
be expressed from the variation of the strain δ = ∂u δu as:
∂
B= (2.2)
∂u
The form in (2.2) is convenient for deriving the strain-displacement matrix as
will be shown in Section 3.4. Introducing B and rewriting the strain variation in
(2.1) to a displacement variation, an internal element nodal force vector, r, can be
Chapter 2. Analysis and optimization 13
This must be balanced by the external element nodal force vector, p, which can
be derived in the standard way by computing equivalent nodal forces from the
distributed loads pb and ps in the external work term. Thus, the residual vector1 ,
R, is defined over all elements, N e , as:
e
N
X
R(u) = rk − pk (2.4)
k=1
For the system to be in static equilibrium the internal and external forces must
balance each other, i.e.
R(u) = 0 (2.5)
which represents the governing system of equations for linear and nonlinear static
problems. Thus, the generic statement “physical laws” in (1.1) can be replaced by
R = 0 to form the basic form of the optimization problem solved in this work:
Objective : min f (a)
a
Subject to : g(a) ≤ G (2.6)
amin ≤ a ≤ amax
R(u, a) = 0
in which u is a fixed point indicating that R(u, a) = 0 has been solved prior to
solving the optimization problem, which constitutes a Nested ANalysis and Design
(NAND) approach.
and used to find the displacement increment ∆u by solving the linear system:
KT ∆u = −R(u) (2.9)
In MUST (2.7) is solved continuously until the ratio kRk/kpk is below some
specified value, typically around 10−8 − 10−6 . At present, MUST encompasses a
number of nonlinear solvers due to Møller (2002), including a full Newton-Raphson
solver, a modified Newton-Raphson solver and a quasi-Newton BFGS solver all
of which can be used with or without line search. Line search gives better global
convergence properties than full N-R and modified N-R directly and has been used
with full N-R when solving for a single equilibrium point. For tracking equilibrium
paths an arc-length solver has been used with modified N-R for the sub-iterations.
If the problem under consideration is geometrically linear the need for (2.4)–(2.10)
can be circumvented and the static problem simplified. Using that s = CBu a
linear stiffness matrix can be defined from the internal work term in (2.1) as:
Z
K= BT CB dV (2.11)
V
The equations (2.3) and (2.10) above form the basis for proceeding with the finite
element formulation and implementation as described in Chapter 3.
must first be established how the performance of the design changes with changes
in the design variables. This is achieved through a design sensitivity analysis
(DSA) in which the gradients of the characteristic properties of the optimization
problem (2.6) are found.
dC(a) du(a)
= pT (2.14)
dai dai
This gradient of the objective will indicate whether a change in a particular design
variable will increase or decrease the performance of the design. In the same way
the constraint gradients, dg(a)/da, indicate how the design variables should be
changed to keep the design within the constraint bounds. In the following we
assume that a design sensitivity analysis has already been performed. Details
concerning computation of the gradients for the specific optimization problems
considered in this work will be treated more extensively in Chapters 4 and 5.
Solving the problem of changing the design variables based on the gradient
information such that the performance is improved (minimization of f ) while
observing the constraints falls to an optimizer.
a2
Convex approximation
g~(a) £ G
~
a opt
a opt
a1
methods are particularly efficient for problems involving many design variables and
few constraints, which makes them ideally suited for topology optimization.
The strategy for moving the asymptotes is a key issue for the success of MMA
and several updated schemes have been suggested to improve its convergence
Chapter 2. Analysis and optimization 17
f (a) ~ f (a)
f f f (a)
~
f (a)
~
a opt
~
a opt
Figure 2.3: Convex MMA approximation to a fictitious design space. In (a) the slope
in a∗ is negative and L is active, in (b) the slope is positive and U is active.
properties, see e.g. Bruyneel et al. (2002) for a review. One of these extended
schemes is the Globally Convergent MMA (GCMMA) algorithm, which has also
been implemented in MUST from FORTRAN77 source code provided by Professor
Krister Svanberg. GCMMA uses a non-monotonic approximation to achieve better
convergence. However, the term “global” is somewhat misleading as GCMMA does
not guarantee convergence to the global optimum solution but just convergence
to a stationary point in the approximated problem from any starting point. The
GCMMA algorithm has been implemented but not used to any great extent in this
work since it tends to converge slower than MMA in terms of number of iterations
used and thus computational time.
Besides controlling the approximation the asymptotes also function as move limits,
which means that the move limits will also gradually tighten. However, early in the
optimization the asymptotes may be too far apart to provide practical boundaries
so an additional move limit strategy should be employed. In the present study a
stationary move limit of typically 2–5% has been used to stabilize the iterations in
the beginning. This has proven a reliable approach and in general, the performance
of the MMA optimizer in this work has been very satisfactory. We have successfully
solved stiffness design problems of more than 740 000 design variables and MMA
used only about 4 seconds per iteration to solve the approximated problem.
Now, the basic aspects of the analysis and optimization process have been discussed
and in the following the particular element technology implemented will be treated
in detail.
18 2.2. Improving the design
3
Multilayered shell finite elements
The framework for shell analysis in MUST was originally developed and imple-
mented in 2000/2001 by Lars R. Jensen, Jens M. Rauhe and Jan Stegmann and
is documented in the joint masters thesis Jensen et al. (2001). During the course
of elaborating the present work the original routines have been reimplemented for
higher computational efficiency and also to accommodate extra features as required
for the implemented optimization procedures. Furthermore, a number of errors
have been corrected in the element routines and a number of features for geometric
handling, pre- and postprocessing have been added. It is inevitable, however, that
some figures and topics presented in the following will resemble those found in
Jensen et al. (2001)1 .
the original printed masters thesis was revised by the author with a number of key new figures
and extended explanations for the elaboration of Stegmann and Lund (2002).
20 3.1. Geometry, kinematics and material
addressed and the solution used in MUST is described. Finally, aspects of element
implementation is discussed in Section 3.4 and in Section 3.5 the performance of
the elements in MUST is demonstrated.
3.1.1 Geometry
It is common practice when dealing with shells to consider the geometry of the
structure as a surface instead of a volume. This is justified because shell structures
are thin compared to the overall size of the structure. The characteristic entity
chosen for describing the shell, the reference surface, is most commonly either
the geometric top, bottom or middle surface of the physical structure. This
geometric representation is still 3D but can be made 2D by introducing curvilinear
coordinates for the surface, Fig. 3.1. In that context it proves convenient to operate
with a covariant vector base as local reference frame for describing the shell.
The covariant base vectors are defined in any point in terms of the position vector,
x = {x, y, z}, and the local curvilinear coordinates, (r, s, t) ≡ ri , as:
∂x ∂x ∂y ∂z
gi = = ; ; (3.1)
∂ri ∂ri ∂ri ∂ri
The “in-plane” covariant base vectors g1 and g2 are tangents of the coordinate
curves r and s, respectively, and span the tangent plane of each point on the
surface. The third covariant vector, g3 , is tangent to t and is generally not normal
to the surface and furthermore, none of the covariant vectors will in general be
mutually orthogonal. This is of no consequence since the vectors are still linearly
independent and thus may serve as coordinate base for the geometric description.
In practice, however, a third vector is defined normal to the shell surface to provide
a meaningful way of expressing the shell thickness (as will be shown shortly). This
Chapter 3. Multilayered shell finite elements 21
g3
g2
g1
r x
s
z
y
x
Figure 3.1: Reference surface (usually the top, bottom or middle of the physical
structure) of some doubly curved shell with global Cartesian reference (x, y, z) and
covariant vectors (g1 , g2 , g3 )
x = N(ri )x (3.2)
with N being the shape functions for the particular element containing x and
x being a vector of nodal coordinates of all nodes, a, in the element, i.e. x =
{· · · |xa , y a , z a | · · · }T . The particular formulation of (3.2) depends on the order
and type of elements chosen, which may vary across the shell geometry.
The use of simple shape functions and (3.2) simplifies matters considerably
compared to having complicated mathematical expressions for the entire geometry.
The covariant base vectors require no additional work since the components in
(3.1) are identical to the components of the Jacobian matrix, J = [g1T , g3T , g3T ]T ,
which can be easily determined as the shape functions are known. Defining the
kinematics is therefore straightforward.
22 3.1. Geometry, kinematics and material
t s t s
8
7 4
4 3
3
r r
5
1
1 6
2
(a) 2 (b)
Figure 3.2: Degeneration of eight-node solid element (a) into four-node shell element
(b). The nodal vectors are the node directors and the shaded surface is the reference
plane. Deleted nodes (ghost nodes) are marked by a “◦”.
3.1.2 Kinematics
In general terms the displacement, u, for any 3D element may be expressed,
analogues to (3.2), as u = Nu where u is the vector of nodal displacements,
i.e. u = {· · · |uax , uay , uaz | · · · }T . The kinematics of a shell element is closely related
to this expression, the difference being that assumptions regarding the structural
behavior of the shell will be built into the kinematic description. To do so the
degenerated solid approach is applied in the following for a four-node element
without any loss of generality.
The starting point is an eight-node volume element and the result is a four-node
shell element, which is geometrically reduced to a surface, here taken to be the
geometric midsurface of the solid element (the shaded area shown in Fig. 3.2).
This is the reference surface of the shell element, so named since it will be used
as geometric and kinematic reference for the element. In Fig. 3.2 the reference
surface is shown to be identical to the midsurface but it may be displaced up or
down within the element volume to serve specific modeling needs. As mentioned,
the mostly used alternatives, which are both supported in MUST, are the geometric
top and bottom of the solid element. The choice of reference surface has no bearing
on the kinematics and can be handled efficiently in the element implementation.
The degeneration procedure shown in Fig. 3.2 basically involves replacing the
original nodes with nodes lying on the reference surface of the shell. In doing
so the element can no longer deform in the transverse direction. At the same
time a vector is introduced in each of the new nodes, pointing from the deleted
bottom node towards the deleted top node (called ghost nodes). The new nodal
vectors in Fig. 3.2 are called node directors and effectively link points opposite
the mid-surface. The node director is allowed to rotate about the node but it
cannot stretch. Together these properties make up two of the Reissner-Mindlin
assumptions, namely zero transverse strain (33 ≡ 0) and “normals remain straight
Chapter 3. Multilayered shell finite elements 23
-a
b v3
v*3
v2 b
v1
Figure 3.3: Rotation of node director, v3 , under deformation to deformed state, v3∗ .
but not necessarily normal”. This also implies that the transverse shear strains will
be continuous across the thickness, which is important in relation to multilayered
structures as discussed in Section 3.2. The last assumption of zero transverse stress
(s33 ≡ 0) will be enforced through the constitutive relation as shown later.
The node directors are very important in that they are used to define the nodal
displacements and rotations. This is done by setting up, in each node, a Cartesian
base called the director coordinate system having the node director as z-axis. The
node director will be denoted v3 and defined as:
g1 × g2
v3 = (3.3)
kg1 × g2 k
This might seem inconsistent with Fig. 3.2 since the ghost nodes need not lie
on a surface normal but in practice (3.3) is the only viable approach since the
coordinates of the ghost nodes are non-existent in the model. However, using
(3.3) directly can lead to discontinuities in the displacements since the surface
normal may change from element to element depending on the mesh generator and
element type. This will in particular be the case when using four-node elements to
model curved geometries. Consequently, a simple algorithm that averages adjacent
node directors for all elements is applied in MUST to ensure continuity. The two
additional base vectors in the director coordinate system are defined from v3 and
the auxiliary vector a as:
a × v3
v1 = ; v2 = v3 × v1 (3.4)
ka × v3 k
where a is the nodal number, A is the number of nodes in the element and h
is the shell thickness. Note that all quantities must be evaluated at node a.
The expression in (3.5) completely defines the kinematics of the shell element in
terms of five degrees of freedom, (ui , α, β), while observing the Reissner-Mindlin
assumptions. The expression (3.5) may be expressed in terms of a modified
shape function operator, N̂, defined as N̂ = N̂(N (r, s, 0), t) such that u = N̂û
where û is defined as a vector containing all the nodal degrees of freedom, i.e.
û = {· · · |uax , uay , uaz , αa , β a | · · · }T . This rather tedious notation will be abandoned
and from this point and on, N and u will always imply “modified shell quantities”
as defined by (3.5).
The strain components follow naturally from (3.5) and may be expressed in terms
of the covariant base vectors as:
1
˜ij = t gi · t gj − 0 gi · 0 gj
2 (3.6)
1 ∂u ∂u ∂u ∂u
= · gi + · gj + ·
2 ∂xj ∂xi ∂xj ∂xi
where ˜ij indicates local strain since (3.6) is stated in the covariant base vectors.
The subscripts t and 0 indicate deformed and initial state, respectively, i.e.
t gi = gi (x + u). The strain may also be obtained from terms of the deformation
gradient, F = 0 ∇t x, which is perhaps a more direct approach:
1 T
˜ij = F F−I (3.7)
2
This can be efficiently evaluated in terms of the displacement gradients, which are
readily available in a finite element framework such that F = 0 ∇t u + I. Which
of (3.6) and (3.7) is most convenient depends on the type of finite element being
implemented.
Chapter 3. Multilayered shell finite elements 25
3.1.3 Material
As briefly mentioned above the third of the Reissner-Mindlin assumptions (s33 = 0)
is enforced through the linear constitutive relation s = C, where s is the second
Piola-Kirchoff stress. This will be addressed in the following.
