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

net/publication/318355034

Concurrent multi-scale design optimization of composite frames with

manufacturing constraints

Article in Structural and Multidisciplinary Optimization · September 2017

DOI: 10.1007/s00158-017-1750-0

CITATIONS READS

30 2,253

4 authors, including:

Jun Yan Zunyi Duan

Dalian University of Technology Northwestern Polytechnical University
120 PUBLICATIONS 1,805 CITATIONS 40 PUBLICATIONS 360 CITATIONS

SEE PROFILE SEE PROFILE

Erik Lund
Aalborg University
140 PUBLICATIONS 4,619 CITATIONS

SEE PROFILE

All content following this page was uploaded by Zunyi Duan on 19 October 2017.

The user has requested enhancement of the downloaded file.

Struct Multidisc Optim
DOI 10.1007/s00158-017-1750-0

RESEARCH PAPER

Concurrent multi-scale design optimization of composite frames

with manufacturing constraints
Jun Yan 1 & Zunyi Duan 1,2,3,4 & Erik Lund 2 & Jingyuan Wang 1

Received: 20 January 2017 / Revised: 24 May 2017 / Accepted: 20 June 2017

# Springer-Verlag GmbH Germany 2017

Abstract This paper presents a gradient based concurrent concurrent optimization model. The multi-scale optimization
multi-scale design optimization method for composite frames model, considering specific manufacturing constraints, pro-
considering specific manufacturing constraints raised from the vides new choices for the design of the composite frame struc-
aerospace industrial requirements. Geometrical parameters of ture in aerospace and other industries.
the frame components at the macro-structural scale and the
discrete fiber winding angles at the micro-material scale are Keywords Concurrent optimization . Composite frames .
introduced as the independent design variables at the two geo- Discrete material optimization . Manufacturing constraints
metrical scales. The DMO (Discrete Material Optimization)
approach is utilized to couple the two geometrical scales and
realize the simultaneous optimization of macroscopic topolo- 1 Introduction
gy and microscopic material selection. Six kinds of
manufacturing constraints are explicitly included in the opti- Frame structures composed of laminated composites like glass
mization model as series of linear inequalities or equalities. or carbon fiber-reinforced polymers (GFRP/CFRP) have been
The capabilities of the proposed optimization model are dem- extensively used in aerospace structure and civil engineering
onstrated with the example of compliance minimization, sub- with excellent performances for the high ratio of stiffness and
ject to constraint on the composite volume. The linear con- weight. For convenience, GFRP/CFRP frame structures are
straints and optimization problems are solved by Sequential simply referred to as composite frames in the following parts
Linear Programming (SLP) optimization algorithm with move of the paper. The composite frame is utilized especially for
limit strategy. Numerical results show the potential of weight aerospace vehicles, load-bearing structure of satellites, spatial
saving and structural robustness design with the proposed stations, transmission towers and wind turbine structures,
where large-scale, high stiffness and low weight are empha-
sized (Schutze 1997; Ibrahim et al. 2000). Therefore, optimi-
* Jun Yan zation of laminated composites has undergone tremendous
[email protected] development in last decades. Some contributions on this topic
have been summarized in the review articles of Ghiasi et al.
1
(2009, 2010) and Bakis et al. (2002). It is worth noting that, in
State Key Laboratory of Structural Analysis for Industrial
Equipment, Department of Engineering Mechanics, International
recent years, Discrete Material Optimization (DMO) of lami-
Research Center for Computational Mechanics, Dalian University of nated composites has achieved considerable progress and
Technology, Dalian 116024, China attracted much attention due to the requirements of
2
Department of Materials and Production, Aalborg University, manufacturing the optimal design of composite structures.
Fibigerstraede 16, DK-9220 Aalborg East, Denmark Based on an extension of the multi-phase topology
3
Harbin Electric Power Equipment Company Limited, Harbin Electric optimization (Sigmund and Torquato 1997) Lund and
Corporation, Harbin 150028, China Stegmann proposed the DMO method for multi-phase optimi-
4
Department of Mechanical Engineering, Korea Advanced Institute of zation of laminated composites (Lund and Stegmann 2005;
Science and Technology, 34141 Daejeon, Republic of Korea Stegmann and Lund 2005). The DMO method has been
 J. Yan et al.

successfully used to dispose many physical problems (Hvejsel approach to perform a hierarchical optimization of laminated
et al. 2011; Lund 2009; Niu et al. 2010; Duan et al. 2014). As composite structures with simultaneous consideration of the
an alternative to the DMO schemes, Bruyneel (2011) intro- fiber orientation and cross-section size/shape. As an immedi-
duced the Shape Functions with Penalization (SFP) scheme ate extension of the original DMO method, Sørensen et al.
based on the shape functions of a quadrilateral first order finite (2014) considered the topology variables of layer thickness
element. Gao et al. (2012) proposed a Bi-valued Coding and fiber orientation as the design variables and presented
Parameterization (BCP) scheme which can be considered as simultaneous optimization of the thickness and fiber
a generalization of the SFP scheme. orientation of laminate composites including certain
Despite remarkable achievements have been made for the manufacturing constraints. Blasques and Stolpe (2012) carried
optimization of laminated composites, there still are two chal- out a framework of simultaneous topology and material opti-
lenging issues for design of laminated composites. The first mization (fiber orientation and laminate thickness) in optimal
one is design for practical applications. This means that certain design of laminated composite beams with specific structural
design guidelines or rules, referred to as manufacturing con- requirements, e.g., compliance, eigenfrequency constraints
straints, should be considered to e.g., reduce the risk of local and structure weight. Liu et al. (2008) proposed the PAMP
failure in the structure. In Bailie et al. (1997), they mentioned (Porous Anisotropic Material with Penalization) model, which
eight weaknesses associated with laminated composites. An has also been extended to multi-scale optimal design of com-
efficient way to prevent these weaknesses from happening is posite material accounting for strength optimization (Yan et al.
to follow a set of fundamental manufacturing constraints 2014), thermo-mechanical loads (Yan et al. 2015a; Deng et al.
based on industrial experiences from tests and analyses. 2013) and frequency optimization (Niu et al. 2009). However,
These manufacturing constraints have been developed to help in the previous references, few works are related with multi-
the designers to exploit the material’s strengths while reducing scale optimization of composite frame structures. Actually, in
the risk of structure and material failure. Costin and Wang order to achieve lightweight designs with specific structural
(1993), Wang and Costin (1992) and Liu et al. (2011) consid- performance (e.g., compliance, frequency) for composite
ered manufacturing constraints in aircraft structural design frame structure, it is an efficient approach to simultaneously
with lamination parameters and numbers of plies with the optimize the ply parameters of the composite material and its
predefined angles (such as [0°, ∓45°, 90°]) as design variables. structural configurations.
Manne and Tsai (1998) utilized plydrop tapering for thickness The present paper proposes a concurrent multi-scale design
optimization of symmetric layups to avoid warping. Recently, optimization model for composite frames with respect to min-
adopting SFP interpolation scheme, Bruyneel et al. (2012) imum structural compliance under the volume constraint. The
realized the optimal stacking sequence design of laminated cross-section of the beam at the macro-scale and the discrete
composites considering several manufacturing constraints. fiber winding angles at the micro-scale are introduced as the
Irisarri et al. (2014) considered an extensive set of design independent design variables to realize the topology and
guidelines to realize the optimal design of laminated compos- stacking sequence optimization at the two geometrical scales
ite structures with ply drops using a stacking sequence table simultaneously. Especially, in the optimization model, six
(SST) method. Sørensen and Lund (2013), Sørensen et al. kinds of specific manufacturing constraints have been explic-
(2014) carried out the serial works on thickness and material itly expressed as series of linear inequalities or equalities, and
choice design of laminated composites with certain the sensitivities of these specific manufacturing constraints are
manufacturing constraints. A more detailed discussion about easy to derive and obtain. At the micro-scale the DMO ap-
design guidelines and their justification is provided in Bailie proach has been applied to achieve the micro-scale material
et al. (1997). interpolation. The extension of the open-source composite
Another challenging issue is how to fully exploit the po- beam analysis tool BECAS developed by Blasques and
tential of composite structures due to the coupling effects be- Lazarov (2012), see also Blasques and Stolpe (2011), has been
tween the macro structures (size, shape of structural compo- adopted to realize high-fidelity analysis of the composite
nents and structural configuration) and material frame structure. In the numerical examples, we implement
microstructure (Yan et al. 2015a, b; Rodrigues et al. 2002) four kinds of optimization models, which consider the effect
(fiber orientation, fiber content and layer thickness). As a kind of single-scale optimization, multi-scale optimization, and dif-
of architecture material, laminated composites offer a good ferent constraint parameters on the optimization results.
opportunity to tailor the material properties. Many researchers The organization of the remainder parts of this paper is as
have carried out corresponding multi-scale lightweight mate- follows. In Section 2, the concurrent multi-scale optimization
rial design of composite structures. Gao et al. (2013) proposed concept and the discrete material optimization model with
a simultaneous optimization of layout design and discrete evaluation of convergence are introduced. Section 3 presents
fiber orientation of laminated structures. Ferreira et al. the explicit mathematical formulation of the manufacturing
(2013) adopted Discrete Material Optimization (DMO) constraints considered in the present paper. Section 4
Concurrent multi-scale design optimization of composite frames

