See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/312472836

The relevance of reliability-based topology optimization in preliminary phases

of aerospace structural design

Conference Paper · January 2017

DOI: 10.2514/6.2017-1937

CITATIONS READS

0 151

3 authors:

Carlos López Aitor Baldomir

Flanders Make University of A Coruña
27 PUBLICATIONS 70 CITATIONS 73 PUBLICATIONS 324 CITATIONS

SEE PROFILE SEE PROFILE

Santiago Hernandez
University of A Coruña
252 PUBLICATIONS 1,018 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Reliability-based design and topology optimization of aerospace components and structures View project

Optimum crossing cable system in multi-span cable-stayed bridges View project

All content following this page was uploaded by Carlos López on 04 September 2017.

The user has requested enhancement of the downloaded file.

 The relevance of reliability-based topology optimization
in preliminary phases of aerospace structural design.

Carlos López, ∗ Aitor Baldomir † and Santiago Herńandez. ‡

Structural Mechanics Group, School of Civil Engineering, Universidade da Coruña, A Coruña, Spain

This paper aims to compare the structural schemes derived from the Deterministic
Topology Optimization (DTO) and Reliability-Based Topology Optimization (RBTO) re-
sults in order to prove which structure performs better in a subsequent Reliability-Based
Design Optimization (RBDO) sizing phase during the design process of an aircraft com-
ponent or structure. The RBTO is performed through a decoupled RBDO algorithm in
order to take profit of the capabilities of an external optimization software when deal-
ing with the topology optimization problem. The results show that the designs based on
RBTO with a target reliability index β T perform better than the designs based on DTO
when both structures are compared in a reliability-based size optimization imposing as
reliability target β T .

I. Introduction
Generally speaking, topology optimization tries to find the optimal structural scheme within a specified
design region for a set of objectives, constraints, loads and boundary conditions. During the past 30 years,
it has become the most active topic in the research framework of structural optimization starting with the
landmark study of Bendsoe and Kikuchi.1 Afterwards several researchers have made important contributions
on the way to adopt this promising discipline in all major industries (automotive,2, 3 civil,4 aerospace,5 etc.).
Deep discussion about topology optimization and methods can be found on the surveys by Rozvany,6 Sigmund
and Maute7 or Deaton,8 while the study of Zhu9 rewiews recent advances of topology optimization techniques
applied to aerospace structures and proposes potential future applications such as dynamic responses design,
shape preserving design, smart structures design, structural features design and additive manufacturing.
In the last decade, topology optimization has emerged as a highly valuable tool for aerospace companies in
pursuit of their goal to reduce the weight of aircraft components and structures.10 Thanks to this discipline,
the Airbus A380 reached a weight reduction of 700 kg per aircraft through the redesign of the fuselage
door intercostals and the leading edge ribs.11 Moreover the recent Airbus A350 benefited from topology
optimization to redesign structural components like the pylon.12 Other major aerospace companies also
employ extensively this discipline. Boeing achieved savings of 25-45% in the wing leading-edge structure,13
while Bombardier saved a 10% in the mass of Rib 11.14
The traditional way of designing a structure through topology optimization consists of performing a De-
terministic Topology Optimization (DTO) and extract from the results a structural configuration that is later
interpreted, defined geometrically and submitted to a size optimization process.14, 15 Industry processes such
as aircraft design usually involve dealing with uncertainties in loads, material properties or manufacturing.
During this design process they are usually included in the latter phase (size optimization), traditionally
through the imposition of partial safety factors or more recently through Reliability-Based Design Opti-
mization (RBDO), which considers the uncertain data as random variables. However, other valid approach
would be to consider such uncertainties from the preliminary design stages combining topology optimization
∗ Ph.D. Candidate, Structural Mechanics Group, School of Civil Engineering, Campus de Elviña s/n, 15071 A Coruña,

carlos.lopez.rodriguez@udc.es
† Associate Professor, Structural Mechanics Group, School of Civil Engineering, Campus de Elviña s/n, 15071 A Coruña,

abaldomir@udc.es, AIAA Member

‡ Professor, Structural Mechanics Group, School of Civil Engineering, Campus de Elviña s/n, 15071 A Coruña, hernan-

dez@udc.es, AIAA Fellow

1 of 19

American Institute of Aeronautics and Astronautics

with statistical and probabilistic design methods, namely using Reliability-Based Topology Optimization
(RBTO16 ). This discipline has seen an increasing research activity since mid-2000 (Maute and Frangopol,17
Guest and Igusa,18 Silva et al.,19 Dunning et al.20 or Tootkaboni21 ). A RBTO approach may be interest-
ing since the structural configuration obtained through topology optimization can depend on the reliability
treshold imposed by the engineers conditioning the next phases of the aircraft design process.
The topology of preliminary baseline models is critical for the upcoming design phases. As stated in López
et al.,22 the results obtained from the DTO and the RBTO may lead to different structural schemes, which
makes difficult to decide which configuration will adapt better to the design conditions of the subsequent
design phase. In this study two structures, emerged from the DTO and RBTO respectively, will undergo
a reliability-based size optimization phase in order to consider the propagation of uncertainties in the next
design phase. Comparing both approaches may help to determine if it is worth to take into account such
uncertainties in early design stages (topology optimization) or by contrast if it is better to include them just
in the sizing phase.
In this research a decoupled RBDO method has been programmed in an in-house computational code,
which is combined with an external optimization solver to assess the RBTO. The RBDO method selected
is the Sequential Optimization and Reliability Assessment (SORA23 ) based on its robustness and easy
combination with external software.24 The target is to obtain from the RBTO a novel structural scheme with
at least the same global stiffness that the layout provided by the DTO approach.22 Afterwards two different
structures, emerged from the DTO and RBTO results, are defined and interpreted as real bar structures
through an engineering process. Finally, both structures undergo a reliability-based size optimization process
considering the same level of reliability than in the RBTO case in order to include randomness in those
uncertain parameters. The results obtained will help to decide if it is worth to consider uncertainties in early
design stages (topology optimization) or by contrast if it is better to include them only in the sizing phase,
establishing which approach provides a better solution.