First, we define the constitute matrix of an orthotropic material, which is the most
general case for the present work:
C11 C12 C13 0 0 0
C22 C23 0 0 0
C33 0 0 0
C=
(3.8)
C44 0 0
Sym. C55 0
C66
Enforcing s33 = 0 simply involves deleting the third row and column in (3.8) and
the most general form of C is then:
Ĉ11 Ĉ12 0 0 0 0
Ĉ22 0 0 0 0
0 0 0 0
C= (3.9)
C44 0 0
Sym. C55 0
C66
where the coefficients are given in terms of the engineering constants as:
E1 E2 ν21 E1
Ĉ11 = Ĉ22 = Ĉ12 =
1 − ν12 ν21 1 − ν12 ν21 1 − ν12 ν21 (3.10)
C44 = G12 C55 = γG23 C66 = γG13
where the modified terms Ĉ11 , Ĉ12 and Ĉ12 can be derived from the general
expressions by setting the transverse Poisson ratios equal to zero (see e.g. Reddy
(2004)). In (3.10) the factor γ is the shear correction factor taken to be 5/6 for both
multi- and single-layered structures. This is somewhat crude but the predictive
capabilities of the formulation are quite good so no further steps have been taken
towards improving this along the lines of e.g. Pai (1995) or Auricchio and Sacco
(1999). For sandwich structures, however, MUST supports an alternative scheme
in which γ is set to 1.0 for the core layer while the transverse shear stiffnesses
are set to zero for the skin layers. This corresponds to assuming that transverse
shear is supported only by the sandwich core and provides very good predictive
capabilities for sandwich structures (Jensen et al., 2001).
From (3.10) it is apparent that only in-plane normal material properties and out-of-
plane shearing properties are taken into account, which is consistent with the shell
assumptions. Each of the expressions above are stated in the orthotropic principal
26 3.1. Geometry, kinematics and material
Fiber direction
v3c
v3b
v3a
c b
Figure 3.4: Conceptual difference between director vector (dotted line) and material
coordinate system (solid lines), which changes through the thickness.
directions, which are of no particular interest. Therefore the need arises for a
coordinate system that uniquely defines the orientation of the orthotropic material
at any point in any layer of the element. This coordinate system is very fittingly
called the material coordinate system and the major difference between this and
the director coordinate system is that the material coordinate system changes
from layer to layer as illustrated in Fig. 3.4 for a curved nine-node element. For
flat (three- and four-node) elements the coincidence of all normal vectors could
be exploited to bypass evaluation of some of these coordinate systems but as
the implemented element routines are generic the same approach is used for all
elements.
The material coordinate system is spanned by the base vectors mi and for the
constitutive properties to be meaningful the system must span the tangent plane
in any given point. As such it is natural to define m3 to be normal as:
g1 × g2
m3 = (3.11)
kg1 × g2 k
which is the same definition as used for the director vector, v3 , the difference
being that (3.11) is updated in all Gauss-points and layers (Fig. 3.4). The in-
plane material base vectors may in principle point in any direction as long as they
are orthogonal to m3 but for practical modeling purposes it proves convenient to
introduce a projection of a defined (or default) global material system onto the
element plane. This procedure has been adopted from the commercial software
packages ANSYS and COSMOS and can be stated as follows.
A global material system is defined in terms of three vectors, di , which are either
taken to be the global Cartesian base vectors (default) or, if defined, read from
user input. These vectors, di , will be projected onto the element plane as follows.
If the angle between d1 and m3 is smaller than 45◦ , d1 is projected as m1 . Else
d2 is projected as m1 . The procedure is outlined in Algorithm 3.1.
default is oriented identically for all elements and has an “intuitive feel” when
handling materials in modeling situations. The user also has the option of defining
global material coordinate systems, which allows the user to control the material
direction explicitly on various parts of the model. The procedure outlined above
provides a well-defined framework for handling orthotropic materials using a single
angle, θ. This angle is defined relative to the local, projected material coordinate
system in any point, as shown in Fig. 3.5.
The constitutive behavior can now be described by (3.9) and an in-plane rotation of
C such that C = TT (θ)CT(θ) where T(θ) is a standard transformation matrix (see
e.g. Cook et al. (1989)). Depending on the element formulation the constitutive
matrix must be transformed once more since the material coordinate system is
usually not the preferred frame for setting up the element stiffness matrix. In
the present work the element stiffness is either expressed in the global Cartesian
frame (i, j, k) or the covariant frame (g1 , g2 , g3 ) depending on what provides
fewest element level computations. The transformations employed are stated
in Cook et al. (1989) for transformation to (i, j, k) and in Heinbockel (2001) for
transformation to (g1 , g2 , g3 ).
m3
2
m2
q 1
m1
Figure 3.5: Orthotropic principal directions (1, 2) with reference θ to the material
coordinate system base vectors mi .
continuous across the thickness as well and furthermore, that transverse strains
(13 and 23 ) are constant through the thickness. Consequently, a single node in
the thickness direction is sufficient for describing the kinematics of the laminate
and consequently, all quantities derived in the previous sections apply directly
to ESL shell models. The downside of using ESL is that interlaminar effects
such as delamination are difficult to predict due to the absence of normal stress
and strain components in the thickness direction. This can be remedied to some
extent by introducing stress recovery as discussed by e.g. Cho and Choi (2001)
in which the equilibrium equations are integrated a posteriori to obtain a better
result for the stress components. Another approach is to introduce a modified
kinematic assumption, allowing for piecewise continuous displacements. This is
known as the zig-zag approach and has become increasingly popular over the last
years, see e.g. Carrera (2003) for a review. However, the unmodified laminate
description employed here is still the most widely used since it provides a good
approximation of the structural stiffness and is computationally less expensive.
Both these properties are desirable in the present context and thus, the method
chosen provides a sufficient and well-established basis for doing global analysis and
optimization.
The notation associated with laminates is shown in Fig. 3.6 where the number of
layers is designated N l and the layer thickness is hl . Each layer is also associated
with a material, Cl , as well as an angle, θl , for orthotropic materials. When mixing
element- and layer-wise quantities superscripts e and l will be used, respectively,
to relate each quantity.
Each layer, l, in the laminate of Fig. 3.6 is described by the constitutive relation:
and consequently, the stresses will be layer-wise continuous since strains are
continuous and Cl changes from layer to layer. The constitutive behavior of
the entire laminate is obtained by integration through the thickness. In Classical
Laminate Theory (CLT) this is achieved explicitly by forming the extension matrix,
Chapter 3. Multilayered shell finite elements 29
Layer N l
Layer N l-1
hl
Layer 3
Layer 2
Layer 1
h
where tl = ± √13 for two-point quadrature. Now, the global tangent stiffness matrix
becomes:
l
Ne
X XN ZZZ
∂BTq ∂sq hl
KT = sq + BTq drdsdtl (3.14)
q=1
∂u q ∂u q h
k=1
k
30 3.3. Unlocking – Assumed Natural Strain
where the ratio hl /h arises from (3.13) since dt = (hl /h)dtl . Similarly, the internal
force vector is obtained as:
l
Ne
X X N ZZZ
hl
r= T
Bq sq drdsdtl (3.15)
q=1
h
k=1
k
In the following the layer-wise form of volume integration in (3.14) and (3.15) will
be implied whenever stating the stiffness matrix or internal force vector in integral
form.
3.3.1 Locking
Locking is usually characterized as either shear locking, membrane locking, volume
locking or thickness locking and the different types are encountered depending on
the element type and conditions of the model. This brief discussion is limited
to the elements in MUST and an exhaustive account is left the literature, e.g.
Cook et al. (1989) or Bischoff (2004).
element order gets higher. Indeed, the SHELL16 element in MUST has proven
very robust and only exhibits locking behavior under extreme conditions such
as when elements get highly curved and distorted, which can cause membrane
locking. The SHELL16 element can therefore serve as reference for other elements
although it is not efficient for optimization problems in general due to its high
computational cost. The SHELL9 element in MUST is less expensive and has
been used extensively with good results. However, the element may exhibit shear
locking when subjected to non-constant bending moments and can also suffer from
membrane locking when curved. The difficulty in using the SHELL9 and SHELL16
elements lies in the fact that the occurrence of locking is hard to predict and can
be difficult to detect unless it is severe. The only viable approach with these
elements is therefore to recognize the problem and perform comparative solutions
to validate the results. For this and other reasons (Section 3.5) the linear MITC
elements remain the choice of preference.
In the present work the ANS method introduced by Dvorkin and Bathe (1984)
as Mixed Interpolation of Tensorial Components (MITC) has been applied
successfully to four- and three-node elements, effectively eliminating problems
with shear locking. The basic idea is to use a new set of points, p = 1 . . . N p ,
for strain evaluation instead of the Gauss points, which would be used in non-
stabilized elements. These new points are called tying points and are chosen such
that the strains evaluated in these points are free from parasitic strains. Having
obtained a set of “correct” strains these are then interpolated across the element,
thus eliminating problems with parasitic strains altogether. The strains in the
tying points, (rp , sp , tp ), are evaluated directly from the strain definition in (3.6),
i.e. ˜ij |p ≡ ˜ij (rp , sp , tp ). Introducing a set of interpolation functions for each
32 3.3. Unlocking – Assumed Natural Strain
C C
s=1
4 3 3
s s
r r
D B D B
1 2 1 2 s = -1
A r = -1 A r=1
Figure 3.7: Three- and four-node elements with MITC tying points (A–D) used for
transverse shear strain evaluation.
strain component, Npij , the assumed strains (AS) are expressed as:
p
N
X
˜AS
ij = Npij (r, s)˜
ij |p (3.16)
p=1
where N p is the number of tying points, p, for the ij’th strain component. The
MITC interpolation functions, Npij , must naturally fulfill the relation:
so that the p’th interpolation function assumes the value 1 in the p’th tying point
and the value 0 in all other tying points. It is therefore natural to choose Lagrange
polynomials of an order appropriate to the number of tying point, i.e. first order
Lagrange polynomials for two tying points etc.
The two expressions (3.16) and (3.17) above are general for all MITC elements,
which include linear, quadratic and cubic elements. The location and number of
the tying points will vary depending on the order of the element but the basic idea is
the same (see Bucalem and Bathe (1993) for details). For the three- and four-node
MITC elements there are four tying points, A–D, used for transverse shear strain
evaluation (two tying points for each strain component). Both transverse shear
strains are interpolated linearly but 13 is interpolated along s between points A
and C and 23 is interpolated along r between points B and D. The four points are
located at the element mid-sides, i.e. A = (0, −1, 0), B = (1, 0, 0), C = (0, 1, 0),
D = (−1, 0, 0), as shown in Fig. 3.7. The reason for choosing these locations for the
tying points is that the element mid-sides are the only points in which the strain
definition (3.6) is able to evaluate zero transverse shear strain in pure bending.
This is because the mid-sides are the only points in which the deformation figure
is correct, i.e. the normal remains normal whereby the transverse shear strains are
zero, Fig. 3.8. However, the choice of tying point for the MITC3 element involves
a problem since the transverse strains become dependent on the node numbering.
Chapter 3. Multilayered shell finite elements 33
(a) (b)
Figure 3.8: Linear shell element subjected to constant bending moment (a) and
corresponding actual deformed shape (b) with correct deformed shape (dotted line).
The assumed transverse shear strains for both the three- and four-node elements
are now found by evaluating the tying point strains, ˜ij |p , and using (3.16) for
N p = 2 with first order Lagrange polynomials:
1 1
˜13 = (1 − s)˜
13 |A + (1 + s)˜
13 |C
2 2 (3.18)
1 1
˜23 = (1 − r)˜23 |D + (1 + r)˜
23 |B
2 2
which provides linearly varying transverse shear stresses in the r, s-plane and
constant stresses in the t-direction. The assumed strains (3.18) are used for
determining the strain-displacement matrix, and as such the stiffness matrix, and
allow the elements to show proper behavior for thin elements.
3.4 Implementation
The shell element library in MUST originally consisted of the MITC4, SHELL3,
SHELL4, SHELL6 and SHELL9 elements with linear and nonlinear capabilities.
These were implemented by Jensen et al. (2001) and later, a linear MITC3 element
was added and tested by Moser (2002) and Poulsen et al. (2003) adapted MUST
to accommodate a SHELL16 element and performed extensive testing. In those
implementations the invariant properties of the energy terms in (2.1) were not
exploited and consequently, a number of essentially redundant transformations
were used. Removing these transformations and formulating the stiffness matrix
in the covariant base vectors instead of the global Cartesian system lowered the
computational cost of forming the global tangent stiffness matrix by approximately
2 Tests performed by Moser (2002) in connection with the original implementation of the linear
MITC3 element.
34 3.4. Implementation
15% and rendered the elements very efficient when comparing to commercial codes.
However, a number of trivial errors existed in the original implementation of the
tangent stiffness matrix of the MITC4 elements. This did not cause noticeable
problems for moderate nonlinear analysis but was quite clear when comparing the
stiffness matrix to a central finite difference approximation. The implementation
strategy was (and still is) to formulate the tangent stiffness matrix analytically,
which results in some rather involved expressions and it is natural to suspect
insufficient differentiation in the derivation of the linearized terms or other errors
in the derivations. Unfortunately, even after spending considerable time on the
problem, a proper solution has not been found. Instead, the MITC3 and MITC4
element routines have been completely re-implemented and furthermore, the
SHELLn element routines have been optimized for higher computational efficiency.