introduces the mathematical formulation of the optimization (2007) suggested that 0° and 90° fiber winding angles in fila-
problem and the structural analysis model. Sensitivity analysis ment winding process should be implemented by 5° and 85°
formulas are presented in Section 5. Section 6 presents the fiber winding angles, respectively. So in this work, we con-
optimal results and the comparison of the single-scale and sider the assembly of [5°, ∓45°, 85°] as a set of candidate com-
multi-scale optimization. Finally, a section with conclusions posite fiber winding angles. The fiber winding angle is as-
closes the paper. sumed to be constant in a given ply.

2.2 Discrete Material Optimization (DMO) approach

2 Concurrent multi-scale optimal design of composite
frame with Discrete Material Optimization (DMO) To ensure the integrity of this paper and the reading conve-
approach nience, the basic idea and formula of the DMO model are
briefly introduced in this section. For the detailed description
2.1 Concurrent optimization of composite frame please refer to the references (Stegmann and Lund 2005;
Hvejsel and Lund 2011; Lund 2009; Niu et al. 2010; Duan
Because frame structures frequently adopt constant circular et al. 2014).
cross-sections and a fixed number of layers in most aerospace The discrete material optimization formulation is imple-
applications, for the ease of derivation and without loss of mented in a finite element framework. In the DMO approach
generality, frames made of tubes with constant circular the element constitutive matrix per layer Dei; j (the superscript e
cross-sections and a fixed number of layers are investigated refers to “element”, and the subscripts mean the i’th tube and
in the present paper. The joints connecting the composite j’th layer) can be expressed as a weighted sum of the consti-
tubes can transfer moments and are assumed infinitely stiff. tutive matrices Di , j , c of the candidate materials (the subscript
At the macroscopic level, the radius of the cross-section is c means the c’th candidate material). In general, for multi-
used as the macro design variable, similar as that in the clas- layer structures, the interpolation method can be implemented
sical size optimization of frame structures. The radius of the layer-wise for each element, i.e., for all layers in all elements.
tube can be recognized as the size and topology variable at the Various parameterization schemes have been developed in
same time. When the radius reaches its lower limit, the tube Lund and Stegmann (2005); Stegmann and Lund (2005).
can be regarded to be deleted from the ground structure and Consequently, the interpolation scheme is written by layers,
thereby realize the structural topology optimization (Bendsøe and the constitutive relation for the j’th layer can be expressed
and Sigmund 2003; Eschenauer and Olhoff 2001). As has as a sum over the number of candidate materials Ncand for the
been shown in the schematic diagram of Fig. 1, the middle layer:
hidden beam component means the macro design variable of
cand
the beam has reached its lower limit and has been deleted from Dei; j ¼ ∑N
c¼1 ωi; j;c Di; j;c ð1Þ
the ground frame structure.
At the microscopic level, the discrete fiber winding angle where each candidate material is characterized by a constitu-
selection is solved by using the DMO approach in the sense tive matrix Di , j , c. The weighting functions ωi , j , c all attain
that the structural constituents are chosen from among a given values between 0 and 1 because no stiffness or mass matrix
set of candidate materials. In most practical applications, the can contribute more than the physical material properties, and
candidate composite ply is restricted to [0°, ∓45°, 90°], which a negative contribution is physically meaningless. In (1), the
are the conventional orientations used in aeronautics (Baker parameterization model can be realized for single-layers and
et al. 2004). Considering the manufacturing process, Mallick layer-wise for multiple layers for a large number of candidate

Fig. 1 Schematic diagram of the

concurrent optimization of a
composite frame composed of
three beams
 J. Yan et al.

materials. In this paper, the generalization of the SIMP multi- been discarded. For each layer, the following inequality is
material interpolation schemes (Hvejsel and Lund 2011) is evaluated according to all weighting functions layer-wise.
used to push the weighting functions towards 0 or 1 and obtain qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a distinct material selection. The weighting functions can be ωi; j;c ≥ ε ω2i; j;1 þ ω2i; j;2 þ ⋯ þ ω2i; j;Ncand ð4Þ
expressed as

p where ε is a tolerance level; typically, ε ∈ [0.95 ~ 0.99]. If
ωi; j;c ¼ xi; j;c ð2Þ inequality (4) is satisfied for any ωi , j , c in the j’th layer, that
layer is flagged as converged. The convergence assessment
where p is a penalty parameter and the design variables xi , j , c
criterion Hε is defined as the ratio between the number of
are the artificial materials density of candidates. If xi , j , c = 1, it
converged layers Nl;tot
c and the total number of layers Nl,tot.
means that the distinct c’th candidate material has been select- lay
ed from a set of candidate materials for the j’th layer of the i’th N is the number of layers in each tube, and it is assumed that
tube. It is worth noting that in the original DMO multi- each tube has the same number of layers in the present paper.
material interpolation, to keep the physically meaning in the Thus, Nl,tot can be expressed as the number of tubes multiplied
case of a mass constraint or eigenfrequency optimization, a by the number of layers in a tube Nl,tot = Ntub ∙ Nlay:
normalized weighting scheme was adopted (Stegmann and Nl;tot
Lund 2005). Hε ¼ c
ð5Þ
Nl;tot
Hvejsel and Lund (2011) formulated multi-material varia-
tions of the SIMP (Bendsoe 1989) and RAMP (Stolpe and If the tolerance level is 95% and fully converged, i.e., Hε =
Svanberg 2001) interpolation schemes and relied on a large 0.95 = 1,
all layers have a single weight factor that contributes
number of sparse linear constraints to enforce the selection of more than 95% to the Euclidian norm of the weight factors.
at most one material in each design subdomain. Inspired by
Hvejsel and Lund (2011), in the present paper, the simple
linear equality constraints on candidate artificial material den-
3 Manufacturing constraints
sity and continuous penalty strategy are adopted. The linear
equality constraints on candidate artificial material densities
Recent years, manufacturing constraints have attracted more
can be expressed as
and more attention in design of laminate composites, e.g.,
∑N
cand
minimum percentage of each orientation constraints (10%
c¼1 xi; j;c ¼ 1 ð3Þ
rule), contiguity constraints, balance constraints, symmetry
In this paper, the equality constraint in (3) is labeled as constraints (e.g., Bruyneel et al. 2012; Seresta et al. 2007;
DMO normalization constraint (DMOnC), which should be Kassapoglou 2013), damage tolerance constraints, ply-drop
realized layer-wise for laminated composites with multiple design constraints (e.g., Irisarri et al. 2014), thickness varia-
layers and solved by SLP (Sequential Linear Programming) tion rate and intermediate constraints (e.g., Sørensen and Lund
method (Fletcher et al. 1998 or Gomes and Senne 2011). The 2013; Sørensen et al. 2014). Taking into account the relevance
details about DMOnC constraint and other manufacturing of the above mentioned manufacturing constraints for the case
constraints will be shown in Section 3. With the micro-scale of composite frames, the variable stiffness design i.e., thick-
parameters of the DMO material interpolation recognized as ness variation and ply-drop problems are not considered in
the micro design variable (i.e., xi , j , c), the manufacturing con- this work and will be left for future work. The benefits of
straints can be explicitly expressed as linear equalities or in- obeying these constraints are obvious in designing laminated
equalities in the form of the micro-scale parameters. Then, composites, such as the following lists.
considering the coupling effect of design variables at the two
geometrical scales and the specific manufacturing constraints, [1] Manufacturing constraints make it possible to exploit the
the concurrent multi-scale optimization of composite frames strengths of the material while mitigating the adverse
can be established for the specified structural loading and effects of the material (e.g., matrix cracking and delam-
boundary conditions. ination; warping under thermal loading; out-of-plane
failure modes).
[2] Manufacturing constraints can furthermore be used to
2.3 Evaluation of convergence of DMO limit the complexity of the optimized design, thereby
making it possible to achieve a higher degree of manu-
A convergence measure given in Stegmann and Lund (2005) facturability (Sørensen et al. 2014).
is adopted to describe whether the optimization has converged [3] Following certain manufacturing constraints, we can
to a satisfactory result, i.e., a single candidate material has greatly improve the robustness of composite structures
been chosen in a specified element and all other materials have and improve service time of equipments.
Concurrent multi-scale design optimization of composite frames