II. Comparison of DTO and RBTO based designs in upcoming stages of

aircraft design
The general DTO problem that will be carried out in this research is shown in Equation 1:
min V (1a)
subject to:
|σ| ≤ σmax (1b)
where V is the total volume of the structural model, σ refers to the Von Mises stress obtained from the
DTO and σmax is the maximum admissible stress considered. The DTO is solved through Altair Optistruct25
because it offers a high efficiency and versatility in solving large topology optimization problems and it is
widely used in aerospace industry.15 The structural scheme derived from the optimization will have a global
stiffness that is evaluated in this software through the weighted compliance.
The strategy followed in this research is explained below. First of all, the DTO is performed as expressed
in equation 1 and the weighted compliance of the optimal design W C DT O is stored in order to set the
minimum level of stiffness that the RBTO design must accomplish. The next step is to consider uncertainty
in the loads and material properties treating them as random variables so as to compute the probabilistic
constraint, which is defined through the limit state function G of the global stiffness of the structure (G =
1−W C/W C DT O ). By convenction, G > 0 denotes a safe region whereas G < 0 is a failure region. Therefore,
the RBDO problem is constructed as stated below:
min V (2a)
subject to:
|σ| ≤ σmax (2b)
P [G < 0] ≤ Pf (2c)
where P [·] is the probability operator and Pf is the imposed probability of failure.
The RBTO problem is solved through a decoupled RBDO algorithm that splits the problem into a
sequence of two steps that are repeated cyclically until convergence:

2 of 19

American Institute of Aeronautics and Astronautics

 • Deterministic Optimizations (DO) phase. In this phase the values of the random variables in the
first optimization cycle are the means of their probability distributions. In the subsequent optimization
cycles, the values of the random variables are actualized to the Most Probable failure Points (MPP)
obtained in the previous Reliability Analysis (RA) step.
• Reliability Analysis (RA) phase. It aims to obtain the probability of failure of the structural
system against a determined limit-state by obtaining the Most Probable failure Point (MPP) of the
structure. The MPP is defined as the most probable values of the random variables when the limit-
state function G = 0 is reached and determines the state of the structure where it is more likely to
fail.
In this way the capabilities of Altair Optistruct solving DTO problems can be fully exploited. The RBDO
algorithm chosen in regards of its robustness and accuracy is the Sequential Optimization and Reliability
Assessment (SORA23 ), which has been programmed in a MATLAB code.26 The RA phase is solved through
the Hybrid Mean Value (HMV27 ) which is an efficient inverse MPP search method. The structural responses
required by the HMV algorithm are obtained through structural analyses of the FE model that are performed
in Altair Optistruct. Further details of this strategy can be found in López et. al.22
Figure 1 shows a flowchart of the process described above. The right part of the chart shows the RBDO
process implemented in the in-house MATLAB code, while in the left part of the chart it is presented the
flowchart of the HMV method implemented in the MATLAB code as well.
Initial Design
Initial Point
uk+1 = M P Pi d0
(u0i = 0) i
xk+1
i = µxi

Deterministic
Optimization
Calculate ∇Gi (d, uki ) (OPTISTRUCT)
Convergence?
(OPTISTRUCT)

d∗

Obtain unitary vector k=1,2,3 Calculate Update FEM with

Reliability Analysis
G (ui,k ) new point u: d∗
αi,k = − kGii (ui,k (HMV-MATLAB)
)k ui,k+1 =β T · αi,k xk+1
i = µxi + σxi · uik+1

k>3
uk+1
i = M P Pi
Calculate ξ>0
ξ = (αi,k −αi,k−1 )·(αi,k−1 −αi,k−2 )

ξ<0
Convergence
no
Calculate new point u: in Objective
α +αi,k−1 +αi,k−2
ui,k+1 =β T · kαi,k
i,k +αi,k−1 αi,k−2 k
Function?

yes

Optimum design

Figure 1. Flowchart of the proposed methodology combining MATLAB and external optimization software.

The target of this research is to assess a comparison of real structural systems arisen from previous DTO
and RBTO processes. In the scientific community it is usual to obtain a structural scheme from the results
provided by the DTO, since it gives the best material distribution for a given set of loads and boundary
conditions. However, DTO does not take into account uncertainty in the design parameters, which will
necessarily have to be considered in subsequent design phases. Therefore it is fair to question if taking into
account those uncertainties in the topology optimization phase might have benefitial effects on the cost of
the structure.
In this research the strategy followed includes two phases. The preliminary stage consists of performing
DTO and RBTO over a specified design region and then interpret and define two structures on the basis of the

3 of 19

American Institute of Aeronautics and Astronautics

DTO and RBTO results, providing structural elements with real dimensions. Afterwards the performance
of both structures will be compared in a following size optimization stage. When DTO is used the uncertain
data is included only in the second size optimization stage leading to a RBDO problem, whereas if RBTO
is considered uncertainty is included in both phases.
Thereby, the results of the DTO and RBTO go through a filtering process carried out in the Hypermesh
module “OSSmooth”,25 which allows to display only the elements with a density ρ above a specified value.
From these results Hypermesh can generate a set of surfaces and lines in order to define truss structural
members that help to build the structure. This engineering interpretation and definition of the structure
is denoted as conceptual design definition. Later on, both structures (denoted as DES-1 for the DTO and
DES-2 for the RBTO approaches, respectively) will undergo a reliability-based size optimization process
considering uncertainty in the loads and material properties for comparing their performance in terms of
weight or volume. Figure 2 shows a flowchart of the process described above.

Initial design Initial design

DTO RBTO β T

DTO results RBTO β T results

Configuration 1 Configuration-2

Interpretation Interpretation

DTO based RBTO based

structure DES-1 structure DES-2

RBDO size β T RBDO size β T

Optimum design Optimum design

DES-1∗ DES-2∗

DES-1∗ vs. DES-2∗

Figure 2. Flowchart of the strategy proposed for the conceptual design definition task.