The MAPLE routines start from the manually derived strain expression, which
are obtained by inserting (3.5) in (3.6). The linear strain-displacement matrix,
Be , is then obtained as:
∂e 1 ∂ ∂N ∂N
Be = = giT u + gjT u
∂u 2 ∂u ∂rj ∂ri
(3.20)
1 T ∂N ∂N
= gi + gjT
2 ∂rj ∂ri
Chapter 3. Multilayered shell finite elements 35
where i and j are set as appropriate for the given strain component (row in Be ),
i.e. i = j = 1 for e1 = e11 and so forth. This leads to the usual definition of the
linear strain-displacement matrix, i.e. δe = Be δu and from the strain definition
(3.6) it follows that e = Be u. It is implied in (3.20) that the covariant vectors are
computed in the appropriate node depending on the degree of freedom being used
from u. The nonlinear contribution is derived as:
∂η 1 ∂ ∂NT ∂N
Bη = = uT u
∂u 2 ∂u ∂rj ∂ri
T
(3.21)
∂N ∂N
= uT
∂rj ∂ri
and thus it follows that δη = Bη δu and from the strain definition in (3.6) that
η = 12 Bη u. Note that the expressions in (3.20) and (3.21) should be considered as
a conceptual aid rather than a model for implementation. Instead, all components
of the B-matrices are derived explicitly row by row as shown in Algorithm 3.2 in
pseudo code. Once the nonlinear strain-displacement matrix is known the second
term of the tangent stiffness matrix in (3.19) can be formed directly while the
first term, usually called the initial stress stiffness matrix and denoted Kσ , can be
determined as shown in Algorithm 3.3. Due to symmetry it is only necessary to
determine the upper or lower triangle of the matrix, which saves quite a number
of lines in the code and improves the computational efficiency. In both Algorithm
3.2 and 3.3 the variable ndof is the number of degrees of freedom for the element,
which is the number of nodes multiplied by the number of degrees of freedom per
node, e.g. ndof = 4 × 5 = 20 for four-node elements.
The code generated by MAPLE is efficient but the practical down-side is that it
consists of many lines of code, rendering debugging somewhat difficult. Also, the
element routines are rather cumbersome to work with in the editor but still, from
a computational and modeling point of view the result is very satisfactory. To
obtain a more editor-friendly implementation the element routines could be stored
in a pre-compiled library, but this has not been implemented.
Compute BTη
Compute s
for i = 1 to ndof do
for j = 1 to i do P6 T
Compute (Kij )σ = k=1 (∂(Bik )η /∂uj )sk
end for
end for
As patch test strategy for choice of mesh and boundary conditions we adopt the
approach of Dvorkin and Bathe (1984) and Lee and Bathe (2004) for the MITC4
and MITC3 elements, respectively. The patch test is successfully passed if the
element can display constant stress when subjected to appropriate boundary
conditions in a non-regular mesh. If the element passes it implies that the
Chapter 3. Multilayered shell finite elements 37
Name: MITC4 patch test - s11 Name: MITC3 patch test - s13
Sxx Sxz
2.000E+005 2.000E+005
2.000E+005 2.000E+005
2.000E+005 2.000E+005
2.000E+005 2.000E+005
2.000E+005 2.000E+005
2.000E+005 2.000E+005
Figure 3.9: Randomly selected patch test results for MITC4 elements (left) and MITC3
elements (right). The same uniform stress results are obtained for all other components.
kinematics are correct since stresses are computed from strains, which are in turn
computed from displacements. As expected both MITC elements pass all patch
tests for the five stress components s11 , s22 , s12 , s23 and s13 . Not all results are
reproduced here but the results for s11 and s13 obtained with the MITC4 and
MITC3 elements, respectively, are shown in Fig. 3.9.
The same results have been obtained for the SHELLn family of elements, but these
have been left out for brevity.
For the single-layer case three reference solutions have been obtained with ANSYS
and ADINA. The ANSYS SHELL181 element is a four-node element formulated
with MITC as the MITC4 element in MUST and is therefore a natural choice.
Furthermore, two quadratic elements have been used – the SHELL93 element in
ANSYS and the MITC9 element in ADINA, which is widely recognized as one of
the most robust elements available. As can be seen from Fig. 3.11 the results from
MUST and the commercial codes correlate very well.
38 3.5. Numerical verification
Polymeric foam
For the multilayer test three reference solutions have been obtained using ANSYS,
ADINA and COSMOS. The element of choice was again the SHELL181 element
in ANSYS but unfortunately the element was unable to converge. ANSYS
has recognized the problem and their Research and Development department is
currently investigating the cause of this problem. Instead the SHELL4L element
in COSMOS has been used as well as quadratic elements in ADINA and ANSYS.
The results are compared in Fig. 3.12 and again, the correlation is very good.
Equally good correlation has been found with solutions obtained using the
quadratic TYPE75 element in MSC.MARC3 and the MITC4 element in ADINA,
but these have been left out for the sake of brevity.
to be less efficient than the arc-length solvers employed by the other packages.
Chapter 3. Multilayered shell finite elements 39
1,0
MUST (MITC4)
0,9
MUST (MITC3)
0,8 MUST (SHELL9)
ANSYS (SHELL181)
0,7 ANSYS (SHELL93)
Load factor, k
0,5
0,4
0,3
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
Normalized center point displacement
1,0
MUST (MITC4)
0,9
MUST (MITC3)
0,8 MUST (SHELL9)
ANSYS (SHELL91)
0,7 COSMOS (SHELL4L)
Load factor, k
0,5
0,4
0,3
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
Normalized center point displacement
The remaining part of the chapter is organized as follows. In Section 4.1 the design
parametrization is discussed and the optimization problem is briefly defined. In
Section 4.2 the evaluation of sensitivities is explained and finally, four numerical
examples for plates and shells are presented in Section 4.3.
Solid 1
Voided
Solid Voided Solid 2
Voided Solid Solid 1
Voided Solid Voided Voided
Figure 4.1: Typical lamina sequences for plate/shell topology optimization: (a)
“classical” topology optimization, (b) symmetric lay-up with voided core, (c) symmetric
lay-up used with voided outer layers and (d) generic scheme.
sl = ρe Cl l (4.1)
This is standard in topology optimization and the only difference for multilayered
structures is that the scaling is employed on specific layers (as indicated by
superscript l in (4.1)) instead of on the entire element. The solid layers are
not scaled and the stiffness of these is calculated using the standard constitutive
relation, i.e. s = C. As in topology optimization the element densities, ρe ,
constitute the design variables and only a single design variable is assigned to any
element. It should therefore be noted that even though several voided layers may
be specified, all voided layers within the same element will be scaled in the same
way.
Choosing in which layers to apply the scaling in order to perform various types of
optimization is the next step and several optimization schemes are available.
K¹0
y d y
K=0
x
d
¹ x
K¹0
K¹0
Figure 4.2: Laminate optimized with scheme (b) to zero stiffness (K = 0) of the core,
which is not equivalent to removing the layer.
optimization no such measures have been implemented, and scheme (a) has not
been utilized.
The scheme in Fig. 4.1(b) seems interesting since core layout design is of
practical importance in many applications involving stiffening with sandwich cores.
However, the scheme lags in that the “removal” of core material by scaling ρe to
zero (or near-zero) does not eliminate its stiffness contribution entirely since the
presence of the solid layer in the numerical integration (3.13) still contributes to
the total stiffness of the laminate. This can be illustrated by considering the
moment of inertia, Izz , of the structure illustrated in Fig. 4.2. Scaling the stiffness
to zero eliminates the contribution to Izz from the core layer (white) but due to
the parallel axis theorem there will still be a contribution to the total moment of
inertia, proportional to d2 . Consequently, scheme (b) does not provide a viable
method for solving the core layout problem and should be replaced by some update
scheme, which removes/adds both the stiffness of layers and their contribution to
(3.13). There is no immediate solution to this problem and no solution has been
sought out in this work.
The two remaining schemes, however, have both been employed in the present
study. The rib stiffener scheme in Fig. 4.1(c) allows formation of symmetric ribs
on both sides of a fixed surface, which has many practical applications and suffer
from none of the deficiencies outlined above. Scheme (c) has been used by e.g.
Lee et al. (2000) and will be used in the first numerical example in Section 4.3.
The general scheme in Fig. 4.1(d) illustrates the point that any lay-up is treatable
in the present formulation and indeed, two of the three numerical examples in (4.3)
employ an asymmetric lay-up to form ribs on one side of a shell structure. However,
care must be taken to obtain a physically sound model when using scheme (d)
and furthermore, introduction of internal voided layers should be avoided unless
specifically required by the physical model being optimized.
The constraint acts on the effective volume, V , defined as the sum of the scaled
volume of all voided layers, N vl , in all elements, N e , i.e.:
e vl
N X
X N
V = ρe Ven (4.3)
e=1 n=1
The volume constraint is included to ensure that the optimizer does not increase
the stiffness (reduce the compliance) by pushing all design variables to 1, which
would be a rather useless result. To reduce the occurrence of intermediate densities
the SIMP methodology is applied to the stiffness scaling whereby (4.1) is actually
computed for each layer as:
sl = ρpe Cl l (4.4)
where the SIMP power p ≥ 1 penalizes densities between 0 and 1 by making
them uneconomical. This helps push the optimal topology towards a black/white
(0/1) design with no or few intermediate values of ρe as described by e.g.
Bendsøe and Sigmund (2003). The power is often chosen statically to p ≥ 3 but
in this work an incremental scheme is also used in which the power, p, is increased
from a starting value to a final value in steps of some chosen size. In this way the
penalization is increased as the design converges, which in our experience results
in more distinct topologies. As the power increases, however, the term ρpe tends to
zero quite rapidly, which may cause numerical problems. To avoid this problem
entirely, a conservative lower bound of ρmin = 1 · 10−3 has been used for all design
variables, as indicated in (4.2).
Solution of the problem in (4.2) is done as described in Section 2.2 using analytical
design sensitivities and the Method of Moving Asymptotes (MMA). This approach
has proven efficient in other works on nonlinear topology optimization (see e.g.
Bruns et al. (2002); Buhl et al. (2000)).
The sensitivities du/dρe are not readily available and hence, it is not possible to
determine the sensitivities of the objective function directly from (4.5). To circum-
vent this problem a few paths are available: (1) finite difference approximation,
(2) direct differentiation method and (3) the adjoint variable method. For the
finite difference to be accurate a central finite difference approximation should be
used but this involves two additional analyses and is therefore highly impractical
from a computational efficiency point of view. The direct differentiation method
is only favorable if the number of constraints is larger than the number of design
variables, which is not the case here. The adjoint variable method is fairly simple
to implement and has proven efficient for large numbers of design variables and
has therefore been chosen for this work. The procedure for forming an adjoint
problem (see e.g. Haug et al., 1986) is described in the following.
Finding the adjoint variables is very simple and fast since it just requires solving
the linear system:
KT λ = −p (4.11)
in which the factored stiffness matrix is readily available from solving the analysis
and the force vector, p, is known. Now, having solved for λ the objective function
sensitivities may be found directly as:
dC ∂R
= λT (4.12)
dρe ∂ρe
Finding the derivative of the residual in (4.12) with respect to ρe can be done
by finding the derivative of the internal force vector defined in (2.4) since
displacements have been assumed to be independent of the design variables, ρ.
The internal element force vector is stated again for reference:
Z
r= BT s dV (4.13)
V
Finding derivative of (4.13) effectively only involves finding the derivative of the
stress vector since the strain-displacement matrix, B, is not dependent on the
design variables. Thus, the residual derivative can be found by computing the
derivative of (4.4) as:
∂sl
= p ρ(p−1)
e Cl l (4.14)
∂ρe
And so, the vector product in (4.12) only yields contributions for the given element,
e, which renders the sensitivity calculation quite efficient.
1000
e
800
600
p d
400
c
200 b
0 a
0 20 40 60 80 100
u
Figure 4.3: Fictitious load-displacement space with selected load points a = (1, 10),
b = (15, 200), c = (25, 350), d = (80, 550) and e = (100, 1000).
where q is the load case number and wq is a weight factor for load case q. The
weights can be chosen arbitrarily but the natural approach would be to choose
them such that the contribution from each load case is approximately equal. This
ensures that no single load is favored over others and should therefore yield the
design that performs best at all loads. In the present work two strategies have
been tested:
kp1 k e T1 u
(p) e1
I : wq = II : wq = (4.16)
kpq k e T
(p)q ueq
Now, using (4.16) the weights for the compliance in (4.15) can be found from
Fig. 4.3 by inserting. The results are listed in Table 4.1. As indicated, the
difference between the two schemes is significant but for large numbers of q this
can be leveled out to some extent by choosing more loads in the lower part of the
spectrum. In general, however, it is more “fool proof” to choose scheme II since
it is hard to predict how the weighting affects the results. There is no “right”
approach to choosing the weighting scheme and a bit of tweaking of the weighting
factors must probably be endured before obtaining satisfactory results.
48 4.3. Numerical examples
Table 4.1: Weights computed with scheme I and II in (4.16) for fictitious load-
displacement space in Fig. 4.3. Values are rounded off to six digits.