The explicit linear equality or inequality manufacturing they are less susceptible to the weaknesses associated with
constraints are presented with respect to micro-scale design highly orthotropic laminates. It is important to note that the
variables (xi , j , c). The linear formulations are highly attractive 10% fiber-dominated guideline is often interpreted differently
from an optimization point of view and possible to achieve for with regard to the ±45∘ plies. Some project directives require
all the manufacturing constraints presented in this paper. there is at least 5% “+45∘” and 5% “−45∘” plies, rather than
10% of +45∘ and −45∘ plies. There are no guidelines that
3.1 Contiguity constraint (CC) establish a rigorous differentiation between these two alterna-
tive minimum 45° ply contents. Other projects have issued
The definition of the contiguity constraint (CC) is that no more guidelines requiring at least 6% (rather than 10%) 85° plies
than a given number of plies, CL, of the same orientation are included when there are at least 20% “±45∘” plies. In most
should be stacked together. The benefit of this manufacturing aerospace application, the 10% rule is frequently adopted, and
constraint is to avoid matrix cracking failure (Sørensen et al. therefore the 10% rule is adopted in the implementation of the
2014). The contiguity constraint with respect to micro-scale present paper.
parameters can be formulated as a linear inequality described The TPR is expressed as
by (6). Let CL ∈ N∗ denote the contiguity limit, then for any lay

i ∈ Ntub , j ∈ Nlay , c ∈ Ncand, it should follow (6), and the loop ∑Nj¼1 xi; j;c ≥ 10% Nlay ð8Þ
should meet the dimension of j + CL ≤ Nlay. Ntub is the number
Equation (8) can be explained as the sum for every candi-
of tubes in the frame structure.
date material density (xi , j , c) in the whole laminate should be
The CC can be expressed as
greater or equal than 10%. That is to say, if the combination of
xi; j;c þ ⋯ þ xi; jþCL;c ≤ CL; j þ CL≤ Nlay ð6Þ a laminate is −45∘ (8%); 5∘ (62%); +45∘ (25%); 85∘ (5%),
then it does not fulfill the 10% rule, because the layer propor-
Table 1 gives an example of a laminate with four layers and tions of the −45∘ and 85∘ candidate materials in the whole
four candidate materials in each layer. To present the contigu- laminate are less than 10%.
ity constraint clearly for this example, the first candidate ma-
terial, i.e., −45∘ and the contiguity constraint with CL = 1 are 3.3 Balance constraint (BC)
considered. Then the contiguity constraint can be expressed as
linear inequalities: Balance constraint means that angle plies (those at any angle
other than 5∘ and 85∘) should occur only in balanced pairs
xi;1;1 þ xi;2;1 ≤ CL; xi;2;1 þ xi;3;1 ≤ CL; xi;3;1 þ xi;4;1 ≤ CL ð7Þ
with the same number of +θ∘and −θ∘ plies (θ∘ ≠ 5∘ , 85∘). For
the set of the 5° , ∓ 45° , 85° candidate materials, any +45∘ ply
Here we note that, the contiguity constraint should be im- should be accompanied by a −45∘ ply. A typical example of
plemented layer-wise for every candidate material, i.e., for all the difference between balanced and unbalanced laminates is
four fiber winding angles considered in this work. shown in Fig. 2. The parameterized linear equality constraint
For example, if a composite tube has twenty layers i.e., with respect to candidate artificial material density xi , j , c can
Nlay = 20, and every layer has four candidate materials i.e., be expressed as (9).
Ncand = 4, the total number of contiguity constraints should The BC is computed as
be calculated as (Nlay − CL)∗Ncand = (20 − 1)∗4 = 76, when
∑Nj¼1 xi; j;þc −∑Nj¼1 xi; j;−c ¼ 0; c≠ð5∘ ∪85∘ Þ
lay lay

the contiguity limit is equal to one i.e., CL = 1. ð9Þ

In (9), xi , j , + c denotes a positive angle, conversely, xi ,

3.2 10% rule (ten percent rule, TPR)
j , − c denotes an accompanied negative angle. In the set of
the 5° , ∓ 45° , 85° candidate materials, xi , j , + c is +45∘ ply,
In the 10% rule, a minimum of 10% of plies of each of the 5∘, and xi , j , − c is −45∘ ply. An example of the balance con-
±45∘ and 85∘ angles is required. The benefit of this constraint straint for a four layer laminate as shown in Table 1 can
is to obtain laminates that are more robust in the sense that
expressed as ∑4j¼1 xi; j;1 −∑4j¼1 xi; j;3 ¼ 0. This equation guar-
Table 1 Example of antees the +45∘ and −45∘ candidates have the same number
laminate with four layers −45∘ 5∘ +45∘ 85∘ of plies in the laminate.
and four candidate
materials in each layer xi , 1 , 1 xi , 1 , 2 xi , 1 , 3 xi , 1 , 4
xi , 2 , 1 xi , 2 , 2 xi , 2 , 3 xi , 2 , 4 3.4 DamTol constraint (DTC)
xi , 3 , 1 xi , 3 , 2 xi , 3 , 3 xi , 3 , 4
xi , 4 , 1 xi , 4 , 2 xi , 4 , 3 xi , 4 , 4
A damage tolerance constraint, abbreviated as DamTol con-
straint, is introduced. It states that the 5∘ ply along the axial
 J. Yan et al.