III. Application examples

To demonstrate the benefits of the strategy proposed in section II two different examples are proposed:
a two-dimensional rectangular domain and a three-dimensional aircraft tail fuselage.

A. Two-dimensional rectangular domain

This subsection presents a two-dimensional rectangular domain of 10 m length and 3 m height that is shown
in figure 3. The external loads are applied at a distance of 2.5 m from the right top node (point A in figure 3)
and are defined as a vertical load of P =1000 kN and an horizontal load of Q =100 kN. The boundary
conditions are placed in the left bottom node (point B in figure 3) which is fixed, and in the right bottom
node (point C in figure 3) which is constrained so that it can only move horizontally.
The 2D domain is discretized by 3000 shell elements, the material is steel with a Young’s modulus of
210 000 MPa and three load cases are considered, being the first one defined by the vertical load P , the
second one by the horizontal load Q and the third by a combination of both loads.
First the Deterministic Topology Optimization (DTO) is performed using Altair OptiStruct. The DTO
problem consists of minimizing the total volume of the structure subject to stress constraints, as exposed in
equation 1. In this case the maximum Von Mises stress is set to σmax =355 MPa. The volume obtained is
V DT O = 7.37 · 107 mm3 which is a 16% of the initial volume global, while the compliance of the structural
layout is W C DT O = 1.93 · 107 mm/N.

4 of 19

American Institute of Aeronautics and Astronautics

 Figure 3. Rectangular domain.

Then the RBTO problem is performed using the strategy exposed in section II (equation 2) being G =
1− W W C
C DT O
. In this case, the probability of failure is set to Pf = 2.86·10−7 , which corresponds to a reliability
index target β T = 5 applying the relation Pf = Φ(−β). Uncertainty is considered in both load magnitudes
and Young’s modulus of the material, becoming the random variables of the problem and following a Normal
distribution whose mean and standard deviation values are presented in Table 1.

Table 1. Statistical moments of the random variables.

Random Variable Distribution µ σ δ

Vertical Load [N] Normal 1.00 · 106 1.00 · 105 0.1

Horizontal Load [N] Normal 1.00 · 105 1.00 · 104 0.1
Young’s module [MPa] Normal 2.07 · 105 1.035 · 104 0.05

After three iterations of the SORA method, the volume obtained is V DT O = 1.53 · 108 mm3 which is a
34% of the initial volume global. Figure 4 shows the results of the DTO and RBTO for β T = 5 overlapped
with the conceptual designs interpreted from both layouts. For the DTO the engineered design is defined as
a statically determined and symmetric 7-bar truss-structure (DES-1) whereas for the RBTO it is a statically
determined 13-bar truss-structure (DES-2).

(a) DTO results and 7-bar conceptual design (b) RBTO β T = 5 results and 13-bar conceptual design

Figure 4. Topology results and conceptual designs based on the DTO and RBTO β T = 5.

As expected, when uncertainty is considered within the optimization process, new members emerge from
the topology solution and thus it can be concluded that the RBTO results may lead to different structural
schemes than DTO results. After both topology solutions undergo the concept design definition process,
a comparison between both designs is carried out in a subsequent size optimization phase that considers
uncertainty in the loads and material properties (RBDO). Figure 5 presents the 7-bar truss and 13-bar truss
structures obtained from the DTO and RBTO cases. The coordinates of the nodes that define this 13-bar
truss are shown in table 2.
In order to assess the comparison between both designs, a RBDO problem is defined as exposed in
equation 2. In this case the design variables are the transversal area of the bars and the random variables

5 of 19

American Institute of Aeronautics and Astronautics

 Table 2. Node coordinates of the 13-bar structure.

NODES 1 2 3 4 5 6 7 8

Horizontal coordinate [m] 0 10 2.5 7.5 5 2.5 1.05 3.3

Vertical coordinate [m] 0 0 3 3 0 0 1.7 1.9

P = 1000 kN

2.5 m
3 4 4

A Q = 100 kN
12 7 9

y
7
3m 8 6 5

11 13 10 8
1 2
1 x 6 2 3
5

5m 5m

(a) 7-bar (DES-1) obtained from DTO (b) 13-bar (DES-2) obtained from RBTO β T =5

Figure 5. Initial designs for size optimization. Obtained by engineering the solution of DTO and RBTO β T =5.

are the loads and Young’s module of the bars. The target is to minimize the weight W by modifying the
transversal areas of the bar trusses, accomplishing the stress constraints and assuring that the structure is
as stiff as the topology layout obtained from the DTO imposing a target reliability index of β T = 5. This is
expressed mathematically as follows:

min W (3a)

subject to:

|σ| ≤ 355 M P a (3b)

P [W C > 1.93 · 107 mm/N ] ≤ Pf (3c)

where Pf = 2.86 · 10−7 (β T = 5). Figure 6 shows the convergence of the objective function for both
RBDO approaches. Table 3 presents a summary of the results and MPP for the optimum 7-bar (DES-1) and
13-bar (DES-2) truss structures, showing that the latter provides the better design leading to a 2% weight
savings.

Table 3. Summary results of size optimization.

