Master MEETING Internship Defense, 2016

Maturation of Topology Optimization Methods: Numerical

strategy for Optimal Design of 3D Aircraft Components
Hamza Ibrahmi1 , Jérôme Chambert1 , Thomas Livebardon2

Abstract
Structural topology optimization obtains an optimal material distribution in a prescribed design domain for some boundary
conditions. This optimized design is found by minimizing the strain energy subject to some constraints modelling technical
specifications using Finite Element analysis. CT INGENIERIE provides its knowledge to develop generic methods for topological
large structures in two dimensions (2D) in regard with Additive Layer Manufacturing requirements. These methods still has
not become a mainstream technology for the small 3D components that belong to large and complex structure like the forward
fuselage frame of an aircraft. The purpose of this paper is to develop such a methodology to fill in this lack. The present work
considers the use of the compliance formulated topology optimization using SIMP (Solid Isotropic Material with Penalization)
approach and detailed topology optimization methods to the design of aircraft small 3D components. It also discusses and
solve the difficulties in obtaining correct loading and boundary conditions for finite element based analysis and optimization of
components that are integral parts of a larger structure. It will then go on to show some results such as the contribution of the 3D
design volume to the global stiffness of the structure and the distribution of loads at the interface between environment/design
volume.
Keywords
Structural topology optimization, aircraft structures, finite element method, design domain, boundary conditions, SIMP method,
3D components, global stiffness
1
FEMTO-ST Department of Applied Mechanics, Besançon, France
2
CT Ingenierie, Toulouse

Contents References 7

Introduction 1
Introduction
1 NUMERICAL METHODS FOR STRUCTURAL TOPOLOGY OP-
TIMIZATION 2
Topology optimization has become an effective tool for
least-weight and performance design, especially in aero-
1.1 Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
nautics and aerospace engineering. The lightening of
1.2 Numerical methods . . . . . . . . . . . . . . . . . . . . . . . 2 aeronautical structures is therefore a major issue. Fur-
2 Construction and validation of the finite element model for thermore, shortened development time-scales in the civil
topology optimization 3 aircraft industry call for an incorporate of higher com-
puter aided optimization methods into the overall compo-
3 Setting up the optimization problem 4
nent design process. Structural optimization has previ-
4 Results and discussion 5 ously been employed to improve aircraft performance. To
4.1 The construction of the 3D design space to the global stiff- meet this need, CT INGENIERIE provides its knowledge
ness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . to develop generic methods and conception rules for topol-
5 ogy structures in regard with ALM requirements. It has
formed a partnership with Centrale Nantes, Constellium
4.2 Distribution of loads at the interfaces between environ-
and Stelia Aerospace to engage in a research and develop-
nement and the design space . . . . . . . . . . . . . . . . . .
ment project called DEFACTO1 program.
6
4.3 Penalization factor . . . . . . . . . . . . . . . . . . . . . . . . 6 However, in the last decade, progresses in advanced
5 Conclusions 6 1
DEveloppement de la Fabrication Additive pour Composants
Acknowledgments 7 TOpologiques.
 Maturation of Topology Optimization Methods: Numerical strategy for Optimal Design of 3D Aircraft Components — 2/7

manufacturing technologies such as Additive Layer Manu- It may be interesting to give a brief literature review
facturing (ALM, also known as 3D printing) freed up engi- of structural topology optimization problems and to make
neers to use topology optimization as efficient approach to a number of preliminary choices [5]:
design low-weight structures. The DEFACTO project (Fig-
ure 1) aims to develop a numerical methodology based on 1. Model: What is the most relevant model in terms of
a 2D structures optimization chain (topology, shape and adequacy, cost, etc.?
parametric) which intend to develop the Airbus A320NEO 2. Design variables: Which parameters are interesting
forward fuselage components with breakthrough design to vary?
philosophy allowed by ALM possibilities. 3. Space of research: What is the limit within we
should vary those parameters?
4. Objective function: What are the goals to achieve?
5. Optimization method: Which method to choose?