Scheme I Scheme II
Weight: wi wi Ci wi wi Ci
The weighting factors naturally enter into the sensitivity expression as well
whereby (4.12) for multiple load cases is written as:
XNq
dC ∂Rq
= wq λTq (4.17)
dρe q=1
∂ρe
The major disadvantage of solving the multiple load case problem is the increased
computational time and memory consumption caused by the need to store multiple
tangent stiffness matrices. The latter issue is not a generic problem but exists in the
current implementation in MUST. For solving problems of practical engineering
interest this currently poses a severe limitation on MUST as memory usage may
easily approach the 4 GB limit on standard PC systems. This will, however, be
remedied in a later version of the software.
The optimal boundary curve may in practice be obtained in a number of ways. One
is to start all analyses from the same conditions, e.g. evenly distributed material
Chapter 4. Nonlinear topology optimization 49
(uniform gray) and u = 0. This strategy may lead to local optimum solutions
so at each load, alternatives should be sought out to see if a better solution can
be found. This constitutes solving for the same load several times with different
initial conditions, e.g. restarting from the previous or subsequent load with state
variables alone or both state variables and topology (this is illustrated in Fig. 4.11).
As such, the generation of the optimality boundary in the load-displacement space
may require a substantial amount of work. Even then, it is not guaranteed that
no better solution can be found at a particular load. This could be the case
for problems exhibiting non monotonic behavior in the load-displacement space,
e.g. snap-through behavior, since the Newton-type solvers are not well suited
for these types of problems. To provide a better understanding of the nonlinear
behavior over the entire load spectrum, the arc-length method has been used on
some optimal designs as illustrated in Figs. 4.12 and 4.14.
The first example is a plate example, which serves as a validation example for the
present formulation since the linear topology solution is well known. The second
and third examples are for a doubly-curved shell structure and are included to
demonstrate the effects of the nonlinearities on a more general structure for single
and multiple load cases, respectively. Finally, a multilayered, curved panel of
orthotropic lamina is included.
Note that in the chapter the examples have been solved exclusively using the
SHELL9 element since the MITC4 element had not been fully reimplemented
when these results were generated. Consequently, checker-boarding does not
pose a problem but a sensitivity filter has still been employed to ensure mesh-
independency (Bendsøe and Sigmund, 2003).
1,0 6
0,9
0,8
Normalized objective
0,7
5
0,6
Power
0,5
0,4
4
0,3
0,2
0,1
0,0 3
0 10 20 30 40 50
Iterations
Figure 4.4: History of objective function (left axis, continuous curve) and SIMP power
(right axis, step curve) for linear optimization of square plate example. Load factor is
0.027.
Figure 4.5 shows the evolution of the topology with increasing load factor, κ, and
two distinct shifts in topology can be observed (the gray areas). As expected
the topologies at small loads are the same for the linear and nonlinear cases, and
correspond well to those obtained by e.g. Lee et al. (2000).
With an increasing load two effects can be noted in Fig. 4.5 due to increasing
stiffening from in-plane stresses; 1) the four corner reinforcements become wider
and extend further into the plate and 2) the center reinforcement is reduced in size
and instead ribs extend towards the boundaries until at maximum load, the corner
and center reinforcements become a single reinforcement. This example both
validates the formulation and gives a first indication that the nonlinear effects are
of increasing importance as the load increases. This corresponds well to the usual
progressive nature of nonlinear effects. At light loading the optimal topologies
are identical, see Fig. 4.5, but show increasing deviation as the load increases.
However, the increase in performance in very small as indicated in Fig. 4.6, where
the topology obtained with a linear solution (κ = 0.001) has been compared to the
topology obtained with a nonlinear solution at maximum load (κ = 1.000). This is
achieved by performing a nonlinear analysis of the optimal topologies and tracing
the equilibrium path over the entire load spectrum using an arc-length solver.
As indicated in Fig. 4.6 the linear solution performs slightly better than the
nonlinear solution in the lower part of the spectrum and vice versa as the load
increases. This is to be expected but the difference in performance is only
approximately 0.45%, which amounts to nothing for practical purposes. This was
not particularly encouraging so we continue to investigate the same phenomena
for shell structures.
Chapter 4. Nonlinear topology optimization
1,0
0,9
0,8
0,7
Load factor, k
0,6
0,5
0,4
0,3
0,2
0,1
0,0
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
Normalized center point displacement
Figure 4.5: Topological evolution of square plate with increasing center point load. The dotted line is the linear load-displacement
51
curve, the solid line is the equivalent nonlinear response. Both curves are for the optimized structure.
52 4.3. Numerical examples
1,0
Boundary of optimal designs
0,9
k = 0.001
0,8 k = 1.000
0,7
Load factor, k
0,6
0,5
0,4
0,3
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
Normalized center point displacement
Solid: Aluminum
Solid: Foam
Solid: Aluminum
Voided: Aluminum
Figure 4.7: Geometry of spherical cap example with actual thickness and distribution
of layers. The model is subjected to a center point load and hinged along the four edges.
1,0 6
0,9
0,8
Normalized objective
0,7
5
0,6
Power
0,5
0,4
4
0,3
Nonlinear, k = 1
0,2
Linear
0,1
0,0 3
0 10 20 30 40 50
Iterations
Figure 4.8: History of objective function (left axis, continuous curve) and SIMP power
(right axis, step curve) for nonlinear optimization of spherical cap example. Normalized
to initial compliance.
The full shell geometry is modeled using a 40 × 40 mesh of SHELL9 elements and
the model is hinged (ui = 0) on the four edge curves. The volume constraint is
Vc = 0.5 and the SIMP power is increased from 3.0 to 6.0 in steps of 1.0 over the
first 30 iterations as shown in Fig. 4.8. As indicated in Fig. 4.8 the optimal design
only has a moderately improved performance over the initial design (uniform grey),
which is due to the moderate thickness of the stiffening layer as mentioned above.
Figure 4.9 shows the evolution of the topology with increasing load factor.
54
1,0
0,9
0,8
0,7
0,6
Load factor, k
0,5
0,4
0,3
0,2
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
Normalized center point displacement
Figure 4.9: Topological evolution of spherical cap with increasing center point load. The dotted line is the linear load-displacement
curve, the solid line is the equivalent nonlinear response. Both curves are for the optimized structure. The gray areas mark the
transitions between topologies.
Chapter 4. Nonlinear topology optimization 55
Figure 4.10: Actual deformed shapes of the three basic topologies of a quarter of the
spherical cap geometry: (a) moderately nonlinear, κ = 0.10, (b) intermediate topology,
κ = 0.37, (c) final configuration, κ = 1.00.
As the load increases the optimal stiffening topology changes five times as marked
by the gray zones in Fig. 4.9. The topology changes over these zones due to
changes in the load carrying mechanisms of the structure as shown for the three
major topologies in Fig. 4.10. The first transition (from (a) to (b) in Fig. 4.10)
occurs because material is needed to reinforce the corners, which are subjected to
local bending. At the same time the center area is highly affected by the load and
must be reinforced as well. The change (from (b) to (c) in Fig. 4.10) is caused by
increasing membrane effects due to the near-flat surfaces extending from the load
towards the hinged boundaries and along the symmetry planes.
It is interesting to note the major jump in topology that occurs when the structure
snaps through (the horizontal gray area in Fig. 4.9). The sudden change in
topology can be seen clearly when starting an optimization from a high load
factor (the post-snap regime, κ > 0.35) and solving for a smaller load factor.
This is illustrated in Fig. 4.11 where the state variables obtained at κ = 0.36 have
been reused as initial guess for solving for κ = 0.35. As can be seen, the optimizer
effectively pushes the design to the pre-snap configuration between iteration 19 and
20. This also indicates that the curve in Fig. 4.9 may be discontinuous between
these two points or, at least, have near-zero slope. This makes good sense since
the pre-snap configuration is by far more attractive in terms of compliance than
the post-snap configuration.
The increased performance of the nonlinear design over the linear solution is
modest since the stiffening layer only contributes moderately to the total stiffness.
However, the structure exhibits noticeably different nonlinear response for the
design obtained with the linear method compared to some of those obtained with
the nonlinear method. Consider the three load-displacement curves in Fig. 4.12,
generated using an arc-length solver for small, medium and heavy load (κ = 0.01,
κ = 0.42 and κ = 1.00, respectively). As expected all three curves stay below
the boundary of optimal designs (obtained from Fig. 4.9) but the response of the
three designs is not the same over the entire load spectrum. At low load factors
the curves are almost indistinguishable but as the load increases the different
topologies show their strengths and weaknesses.
56 4.3. Numerical examples
1,0 6
0,9
0,8
Normalized objective
0,7
5
0,6
Power
0,5
0,4
4
0,3
0,2
0,1
0,0 3
0 25 50 75 100 125 150
Iterations
Figure 4.11: History of objective function (left axis, continuous curve) and SIMP power
(right axis, step curve) for nonlinear optimization of spherical cap example, restarted from
state variables at κ = 0.36 and solved for κ = 0.35.
1,0
Boundary of optimal designs
0,9 k = 0.01
0,8 k = 0.42
k = 1.00
0,7
Load factor, k
0,6
0,5
0,4
0,3
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
Normalized center point displacement
Solid: Aluminum
Solid: Foam
Solid: Aluminum
Voided: Aluminum
Figure 4.13: Geometry of spherical cap (multiple load cases) example with actual
thickness and distribution of layers. The model is subjected to a center point load and
hinged along the four edges.
In Fig. 4.12 it is noteworthy that the result for κ = 0.42, although not optimal in
all configurations, has a superior ability to sustain load during snap-through. This
is another interesting aspect of utilizing nonlinear optimization for such problems.
The multiple load case optimizations are solved at κ values of 0.01, 0.17, 0.33,
0.67, 1.0 (5 load cases) and 0.03, 0.10, 0.17, 0.33, 0.40, 0.47, 0.53, 0.60, 0.67, 0.83
(10 load cases). For the 10 load case optimization the weights are chosen as a
scaling of the load (scheme I in (4.16)) while the 5 load case optimization is scaled
using the estimated compliance values at each load (scheme II in (4.16)). The
estimated values are obtained by running an initial analysis of the non-optimized
structure. Consequently, the weighting is not entirely uniform as indicated in
Table 4.1, page 48, but uniform to an order of magnitude, which is sufficient to
avoid severe favoring of individual load cases. The choice of load factors, κ, is the
result of an iterative process and is used primarily to seek out different topologies,
testing their nonlinear response, and deciding which ones perform satisfactorily.
This procedure is somewhat “unscientific” and illustrates the tweaking involved in
using the weighted sum formulation.
Using the approach described above should result in the topology (or topologies)
that constitutes a “good compromise” between the very different topologies ob-
tained at distinct loads, given the criteria defined in the weighted sum formulation.
In Fig. 4.14 the obtained topologies for 5 and 10 load cases are shown together with
reference topologies obtained using a linear and nonlinear solution, respectively.
An arc-length solver has been used to determine the nonlinear response of both the
topology obtained from a linear solution, the topology obtained from a nonlinear
solution at κ = 1 and the topologies obtained with 5 and 10 load cases.
Note that the curves in Fig. 4.14 represent transverse center point displacement
in the thickness direction. The step in the curves for the nonlinear solution
(κ = 1.00) and the 5 load case solution indicates that the geometry is deforming
with components perpendicular to the transverse direction, which is a well-known
behavior of such structures. The reason that this is observed here and not in
the previous example is the reduced thickness, which reduces the stiffness of the
shell in a way that allows it to snap in other directions before actually snapping
through.
0,9
0,8
0,6
Load factor, k
0,5
0,4
0,3
0,2
0,1
0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
Normalized center point displacement
Figure 4.14: Nonlinear response of optimized topologies for nonlinear optimization with multiple loads of spherical cap example. The
curves represent the response of the optimal topologies obtained with the linear solution and the nonlinear solution for κ = 1.00 as
59
well as the results of solving with 5 and 10 load cases (LC).
60 4.3. Numerical examples
Figure 4.15: Geometry of cylindrical shell example shown with actual thickness and
distribution of layers. Fiber angles are taken with respect to the straight edges of the
shell. The model is subjected to a center point load and hinged at the four corners.
The shell is modeled using a mesh of 50 × 50 SHELL9 elements and the model
is supported in three directions (ui = 0) at the four corner nodes. This renders
the model very flexible and snap-through occurs at a relatively low load factor as
shown in Fig. 4.16, which also shows the change in topology with increasing load.
In Fig. 4.16 the cross-shaped topology obtained for large displacements coincide
with the primary material directions of the biax reinforcement layer and as such
correspond well to the anticipated design.
Chapter 4. Nonlinear topology optimization
1,0
0,9
0,8
0,7
0,6
Load factor, k
0,5
0,4
0,3
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
Normalized center point displacement
Figure 4.16: Topological evolution of cylindrical shell with increasing center point load. The dotted line is the linear load-displacement
curve, the solid line is the equivalent nonlinear response. Both curves are for the optimized structure. The gray areas mark the
61
transitions between topologies.
62 4.4. Summary and conclusions
Figure 4.17: Optimal stiffening topology for cylindrical shell, solved for 10 load factors
(0.01, 0.04, 0.08, 0.12, 0.20, 0.40, 0.56, 0.64, 0.80, 1.00).