RBDO 7-bar β T = 5 RBDO 13-bar β T = 5

Objective Function [kN] 9.28 9.10

MPP Vertical Load [N] 1.464 · 106 1.465 · 106
MPP Horizontal Load [N] 1.461 · 105 1.467 · 105
MPP Combination [N] 1.463 · 106 1.464 · 106
MPP Young’s module [MPa] 1.878 · 105 1.880 · 105

Table 4 shows the optimum design variable values for both structures being their initial values xi =
1.0 · 102 mm2 and their upper and lower bound ub = 1.0 · 105 mm2 and lb = 1.0 · 10−2 mm2 , respectively.
The results show that for a reliability index of β T = 5 the 13-bar truss structure is ligther that the
7-bar truss structure. The comparison between these structures is interesting from the point of view of
the performance of DTO and RBTO based designs under uncertainty. It is somehow intuitive to say that

6 of 19

American Institute of Aeronautics and Astronautics

 Obj. Function 7-bar
Obj. Function 13-bar
9.4

12
Obj. Function 7-bar 9.3

11 Obj. Function 13-bar 9.2

10

Obj. Function W [kN]

9.1

9 9
2
Number of Iterations
3

8
7
6
5 5 12 12

4 4.5
4
11
10
11
10
3.5 9 9

3 3
2.5
8
7
8
7
6 6

2 2
1.5
5
4
3
5
4
3
1
1 0.5
0
Obj. Function 7-bar
Obj. Function 13-bar
2
1
0
Obj. Function 7-bar
Obj. Function 13-bar
2
1
0
Obj. Function 7-bar
Obj. Function 13-bar

0 1 2 3 4 5 6 7 8 9 10 11 12 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
0
0 1 2 3
Number of Iterations

Figure 6. Evolution of the objective function in 7-bar and 13-bar truss structures.

Table 4. Values of the design variables in 7-bar and 13-bar structure.

Design Variables RBDO 7-bar RBDO 13-bar Design Variables RBDO 7-bar RBDO 13-bar
β T = 5(·103 ) β T = 5(·103 ) β T = 5(·103 ) β T = 5(·103 )

x1 [mm2 ] 1.764 1.295 x8 [mm2 ] - 2.457

x2 [mm2 ] 5.297 1.972 x9 [mm2 ] - 0.802
x3 [mm2 ] 2.753 5.246 x10 [mm2 ] - 0.575
x4 [mm2 ] 3.529 2.877 x11 [mm2 ] - 2.466
x5 [mm2 ] 8.261 8.194 x12 [mm2 ] - 2.349
x6 [mm2 ] 2.753 2.457 x13 [mm2 ] - 0.697
x7 [mm2 ] 2.753 1.938

DTO based designs will lead to the lightest structures after performing a posterior size optimization process.
Beyond we conclude that the RBTO based design for β T = 5 (13-bar truss) is lighter than the DTO based
design (7-bar truss) when applying size optimization for that particular safety treshold. Thus it has been
proven that RBTO is a valuable tool in preliminary structural design phases since it provides novel designs
that may perform better under reliability constraints than traditional DTO based designs. Figure 7 shows
an intuitive scheme explaining the comparison we have carried out between DTO and RBTO based designs.
We have proven that the final volume of approach 2 (W2∗ ) is lower than the one provided by approach 1
(W1∗ ).

B. Aircraft tail fuselage

This structural model was previously studied in López et al.22 It shows a three-dimensional aft fuselage of
10 000 mm length composed by a connection of two non-concentric truncated cones, with greater radios of
2100 mm and 1750 mm and lower radios of 1750 mm and 450 mm respectively. Some views of the aft fuselage
FE model are presented in figure 8.
The fuselage is loaded in the rear part with a combination of six load cases acting in the interface of the
FE model with the tail cone and with the vertical and horizontal tail planes (VTP-HTP). These loads are
symmetric in pairs with respect to x − z plane and are exposed in table 5. The FE model is composed by
a three-dimensional domain that is the design region of the topology optimization problem, shell elements
that simulate both the skin of the fuselage and the connection with the HTP-VTP and rigid elements that
act as load transmissors, leading to a 75978 FE mesh.
The DTO problem of equation 1 is solved, setting as maximum Von Mises stress σmax =300 MPa. The

7 of 19

American Institute of Aeronautics and Astronautics

 DESIGN REGION BASELINE DESIGN WEIGHT OPTIMUM DESIGN
Topology Size
optimization optimization
RBDO
DTO ( E T 5)
W1*
Interpretation

W2* W1*
RBTO RBDO
( E T 5) ( E T 5)
W2*
Interpretation

Figure 7. Conceptual scheme of the comparison between DTO and RBDO based designs.

Figure 8. Aft fuselage.

global compliance of the topology solution is W C DT O = 4.56 · 107 mm/N. Following the strategy proposed
in section II the RBTO problem of equation 2 with a target reliability index of β T = 3 (Pf = 1.35 · 10−3 )
is proposed, being G = 1 − W W C
C DT O
. The random variables are the loads and the material properties, being
their distribution type and nominal values presented in table 6. Figure 9 shows the topology results of both
approaches and the conceptual designs interpreted after the conceptual design definition task carried out
leaning on Hypermesh, which as exposed in Section II generates a set of surfaces from the topology results
allowing to create lines and splines following the curvature of the skin according to the structural members
emerged. The results of this engineering interpretation process give birth to two different structures of rigid
nodes denoted as DES-1 (DTO layout) and DES-2 (RBTO layout).
The definition of the beams (1D entity) starting from the lines drawn in Hypermesh is carried out thanks
to the “Engineering Solutions - Aerospace” profile of the Altair Hypermesh 14.0 release,28 which enables to
generate beams and give them the proper orientation from the lines that are contained in the skin surface.
These lines split the skin surface in a set of smaller surfaces (2D entities) that are meshed separately with
shell elements according to the number of beam elements desired. Then both the beam and shell elements
are merged in order to connect the movements of both entities. The FE model employed for size optimization
includes the 1D beams emerged from the conceptual design as well as the 2D shell elements that simulate the
fuselage skin. The average element size is about 100 mm, leading to 1255 beam elements defined as CBEAM
and 22100 shell elements defined as CTRIA3 due to the irregular shape of the smaller surfaces emerged after
the splitting of the skin surface. The beams have a cross-section of T-profiles, that are assigned to each
1D element following the orientation of the lines created before. Figure 10 shows some detailed view of the
T-profile beams merged to the fuselage skin.
The thickness of the aluminium skin is of 2.0 mm, while the initial dimensions of the beam profiles are