To formulate the structural optimization problem, an ob-

jective function, design variables and constraints need to
be defined as follows :

1. Objective function: represents an objective that

could either be minimized or maximized. A typi-
cal objective could be the stiffness or volume of a
structure. The objective function is subjected to de-
sign variables and to associate constraints to steer
the optimization to a sought solution.
2. Design variables for topology optimization: Vari-
ables related to the FEM model and modified by
the optimization algorithm. In mechanics, these
variables can represent the dimensions, the shape,
the material density of each element, the direction
of the fibers in a composite fold, etc.
3. • Associated constraints: Allows to define lower or
Figure 1. DEFACTO methodology. upper bound constraints of the optimization study
of responses previously defined. It is corresponding
to all the restrictions imposed by the optimization
Whose topology optimization highlights the main paths
problem that may depend on the design variables
of effort that allow the structure to have sufficient rigid-
(Volume, displacement, etc .).
ity for some boundary conditions and mechanical loads.
An automatic stiffener detection tool then makes it pos-
1.2 Numerical methods
sible to precisely identify the corresponding lattice. This
The advances in computer science have made it possi-
methodology outcomes optimized components which are
ble to improve the performance of numerical simulation
directly substitutable. It allows promising performance
methods but also to implement new calculation processes.
improvement in terms of weight reduction and mechanics.
Topology optimization of structures methods are among
Unfortunately, this process is mainly dedicated to large
these new processes. It has evolved over the past two
aircraft structures for which two-dimensional optimiza-
decades and it is nowadays the most active and broadly
tions are efficient. Consequently, this methodology is not
studied branch in structural optimization [8, 7].
directly usable for massive components in which reducing
There are many different numerical methods for the
dimension is not feasible and suitable especially for small
optimal design of structure such as the homogenization
components.
method, level set method, topological gradient method,
etc. This paper will focus on the homogenization method
1. NUMERICAL METHODS FOR STRUCTURAL named SIMP (Solid Isotropic Material with Penalization)
TOPOLOGY OPTIMIZATION method for three reasons: Firstly, it is efficient and simple
to implement [3]. Secondly, this method is available in the
1.1 Generalities Altair’s structural solver OptiStruct [1] and well-known
Topology optimization is an effective design method for a in CT INGÉNIERIE as well as in the scientific community
variety of engineering applications [10]. This is based on [2, 4, 9].
a mathematical formulation using finite element method This method has been introduced in 1989 by A. Allaire
(FEM) to find optimal material distribution in a domain [2], which, the material density of each element should
for satisfying a set of constraints. take a value of either 0 or 1, defining the element as being
 Maturation of Topology Optimization Methods: Numerical strategy for Optimal Design of 3D Aircraft Components — 3/7

either void or solid, respectively. These represent the 6. if convergence criteria are not satisfied repeat the
state of void and solid, respectively. Intermediate values loop else plot results
of density represent fictitious material.
The optimal solution of problems involves large gray
areas of intermediate densities in the structural domain.
Therefore, techniques need to be introduced to penalize
intermediate densities and to force the final design to be
represented by densities of 0 or 1 for each element. The
penalization technique used is the power law represen-
tation of elasticity properties which can be expressed for
any solid 3-D or 2-D element as follows:

e = ρpK
K (1)

where K e and K represent the penalized and the real stiff-

ness matrix of an element, respectively, ρ is the density
and p the penalization factor which is always greater
than 1.
The interpolation curves with different values of penalty
factor are shown in Figure 2. As shown, with increase
in the value of penalty factor p, the interpolation curve
becomes much steeper near the point (1,1), whereas the in-
terpolation curve becomes much smoother near the point
(0,0). Consequently, the intermediate density is quickly
penalized.

Figure 3. Topology optimization iterative local search

algorithm.

