Professional Documents
Culture Documents
CLopez Et Al SciTech2017
CLopez Et Al SciTech2017
net/publication/312472836
CITATIONS READS
0 151
3 authors:
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
All content following this page was uploaded by Carlos López on 04 September 2017.
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,
1 of 19
2 of 19
Deterministic
Optimization
Calculate ∇Gi (d, uki ) (OPTISTRUCT)
Convergence?
(OPTISTRUCT)
d∗
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
DTO RBTO β T
Interpretation Interpretation
Figure 2. Flowchart of the strategy proposed for the conceptual design definition task.
4 of 19
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.
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
NODES 1 2 3 4 5 6 7 8
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:
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 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
12
Obj. Function 7-bar 9.3
10
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.
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 )
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∗ ).
7 of 19
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.
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
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
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)
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:
For example, the design variable x31 corresponds to the flange thickness DIM3 of the top beams (Beam1),
10 of 19
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:
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
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 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
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.
Design RBDO DES-1 RBDO DES-2 Design RBDO DES-1 RBDO DES-2
Variables βT = 3 βT = 3 Variables βT = 3 βT = 3
13 of 19
DIM3 = 1,06 mm
DIM3 = 3 mm
DIM2 = 80 mm
DIM2 = 28,7 mm
DIM4 = 3 mm
DIM4 = 1,05 mm
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
Figure 15. Initial design (left) and optimum design (right) of the top beams (Beam1) T-profile of DES-1 and
DES-2.
DIM3 = 1,26 mm
DIM3 = 3 mm
DIM2 = 80 mm
DIM4 = 3 mm
DIM4 = 1,28 mm
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
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
DIM3 = 3 mm
DIM3 = 8 mm
DIM2 = 80 mm
DIM4 = 3 mm
DIM2 = 300 mm
DIM4 = 4,5 mm
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
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
DIM3 = 1,03 mm
DIM3 = 3 mm
DIM2 = 80 mm
DIM2 = 28,41 mm
DIM4 = 3 mm
DIM4 = 1,04 mm
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
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
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
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
DIM3 = 3 mm
DIM3 = 3,1 mm
DIM2 = 80 mm
DIM4 = 3 mm
DIM2 = 300 mm
DIM4 = 1,98 mm
DIM1 = 80 mm DIM1=20 mm
DIM3=1 mm
DIM3 = 3 mm
DIM2 = 80 mm
DIM2=20 mm
DIM4=1 mm
DIM4 = 3 mm
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
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.
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
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
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
18 of 19
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
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
- 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
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
19 of 19