8 of 19

American Institute of Aeronautics and Astronautics

 Table 5. Loads applied to the aircraft rear fuselage FEM

Load Case Loads in VTP-HTP (N and N · mm)

Fx Fy Fz Mx My Mz

LC 1 0 −1 · 105 0 2.5 · 108 0 1.5 · 108

LC 2 0 1 · 105 0 −2.5 · 108 0 −1.5 · 108
LC 3 −1 · 104 0 −1 · 105 0 4 · 108 0
LC 4 1 · 104 0 1 · 105 0 −4 · 108 0
LC 5 2 · 104 0 1 · 105 0 −5 · 108 0
LC 6 −2 · 104 0 −1 · 105 0 5 · 108 0

Load Case Loads in rear fuselage (N)

Fx Fy Fz

LC 1 0 −8 · 103 −12 · 103

LC 2 0 8 · 103 12 · 103
LC 3 0 0 −28 · 103
LC 4 0 0 28 · 103
LC 5 −8 · 103 0 8 · 103
LC 6 −8 · 103 0 −8 · 103

Table 6. Distribution and statistical moments of the random variables.

Random Variable Distribution µ σ δ

Load Multiplier factor φi (i=1,...,6) Normal 1.00 0.1 0.1

Young’s module [MPa] Log-Normal 7.4 · 104 3.7 · 103 0.05

Figure 9. Topology results and conceptual designs of the DTO (left) and RBTO β T = 3 (right) aft fuselage.

detailed in Figure 11. The web and flange length is set to 80.0 mm, while their thicknesses are set to 3.0 mm.

The beams of the internal skeleton are organized in six types of beams (top, upper-side, side, lower-side,
bottom and rear). Figure 12 shows side views of both structures (DES-1 and DES-2) where it can be noticed
the type of beam that belongs to each structural member, as well as the internal skeleton of both DES-1

9 of 19

American Institute of Aeronautics and Astronautics

 Figure 10. Detailed views of the T-profile beams attached to the fuselage skin.

Figure 11. Transversal dimensions of the T-profile beams.

and DES-2. Table 7 recaps the relation between each type of beam with the scheme of Figure 12.

Table 7. Organization of the types of beams within the tail-fuselage designs DES-1 and DES-2.

Number Type of Beam Color (see Fig. 12) Number Type of Beam Color (see Fig. 12)

Beam1 Top beams Green Beam4 Side beams Black

Beam2 Rear beams Pink Beam5 Bottom beams Yellow
Beam3 Upper-side beams Blue Beam6 Lower-side beams Red

The design variables of the size optimization problem are the four general dimensions of the T-profiles
(DIM1 - DIM4) for each of the six types of beams defined above in order to simplify the problem, leading to
24 design variables. The numeration of the design variable follows the formula:

xij = DIMi · Beamj (i = 1, ..., 4; j = 1, ..., 6) (4)

For example, the design variable x31 corresponds to the flange thickness DIM3 of the top beams (Beam1),

10 of 19

American Institute of Aeronautics and Astronautics

 1
1

3 2 2
3
4
4

5
6 6 5

(a) DES-1 from DTO layout (b) DES-2 from RBTO β T = 3 layout

(c) DES-1 internal T-profile skeleton (d) DES-2 internal T-profile skeleton

Figure 12. Initial designs for size optimization of the aft fuselage.

while the design variable x15 corresponds to the web lenght DIM1 of the bottom beams (Beam5). Seven
random variables were defined, being six of them the load multiplier factors of the load cases and the other
one the Young’s modulus of the aluminium beams (Table 6).
Both designs DES-1 and DES-2 undergo a reliability-based size optimization problem similar to the
exposed in Equation 3. The RBDO problem aims to minimize the weight W of the structure, which must
fulfill the stress and compliance constraints. The maximum allowable stress of the skin is set to σmax,sk =
300.0 MPa, while the stress of the beams is set to σmax,b = 400.0 MPa since they usually absorb a larger rate
of stress. Moreover the designs obtained must be as stiff as the topology layout obtained from the DTO,
with a target reliability index of β T = 3, which is achieved by setting the upper bound of the probabilistic
constraint to W C DT O = 4.56 · 107 mm/N. The problem is formulated mathematically as:

min W (5a)

subject to:

|σsk | ≤ 300 M P a (5b)

|σb | ≤ 400 M P a (5c)
7
P [W C > 4.56 · 10 mm/N ] ≤ Pf (5d)

where the probability of failure is set to Pf = 1.35 · 10−4 (β T = 3). Figure 13 presents the convergence
of the objective function in both DES-1 and DES-2, being noteworthy that the RBTO based design DES-2
gives a lower weight after the RBDO process. Table 8 presents the RBDO results from both DES-1 and
DES-2 designs as well as their respectives MPPs.
The configuration of DES-1 achieved a weight of WDES−1 = 8.16 kN, whereas DES-2 reached to WDES−2
= 7.35 kN, leading to a 9% lighter structure than DES-1. Figure 14 shows a similar scheme to the one of

11 of 19

American Institute of Aeronautics and Astronautics

 9
8.8 Obj. Function DES-1
8.6 Obj. Function DES-2
8.4

Obj. Function W [kN]

8.2
8
7.8
7.6
7.4
7.2 13 16 10
Obj. Function DES-1
7 12
Obj. Function DES-1
Obj. Function DES-2
15
14
Obj. Function DES-2
9.5
Obj. Function DES-1
Obj. Function DES-2

6.8 11 13
12
9

6.6 10

9
11
10
8.5

8
6.4 8
9
7.5
8
6.2 7
0 1 2 3 4 5 6 7 8 9 10 11
7
0 2 4 6 8
7
0 1 2 3 4 5 6 7 8 9 10

6
0 1 2 3
Number of Iterations

Figure 13. Evolution of the objective function in DES-1 and DES-2

Table 8. Summary results of reliability-based size optimization.