As in the previous example the problem has also been solved for multiple load
factors to obtain a compromise design that will perform well throughout the load
spectrum. The 10 load factors considered are 0.01, 0.04, 0.08, 0.12, 0.20, 0.40,
0.56, 0.64, 0.80, 1.00, which yields the design shown in Fig. 4.17 using scheme I
from (4.16). Again, the multiple load factor design is a compromise between the
different topologies, which is evident when comparing Fig. 4.16 to Fig. 4.17.
problem solvable without the need for any additional numerical tricks. Multiple
load cases are handled in a weighted sum formulation and allows us to solve for
“best compromise” designs in a wide load spectrum.
Four numerical examples demonstrate the capabilities of the method and show
the effects of the nonlinear terms. The performance of the nonlinear designs over
the linear solutions is better in all examples (lower compliance) but the difference
in compliance may be relatively small. This is the case in the first spherical cap
example where the optimal distribution of an isotropic reinforcement material is
found for a single and relatively thin layer. If the thickness of the reinforcement
layer is increased substantially the difference in compliance also increases as
demonstrated in the second spherical cap example. Here, a major difference in
the nonlinear response for topologies obtained at low and high loading can also
be observed. Furthermore, the example shows the “best compromise” response
obtained when solving for 10 different load factors simultaneously. This indicates
that a multiple load case approach should be taken for structures operating at both
low and high loads. Finally, a cylindrical shell example shows the reinforcement
pattern of two opposite orthotropic biax layers. The result correspond well to the
anticipated result in that the reinforcement coincides with the principal material
directions of the biax.
In all the examples a significant change in topology is observed for increasing loads.
Also, the results are consistent with the progressive nature of nonlinear effects in
that a gradual increase of loads yields a gradual change in topology (starting from
the linear and nonlinear topologies being identical and tending towards increasing
deviation of the nonlinear topologies). A common characteristic of the large
displacement topologies is that they are “global” in nature where the linear results
tend to involve local reinforcements. This is to be expected since an increasing
portion of the geometry becomes affected when the load increases. An interesting
observation is that the manufacturability of the large displacement results seems to
be higher since they are continuous over the geometry. For laminated composites
this is essential since application of smaller reinforcement patches would be a highly
impractical approach.
64 4.4. Summary and conclusions
5
Discrete material optimization
Displacements will in this chapter be assumed small and material behavior linear
but the proposed parametrization could easily be used for nonlinear problems as
well. For the linear case the element stiffness matrix, Ke , is defined as in (2.11)
but stated here again for reference:
Z
Ke = BT CB dV (5.1)
V
which is summed over all elements and used for solving the linear static equilibrium
equations:
Ku = p (5.2)
Again, the compliance, C, is used as objective function and may for the linear case
be stated as:
C = uT p = uT Ku = 2U (5.3)
where U is the total strain energy. These equations form the basis for implementing
the proposed parametrization and will be addressed in more detail in Section 5.2.
The rest of the chapter is organized as follows. Section 5.1 discusses orientational
optimization methods to place DMO in a context, Section 5.2 introduces the
DMO parametrization, Section 5.3 takes a closer look at the element level
66 5.1. Orientation optimization with orthotropic materials
q1
qm
qNe
parametrization and Section 5.4 states the optimization problem. In Section 5.5 a
number of 2D and 3D examples are presented to illustrate the capabilities of the
DMO method.
Objective : min C(θ) = pT u
θ
(5.4)
Subject to : Ku = p
θmin ≤ θ ≤ θmax
where θmin and θmax contain the lower and upper bounds on the design variables,
respectively (typically −90◦ and +90◦ ). The design sensitivities (compliance
sensitivities) can be obtained as the derivative of (5.3) and when assuming design
independent loads, the following expression may be derived (see e.g. Pedersen
(1991) or Masur (1970)).
dC dK ∂Kem e
= −uT u = −(uem )T u (5.5)
dθm dθm ∂θm m
This expression is valid if θm only pertains to a single element. Later we will
introduce so-called patch variables which affect several elements and in such cases
Chapter 5. Discrete material optimization 67
10
y q1
5
y
x
x
P
P P
2
q2
P
(a) (b)
Figure 5.2: Example geometry and boundary conditions (a) and optimum solution (b).
the sensitivity dC/dθm must be evaluated as a sum over all elements affected
by the variable. The expression in (5.5) is common to both CFAO and topology
optimization and is in practice quite efficient to evaluate either analytically or more
commonly for CFAO by a finite difference approximation for general geometries.
The major difficulty faced when using this type of formulation is that the global
design space becomes non-convex. To illustrate this the simple problem in Fig. 5.2
has been constructed and solved using a continuous formulation and the Method
of Moving Asymptotes (MMA) by Svanberg (1987). The result is the design space
shown in Fig. 5.3(a), where four extrema can be found – one global and three local
(see also Table 5.1). To obtain the global optimum solution the initial guess must
be within the non-shaded part of the design space (the safe domain) as illustrated
in Fig. 5.3(b). In a simple case like this it is fairly easy to determine how to start
the optimization but for complicated geometries with multiple layers it becomes
almost impossible to state the problem a priori within the safe domain. It follows
that for generic problems the results obtained in this manner with CFAO will often
be sub-optimal albeit better than the initial design.
Table 5.1: Extremum values for the design space in Fig. 5.3.
This is, of course, not a new realization and several methods have already been
proposed to circumvent the problem of local optimum solutions. These methods
roughly fall within one of four categories:
-90
-60
Normalized ocjective
(compliance)
1,80 1,72-1,80
-30
1,72 1,64-1,72
1,56-1,64
Element 1
1,64
1,56 1,48-1,56 0
1,48 1,40-1,48
1,40 1,32-1,40
-90 1,32 1,24-1,32 Optimum 30
1,24 1,16-1,24
-60
1,16 1,08-1,16
-30 1,08
1,00-1,08 60
1,00
0
-90
Element 2 -60
30 90
-30
-90 -60 -30 0 30 60 90
60 0
30 Element 2
Element 1
90 60
(a) 90 (b)
Figure 5.3: Normalized objective function for CFAO test example. The red dots mark
the local extrema and the shaded area marks values that will lead to local optimum
solution.
limitation in the applicability of the methods but for simpler geometries the
methods have been applied very successfully and often serve as benchmark
solutions for purely numerical methods. The major contributor to optimality
criterion methods in orientation design is Pedersen (1991) who has used
the method to optimize 2D continua, beams and plates. Luo and Gea
(1998) used a similar approach to solve for optimal orientation of plate bead
stiffeners.
2. Mathematical programming techniques are purely numerical and are con-
cerned with tuning the optimizer itself rather than reformulating the problem
or change the parametrization. Among others Bruyneel and Fleury (2002)
and Moita et al. (2000) have applied such methodology with success. Still,
these methods do not ensure convergence to the globally optimum solution.
3. Parametrization methods start from the realization that the problem of
local optimum solutions is inherent in the CFAO method. The goal of
the methods is to change the parametrization and obtain a convex design
space. Such methods include the Lamination Parameter Method introduced
by Tsai and Pagano (1968) and used by e.g. Miki and Sugiyama (1993) for
orientation optimization of plates and Foldager et al. (2001) for plates and
cylindrical shells. The method requires closed-form analytical formulation of
the feasible domain of the lamination parameters which has so far only been
achieved for relatively simple geometries.
4. Evolutionary techniques, attributed to Holland (1975), are fundamentally
search methods but employ ideas derived from genetics and Darwin’s “sur-
vival of the fittest” principle to more effectively select the best solution from
Chapter 5. Discrete material optimization 69
Common to the parametrization used in the works stated above is that materials
70 5.2. The discrete material optimization method
are assumed to be isotropic, the elements are single layered and the maximum
number of phases involved is four – three distinct materials and void. Furthermore,
the structures considered have been either 2D continua or plates. The DMO
formulation extends the scope of application by considering multiple phases and
3D structures and furthermore by allowing the structures to be multilayered and
of orthotropic materials. However, the method is still limited to operate on a fixed
design domain, i.e. thicknesses and shape are not changed during optimization.
It follows that the number of candidate materials is also the number of element
design variables and if N e is the number of elements, the total number of design
PN e e
variables, N dv , for single layered structures is N dv = i=1 ni . Note that
the “classical” topology optimization formulation having one design variable per
element is obtained by setting ne = 1 in (5.6).
The weights, wi , in (5.6) must have values between 0 and 1 as no matrix can
contribute more than the physical material properties and a negative contribution
is physically meaningless. In this way, as in classical topology optimization,
the weights on the constitutive matrices become “switches” that turn on and off
stiffness contributions such that the objective is minimized and a distinct choice of
candidate material is made. This underlines that the DMO method relies heavily
on the ability of the optimizer to push all weights to the limit values. Any element
having intermediate values of the weights must be regarded as undefined since
the constitutive properties are non-physical. For the same reason any element
having more than a single weight of value 1 must be considered undefined as well.
Consequently, the single most important requirement for the DMO method is that
every element must have one single weight of value 1 and all other weights of value
0. Failing to comply with this essentially renders the results meaningless. It follows
that the choice of weighting functions, wi , is very important for the performance of
Chapter 5. Discrete material optimization 71
-90
-60
qi 24.2° 1,72-1,80
-30
y 1,64-1,72
1,56-1,64
Element 1
1,48-1,56 0
x 1,40-1,48
1,32-1,40
1,24-1,32 30
qi -41.6°
1,16-1,24
1,08-1,16
1,00-1,08 60
90
-90 -60 -30 0 30 60 90
Element 2
(a) (b)
Figure 5.4: Normalized objective function for DMO test example. (a) The candidate
materials at 12 angles, θi . (b) The yellow dots mark possible combinations of candidate
materials and the large and small white dots mark optimum solutions obtained with
DMO and CFAO, respectively.
the DMO method and several formulations have been developed and evaluated (as
described shortly). The initial values of the design variables, xi , may in principle
be any set of numbers between 0 and 1 but in general the values should be chosen
such that the initial weighting is uniform, i.e. wi = wj for all i, j = 1 . . . ne . This
provides the most “fair” starting guess since no materials are favored a priori.
To illustrate the methodology for fiber angle optimization we solve the example
in Fig. 5.2 using DMO1 with the same orthotropic material oriented at 12
different angles (0◦ , ±15◦ , ±30◦ , ±45◦ , ±60◦ , ±75◦ , 90◦ ) as the candidate materials
(Fig. 5.4(a)). The possible material constitutions for the structure are then all
combinations of the 12 candidate materials for two elements, e.g. 12 × 12 = 144
combinations in all (±90◦ are identical). These are marked by yellow dots in
Fig. 5.4(b) where the obtained optimum solution 30◦ /−45◦ is marked by the
large white dot. This solution is the “best fit” to the global optimum solution
24.2◦ /−41.6◦ obtained with CFAO (see Table 5.1). The normalized compliance of
the DMO solution is 1.0040, which is very close to the global optimum solution.
In this formulation each design variable scales only one constitutive matrix and has
no influence on any of the other matrices. To push the design variables towards
0 and 1 the SIMP method has been adopted by introducing the power, p, as a
penalization of intermediate values of xi (see e.g. Bendsøe and Sigmund (2003)
for details). The method in (5.7) is not very efficient as it fails to push the design
variables to the limit values for all the cases tested.
As can be seen from (5.8) the SIMP methodology is still encompassed in this
formulation. However, several implementations of the power, p, have been
Chapter 5. Discrete material optimization 73
w1 w1 w1
1 1 1
0 0 0
0 0 0
0.2 0.2 0.2
0.4 0.4 0.4
0.6 0.6 0.6
0.8 0.6 0.4 0.2 0 0.8 0.6 0.4 0.2 0 0.8 0.6 0.4 0.2 0
x2 11 0.8 x2 11 0.8 x2 11 0.8
x1 x1 x1
p=1 p=3 p = 15
w2 w2 w2
1 1 1
0 0 0
0 0 0
0.2 0.2 0.2
0.4 0.4 0.4
0.6 0.6 0.6
x2 0.8 0.4 0.2 0 x2 0.8 0.4 0.2 0 x2 0.8 0.4 0.2 0
11 0.8 0.6 11 0.8 0.6 11 0.8 0.6
x1 x1 x1
Figure 5.5: Weight functions, w1 and w2 , for two materials, computed with DMO
scheme 3. Top and bottom row represent w1 and w2 , respectively.
suggested but here we follow the strategy suggested by Gibiansky and Sigmund
(2000). The formulation in (5.8) is not very general since it can only handle two
materials, so the scheme has been adapted to multiple materials in MUST.
where the limits 0 ≤ xi ≤ 1 have been excluded for brevity. The expression in (5.9)
becomes tedious to write out for larger number of variables but can be generalized
for any number of candidate materials:
e
nX −1 i−1
Y e
nY −1
Ce = (xe0 )p [1 − (xei )]p (xej )p Ci + (xej )p Cne (5.10)
i=1 j=1 j=1
| {z } | {z }
wi wne
P
where we have that wi = 1. To accommodate fiber angle optimization the
scheme can in MUST be used either with or without the topology optimization
variable, x0 . DMO scheme 3 has proven efficient for up to three phases but when
the number of phases is greater the formulation tends to get stuck in local optima.