2. Construction and validation of the finite

element model for topology optimization
This part of study deals with construction of the FEM
model (the Airbus A320NEO passenger door frame) repre-
sents the detailed environment (DFEM) which is located
at the front end of the aircraft. In this structure, frames
FR20 and FR21 have 7 abutments on which are attached
7 intercostal. One of those intercostal will modeling in a
Figure 2. The penalty factor influence.
3D part represents the design space on which the topology
optimization will applied. The structural elements are
The steps of a topology optimization algorithm are
detailed in Figure 4.
described below and further illustrated in Figure 3 :

1. Initialization
(a) define design domain
(b) apply loading and boundary condition
(c) distribute randomly material inside the do-
main
2. compute using finite elements method (FEM) the Figure 4. The A320NEO passenger door frame.
structure
3. run sensitivity analysis A first step is to isolate both frames FR20 and FR21,
4. run low pass filtering (eliminate checkerboards ef- as well as the skin of the frame, and create the 3D vol-
fects) ume representing the numerical domain of the targeted
5. update design variables intercostal for the optimization.
 Maturation of Topology Optimization Methods: Numerical strategy for Optimal Design of 3D Aircraft Components — 4/7

The second step is the conversion from the 3D CAD Applying constraints is the next step of the study.
model to a thin-walled 2D shell model by extracting the Static analyses of the main constraints were previously
outer skin of the two frames and the skin of the passage performed on the initial GFEM to determine which load
door. This method allows keeping the shape of the struc- cases were relevant for passage door frame. This allows
ture and aims to get a simpler geometry to ease finite classifying every load case into 13 main families well-
element calculations. Indeed, this step leads to a signif- representing the complete set of loads. Unfortunately, as
icant decrease in the total number of DOF2 of the FE these loads are applied on the boundaries of the aircraft
model. (ie the GFEM), it is mandatory to propagate these loads
The three structures (skin, frames and 3d volume) are up to the boundaries of the DFEM. Hence, a static conden-
then meshed according to the geometric complexity, the sation is performed on the GFEM to take into account the
various types of elements used, and the mesh size: solid stiffness of the whole structure as well as put accurate
elements (pentagon, tetrahedron 3 DOF), shell elements loads on DFEM boundary nodes.
(triangle, square: 6 DOF), and Beam3 or Rod4 elements. Similarly, an equivalence (finds duplicate nodes of
The total number of generated elements for the model is selected components or elements within a tolerance) has
645860. been achieved between the GFEM/DFEM interface nodes
The next step is to create a database of mechanical to ensure the continuity of the degrees of freedom between
and physical properties (Young’s modulus, densities . . . ) the frame and the rest of the structure for which the
into a database, and assign them to the different elements meshes are not conformal. The final finite model of the
that built up the model. As well as the properties that A320NEO passenger door frame is shown in Figure 5.
includes the element types (3D Solid, 2D SHELL, 1D
ROD/BEAM) and the different dimensions (thicknesses,
depths of the pockets, etc.). Finally, 32 properties for both
frames, 2097 properties for the skin and one property for
the 3d volume, are created.
Once the model parts have been meshed, they must be
assembled in accordance with reality. There are alterna-
tive methods to simplify the assembly point by combining
two types of elements: springs (CBUSH) and the rigid Figure 5. Finite Element Model of the A320NEO
links (RBE3). However, special care must be taken in passenger door frame.
the modeling of the assembly point to well-represent its
behavior. The equivalent spring element used allows mod-
eling the fixation by a flexible body and contains 4 nodes
at its ends with 6 DOF per node (3 Translations, 3 Rota- 3. Setting up the optimization problem
tions). Indeed, tensile / compression and shear stiffness The compliance has been introduced as an objective re-
properties calculated using Huth formulas [6] can be as- sponse in this study. A possibility to maximize the global
sociated with them. stiffness of a structure is to minimize its compliance that
defined as the strain energy (positive energy) of the FE
E c .S
K1 = (2) solution which yields higher stiffness when minimized .
L
The compliance is defined as:

C E ρ = F tU
¡ ¡ ¢¢
(4)
K2 = K3 =
1 with, C compliance, E Young’s modulus ( N / mm2 ), ρ is the
¡ e 1 + e 2 ¢ b2 ³ ´ (3) density, F force ( N ), and U displacement ( mm),where U
1
2.D m e 1 .E 1 + m.e12 .E 2 + 1
e 1 .E c + m.e12 .E c
solves the equilibrium equation

with, e 1 thickness of plate 1, e 2 thickness of plate 2, E 1

K E ρ U =F
¡ ¡ ¢¢
(5)
Young modulus of plate 1, E 2 Young modulus of plate 2, E c
Young modulus of connector, b 1 et b 2 coefficient depending with, K is the stiffness matrix, so
on the connector type, m = 1 for single shear and 2 for
double shear and D the diameter of the connector. C E ρ = U t KU.
¡ ¡ ¢¢
(6)
2
Degree Of Freedom
3
This optimization study uses volume fraction and com-
Bar elements are 1D elements with 2 nodes used to model axial,
bending, and torsion behavior.
pliance responses. Volume fraction refers to the percent
4
Rod elements are 1D elements with 2 nodes used to model axial of initial design space to be maintained in the final solu-
behavior only. tion. An upper bound constraint of 30% for the designable
 Maturation of Topology Optimization Methods: Numerical strategy for Optimal Design of 3D Aircraft Components — 5/7

volume can be chosen. Finally, the optimization problem

may be stated as:

min ρ C (ρ ) = U tK E ρ U
 ¡ ¡ ¢¢


V (ρ )

 

= f = 0. 3

 

V0 (7)


avec
K E ρ U =
 ¡ ¡ ¢¢
F



 
 
ρ∈ [0; 1]
 

where f volume fraction of the optimal topology, V (ρ )

total volume at current iteration based on density and V0
the initial design volume.

4. Results and discussion

4.1 The construction of the 3D design space to the global
stiffness
In the case of subcomponents as part of a global large
structure, the contribution to the total compliance of sub-
components can be very small. Figure 6 shows that there
is a big difference in order of magnitude between global
(106 ) and local (102 ) compliance.

Figure 7. Example of the topology optimization of the

Airbus A320 cockpit floor.

is relatively close. However, it did not give concrete and

satisfactory results on this project because of the huge
difference. This leads us to introduce a new approach to
be tried which allows “to bring together” the variations of
two compliances mentioned above. As a second step, the
compliance product is presented on a logarithmic scale:
this is helps to improve the visualizing the range of varia-
tion of the product of the two compliances. This method is
adapted to realize the orders of magnitude in this applica-
tion. Although this method demonstrates well the small
results variations, it belittles the huge results variations
Figure 6. Comparison global/local compliance (order of
between the iterations. Moreover, this method is very
magnitude).
expensive to evaluate as it is very nonlinear as illustrated
in Figure 8.
Studies have been carried out have studies have shown
the importance of considering the contribution of the sub-
component to the overall stiffness of a structure. Other-
wise, material distribution problems within the design
domain volume may appear as shown in Figure 7..
It is therefore necessary to find a solution to increase
the sensitivity of optimization algorithm to local modi-
fications in the design domain. Several methods have
been suggested and presented: firstly an objective offset
is applied. The compliance considered here as objective is
the difference between the overall compliance of the struc-
ture and a constant value C 0 representing the compliance
of the whole structure without the design volume. This Figure 8. Convergence curve of the objective compliance
method has been well tested and verified on specific cases function with a function LOG (Compliance / Iterations).
when the difference between global and local compliance
 Maturation of Topology Optimization Methods: Numerical strategy for Optimal Design of 3D Aircraft Components — 6/7

4.2 Distribution of loads at the interfaces between environ-

nement and the design space
Before starting explain the present work done on this part,
it is interesting to show some results previously obtained.
Figure 9 shows the topological optimization results of the
3D design domain from the previous assembly modeling.

Figure 10. Discontinuities in the material distribution.