Fig. 2 Illustration of balanced

and unbalanced symmetric
laminates

direction cannot be selected in the inner and outer layer. This 3.6 DMO normalization constraint (DMOnC)
manufacturing constraint is very reasonable for composite
frames, because it is not easy to wind the fiber on the inner As has been mentioned, in order to keep the physical meaning
and outer surface of the tube with 5∘ fiber along the axial in the case of a mass constraint or eigenfrequency optimiza-
direction. Furthermore, a composite tube with the layer of 5∘ tion, the sum of the candidate artificial materials density in the
can easily delaminate, which should be avoided. So this con- same layer should be equal to one, which should be realized
straint can be expressed as the artificial density of 5∘ candidate layer-wise for laminated composites with multiple layers.
material in the outer surface is zero and the same as in the The DMOnC normalization constraint is expressed as
inner surface i.e., xi;1;c ¼ 0; xi;Nlay ;c ¼ 0; ðc∈½5∘ Þ. The sepa- cand

rated two equality constraints can be compounded as one ∑N

c¼1 xi; j;c ¼ 1 ð12Þ
equality constraint as (10).
The DTC constraint is defined as
xi;1;c þ xi;Nlay ;c ¼ 0; c∈½5∘  ð10Þ 4 Mathematical formulation of the optimization
problem and the structural analysis model
As an alternative strategy, the DamTol constraint can
also be realized through micro-scale DMO material inter- 4.1 Mathematical formulation of the optimization problem
polation strategy. That is to say, the candidate materials set
of the outer and inner surfaces does not contain the 5∘ We consider the concurrent multi-scale optimization of com-
candidate material. For a four layer laminate as shown in posite frame structures with the objective of minimizing the
Table 1, the damage tolerance constraint can be expressed structural compliance under specific manufacturing and total
as xi , 1 , 2 + xi , 4 , 2 = 0. volume constraints. The details of the specific manufacturing
constraints have been presented in Section 3. The macro-scale
inner tube radius (ri) of the circular cross-section and micro-
3.5 Symmetry constraint (SC)
scale artificial materials density (xi , j , c) related to the discrete
fiber winding angles are introduced as the independent design
Whenever possible, the winding sequence should be sym-
variables to realize the topology and stacking sequence opti-
metric about the mid-plane, which in the case of the com-
mization of the two geometrical scales simultaneously.
posite tube in the present paper specifically refers to the
Considering manufacturing constraints, the optimization for-
average radius plane of the tube. There are two reasons
mulation can be listed as
why this guideline is representative of a good practice:
 
(1) to uncouple bending and membrane response, and (2) Find X ¼ ri ; xi; j;c ð13Þ
to prevent warping under thermal loading. Clearly, this
guideline cannot always be rigorously enforced such as in Min C ¼ U T KU ð14Þ
8


zones where thickness is tapered. However, any asymme- >
> S:T : K De hri ; xi; j;c U ¼ i F
>
>
try existence due to manufacturing constraints should be >
> V ð r i Þ ¼ ∑N
tub
π t tot 2
þ 2r t tot
Li ≤ V
>
> i¼1 i i i
minimized. <
Manufacturing Constraints ðCC; TPR; BC; DTC; SC; DMOnC Þ
This constraint can be expressed as follows. > ri ∈½rmin ; rmax 
>
>
>
> xi; j;c ∈½0; 1 .
xi; j;c ¼ xi;Nlay − jþ1;c ð11Þ >
>
>
: lay
i ¼ 1; 2; …; N ; j ¼ 1; 2; …; N
tub
; c ¼ 1; 2; …; Ncand
2
It should be noted that, to guarantee the symmetry of the ð15Þ
layers, the fiber winding thickness is fixed at constant thick-
ness, even though fiber winding thickness is not considered as where xi , j , c is the artificial density of DMO candidate mate-
design variable. rials, ri is the inner radius of the composite tube, and t tot
i is the
Concurrent multi-scale design optimization of composite frames

   
total layer thickness of the tube. The subscripts i, j and c max 0; xui; j;c −δuxi; j;c ≤ xuþ1
i; j;c ≤ min xi; j;c þ δxi; j;c ; 1 ∀i; j ð16bÞ
u u
denote the number of tube, layer and candidate material, re-
spectively. Ntub, Nlay and Ncand denote the total number of
tubes, layers and candidate materials, respectively. V  is the Let O(u) denote the oscillation indicator for iteration (u)
allowable volume of the macro-design domain and Li is the such that
length of the tube. rmin (rmin = 0.1mm in the present paper) is a
i −ri
ru−1 u−2
small positive value to avoid singularity of the stiffness matrix Oðrui Þ ¼ ð17aÞ
rui −ru−1
i
during optimization iterations. rmax is the upper bound of the
inner radius. As mentioned, in the optimization model, the
i; j;c −xi; j;c
xu−1 u−2
DamTol constraint (DTC) and symmetry constraint (SC) are Oðxui; Þj;c ¼ ð17bÞ
realized through the micro-scale DMO material interpolation xui; j;c −xu−1
i; j;c
strategy and the association of design variables. Because of
the symmetry constraints applied on the micro-scale design As mentioned, the move limits δuri and δuxi; j;c are being
variables, only half of layers are considered as design vari- adjusted according to a certain criterion. The reduction or
lay
.
ables, and the range of number of layers is j ¼ 1; 2; …; N expansion of the move limits depends on the oscillation
2:
indicator. If Oðrui Þ or Oðxui; Þj;c is less than 0, then δuri ¼ δuri ∙β2 ,
. .
δu
δuxi; j;c ¼ δuxi; j;c ∙β 2 , else δuri ¼ δri , δuxi; j;c ¼ xi; j;c . Here, β
u
4.2 Structural analysis of composite frame
β β

The response of the composite frames is analyzed based on an is the move limit expansion or recovery factor. In this
extension of the beam finite element tool called BEam Cross work, β = 0.7 is found appropriate according to our numer-
section Analysis Software (BECAS) developed by Blasques ical experiences.
and Lazarov (2012). BECAS is an analysis tool of cross sec-
tions for anisotropic and inhomogeneous beam sections with 4.4 Continuation strategy
arbitrary geometry. It has been successfully used by Blasques
and Stolpe (2012) and Blasques (2014) to develop the multi- In order to obtain discrete designs at the micro-scale, i.e., a
material topology optimization of wind turbine blades with distinct selection of one of the candidates in every layer, a
respect to static and frequency design. In the present paper, continuation strategy for the penalization parameter p in
the BECAS analysis tool is extended to be applied to the (2) is adopted in this paper. The initial penalty parameter
composite frame structures combined with DMO discrete ma- p is set as p = 1. It has been shown by Hvejsel and Lund
terial interpolation scheme. For the detailed description about (2011) that the value of the penalty factor p larger than p =
BECAS, please refer to the references (Blasques and Lazarov 3 will not help too much to penalize intermediate values of
2012; Blasques and Stolpe 2012; Blasques 2014). the design variables. So in the present paper the power p is
linearly increasing with a slope of 0.1 of every ten itera-
4.3 SLP and move limit strategy tions from 1 to 3. Numerical examples show that this ap-
proach is effective.
The optimization problems to solve contain many linear con-
straints, which can be efficiently handled using a SLP
(Sequential Linear Programming) approach. Thus, SLP is ap- 5 Design sensitivity analysis
plied in this paper, and the approach is implemented in the
Matlab environment. Without precautions, a SLP approach is In order to perform gradient-based optimization, gradients
generally subject to oscillating function and design variable should be obtained efficiently. Due to its ease of derivation
values, and a move limit strategy is required to accommodate and implementation, the semi-analytical method (SAM)
inevitable oscillations. Let δuri and δuxi; j;c denote the move limits (Lund 1994; Blasques and Stolpe 2011) is adopted instead
for variables ri and xi , j , c, respectively. Let u denote the itera- of deriving and implementing analytical sensitivities in this
tion number. The initial move limits are set to δ0ri ¼ 0:01 and work. The SAM is computationally efficient and thus often
δ0xi; j;c ¼ 0:1. With the optimization iteration, the macro and used for the sensitivity analysis of finite element models.
This section only presents the compliance sensitivity anal-
micro move limits are changed according to a certain criterion
ysis with respect to micro design variable xi , j , c. The sen-
described below. Then the move limit strategy can be
sitivity of the compliance with respect to macro-scale de-
expressed as
sign variable ri can be obtained in a similar procedure.
   