This is particularly the case when the remaining choice is between two almost
74 5.3. Element level parametrization
w2 w2 w2
1 1 1
0 0 0
0 0 0
0.2 0.2 0.2
0.4 0.4 0.4
x2 0.60.8 0.2 0 x2 0.60.8 0.2 0 x2 0.60.8 0.2 0
11 0.8 0.6 0.4 11 0.8 0.6 0.4 11 0.8 0.6 0.4
x1 x1 x1
p = 1, q = 1 p = 1, q = 3 p = 1, q = 15
w2 w2 w2
1 1 1
0 0 0
0 0 0
0.2 0.2 0.2
0.4 0.4 0.4
0.6 0.6 0.6
x2 0.8 0.4 0.2 0 x2 0.8 0.4 0.2 0 x2 0.8 0.4 0.2 0
11 0.8 0.6 11 0.8 0.6 11 0.8 0.6
x1 x1 x1
p = 3, q = 3 p = 3, q = 9 p = 3, q = 15
Figure 5.6: Weight function, w2 , for two materials, computed with DMO scheme 3. Top
and bottom row represent w2 with q-penalization and w2 with combined pq-penalization,
respectively.
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 0
0.2 0.2
0.4 0.4
0.6 0.6
0.8 0.6 0.4 0.2 0 0.8 0.6 0.4 0.2 0
x2 11 0.8 x2 11 0.8
x1 x1
p = 1, q = 3 p = 3, q = 3
Figure 5.7: Sum of weight functions, w1 and w2 , for two materials, computed with
DMO scheme 3.
equally good materials. This can be explained by considering a plot of the weights
for two materials as shown in Fig. 5.5. Since the two weights add up to 1 any spot
in the design space is equally good and the optimizer cannot figure out where to
go since nothing forces it to choose. This behavior has been observed for elements
in pure shear where the same orthotropic material at ±45◦ is equally good. To
counter this a modified SIMP scheme has been tried in which a second power, q, is
introduced on the weights as [1 − (xei )p ]q and (xej )pq . This changes the behavior
of the weight as illustrated for the second weight, w2 , in Fig. 5.6 while the first
weight, w1 , is unaffected. Using the pq-power scheme the weights no longer add
up to one and the middle region of the space becomes unfavorable as shown in
Fig. 5.7. This improves the capabilities of the method in some cases but requires
some tweaking of the power q to converge to the optimum solution. Consequently,
interpolation scheme 3 in (5.10) is not generally employed.
Chapter 5. Discrete material optimization 75
w1 w1 w1
1 1 1
0 0 0
0 0 0
0.2 0.2 0.2
0.4 0.4 0.4
x2 0.60.8 0 x2 0.60.8 0 x2 0.60.8 0
0.8 0.6 0.4 0.2 0.8 0.6 0.4 0.2 0.8 0.6 0.4 0.2
11 11 11
x1 x1 x1
w2
p=1 w2
p=3 w2
p = 15
1 1 1
0 0 0
0 0 0
0.2 0.2 0.2
0.4 0.4 0.4
x2 0.60.8 0 x2 0.60.8 0 x2 0.60.8 0
0.8 0.6 0.4 0.2 0.8 0.6 0.4 0.2 0.8 0.6 0.4 0.2
11 11 11
x1 x1 x1
Figure 5.8: Weight functions, w1 and w2 , for two materials, computed with DMO
scheme 4. Top and bottom row represent w1 and w2 , respectively.
The difference from (5.7) is the term (1 − xej ), which is included so that an increase
in xi automatically involves a decrease in all other weights. The behavior of the
interpolation for two materials is illustrated in Fig. 5.8.
The introduction of the (1 − xej ) term helps drive the design towards 0/1 and the
method has proven quite effective for the problems tested. The difference between
(5.10) and (5.11) becomes clearer when writing out the expression for e.g three
materials (phases) and comparing to (5.9):
The interpolation is quite simple to implement in a general way from (5.11) and
the weight factors can be computed efficiently for both analysis and sensitivity
analysis as illustrated in Algorithm 5.1.
76 5.3. Element level parametrization
Algorithm 5.1: Pseudo code for computing weight factors with DMO scheme 4.
wi = 1 for all i = 1 . . . ne . Initialization
if (analysis) then
for i = 1 to ne do
wi = xpi
for j = 1 to ne do
if (i 6= j) wj = wj × (1 − xpi )
end for
end for
else if (sensitivity analysis) then
for i = 1 to ne do
if (k = i) then . k is the active variable, i.e. ∂wi /∂xk
wi = p xp−1
k . From xpk term
else
wi = −p xp−1 k . From (1 − xpk ) term
p
wi = wi × xi . Constant term
end if
for j = 1 to ne do
if ((i 6= j) and (i 6= k)) wj = wj × (1 − xpi )
end for
end for
end if
The structure of Algorithm 5.1 is identical for the other interpolation schemes in
MUST and consequently, pseudo codes for these have not been included.
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
01 01
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.4 0.2 0 0.2 0.4 0.2 0
x2 01 0.8 0.6 x2 0 1 0.8 0.6
x1 x1
p=1 p=3
Figure 5.9: Sum of weight function, w1 and w2 , for two materials, computed with DMO
scheme 4.
w1 w1 w1
1 1 1
0 0 0
0 0 0
0.2 0.2 0.2
0.4 0.4 0.4
x2 0.60.8 0.4 0.2 0 x2 0.60.8 0.4 0.2 0 x2 0.60.8 0.4 0.2 0
11 0.8 0.6 11 0.8 0.6 11 0.8 0.6
x1 x1 x1
p=1 p=3 p = 15
w2 w2 w2
1 1 1
0 0 0
0 0 0
0.2 0.2 0.2
0.4 0.4 0.4
x2 0.60.8 0.4 0.2 0 x2 0.60.8 0.4 0.2 0 x2 0.60.8 0.4 0.2 0
11 0.8 0.6 11 0.8 0.6 11 0.8 0.6
x1 x1 x1
Figure 5.10: Weight functions, w1 and w2 , for two materials, computed with DMO
scheme 5. Top and bottom row represent w1 and w2 , respectively.
from scheme 4 (5.11) and complementary to this we have the scaled version:
e e
n
X n
Y
ŵi
Ce = x0 Pne Ci where ŵi = (xei )p 1 − (xej )p (5.15)
ŵk
i=1 | k=1
{z } j=1;j6=i
wi
from scheme 5 (5.13). These two schemes have been implemented recently and
therefore not tested extensively yet, but this will be part of our future work.
1 1 2 2 1
2
1 1 2 2 3
4
4
4 4 3 3 3
2
4 4 3 3 1
(a) (b)
Figure 5.11: Four patch variables (1–4) used to reduce the total number of design
variables by collecting elements over the structure (a) or layers in an element (b).
where l again denotes “layer” and thus, nl is the number of candidate materials
for the layer.
The number of element design variables, ne , for multi layered elements is then
the sum of the number of design variables per layer, nl , over all layers, N l , i.e.
P l l
ne = N k=1enk . As before, the total number of design variables in the problem is
P N
dv
N = i=1 nei , which for multilayered structures implies a significant increase in
the total number of design variables. To counter this, patches of design variables
are introduced.
The total number of design variables, N dv , when using patch design variables
80 5.4. The optimization problem
PN p
is N dv = i=1 npi where np is the number of candidate materials for the patch
variable and N p is the number of patches. This can provide a significant reduction
in the number of design variables but at the same time requires some extra effort
and insight on the part of the engineer.
where m is the mass of the structure and mc is the allowable mass, analogous to
the constraint on volume in topology optimization, (4.2). The mass constraint is
compared to the weighted mass of all, nl , candidate materials in all layers, N l ,
over all elements, N e , i.e.:
e l l
N X
X N X
n
m= (wi ρi Vi )le (5.18)
e=1 n=1 i=1
This constraint is not active when doing pure fiber angle optimization since a
change in fiber angle involves no change in mass and consequently, the constraint
may be left out entirely (which in MMA is achieved by setting the constraint
bound very high). When doing multi material optimization the mass constraint is
important since it effectively determines the amount of light material in the final
structure. Note that in (5.18) the density, ρi , is the physical material density in
[kg/m3 ] of material i and not a scaling as was the case in Chapter 4.
Obtaining the derivative of the residual (or the stiffness matrix) consists of
computing the derivative of the internal force vector (2.3) since the forces are
assumed independent of the design variables. In turn, the gradient is the derivative
of the weighting functions introduced earlier in Section 5.3. These are relatively
simple polynomial expressions that allow the sensitivity analysis to be implemented
analytically in a general and simple way. The weighting function derivative is also
used to compute the constraint function sensitivities directly from (5.18).
where w is a scaling parameter introduced to ensure that the penalty term does not
dominate the problem. The value of w can be set in various ways e.g. according to
the compliance value, the largest sensitivity or some other measure. The penalty
function is designed to automatically decrease as the design variables tend towards
0/1 by encompassing the term (1 − xpi ).
An elaborate framework has been implemented in which the penalty function can
be introduced either depending on the iteration number or the level of convergence
reached, i.e. when h reaches a preset value. Furthermore, the penalization can
be gradually increased or decreased as the optimization progresses by additional
scaling of w. However, the method requires an extensive amount of tweaking
and for the tested cases it was not able to significantly improve the overall
convergence of the optimization. Eventually the method was abandoned in favor
of the combined interpolation scheme in which DMO scheme 4 is used for the
stiffness and scheme 5 for the physical constraints, see Section 5.3.5.
Figure 5.12: Optimal fiber angle distribution in cantilever beam with uniformly
distributed top load.
Figure 5.13: Optimal fiber angle distribution in cantilever beam with uniformly
distributed top load. Solved using 48 patches of 4 × 4 elements.
To illustrate the patch variable methodology the problem has been solved using
48 patches of 4 × 4 elements, which reduces the number of design variables to 576
and reduces the runtime by approximately 16%. The resulting optimal fiber angle
distribution is shown in Fig. 5.13.
symmetry
Figure 5.14: Optimal material and fiber angle distribution in beam subjected to four-
point bending (symmetric). Black indicates glass/epoxy, light gray indicates foam and
intermediate values represent unconverged elements.
polymeric foam material having E = 125 MPa and νxy = 0.30 is used. This
results in 5 design variables per element and thus 3, 840 in total. The mass
constraint is set to mc = 4 kg which effectively means that the foam must account
for roughly 74% of the material usage when the densities of glass/epoxy and
foam are 1900 kg/m3 and 100 kg/m3 , respectively. The optimization converges
monotonically to h95 = 0.96 in 125 iterations taking just over 30 minutes.
The result of the optimization is shown in Fig. 5.14. The edge of the optimum
geometry (marked by black) is very similar to results obtained with classical
topology optimization techniques. The difference is that instead of obtaining a
frame-like structure, the DMO method uses the polymeric foam material (light
gray in Fig. 5.14) to form a sandwich structure. If the mass constraint is loosened
the DMO method will tend toward distributing stiff material in a frame structure
as well. The intermediate densities found in the area below the point load is a
local effect and roughly outline would-be bars. Tightening the mass constraint
will reduce this effect but the intermediate densities have been allowed here in
order to illustrate the relationship of the DMO methods with classical topology
optimization.
Glass/epoxy (+45,0,-45,90)
} Glass/epoxy (+45,0,-45,90)
Figure 5.15: Geometry of hinged spherical cap example with actual thickness and
distribution of layers.
20 1,0
18 0,9
Figure 5.16: Convergence of objective function (left axis) and DMO convergence ratio
(right axis) for spherical cap example.
(1) (2)
(3) (4)
Figure 5.17: Optimal material distribution and orientation in layers 1–4 of spherical
cap. White and black represents foam and glass/epoxy, respectively.
The convergence is shown in Fig. 5.16 and the resulting material distribution
and fiber orientations are shown for layers 1–4 in Fig. 5.17 and for layers 5–8
in Fig. 5.18. The layers are numbered from the outside of the shell, i.e. layer 8 is
on the center of curvature side.
As shown in Figs. 5.17 and 5.18 the fiber angle optimization problem has been
solved for the skin layers (layer 1 and 8) and the combined material distribution
and orientation problem has been solved for the internal layers (layers 2–7).
The result for the skin layers resemble those found in the literature for plates
Chapter 5. Discrete material optimization 87
(5) (6)
(7) (8)
Figure 5.18: Optimal material distribution and orientation in layers 5–8 of spherical
cap. White and black represents foam and glass/epoxy, respectively.
under similar boundary conditions (e.g. Pedersen (1991)). A solution for the
material distribution problem has not been reported in the literature but the
solution corresponds well to known reinforcement techniques for sandwich panels
(Bozhevolnaya and Lyckegaard, 2005).
In the final solution a cone has been formed through the thickness of the shell
to support the local, concentrated load. At the bottom of the cone (the center
of curvature side) a wider reinforcement is obtained to distribute the transverse
load over a larger area, thus reducing the local deformation in the lower skin. The
88 5.5. Numerical examples
Figure 5.19: Internal material distribution of glass/epoxy material at the center (real
geometry is 3D and curved, see Fig. 5.15). The thickness has been magnified 10 times to
better illustrate the cone-like shape.
The hinged spherical cap example agrees well with the expected results and, to the
extent comparison is possible, with existing results in the literature. The example
demonstrates that the DMO method is capable of simultaneously solving the fiber
angle problem and the material layout problem. The latter is in fact solved both
over the surface of the structure as well as in the thickness direction.