RBDO DES-1 β T = 3 RBDO DES-2 β T = 3

Objective Function [kN] 8.16 7.35

MPP LC 1-2 [N] 1.10 1.13
MPP LC 3-4 [N] 1.11 1.10
MPP LC 5-6 [N] 1.15 1.12
MPP Eb [MPa] 7.153 · 104 7.116 · 104

Figure 7 where the comparison between DTO and RBTO based designs (DES-1 and DES-2) for the rear
fuselage is explained.
The results confirm the benefits of performing RBTO in early design stages. When DTO and RBTO
based designs, which are obtained throughout a conceptual design definition task, are compared in subsequent
design phases such as a size RBDO process setting as probability of failure the same that in the RBTO, the
structure defined from the RBTO layout is ligther than the one defined from the DTO layout.
Table 9 shows the optimum design variable values. For the flange and web length (DIM1 - DIM2) the
initial values are the original dimensions (80.0 mm), while the lower and upper bound are 20.0 mm and
300.0 mm respectively. For the flange and web thickness (DIM3 - DIM4) the initial values are 3.0 mm, being
the lower bound of 1.0 mm and the upper bound of 8.0 mm.
The constraints that are active in both RBDO problems are the stresses of the rear, side and lower-side
beams. From these results it can be said that the RBTO β T = 3 based design (DES-2) provides a better
distribution of material than the DTO based design (DES-1) if a subsequent size RBDO problem imposing
that particular safety target β T = 3 is conducted, since it fulfills the stress and stiffness constraints with a
lighter design. Figures 15 - 20 shows a same scale comparison between the initial and optimum design of
each of the six types of beams of DES-1 and DES-2, aiming to provide a better understanding of the results
obtained from the optimization process.

12 of 19

American Institute of Aeronautics and Astronautics

 DESIGN REGION BASELINE DESIGN WEIGHT OPTIMUM DESIGN
Topology Size
optimization optimization
RBDO
DTO ( E T 5)
W1*
Interpretation

W2* W1*
RBTO RBDO
( E T 5) ( E T 5)
W2*
Interpretation

Figure 14. Conceptual scheme of the comparison between DTO and RBTO based designs in the rear fuselage.

Table 9. Values of the design variables in the tail-fuselage structures.

Design RBDO DES-1 RBDO DES-2 Design RBDO DES-1 RBDO DES-2
Variables βT = 3 βT = 3 Variables βT = 3 βT = 3

x11 [mm] 28.35 29.07 x14 [mm] 27.46 95.15

x21 [mm] 28.72 23.74 x24 [mm] 28.41 117.8
x31 [mm] 1.06 1.44 x34 [mm] 1.03 1.22
x41 [mm] 1.05 1.18 x44 [mm] 1.04 1.33
x12 [mm] 189.3 139.4 x15 [mm] 41.7 46.27
x22 [mm] 300.0 300.0 x25 [mm] 48.9 74.75
x32 [mm] 8.0 3.02 x35 [mm] 1.65 1.40
x42 [mm] 4.49 3.81 x45 [mm] 1.84 1.03
x13 [mm] 33.57 20.0 x16 [mm] 108.5 20.0
x23 [mm] 35.27 33.89 x26 [mm] 300.0 20.0
x33 [mm] 1.26 1.00 x36 [mm] 3.10 1.00
x43 [mm] 1.28 1.00 x46 [mm] 1.97 1.00

13 of 19

American Institute of Aeronautics and Astronautics

 DIM1 = 80 mm DIM1 = 28.35 mm

DIM3 = 1,06 mm
DIM3 = 3 mm
DIM2 = 80 mm

DIM2 = 28,7 mm
DIM4 = 3 mm
DIM4 = 1,05 mm

(a) Initial vs. optimum design of top beams (Beam1) in DES-1

DIM1 = 80 mm DIM1=29,07 mm

DIM3=1,44 mm
DIM3 = 3 mm
DIM2 = 80 mm

DIM2=23,74 mm
DIM4 = 3 mm
DIM4=1,22 mm

(b) Initial vs. optimum design of top beams (Beam1) in DES-2

Figure 15. Initial design (left) and optimum design (right) of the top beams (Beam1) T-profile of DES-1 and
DES-2.

DIM1 = 80 mm DIM1 = 33,57 mm

DIM2 = 33,27 mm

DIM3 = 1,26 mm
DIM3 = 3 mm
DIM2 = 80 mm

DIM4 = 3 mm
DIM4 = 1,28 mm

(a) Initial vs. optimum design of upper-side beams (Beam3) in DES-1

DIM1 = 80 mm DIM1=20 mm
DIM3=1 mm
DIM3 = 3 mm
DIM2 = 80 mm

DIM2=33,9 mm

DIM4 = 3 mm DIM4=1 mm

(b) Initial vs. optimum design of upper-side beams (Beam3) in DES-2

Figure 16. Initial design (left) and optimum design (right) of the upper-side beams (Beam3) T-profile of DES-1
and DES-2.

14 of 19

American Institute of Aeronautics and Astronautics

 DIM1 = 189,3 mm
DIM1 = 80 mm

DIM3 = 3 mm

DIM3 = 8 mm
DIM2 = 80 mm
DIM4 = 3 mm

DIM2 = 300 mm
DIM4 = 4,5 mm

(a) Initial vs. optimum design of rear beams (Beam2) in DES-1

DIM1=139,4 mm
DIM1 = 80 mm
DIM3 = 3 mm

DIM3 = 3,02 mm
DIM2 = 80 mm

DIM4 = 3 mm

DIM2=300 mm

DIM4=3,81 mm

(b) Initial vs. optimum design of rear beams (Beam2) in DES-2

Figure 17. Initial design (left) and optimum design (right) of the rear beams (Beam2) T-profile of DES-1 and
DES-2.