Using SIMP method, a decrease of 53% in the compli-

ance value is noticed, which means a maximization of the
structure’s stiffness, while keeping the same final mass
is equal to 30% of the initial mass (fractional volume =
Figure 9. Topology optimization results of previous constant = 0.3).
assembly models (density distribution). To conclude, it is fundamental to say that on parts’
optimization in 3D as part of a global structure and joined
Initially, modeling of bolt type have been used to take together by connections of punctual type (bolts, rivets
into account the junctions between the optimization vol- etc..), it is necessary to divide up the efforts by transition
ume and its environment as shown in Figure 10. Ar- zone. Thus, it helps to put a plate to the interface in order
rows in yellow represent the load inputs (punctual con- to facilitate the model’s construction.
nections between the optimization volume and the rest
of the model). The area adjacent to the connections is a
critical zone in calculation and highly submitted to stress
5. Conclusions
concentrations. So it’s creates punctual load inputs and The present work illustrates how the developed topology
forces topology optimization to place extremely localized optimization method has been successfully used to sug-
clusters of material at the interfaces as illustrated in Fig- gest good initial design for aircraft small 3D particular
ure 10. Which later, makes the junction’s implementation components that are integral parts of a larger structure.
impossible (missing support), and leads to discontinuous Returning to the hypothesis posed at the beginning of
material distribution. this study, it is now possible to state the important in-
This result illustrates the important influence of the fluence of the boundary conditions on the density field.
boundary conditions on the density field. It seems, thus, But also, the importance of considering the contribution
essential to allow a homogeneous distribution of loads of the sub-component to the overall stiffness of a struc-
entering the optimized part. The envisaged solution aims ture. The distribution of loads at the interface between
at including a plate between the design domain and the environment/design volume is also interesting to rectify.
environment as illustrated in Figure 11. Taken together, these results suggest that is perhaps, a
minimum member size constraint can be used to obtain a
4.3 Penalization factor well-defined design.
With Altair HyperMesh software, it is possible to alter the
parameter at the density power to increase the penalty.
Acknowledgments
For this simulation, a topology optimization calculation
was realized by choosing a value equal to p = 3. The I would like to express my deep gratitude to Thomas Live-
topology obtained is shown below in Figure 12. bardon and Jerome Chambert, my research supervisors
 Maturation of Topology Optimization Methods: Numerical strategy for Optimal Design of 3D Aircraft Components — 7/7

a homogenization method. Computer Methods in

Applied Mechanics and Engineering, 71(2):197–224,
November 1988.
[5]
N BOUHADDI. CONCEPTION ROBUSTE : COURS
D’OPTIMISATION.
[6]
Eric Paroissien. Contribution aux assemblages hy-
brides (boulonnés/collés) - Application aux jonctions
aéronautiques. page 288, 2006.
[7]
J. París, S. Martínez, F. Navarrina, I. Colominas, and
M. Casteleiro. Topology optimization of structures
with stress constraints: Aeronautical applications.
IOP Conference Series: Materials Science and Engi-
neering, 10(1):012206, 2010.
[8]
José París, S Martínez, F Navarrina, Ignasi Colom-
inas, and M Casteleiro. Structural optimization of
high voltage transmission line towers considering
continuum and discrete design variables. In WIT
Transactions on the Built Environment, volume 125,
Figure 11. Topology optimized design domain ( pages 59–69, June 2012.
Intercostal). [9]
G. I. N. Rozvany, M. Zhou, and T. Birker. Gener-
alized shape optimization without homogenization.
Structural optimization, 4(3-4):250–252, September
1992.
[10]
Benliang Zhu, Xianmin Zhang, and Sergej Fatikow.
Structural topology and shape optimization using a
level set method with distance-suppression scheme.
Computer Methods in Applied Mechanics and Engi-
Figure 12. Impact of penalization factor on the density neering, 283:1214–1239, January 2015.
field inside the design domain (Intercostal).

for their patient guidance, enthusiastic encouragement

and useful advices of this research work. I would also
like to extend my thanks to Matthieu Merle R&D and
Innovation Business Unit Manager and the engineers of
R&D department of the CT INGENIERIE group for their
goodwill and help.
So long and thanks for anything.

References
[1]
Altair Connect - Base de connaissances.
[2]
Grégoire Allaire, Eric Bonnetier, Gilles Francfort,
and François Jouve. Shape optimization by the
homogenization method. Numerische Mathematik,
76:27–68, March 1997.
[3]
S. R. M. Almeida, G. H. Paulino, and E. C. N. Silva.
A simple and effective inverse projection scheme
for void distribution control in topology optimiza-
tion. Structural and Multidisciplinary Optimization,
39(4):359–371, October 2009.
[4]
Martin Philip Bendsøe and Noboru Kikuchi. Gener-
ating optimal topologies in structural design using