Assembly
Main spar Mf
Me Trailing edge
Figure 5.20: Composition of a wind turbine blade. The blade is subjected to flapwise
bending, Mf , and edgewise bending, Me . Courtesy of Lennart Kühlmeier, Vestas Wind
Systems A/S.
Chapter 5. Discrete material optimization 89
Figure 5.21: Finite element model used for maximum stiffness design of the load
carrying main spar.
The main spar carries most of the flapwise bending loads, Mf , whereas the shell
carries most of the edgewise bending loads, Me . In this study the main spar is
subjected to the most critical load case, which is the flapwise bending load that
arises when the turbine has been brought to a standstill due to high wind, and
the blade is hit by a 50 year extreme wind. In the model used, only the main
spar is considered, i.e., the two shell parts in Fig. 5.20 are removed to reduce
the complexity of the model. In order to account for the stiffness contribution of
the shells to the stiffness of the main spar, the thicknesses of the shells in direct
contact with the main spar are added to the top of the flanges of the main spar,
see Fig. 5.22. Thus, the local stiffness contribution is included but the support
conditions from the shells are ignored.
The employed finite element model shown in Fig. 5.21 was generated using ANSYS,
and the mesh consists of 9 600 MITC4 shell elements with 16 layers. The main spar
has a total length of 15 meters and the flapwise bending is applied using two nodal
forces at the end as shown in Fig. 5.21. The model is clamped at the root end,
i.e. all displacements and rotations are fixed. With these boundary conditions the
dominant state will be bending, which results in tension/compression in the top
and bottom flanges and shear in the wedges. Furthermore, due to the geometry
of the spar, which twists its cross section along the length, the spar will also be
subjected to torsion when bend at the tip. These basic considerations will be used
as a guideline for interpreting the results of the optimization.
The design parametrization proceeds in two ways as follows. 1) The main spar is
divided into 77 patches with identical lay-up and thickness to reduce the number
of design variables, Fig. 5.23. The choice of patches follows the constitution of
the main spar in that elements with identical lay-up and thicknesses are collected
in patches. The main spar model was a priori split up into discrete regions with
90 5.5. Numerical examples
Figure 5.22: A closer look at the constitution of the main spar model with discrete
thickness jumps along the length of the spar (left) and over the cross section (right).
Figure 5.23: Main spar model with distribution of patches. Not all patches are shown
due to limitations in the color resolution.
identical thickness and lay-up as indicated in Fig. 5.22. 2) All elements in the
main spar are assigned design variables in all layers. This renders the model
much more complex but allows for a more detailed material selection. For both
parametrizations there are two materials in play, a stiff glass/epoxy material and
a softer isotropic foam core material. To govern the amount of light material
used a mass constraint is included, which allows for 80% of glass/epoxy and 20%
foam material. Due to a confidentiality agreement with Vestas Wind System A/S
we cannot disclose the exact material properties, mass constraint, geometry and
thickness distribution. For simplicity and to keep the number of design variables
relatively low, the glass/epoxy material can only be oriented at 0◦ , ±45◦ , and 90◦ .
Furthermore, only the interior layers are allowed to consist of foam material since
foam is not a realistic choice for the skin layers. Consequently, the distribution
of design variables for the 16 layers is [4, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4], which
brings the number of design variables per element to 78 and thus the total number
of design variables is 6 006 for the patch variable model and 748 800 for the full
Chapter 5. Discrete material optimization 91
1,0
60 Objective (patch) 0,9
Figure 5.24: Convergence of objective function (left axis) and DMO convergence ratio
(right axis) for wind turbine main spar example with patched and full variables.
variable model.
Both models were solved on a desktop PC. The patch variable model used 131
iterations to reach h99.5 = 0.975 taking approximately 33 hours of computational
time while the full variable model used just 81 iterations to reach h99.5 = 0.997
in 21 hours. The vast majority of time, around 98%, is spent on the design
sensitivity analysis and increasing this performance is going to be addressed in
future work. Again the MMA algorithm proves very efficient and solves the
approximate subproblem of the full variable model in just under 4 seconds. The
convergence is shown for the first 50 iterations in Fig. 5.24 where the subsequent
iterations have been left out for clarity. The objective of the full variable model is
approximately 5% lower than for the patch variable model.
The result of the optimization using patch variables is shown for all 16 layers in
Figs. 5.26, 5.27, 5.28 and 5.29 – 4 layers at a time. Similarly, the result of the
optimization using full variables is shown in Figs. 5.30, 5.31, 5.32 and 5.33. The
optimized material directions for the glass fiber are shown for all 16 layers, and
if no material direction is indicated it is implied that soft core material has been
chosen for that particular layer of a given patch/element. Note that layer 1 is the
inner (bottom) layer and layer 16 the outer (top) layer.
The two models yield very similar results and, as expected, most of the foam
core material has been put in the webs in the internal layers 2–15 close to the
root of the main spar in order to increase the bending stiffness where the bending
moment is strong. Looking at the two models in detail it is apparent that the
full variable model chooses to use foam material in all internal layer of the web,
distributed continuously over approximately the first third of the main spar. In
contrast the patch variable model only places foam in layers 2–14, and distributes
92 5.5. Numerical examples
(a) (b)
Figure 5.25: Optimal material directions for the full variable main spar model. Shown
at the tip for layer 1 (a) and layer 8 (b).
it in two distinct regions rather than continuously. This illustrates that choice of
material and orientation for each patch is a compromise between all elements in
the patch while in full variable model, the properties of each element can be set
individually. The orientation of the glass/epoxy material also corresponds well to
what was expected from the basic considerations made earlier regarding the load
carrying mechanisms of the spar. In the flanges all layers are dominated by 0◦
orientation, i.e. along the length of the spar, to account for bending while the
webs are dominated by 45◦ to account for shear. In the full variable model the
pattern of 45◦ elements resemble the distribution obtained for the cantilever beam
(see Fig. 5.12) in that there is a gradual occurrence of 45◦ elements, starting with
the elements in the center of the web.
One of the characteristics of the full variable model is that local effects are affecting
the results much more than for the patch variable model, which is not surprising
since the patch variables smear out the design. The phenomenon is most easily
seen by considering the tip of the main spar, Fig. 5.25. Here, the cross section
has obviously been reinforced by both introducing fibers in the circumferential
direction and foam material in the middle layers. Both of these measures are
taken in order to carry the two point loads applied to the cross section. For design
purposes such patterns may in many cases be considered as noise since point loads
are often used for modeling convenience and do not represent a physical occurrence.
Consequently, the patch variable solution is by far the easiest to realize, from a
manufacturers point of view since, it encompasses large and well defined regions
as opposed to the complex solution of the full variable model.
The results presented here are somewhat crude in that only five candidate materials
have been used, and the natural next step would be to expand the design space
and allow for a larger variation of fiber angles in order to obtain a more detailed
design. However, Figs. 5.26 to 5.33 still illustrate very well the potential of the
DMO method for designing a wind turbine blade main spar for maximum integral
stiffness. As such it has been demonstrated that DMO indeed has potential as a
practical design tool for industrial applications, which has been one of the primary
goals of developing the method.
Chapter 5. Discrete material optimization 93
Layer 1)
Layer 2)
Layer 3)
Layer 4)
Figure 5.26: Optimized material directions for the GFRP material in layers 1 to 4.
Patch variable model with 77 patches and 6 006 design variables.
94 5.5. Numerical examples
Layer 5)
Layer 6)
Layer 7)
Layer 8)
Figure 5.27: Optimized material directions for the GFRP material in layers 5 to 8.
Patch variable model with 77 patches and 6 006 design variables.
Chapter 5. Discrete material optimization 95
Layer 9)
Layer 10)
Layer 11)
Layer 12)
Figure 5.28: Optimized material directions for the GFRP material in layers 9 to 12.
Patch variable model with 77 patches and 6 006 design variables.
96 5.5. Numerical examples
Layer 13)
Layer 14)
Layer 15)
Layer 16)
Figure 5.29: Optimized material directions for the GFRP material in layers 13 to 16.
Patch variable model with 77 patches and 6 006 design variables.
Chapter 5. Discrete material optimization 97
Layer 1)
Layer 2)
Layer 3)
Layer 4)
Figure 5.30: Optimized material directions for the GFRP material in layers 1 to 4. Full
variable model with 748 800 design variables.
98 5.5. Numerical examples
Layer 5)
Layer 6)
Layer 7)
Layer 8)
Figure 5.31: Optimized material directions for the GFRP material in layers 5 to 8. Full
variable model with 748 800 design variables.
Chapter 5. Discrete material optimization 99
Layer 9)
Layer 10)
Layer 11)
Layer 12)
Figure 5.32: Optimized material directions for the GFRP material in layers 9 to 12.
Full variable model with 748 800 design variables.
100 5.5. Numerical examples
Layer 13)
Layer 14)
Layer 15)
Layer 16)
Figure 5.33: Optimized material directions for the GFRP material in layers 13 to 16.
Full variable model with 748 800 design variables.
Chapter 5. Discrete material optimization 101
The material selection is made at the element level and several options for the
weighted sum formulation have been suggested but the most successful is the
combined scheme in which DMO scheme 4 (Section 5.3.4) is used for stiffness
interpolation and scheme 5 (Section 5.3.5) is used for the physical constraint.
This combined scheme has been used in the four numerical examples presented
in Section 5.5. The first example was the cantilever beam, which served as
a validation of the DMO methods ability to solve fiber angle optimization
problems. The results were compared to known solutions in the literature and
good correlation was obtained. The second example was a four-point beam bending
problem which was solved for material layout and orientation and the result could
in part be validated with known solutions from topology optimization. Good
correlation was obtained and the result agreed well with the known physics of the
problem. The third example was a multilayered spherical shell, which was solved
for material layout and orientation both over the surface and in the thickness
direction. This combined example has not been studied before in the literature
but the results for material orientation can be compared to some extent to plane
solutions and the DMO solution showed similar patterns. The material layout
problem was solved and showed a cone-like reinforcement through the thickness,
which has been verified as a known technique for sandwich structures. The final
example was the wind turbine blade main spar, which represented a realistic
industry relevant problem. The problem was optimized for material layout and
orientation and the results correlated well with what was expected based on the
load carrying mechanisms of the structure. Furthermore, the use of patch design
variables illustrated that relatively simple optimal solutions can be obtained and
thus, the manufacturability of the optimal design obtained can be quite high.
None of the tested examples showed behavior similar to that of CFAO methods
where the optimization will distinctively get stuck in a local optimum. This
indicates that the method is more robust than existing methods and in general,
the results obtained with the discrete material optimization method were very
encouraging. Furthermore, the DMO method shows promising potential for
application to industry relevant problems.
102 5.6. Summary and conclusions
6
Conclusions
The performance of the MITC3, MITC4 and SHELL9 elements was demonstrated
as these constitute the working horses among the shell elements in MUST. The
three elements passed the patch tests and showed good predictive capabilities
as well as high computational efficiency compared to commercial finite element
packages. As such, the first objective of the project was reached.
104
Several numerical examples with plates and shells were presented and illustrated
that the importance of the nonlinear effects is problem dependent. For all examples
the change to the optimal topology was dramatic but in some cases the actual
performance gain was very small. However, in some examples the performance
increase was significant and furthermore, the use of multiple load cases provided
designs that showed good performance at multiple load levels. The results were
consistent with the progressive nature of nonlinear effect in that a gradual increase
in load lead to a gradual change in optimal topology, i.e. for very light loading
the linear and nonlinear results were identical while they became increasingly
different with higher loading. An interesting observation was that the topologies
obtained at high load with the nonlinear formulation were “global” in the sense
that they covered a large, continuous area over the geometry. This is interesting
in regard to manufacturing since such patterns are much easier to realize than the
local patterns usually associated with topology optimization. Including nonlinear
effects can therefore, in most cases, be considered an improvement due to the
potential performance increase and improved manufacturability of the optimal
designs obtained. This marks the successful arrival at the second goal of the
present work.
candidate materials, which can be both different materials and the same material
at different orientations. The element level parametrization is a weighted sum
formulation of the constitutive equation in which the design variables switch on
and off contributions from individual materials. An absolute necessary condition
for the method is that a single material is ultimately chosen and consequently, that
all elements/layers have a single weight of value one while all other weights are
zero. The success of the method is therefore dependent on the optimizers ability to
push the weights to 0 and 1 and to assist it, several weighting schemes have been
tested and implemented. The most successful has been found to be the combined
DMO scheme where scheme 4 is used for stiffness and scheme 5 for the physical
constraints.
In the tested examples local optima could not be distinctively identified when
using discrete material optimization as it is the case when running CFAO and in
general, the obtained results were very encouraging and the DMO method shows
promising potential for application to problems of industrial relevance. This brings
home the last goal of the present work.