15 of 19

American Institute of Aeronautics and Astronautics

 DIM1 = 80 mm DIM1 = 27,46 mm

DIM3 = 1,03 mm
DIM3 = 3 mm
DIM2 = 80 mm

DIM2 = 28,41 mm
DIM4 = 3 mm
DIM4 = 1,04 mm

(a) Initial vs. optimum design of side beams (Beam4) in DES-1

DIM1 = 80 mm DIM1=95,15 mm

DIM3=1,22 mm
DIM3 = 3 mm
DIM2 = 80 mm

DIM2=117,8 mm
DIM4 = 3 mm

DIM4=1,33 mm

(b) Initial vs. optimum design of side beams (Beam4) in DES-2

Figure 18. Initial design (left) and optimum design (right) of the side beams (Beam4) T-profile of DES-1 and
DES-2.

DIM1 = 41,7 mm
DIM1 = 80 mm

DIM3 = 1,65 mm
DIM3 = 3 mm

DIM2 = 48,9 mm
DIM2 = 80 mm

DIM4 = 3 mm
DIM4 = 1,84 mm

(a) Initial vs. optimum design of bottom beams (Beam5) in DES-1

DIM1 = 80 mm DIM1=46,27 mm
DIM3=1,4 mm
DIM3 = 3 mm

DIM2=74,75 mm
DIM2 = 80 mm

DIM4 = 3 mm
DIM4=1,03 mm

(b) Initial vs. optimum design of bottom beams (Beam5) in DES-2

Figure 19. Initial design (left) and optimum design (right) of the bottom beams (Beam5) T-profile of DES-1
and DES-2.

16 of 19

American Institute of Aeronautics and Astronautics

 DIM1 = 80 mm DIM1 = 108,5 mm

DIM3 = 3 mm

DIM3 = 3,1 mm
DIM2 = 80 mm
DIM4 = 3 mm

DIM2 = 300 mm
DIM4 = 1,98 mm

(a) Initial vs. optimum design of lower-side beams (Beam6) in DES-1

DIM1 = 80 mm DIM1=20 mm

DIM3=1 mm
DIM3 = 3 mm
DIM2 = 80 mm

DIM2=20 mm
DIM4=1 mm
DIM4 = 3 mm

(b) Initial vs. optimum design of lower-side beams (Beam6) in DES-2

Figure 20. Initial design (left) and optimum design (right) of the lower-side beams (Beam6) T-profile of DES-1
and DES-2.

These results show that the main differences between the T-profiles of DES-1 and DES-2 arise in the
side beams (Beam4) and in the lower-side beams (Beam6). The side beams of DES-2 are bigger than in the
DES-1, which is something expected since they cover a wider area of the side of the fuselage and consequently
absorb a large ratio of load. In contrast, the lower-side beams (Beam6) of DES-1 are much bigger (indeed the
Beam6 of DES-2 go to the minimum values) since they need to bear most of the load from the VTP-HTP,
which is mostly held by the side beams (Beam4) in DES-2. The rest of the beams have similar dimensions
except the web length and thicknesses of the rear beams (Beam2) of DES-1. These beams in DES-1 are
larger than in DES-2 as a consequence of the topology results, which show a rear area less filled of material
in DES-1. This means that the rear part of DES-1 should have a lower amount of bars than in DES-2,
causing that these bars must bear a larger rate of load and consequently increase their dimensions.

IV. Conclusions
This research illustrates how Reliability-Based Topology Optimization (RBTO) may be useful in the
design of aircraft components. The structural layout of a topology optimization problem can be different
depending on the safety level required, which influences the subsequent phases of the design process. In this
study the layouts of a Deterministic Topology Optimization (DTO) and a RBTO were engineered and the
resulting structures were compared againts a subsequent phase of reliability-based size optimization. Two
different examples has been proposed: a simple 2D rectangular domain and a more complex 3D tail fuselage.

17 of 19

American Institute of Aeronautics and Astronautics

 In the first example the structural layout obtained from the DTO led to a 7-bar truss structure through
the concept design definition process, while the RBTO for a reliability target of β T = 5 led to a 13-bar
truss structure. Both structural schemes had the same compliance values according to the constraints of the
RBTO problem. Afterwards both structures were optimized considering the same reliability index target and
the results showed that the 13-bar structure provides better results than the 7-bar, even though the initial
design of the 7-bar structure is lighter. This research shows that if the structure is going to be optimized
later for a determined reliability treshold, the best structure will be the one obtained from the RBTO process
with that reliability level. Altough this fact might seem somehow intuitive, it actually boosts the relevance
of RBTO throughout the structural design process.
This idea is enforced with the second example, where the layouts of the DTO and RBTO for a safet target
of β T = 3 led to completely different structural schemes, although they had the same weighted compliance.
After going through the concept design definition, both structures went through a reliability-based size
optimization process with a reliability target of β T = 3, showing again that the best design was the one
interpreted from the RBTO layout (DES-2), although the lighter initial design is the structure from DES-1.
Therefore this study remarks the importance of RBTO in preliminary design phases of aircraft structures,
because if the structures interpreted from the topology optimization will later undergo a reliability-based
size optimization process for a certain safety target β T , the most efficient design will come from the RBTO
for that particular β T instead from the traditional DTO. In fact, it highlights how important is to know
the behavior of the component that is being designed, regarding to determine a priori the design conditions,
which are the limit-states that the structure must accomplish as well as the probabilities of failure for these
limit-states. If any of these parameters is altered, the optimal structural layout from the DTO/RBTO and
consequently the optimal structures arisen may change.

Acknowledgments
The research leading to this results is part of the project DPI2013-41893-R from the Spanish Ministry
of Economy and Competitiveness. The authors acknowledge the Galician Government for grant GRC2013-
056, and the first author also acknowledges the program “Formación del Profesorado Universitario” from the
Spanish Ministry of Education, Culture and Sports for his sponsorship under FPU grant AP2012-1735.