Assume the applied static loads are design independent,
max rmin ; rui −δuri ≤ ruþ1
i ≤ min rui þ δuri ; rmax ∀i ð16aÞ
then the sensitivity of the objective function (i.e., the
 J. Yan et al.

structure compliance C) in (14) with respect to the micro- Table 2 Material

properties of the E11 143 GPa
scale design variable xi , j , c is given as
uni-directional glass
reinforced epoxy E22 = E33 10 GPa
∂C ∂U Te ∂K e ∂U e
ð18Þ
ele
¼ ∑N K e U e þ U Te Ue þ Ke G12 6 GPa
∂xi; j;c e¼1
∂xi; j;c ∂xi; j;c ∂xi; j;c
G13 5 GPa
where Ue is the displacement vector of element e. Ke is the G23 3 GPa
corresponding element stiffness matrix of element e. υ12 0.3
Furthermore, making use of the equilibrium conditions υ13 0.2
KU = F and assuming design independent loads, (18) can be υ23 0.52
simplified (see Bendsøe and Sigmund 2003) as ρ 1800 kg/m3

∂C ele
T ∂K e
¼ −∑N
e¼1 U e Ue ð19Þ
∂xi; j;c ∂xi; j;c The manufacturing constraints in the present paper are for-
mulated as series of linear inequalities or equalities. Thus, the
It is possible to rewrite (19) using the element stiffness sensitivities of all manufacturing constraints are given explic-
matrix given by K e ¼ ∫Ωe BT De BdΩe where B is the strain- itly and are easy to derive and implement.
displacement matrix and Ωe is the volume of the e’th finite
element:


Nele T e T ∂De xi; j;c ; ri
∂C 6 Numerical examples
¼ −∑e¼1 U e ∫Ω B BdΩe U e ð20Þ
∂xi; j;c ∂xi; j;c
In this section, the classical 10-beam and large-scale thirty-
∂De ðxi; j;c ;ri Þ five-beam composite frame structures have been investigated,
In the current implementation, the sensitivities ∂xi; j;c the 10-beam frame structure is Example 1. In these two nu-
are determined by central differences. The SAM approach merical examples, considering the engineering practical appli-
is computationally more efficient than the OFD (Overall cation, we assume every composite tube has the same number
Finite Difference) method, because the factorization of of layers i.e., Nlay = 20, and every layer has a constant thick-
global stiffness matrix, which is the most time consuming
i =N ¼ 0:1mm, such that the total thickness t tot
lay
ness i.e., t tot i is
part in the computation, is only calculated once for N de- 2mm.
sign variables. In the OFD using forward differences, the The fiber candidate materials are glass fiber reinforced ep-
stiffness matrix needs to be assembled and factored N + 1 oxy with orthotropic properties as shown in Table 2.
times for N design variables. Thus the semi-analytical In order to clearly demonstrate the optimization problem,
method is much more efficient. Then the sensitivities of four optimization models labeled as CMsMC1, CMsMC2,
∂De ðxi; j;c ;ri Þ
∂xi; j;c are calculated as MACs and MICsMC are studied in Example 1. For




 CMsMC1 and CMsMC2 models, to investigate the effects of
∂De xi; j;c ; ri De xi; j;c þ Δxi; j;c ; ri −De xi; j;c −Δxi; j;c ; ri the contiguity constraint parameter CL (contiguity limit), in
≈
∂xi; j;c 2Δxi; j;c CMsMC1 optimization model the CL is set as CL = 1, and in
ð21Þ CMsMC2 optimization model the CL is set as CL = 2. The
optimization model of single macro-scale (tube inner radius
where Δxi , j , c is a small perturbation parameter of the micro- ri) without considering manufacturing constraints is labeled as
scale design variable. The sensitivity of the compliance with MACs, and the optimization model of single micro-scale (can-
respect to macro-scale design variable ri can be obtained in a didate material density xi , j , c) considering manufacturing
similar semi-analytical procedure. For each of the macro-scale
design variables ri, perturbed finite element meshes are gen-
erated for the BECAS cross sectional analysis tool, and then
∂De ðxi; j;c ;ri Þ
the sensitivities ∂ri are obtained by central difference
approximations. The global volume constraint in (15) is only a
function of the macro radius design variables. The sensitivity
of the volume with respect to the radius ri of the frame is easily
obtained as
∂V ðri Þ
¼ 2πt tot
i Li ð22Þ
∂ri Fig. 3 10-beam composite frame structure
Concurrent multi-scale design optimization of composite frames

Table 3 Optimization results of 10-beam composite frame structure

Optimized macro Objective function Convergence measure

Models
structure value .

CMsMC1 0.4885 100%

CMsMC2 0.4336 100%

MACs 0.6759 Not relevant

MICsMC 0.7187 100%

constraints is labeled as MICsMC. From the above four opti- any number between 0 and 1, but in general the values should
mization models, we can gain insight into the interaction ef- be chosen such that the initial weight is uniform, i.e., ωi , j ,
c = ωi , j , k (k ≠ c) for all i, j, c, k=1 , 2⋯ N
cand
fects of the structure and the layup of composite material, and . In this way no
the benefits of the concurrent multi-scale optimization. candidate material is favored a priori. With consideration of
In MACs model, fiber winding angles of all the layers are the DamTol constraint, the initial outer layer values are xi , 1 ,
fixed at a constant angle θi , j = 14.6° and the initial value of the c = 0.33, and other layer values are xi , j ≠ 1 , c = 0.25. In the con-
inner radius is 25 mm i.e., rinit = 25mm. The MACs model current multi-scale optimization models (CMsMC 1 and
thereby has the same initial objective function as the other CMsMC2) the initial macro- and micro-scale design variables
models. Here, rmin is the lower limit of the inner radius of are similar with those in MACs and MICsMC, respectively.
the tube, and as mentioned previously we adopt r min =
0.1mm and rmax = 0.1m in the Example 1. When the radius 6.1 Example 1
reaches this limit, we assume that the tube can be deleted. θi , j
is the fiber winding angle of the j’th layer in the i’th tube. In The loading/boundary conditions and geometric sizes with
MICsMC optimization, with consideration of all the tube number are shown in Fig. 3.
manufacturing constraints mentioned in Section 3, the fiber With consideration of all the manufacturing constraints in
winding angles are considered as the design variables. ri is Section 3, the number of design variables in every tube is 40 in
fixed at a constant value i.e., ri = 25mm. The initial values of CMsMC1 and CMsMC2 optimization models, for the exam-
the micro-scale design variables, xi , j , c, may in principle be ples presented in this paper, which contains 1 sizing design
variable (ri) and 39 candidate material density design variables
(xi , j , c). It should be noted that, considering the symmetry
constraints, the number of candidate material density design
 lay . *
variables (xi , j , c) is N Ncand , and the DamTol con-
2
straint is realized through micro-scale DMO material interpo-
lation strategy. Then, the real number of micro-scale design
 lay . *
variables is N Ncand −1, that is 10*4–1 = 39. Then the
2

Table 4 The optimized radius of tubes of MACs optimization model

(θi , j = 14.6°)

Beam number 1 2 3 4 5

Radius, m 0.0379 0.0356 0.0776 rmin rmin

Beam number 6 7 8 9 10
Radius, m rmin 0.0485 rmin 0.0507 rmin
Fig. 4 Iteration history of the objective function of Example 1
 J. Yan et al.