Adali, S., A. Richter, V.E. Verijenko and E.B. Summers (1995), ‘Optimal design of
hybrid laminates with discrete ply angles for maximum buckling load and minimum
cost’, Composite Structures 32, 409–415. 4
Ahmad, S., B.M. Irons and O.C. Zienkiewicz (1970), ‘Analysis of thick shell and thin shell
structures by curved finite elements’, International Journal for Numerical Methods in
Engineering 4. 3
Alfano, G., F. Auricchio, L. Rosati and E. Sacco (2001), ‘Mitc finite elements
for laminated composite plates’, International Journal for Numerical Methods in
Engineering 50, 707–738. 1.2.1
Barut, A., E. Madenci, A. Tessler and J.H. Starnes Jr. (2000), ‘New stiffened shell element
for geometrically nonlinear analysis of composite laminates’, Computers and Structures
77(1), 11–40. 1.2.1
Bathe, K.J., O. Guillermin, J. Walczak and H.Y. Chen (1997), ‘Advances in nonlinear
finite element analysis of automobiles’, Computers and Structures 64, 881–891. 1.2.1
Belblidia, F., J.E.B. Lee, S. Rechak and E. Hinton (2001), ‘Topology optimization of
plate structures using a single- or three-layered artificial material model’, Advances in
Engineering Software 32(2), 159–168. 1.2.2, 4.1.1
Belytschko, T., W.K. Liu and B. Moran (2000), Nonlinear finite elements for continua
and structures, first edn, John Wiley and Sons. 2.1
Bendsøe, M.P. and O. Sigmund (2003), Topology optimization – theory, methods and
applications, second edn, Springer. 1.1, 1.2.2, 4.1, 4.1.2, 4.3, 5.3.1
Bletzinger, K.U., M. Bischoff and E. Ramm (2000), ‘A unified approach for shear-
locking-free triangular and rectangular shell finite elements’, Computers and Structures
75(3), 321–334. 1.2.1, 3.3.2
Bonet, J. and R.D. Wood (1997), Nonlinear continuum mechanics for finite element
analysis, first edn, Cambridge University Press. 2.1, 3.1
Brank, B. and E. Carrera (2000), ‘A family of shear-deformable shell finite elements for
composite structures’, Computers and Structures 76, 287–297. 1.2.1
Bruns, T.E. and D.A. Tortorelli (2001), ‘Topology optimization of non-linear elastic
structures and compliant mechanisms’, Computer Methods in Applied Mechanics and
Engineering 190, 3443–3459. 4.1.1
Bruns, T.E., O. Sigmund and D.A. Tortorelli (2002), ‘Numerical methods for the
topology optimization of structures that exhibit snap-through’, International Journal
for Numerical Methods in Engineering 55, 1215–1237. 1.2.2, 4.1.2
Bruyneel, M., P. Duysinx and C. Fleury (2002), ‘A family of mma approximations for
structural optimization’, Structural and Multidisciplinary Optimization 24, 263–276.
2.2.2
Bucalem, M.L. and K.J. Bathe (1993), ‘Higher-order mitc general shell elements’,
International Journal for Numerical Methods in Engineering 36, 3729–3754. 3.3.2
Buhl, T., C.B.W. Pedersen and O. Sigmund (2000), ‘Stiffness design of geometrically
nonlinear structures using topology optimization’, Structural and Multidisciplinary
Optimization 19, 93–104. 1.2.2, 4.1.1, 4.1.2
Carrera, E. (2003), ‘Historical review of zig-zag theories for multilayered plates and
shells’, Applied Mechanics Reviews 56(3), 287–308. 1.2.1, 3.2
Chapelle, D. and K.J. Bathe (2003), The finite element analysis of shells – fundamentals,
Springer. 3.1
Cho, M. and Y.J. Choi (2001), ‘A new postprocessing method for laminated composites
of general lamination configurations’, Composite Structures 54(4), 397–406. 3.2
Bibliography 109
Cook, R.D., D.S. Malkus and M.E. Plesha (1989), Concepts and applications of finite
element analysis, third edn, Wiley and Sons. 2.1, 3.1.3, 3.3.1
Diaz, A.R., R. Lipton and C.A. Soto (1995), ‘A new formulation of the problem of
optimum reinforcement of reissner-mindlin plates’, Computer Methods in Applied
Mechanics and Engineering 123, 121–139. 1.2.2
Dvorkin, E.N. and K.J. Bathe (1984), ‘A continuum mechanics based four-node shell
element for general nonlinear analysis’, Engineering Computations 1, 77–88. 1.2.1,
3.3.2, 3.5.1
Fleury, C. and V. Braibant (1986), ‘Structural optimization: a new dual method using
mixed variables’, International Journal for Numerical Methods in Engineering 23, 409–
428. 2.2.2
Foldager, J., J.S. Hansen and N. Olhoff (1998), ‘A general approach forcing convexity of
ply angle optimization in composite laminates’, Structural Optimization 16, 201–211.
1.2.3
Foldager, J., J.S. Hansen and N. Olhoff (2001), ‘Optimization of the buckling load for
composite structures taking thermal effects into account’, Structural and Multidisci-
plinary Optimization 21, 14–31. 3
Gea, H.C. and J. Luo (2001), ‘Topology optimization of structures with geometrical
nonlinearities’, Computers and Structures 79, 1977–1985. 1.2.2
Gibiansky, L.V. and O. Sigmund (2000), ‘Multiphase composites with extremal bulk
modulus’, Journal of the Mechanics and Physics of Solids 48, 461–498. 5.2, 5.3.2,
5.3.2
Haug, E.J., K.K. Choi and V. Komkov (1986), Design sensitivity analysis of structural
systems, Academic Press. 4.2
Heinbockel, J.H. (2001), Introduction to tensor calculus and continuum mechanics, first
edn, Trafford Publishing. 3.1, 3.1.3
Holland, J.H. (1975), Adaptation in natural and artificial systems, University of Michigan
Press. 4
Hughes, T.J.R. (2000), The finite element method, second edn, Dover Publications. 2.1,
3, 3.1
Hughes, T.J.R. and T.E. Tezduyar (1981), ‘Finite elements based upon mindlin plate
theory with particular reference to the four-node bilinear isoparametric element’,
Journal of Applied Mechanics 43(3), 587–596. 1.2.1
110 Bibliography
Jakobsen, L.A. (2002), A finite element approach to analysis and sensitivity analysis of
time dependent fluidstructure interaction systems, Ph.d. thesis, Institute of Mechanical
Engineering, Aalborg University, Denmark. 1.3.1
Jensen, L.R., J.M. Rauhe and J. Stegmann (2001), Finite elements for geomet-
ric non-linear analysis of composite laminates and sandwich structures, Mas-
ter’s thesis, Institute of Mehanical Engineering, Aalborg University, Denmark,
www.ime.aau.dk/ js/masters. 3, 3.1, 3.1.3, 3.4, 3.4.2
Jones, R.M. (1998), Mechanics of composite materials, second edn, Taylor and Francis.
3, 3.2
Kraus, H. (1967), Thin elastic shells, first edn, Wiley and Sons. 3
Krog, L.A. and N. Olhoff (1999), ‘Optimum topology and reinforcement design of disk
and plate structures with multiple stiffness and eigenfrequency objectives’, Computers
and Structures 72, 535–563. 1.2.2
Kumar, W.P. Prema and R. Palaninathan (1997), ‘Finite element analysis of laminated
shells with exact through-thickness integration’, Computers and Structures 63, 173–
184. 3.2.1
Lee, P.S. and K.J. Bathe (2004), ‘Development of mitc isotropic triangular shell finite
elements’, Computers and Structures 82(11-12), 945–962. 3.3.2, 3.5.1
Lee, S.J., J.E. Bae and E. Hinton (2000), ‘Shell topology optimization using the layered
artificial material model’, International Journal for Numerical Methods in Engineering
47, 843–867. 1.2.2, 4.1.1, 4.1.1, 4.3.1, 4.3.1
Luo, J.H. and H.C. Gea (1998), ‘Optimal bead orientation of 3d shell/plate structures’,
Finite Elements in Analysis and Design 31, 55–71. 1.2.3, 1
Mackerle, J. (2002), ‘Finite- and boundary-element linear and nonlinear analyses of shells
and shell-like structures – a bibliography (1999-2001)’, Finite Elements in Analysis and
Design 38, 795–782. 1.2.1
Mackerle, J. (2003), ‘Topology and shape optimization of structures using fem and bem
– a bibliography (1999-2001)’, Finite Elements in Analysis and Design 39, 243–253.
1.2.2
Bibliography 111
MacNeal, R.H. (1998), ‘Perspective on finite elements for shell analysis’, Finite Elements
in Analysis and Design 30, 175–186. 1.2.1
Masur, E.F. (1970), ‘Optimum stiffness and strength of elastic structures’, Journal of the
Engineering Mechanics Division, ASCE 96, 621–640. 5.1
Miki, M. and Y. Sugiyama (1993), ‘Optimum design of laminated composite plates using
lamination parameters’, AIAA Journal 31, 921–922. 1.2.3, 3
Moita, J. S., J. Infante Barbosa, C. M. Mota Soares and C. A. Mota Soares (2000),
‘Sensitivity analysis and optimal design of geometrically non-linear laminated plates
and shells’, Computers and Structures 79, 407–420. 1.2.3, 2
Noor, A.K. (1999), ‘Computational structures technology: leap frogging into the twenty-
first century’, Computers and Structures 73, 1–31. 1.2.1
Noor, A.K., W.S. Burton and C.W. Bert (1996), ‘Computational models for sandwich
panels and shells’, Applied Mechanics Reviews 49, 155–199. 1.2.1
Ochoa, O.O. and J.N. Reddy (1992), Finite Element Analysis of Composite Laminates,
first edn, Kluwer Academic Publishers. 1.2.1
Pai, P.F. (1995), ‘A new look at shear correction factors and warping functions of
anisotropic laminates’, International Journal of Solids and Structures 32(16), 2295–
2313. 3.1.3
Pedersen, P. (1991), ‘On thickness and orientational design with orthotropic materials’,
Structural and Multidisciplinary Optimization 3, 69–78. 1.2.3, 5.1, 1, 5.5.1, 5.5.3
Poulsen, J., T.L. Nielsen and P.D. Pedersen (2003), Development of tools for finite
element based analysis and optimization of vestas wind turbine blades, Master’s thesis,
Institute of Mechanical Engineering, Aalborg University, Denmark. 3.4
Razzaque, A. (1986), ‘The patch test of elements’, International Journal for Numerical
Methods in Engineering 22(1), 63–71. 3.5.1
Reddy, J.N. (2004), Mechanics of laminated composite plates and shells, second edn, CRC
Press. 1.2.1, 3, 3.1.3
Riche, R. Le and R.T. Haftka (1993), ‘Optimization of laminate stacking sequence for
buckling load maximization by generic algorithm’, AIAA Journal 31, 951–956. 4
Rozvany, G.I.N., M. Zhou and T. Birker (1992), ‘Generalized shape optimization without
homogenization’, Structural Optimization 4, 250–252. 1.2.2
Sigmund, O. (1997), ‘On the design of compliant mechanisms using topology optimiza-
tion’, Mechanics of Structures and Machines 25, 493–524. 5.2
Sigmund, O. and S. Torquato (1997), ‘Design of materials with extreme thermal expan-
sion using a three-phase topology optimization method’, Journal of the Mechanics and
Physics of Solids 45, 1037–1067. 1.2.2, 5.2, 5.3.2
Sokolnikoff, I.S. (1956), Tensor analysis – Theory and aplications, second edn, Wiley and
Sons. 3.1, 3.1.2
Soto, C.A. and A.R. Diaz (1993), ‘On the modelling of ribbed plates for shape
optimization’, Structural Optimization 6(3), 175–188. 4.1.1
Svanberg, K. (1987), ‘The method of moving asymptotes - a new method for structural
optimization’, Numerical Methods in Engineering 24, 359–373. 2.2.2, 5.1
Bibliography 113
Sze, K.Y., W.K. Chan and T.H.H. Pian (2002), ‘An eight-node hybrid-stress solid-shell
element for geometric non-linear analysis of elastic shells’, Computers and Structures
55(7), 853–878. 1.2.1
To, C.W.S. and M.L. Liu (2001), ‘Geometrically nonlinear analysis of layerwise
anisotropic shell structures by hybrid strain based lower order elements’, Finite
Elements in Analysis and Design 37, 1–34. 1.2.1
Tsai, S.W. and N.J. Pagano (1968), Invariant properties of composite materials, in S.
W. Tsai et al, ed., ‘Composite Materials Workshop’, Vol. 1 of Progress in Material
Science, Technomic Publishing Co., Stamfort, Connecticut, pp. 233–253. 3
Wang, M.Y. and X. Wang (2004), ‘”Color” level sets: a multi-phase method for
structural topology optimization with multiple materials’, Computer Methods in
Applied Mechanics and Engineering 193, 469–496. 1.2.2, 5.2
Yang, H.T.Y., S. Saigal, A. Masud and R.K. Kapania (2000), ‘A survey of recent shell
finite elements’, International Journal for Numerical Methods in Engineering 47, 101–
127. 1.2.1
Yuge, K., N. Iwai and N. Kikuchi (1999), ‘Optimization of 2-d structures subjected to
nonlinear deformations using the homogenization method’, Structural Optimization
17(4), 286–299. 1.2.2
Zhou, M. and G.I.N. Rozvany (1991), ‘The COC algorithm - Part II. Topological,
geometrical and generalized shape optimization’, Computer Methods in Applied
Mechanics and Engineering 89, 309–336. 1.2.2
Zienkiewicz, O.C. and R.L. Taylor (1991), The finite element method, Vol. 2, fourth edn,
McGraw-Hill. 2.1, 3.2.1, 3.4.2
Index
Patch test, 37
Reference surface, 22
Residual
definition, 13
linearization, 13
Thickness integration, 29
Topology optimization, 5, 41
multilayered, 42
multiphase, 69
Multiple load cases, 46
Parametrization, 42