References
1 Bendsoe, M.P. and Kikuchi N., “Generating optimal topologies in optimal design using a homogenization method.”, Com-

puter Methods in Applied Mechanics and Engineering,, Vol. 71, 1988, pp. 197-224.
2 Yang, R.J. and Chahande, A.I., “Automotive applications of topology optimization”, Structural Optimization,, Vol. 9, No.

3-4, 1995, pp. 245-249.

3 Yang, R.J., “Topology optimization of vehicle structures”, Proceedings of the 1997 Conference on Optimization in Industry,

Palm Coast, FL, USA, 1997, pp. 233-242.

4 Mijar, A.R., Swan, C.C., Arora, J.S. and Kosaka, I., “Continuum topology optimization for concept design of frame bracing

systems”, Journal of Structural Engineering, Vol. 124, No.5, 1998, pp. 541-550.
5 Iuspa, L., Scaramuzzino, F. and Petrenga, P., “Topological optimization of an aircraft engine mount via bit-masking

oriented genetic algorithms”, Proceedings of the Eight International Conference on Civil and Structural Engineering Computing,
Vienna, Austria, 2001, pp. 239-240.
6 Rozvany, G.I.N., “Aims, scope, methods, history and unified terminology of computer-aided topology optimization in

structural mechanics”, Structural and Multidisciplinary Optimization,, Vol. 21, No. 2, 2001, pp. 90-108.
7 Sigmund, O. and Maute, K., “Topology optimization approaches - A comparative review”, Structural and Multidisciplinary

Optimization,, Vol. 48, No. 6, 2013, pp. 1031-1055.

8 Deaton, J. and Grandhi, R.V., “A survey of structural and multidisciplinary continuum topology optimization: Post 2000”,

Structural and Multidisciplinary Optimization,, Vol. 49, No. 1, 2014, pp. 1-38.
9 Zhu, J. and Zhang, W. and Xia, L. “Topology Optimization in Aircraft and Aerospace Structures Design”, Archives of

computational Methods in Engineering, 2015, pp. 1-28.

10 Krog, L., Tucker, A., “Application of topology, sizing and shape optimization methods to optimal design of aircraft

components”, Altair Engineering Conference, 2002.

11 Grihon, S. and Krog, L. and Tucker, A. and Hertel, K. “AA380 weight savings using numerical structural optimization”,

20th AAAF colloquium on material for aerospace applications, Paris, France, 2004.
12 Remouchamps, A., Bruyneel, M., Fleury, C. and Grihon, S., “Application of a bi-level scheme including topology opti-

mization to the design of an aircraft pylon”, Structural and Multidisciplinary Optimization,, Vol. 44, No. 6, 2011, pp. 739-750.
13 Wang, Q., Choi, K.K. and Zhou, C., “New topology optimization method for wing leading-edge ribs”, AIAA Journal of

Aircraft, Vol. 48, No. 5, 2011, pp. 1741-1748.

18 of 19

American Institute of Aeronautics and Astronautics

 14 Buchanan, S., “Development of a Wingbox Rib for a Passenger Jet Aircraft using Design Optimization and Constrained

to Traditional Design and Manufacture Requirements”, Altair Engineering CAE Technology Conference, Michigan, 2011, pp.
1-8.
15 Krog, L., Tucker, A., Kemp, M. and Boyd, R., “Topology optimization of aircraft wing box ribs”, AIAA Journal of

Aircraft, Vol. 3, 2004, Albany, New York, pp. 2020-2030.

16 Kharmanda, G., Olhoff, N., Mohamed, A. and Lemaire, M., “Reliability based topology optimization” Structural and

Multidisciplinary Optimization, Vol. 26, No. 5, 2004, pp. 295-307.

17 Maute, K. and Allen, M., “Reliability-based design of MEMS mechanisms by topology optimization” Computers and

Structures, Vol. 81, No. 8-11, 2004, pp. 813-824.

18 Guest, J.K. and Igusa, T., “Structural optimization under uncertain loads and nodal locations” Computer Methods in

Applied Mechanics and Engineering, Vol. 198, 2008, pp. 116-124.

19 Silva, M. and Tortorelli, D.A. and Norato, J. and Ha, C. and Bae, H.R., “Component and system reliability-based topology

optimization using a single-loop method” Structural and Multidisciplinary Optimization, Vol. 41, No. 1, 2004, pp. 87-106.
20 Dunning, P.D. and Kim, H.A. and Mullineux, G., “Introducing loading uncertainty in topology optimization” AIAA

Journal, Vol. 49, No. 4, 2011, pp. 760-768.

21 Tootkaboni, M. and Asadpoure, A. and Guest, J.K., “Topology optimization of continuum structures under uncertainty

- A Polynomial Chaos approach” Computer Methods in Applied Mechanics and Engineering, Vol. 201-204, 2012, pp. 263-275.
22 López, C., Baldomir, A. and Hernández, S., “Deterministic versus reliability-based topology optimization of aeronautical

structures”, Structural and Multidisciplinary Optimization, Vol.53, No.4, 2016, pp.907-921.

23 Du, X. and Chen, W., “Sequential Optimization and Reliability Assessment Method for Efficient Probabilistic Design”,

Journal of Mechanical Design, Vol.126, No.2, 2004, pp.225-233.

24 Aoues, Y. and Chateauneuf, A., “Benchmark study of numerical methods for reliability-based design optimization” Struc-

tural and Multidisciplinary Optimization,, Vol. 41, No. 2, 2010, pp. 277-294.
25 Altair HyperWorks 12.0 User Manual, Altair Engineering Inc., 2013.
26 MATLAB R2013a Documentation, MATLAB, 2013.
27 Youn, B.D., Choi, K.K. and Park, Y.H., “Hybrid Analysis Method for Reliability-Based Design Optimization”, Journal

of Mechanical Design, Vol.125, No.2, 2003, pp.221-232.

28 Altair HyperWorks 14.0 release User Manual, Altair Engineering Inc., 2016.

19 of 19

American Institute of Aeronautics and Astronautics

View publication stats