Table 5 The detailed optimization results of MICsMC model (ri = 25mm) optimization problem with linear constraints. This makes it
Beam number Fiber winding angle, ° possible to solve the concurrent multi-scale optimization mod-
el considering specific manufacturing constraints proposed in
1 (−45/5/45/5/45/5/85/5/−45/5)sa the present paper.
2 (−45/5/45/5/45/5/85/5/−45/5)s
3 (−45/5/45/5/45/5/85/5/−45/5)s
4 (−45/5/45/5/45/5/85/5/−45/5)s 6.2 Optimization results of example 1
5 (45/5/−45/5/85/5/45/5/−45/5)s
6 (−45/5/45/5/45/5/85/5/−45/5)s The comparison of the macro-scale optimized configurations,
7 (−45/5/45/5/45/5/85/5/−45/5)s the value of the objective function and DMO convergence
8 (−45/5/45/5/45/5/85/5/−45/5)s measure of the above four different optimization models are
9 (−45/5/45/5/45/5/85/5/−45/5)s presented in Table 3. For MACs model, which only considers
10 (−45/5/45/5/45/5/85/5/−45/5)s the inner tube radius ri as design variables, the convergence
measure Hε = 0.95 given by (5) is not relevant. The iteration
a
s represents symmetric layers history is illustrated in Fig. 4. The values of objective func-
tions C in (14) are normalized with respect to the initial ob-
total number of design variables is 400 for this 10-beam ex- jective function, and the four models have the same initial
ample. There are 7 kinds of constraints, one is volume con- value of the objective function.
straint, and the others are manufacturing constraints. The sym- The detailed optimal results of macro and micro-scale de-
metry constraint is realized by the technique of design variable signs are provided in Tables 4, 5, and 6. Table 4 presents the
linking. So in each tube, there are 36 CC constraints when macro-scale optimized results of the inner tube radius with the
contiguity limit is 1, 4 TPR (10% rule) constraints, 10 micro-scale variables fixed at θi , j = 14.6°. Table 5 presents the
DMOnC constraints and 1 BC constraint. Generally, the com- micro-scale optimization results of the MICsMC model when
putational effort for optimization is non-linearly increasing as the macro variables are fixed at ri = 25mm. Table 6 presents
the constraint number increases. For the 10- beam example in the two-scale optimization results of the CMsMC 1 and
the present study, each tube has 51 constraints leading to the CMsMC2 models. In Tables 4 and 6, the label rmin denotes
total 510 constraints which will be larger for a frame com- that the macro radius has reached its lower limit. It is worth
posed with more beams. But for present paper’s concurrent noting that when the radius reaches its lower limit, the layer
multi-scale optimization model with considering specific thicknesses are still existing with very little total thickness
manufacturing constraints, all these specific manufacturing i ¼ 2mm. Therefore, we calculate the optimum structural
t tot
constraints have been simplified as explicit linear constraints, compliance with and without the minimum radius tubes (r =
and then the sensitivity of these constraints with respect to the rmin) for CMsMC1, CMsMC2 and MACs models, respective-
micro-scale design variable xi , j , c will be easily and explicitly ly. The analysis results are shown in Table 7. The values of the
obtained. Meanwhile, the SLP optimization algorithm compliance in the table are true values and have not been
(Mehrotra 1992; Zhang 1998) can efficiently solve the normalized with respect to the initial objective function.

Table 6 The detailed optimization results of CMsMC1 and CMsMC2 model of Example 1

Beam number Optimized macro variables ri, m Fiber winding angle, °

CMsMC1 model CMsMC2 model CMsMC1 model CMsMC2 model

1 0.0395 0.0388 (−45/5/45/5/45/5/85/5/−45/5)s* (45/5/5/85/5/5/−45/5/85/5)sa

2 0.0345 0.0343 (−45/5/45/5/45/5/85/5/−45/5)s (85/5/5/85/5/5/−45/5/5/45)s
3 0.0749 0.0749 (−45/5/45/5/45/5/85/5/−45/5)s (45/5/5/85/5/5/−45/5/85/5)s
4 rmin rmin (45/5/45/5/85/5/−45/5/−45/5)s (45/5/5/85/5/5/−45/5/85/5)s
5 rmin rmin (45/5/45/5/85/5/−45/5/−45/5)s (45/5/5/85/5/5/−45/5/85/5)s
6 rmin rmin (45/5/45/5/85/5/−45/5/−45/5)s (45/5/5/85/5/5/−45/5/85/5)s
7 0.0469 0.0487 (−45/5/45/5/45/5/85/5/−45/5)s (85/5/5/85/5/5/−45/5/5/45)s
8 0.0011 0.0008 (−45/5/45/5/45/5/85/5/−45/5)s (45/5/5/85/5/5/−45/5/85/5)s
9 0.0504 0.0510 (45/5/45/5/85/5/−45/5/−45/5)s (45/5/5/85/5/5/−45/5/85/5)s
10 rmin rmin (−45/5/45/5/45/5/85/5/−45/5)s (85/5/5/85/5/5/−45/5/5/45)s
a
s represents symmetric layers
Concurrent multi-scale design optimization of composite frames

Table 7 Comparison of compliance values including and excluding because the contiguity limit is 2 in the CMsMC2 model, which
minimum radius tubes
relaxes the constraints on micro-scale design variables and
Models Compliance of the optimum structure Deviation enlarges the design domain compared to the CMsMC1 model.
Table 3 shows the optimized configurations of macro-
Without the rmin tubes With the rmin tubes structures based on the four optimization models. The
CMsMC1 2488.1 2454.3 1.36%
CMsMC1, CMsMC2 and MACs models almost have the same
optimized macroscopic structure configuration in which the
CMsMC2 2143.5 2117.0 1.24%
design variables of tubes 4, 5, 6 and 10 have reached the lower
MACs 2719.9 2686.8 1.21%
limit of their cross-sectional radius. The optimized macro con-
figurations comply with the loading condition from the view
From Table 7 we observe that, for three kinds of optimization of structural analysis. An interesting observation is that, in
models, the largest difference of the compliance with and CMsMC1 and CMsMC2 models, tube 8 is maintained with a
without the minimum radius tubes in the models is 1.36% of small radius value, which indicates that the material distribut-
CMsMC1 model. That means the contribution from the min- ed on the eighth tube can further improve the structural per-
imum radius tubes to the overall structure stiffness is very formance, while in MACs model tube 8 is deleted from the
small, and thus these minimum radius tubes can be deleted ground structure. It also reflects the impact of micro-scale
from the ground structure to realize topology optimization on design variables on macro-scale structural topology. Here it
macro-scale. should be noted that, in MACs optimization model, the micro
fiber winding angles are fixed at θi , j = 14.6° to guarantee the
MACs model has the same initial objective function as other
6.3 Discussion of example 1 models. Of course, with different fixed micro fiber winding
angles, the optimized macro configuration of the MACs mod-
From the optimization iteration history shown in Fig. 4, with el will be different, but it is impossible for engineers to give
the same material volume, the objective function values of the the optimal initial fiber winding angles directly for a complex
four types of optimization model i.e., CMsMC1, CMsMC2, structure which shows the necessity of an optimization model
MACs, and MICsMC with respect to the initial values de- for the composite structure.
crease by 51.15%, 56.64%, 32.41% and 28.13%, respectively. Observing the micro-scale design variables from Tables 5
We can intuitively observe that when the minimum compli- and 6, only fiber winding angles of the first ten layers are listed
ance is applied as the objective function, the values of the due to the symmetry constraints adopted for the micro-scale
objective function of concurrent multi-scale optimization are design variables. Firstly, because the outer and inner layers do
significantly better than those from the single-scale optimiza- not contain the 5∘ candidate material, there is no 5∘ ply placed
tions. Furthermore, the structural performance improvements in the outer layer of the laminate in Tables 5 and 6. Secondly,
are approximately 24 ~ 28%. This is reasonable, because when contiguity limit equals one (CL = 1), all arbitrary conti-
CMsMC1 and CMsMC2 models can account for the coupled guity layers have different fiber winding angles, as shown in
effects of the macro-structure topology together with the the micro-scale design variables of the MICsMC and
micro-material selection. Conversely, the single-scale MACs CMsMC1 model. That means all the adjacent fiber winding
and MICsMC models can only improve the performance of angles are different. With CL = 2 in CMsMC2 model, more 5∘
the structure from the macro- or micro-scale. Thus the inter- ply layers are selected than in CMsMC1 and MICsMC
action between the structure and material cannot be consid- models. CMsMC2 model relaxes the constraints on micro-
ered. In this numerical example, from the value of the objec- scale design variables and enlarges the design domain. 5∘
tive function, the MACs and MICsMC models result in quite ply layers are beneficial to improve the axial stiffness of the
similar structural stiffness with completely different topolo- structure with respect to the loading case in the present exam-
gies. It should be noted specially that in the concurrent ple, then the CMsMC2 model can obtain a lower objective
multi-scale optimization model, the CMsMC2 model can ob- function value. However, the larger CL may lead to crack
tain a better design than that from the CMsMC1 model. This is propagation in the laminate ultimately. So in most engineering

Fig. 5 Micro-scale fiber winding

angle with manufacturing
constraints

(a) CMsMC1 model (b) CMsMC2 model

 J. Yan et al.

Fig. 6 35-beam composite frame

structure

applications the CL is settled as 2 ~ 3, especially in aerospace 6.5 Optimization results and discussion of example 2
engineering. Finally, if without the 10% rule constraints, the
85∘ fiber winding angles may not appear. The 10% rule effec- The iteration history of CMsMC2 model is illustrated in Fig. 7.
tively avoids a single fiber angle to dominate excessively to The value of objective function C in (14) is also normalized
make the laminate more robust in the sense that they are less with respect to the initial objective function. Table 8 presents
susceptible to the weaknesses associated with highly the detailed two-scale optimization results of the CMsMC2
orthotropic laminates. model. Figure 8 shows the optimized macro-scale topology
Take the third tube as an example. Figure 5 gives the de- configuration of the CMsMC2 model.
scription of the different manufacturing constraints on micro- From the iteration history of the objective function shown
scale design variables. in Fig. 7, with respect to the initial values, the objective func-
tion values of CMsMC2 model has decreased by 69.87%.
6.4 Example 2 Figure 8 shows that the CMsMC2 model obtains a symmetri-
cal optimized macroscopic structure configurations.
The loading/boundary conditions and geometric sizes with From Table 8, the micro-scale fiber winding angles are
tube number are shown in Fig. 6, and L = 1m. Example 1 strictly following the specific manufacturing constraints for
has detailed discussions on the pros and cons of the this 35-beam frame example. The detailed discussion about
CMsMC 1 , CMsMC 2 , MACs and MICsMC models. the manufacturing constraints is the same with that for
Therefore, Example 2 only considers the multi-scale
CMsMC2 model to further verify the effectiveness of the con-
current multi-scale design optimization method for composite
frames considering specific manufacturing constraints.
For CMsMC2 model in the Example 2, the number of
design variables in each tube is 40, which is the same with
that in CMsMC2 in Example 1. Then the total number of
design variables is 1400 for this 35-beam example. The sym-
metry constraint is realized by the technique of design variable
linking. So in each tube, there are 32 CC constraints when
contiguity limit is 2, 4 TPR (10% rule) constraints, 10
DMOnC constraints and 1 BC constraint. Then the number
of specific manufacturing constraint is 47 in each tube, which
leads to the total 1645 manufacturing constraints. Besides the
upper bound of the inner radius is rmax = 0.075 m in the
Example 2, all the parameters, initial value of design variables
and the material properties are the same with the CMsMC2 Fig. 7 Iteration history of the objective function of Example 2 of 35-beam
model in Example 1. composite frame structure with contiguity limit of 2
Concurrent multi-scale design optimization of composite frames

Table 8 The detailed optimization results of the CMsMC2 model of Example 1. And we could clearly observe that, the proposed
Example 2
concurrent multi-scale design optimization method for com-
Beam Optimized macro variables Fiber winding angle, ° posite frame is effective to solve a problem with relatively
number ri, m larger number of constraints.
As seen from the above discussion and summary in two
1 0.0378 (45/5/5/85/5/5/−45/5/85/5)sa
2 0.0166 (85/5/5/85/5/5/−45/5/5/45)s examples, the proposed concurrent multi-scale optimization
3 rmin (45/5/5/85/5/5/−45/5/85/5)s model for composite frames can efficiently realize the optimi-
4 0.0384 (45/5/5/85/5/5/−45/5/85/5)s zation on two geometrical scales and obtain better results than
5 rmin (45/5/5/85/5/5/−45/5/85/5)s
6 0.0173 (85/5/5/85/5/5/−45/5/5/45)s that from a single-scale optimization. Based on the DMO ap-
7 rmin (85/5/5/45/5/85/5/5/−45/5)s proach, some specified important manufacturing constraints
8 0.0403 (85/5/5/85/5/5/−45/5/5/45)s have been mathematically expressed and numerically solved,
9 rmin (45/5/5/85/5/5/−45/5/85/5)s
10 rmin (85/5/5/85/5/5/−45/5/5/45)s
and the optimization results show that the manufacturing con-
11 rmin (85/5/5/85/5/5/−45/5/5/45)s straints are strictly observed.
12 rmin (45/5/5/85/5/5/−45/5/85/5)s
13 0.0400 (85/5/5/85/5/5/−45/5/5/45)s
14 0.0557 (85/5/5/85/5/5/−45/5/5/45)s
15 0.0750 (85/5/5/45/5/85/5/5/−45/5)s
16 rmin (85/5/5/85/5/5/−45/5/5/45)s 7 Conclusion
17 0.0583 (85/5/5/85/5/5/−45/5/5/45)s
18 rmin (85/5/5/85/5/5/−45/5/5/45)s
19 rmin (45/5/5/85/5/5/−45/5/85/5)s Concurrent multi-scale design optimization of composite
20 0.0750 (45/5/5/85/5/5/−45/5/85/5)s frames is established in this paper using the Discrete
21 rmin (85/5/5/85/5/5/−45/5/5/45)s Material Optimization (DMO) approach with consideration
22 rmin (85/5/5/85/5/5/−45/5/5/45)s
23 0.0750 (85/5/5/45/5/85/5/5/−45/5)s of several different manufacturing constraints. The capabili-
24 rmin (85/5/5/85/5/5/−45/5/5/45)s ties of the proposed method are demonstrated for compliance
25 rmin (85/5/5/85/5/5/−45/5/5/45)s minimization subject to a constraint of specified composite
26 0.0749 (45/5/5/85/5/5/−45/5/85/5)s
27 0.0592 (85/5/5/85/5/5/−45/5/5/45)s volume. Furthermore, an extensive set of design guidelines
28 rmin (45/5/5/85/5/5/−45/5/85/5)s referred to as manufacturing constraints are considered in the
29 rmin (85/5/5/85/5/5/−45/5/5/45)s optimization model. These manufacturing constraints are ex-
30 0.0550 (85/5/5/85/5/5/−45/5/5/45)s
31 rmin (85/5/5/85/5/5/−45/5/5/45)s
plicitly expressed as series of linear inequalities or equalities
32 0.0750 (85/5/5/45/5/85/5/5/−45/5)s in the optimization model and efficiently solved by a SLP
33 0.0750 (85/5/5/45/5/85/5/5/−45/5)s optimization algorithm with move limit strategy and semi-
34 0.0750 (85/5/5/45/5/85/5/5/−45/5)s
35 0.0750 (85/5/5/45/5/85/5/5/−45/5)s
analytical sensitivity analysis. With consideration of the de-
sign guidelines, it can help to reduce the risk of structural and
a
s represents symmetrical layers material failure and the complexity of the optimal design.

Fig. 8 Optimized topology

configuration of CMsMC2 model
 J. Yan et al.

Numerical results show that the concurrent multi-scale op- Duan ZY, Yan J, Zhao GZ (2014) Integrated optimization of the material
and structure of composites based on the Heaviside penalization of
timization of composite frames can further explore the poten-
discrete material model. Struct Multidiscip Optim 51(3):721–732
tial of macro-structure and micro-material to achieve light Eschenauer HA, Olhoff N (2001) Topology optimization of continuum
weight design of composite frames. The two-scale optimiza- structures: A review. Appl Mech Rev 54(4):331–390
tion model provides a new choice for the design of composite Ferreira RTL, Rodrigues HC, Guedes J, Hernandes JA (2013)
frames in aerospace and other industries. In future work, the Hierarchical optimization of laminated fiber reinforced composites.
Compos Struct 107:246–259
concurrent multi-scale optimization of composite frame struc- Fletcher R, Leyffer S, Toint PL (1998) On the global convergence of an
tures with variable cross-section, thickness and frequency SLP-filter algorithm. Numerical Analysis Report NA/183,
constraint problem will be explored. University of Dundee, UK, August
Gao T, Zhang W, Duysinx P (2012) A bi-value coding parameterization
scheme for the discrete optimal orientation design of the composite
Acknowledgments Financial supports for this research were provided laminate. Int J Numer Methods Eng 91(1):98–114
by the National Natural Science Foundation of China (No. 11372060 and Gao T, Zhang WH, Duysinx P (2013) Simultaneous design of structural
11672057), Program (LJQ2015026) for Excellent Talents at Colleges layout and discrete fiber orientation using bi-value coding parame-
and Universities in Liaoning Province, the 111 project (B14013), terization and volume constraint. Struct Multidiscip Optim 48(6):
the Fundamental Research Funds for the Central Universities 1075–1088
(DUT16ZD215), and the Program of BK21 Plus. These supports are Ghiasi H, Pasini D, Lessard L (2009) Optimum stacking sequence design
gratefully acknowledged. of composite materials Part I: Constant stiffness design. Compos
Struct 90(1):1–11
Ghiasi H, Fayazbakhsh K, Pasini D, Lessard L (2010) Optimum stacking
sequence design of composite materials Part II: Variable stiffness
References design. Compos Struct 93(1):1–13
Gomes FA, Senne TA (2011) An SLP algorithm and its application to
topology optimization. Comput Appl Math 30(1):53–89
Bailie JA, Ley RP, Pasricha A (1997) A summary and review of compos-
Hvejsel CF, Lund E (2011) Material interpolation schemes for unified
ite laminate design guidelines. Technical report NASA, NAS1–
topology and multi-material optimization. Struct Multidiscip
19347. Northrop Grumman-Military Aircraft Systems Division
Optim 43(6):811–825
Baker AA, Dutton SE, Kelly DW (2004) Composite materials for aircraft
Hvejsel CF, Lund E, Stolpe M (2011) Optimization strategies for discrete
structures, 2nd edn. American Institute of Aeronautics and
multi-material stiffness optimization. Struct Multidiscip Optim
Astronautics
44(2):149–163
Bakis CE, Bank LC, Brown VL, Cosenza E, Davalos JF, Lesko JJ,
Ibrahim S, Polyzois D, Hassan S (2000) Development of glass fiber
Machida A, Rizkalla SH, Triantafillou TC (2002) Fiber-reinforced
reinforced plastic poles for transmission and distribution lines. Can
polymer composites for construction-state-of-the-art review. J
J Civ Eng 27(5):850–858
Compos Constr 6(2):73–87
Irisarri FX, Lasseigne A, Leroy FH, Riche RL (2014) Optimal design of
Bendsoe MP (1989) Optimal shape design as a material distribution laminated composite structures with ply drops using stacking se-
problem. Struct Multidiscip Optim 1(4):193–202 quence tables. Compos Struct 107:559–569
Bendsøe M, Sigmund O (2003) Topology optimization-Theory, Methods Kassapoglou C (2013) Design and analysis of composite structures: with
and Applications, 2nd edn. Springer-Verlag, Berlin Heidelberg applications to aerospace structures, 2nd edn. Sons, John Wiley &
Blasques JP (2014) Multi-material topology optimization of laminated Liu L, Yan J, Cheng GD (2008) Optimum structure with homogeneous
composite beams with eigenfrequency constraints. Compos Struct optimum truss-like material. Comput Struct 86(13):1417–1425
111:45–55 Liu DZ, Toroporov VV, Querin OM, David CB (2011) Bilevel optimiza-
Blasques J, Lazarov B (2012) User's manual for BECAS: a cross section tion of blended composite wing panels. J Aircr 48(1):107–118
analysis tool for anisotropic and inhomogeneous beam sections of Lund E (1994) Finite element based design sensitivity analysis and opti-
arbitrary geometry. Risø DTU–National Laboratory for Sustainable mization. Institute of Mechanical Engineering, Aalborg University,
Energy Denmark
Blasques JP, Stolpe M (2011) Maximum stiffness and minimum weight Lund E (2009) Buckling topology optimization of laminated multi-
optimization of laminated composite beams using continuous fiber material composite shell structures. Compos Struct 91(2):158–167
angles. Struct Multidiscip Optim 43(4):573–588 Lund E, Stegmann J (2005) On structural optimization of composite shell
Blasques JP, Stolpe M (2012) Multi-material topology optimization of structures using a discrete constitutive parametrization. Wind
laminated composite beam cross sections. Compos Struct 94(11): Energy 8(1):109–124
3278–3289 Mallick PK (2007) Fiber-reinforced composites: materials, manufactur-
Bruyneel M (2011) SFP - a new parameterization based on shape func- ing, and design. CRC press
tions for optimal material selection: application to conventional Manne PM, Tsai SW (1998) Design optimization of composite plates:
composite plies. Struct Multidiscip Optim 43(1):17–27 Part II—structural optimization by plydrop tapering. J Compos
Bruyneel M, Beghin C, Craveur G, Grihon S, Sosonkina M (2012) Mater 32(6):572–598
Stacking sequence optimization for constant stiffness laminates Mehrotra S (1992) On the implementation of a primal-dual interior point
based on a continuous optimization approach. Struct Multidiscip method. SIAM J Optim 2(4):575–601
Optim 46(6):783–794 Niu B, Yan J, Cheng GD (2009) Optimum structure with homogeneous
Costin DP, Wang BP (1993) Optimum design of a composite structure optimum cellular material for maximum fundamental frequency.
with manufacturing constraints. Thin-Walled Struct 17(3):185–202 Struct Multidiscip Optim 39(2):115–132
Deng JD, Yan J, Cheng GD (2013) Multi-objective concurrent topology Niu B, Olhoff N, Lund E, Cheng GD (2010) Discrete material optimiza-
optimization of thermoelastic structures composed of homogeneous tion of vibrating laminated composite plates for minimum sound
porous material. Struct Multidiscip Optim 47(4):583–597 radiation. Int J Solids Struct 47(16):2097–2114
Concurrent multi-scale design optimization of composite frames

Rodrigues H, Guedes JM, Bendsoe M (2002) Hierarchical optimization Stolpe M, Svanberg K (2001) An alternative interpolation scheme for
of material and structure. Struct Multidiscip Optim 24(1):1–10 minimum compliance topology optimization. Struct Multidiscip
Schutze R (1997) Lightweight carbon fibre rods and truss structures. Optim 22(2):116–124
Mater Des 18(4–6):231–238 Wang BP, Costin DP (1992) Optimum design of a composite struc-
Seresta O, Gurdal Z, Adams DB, Watson LT (2007) Optimal design of ture with three types of manufacturing constraints. AIAA J
composite wing structures with blended laminates. Compos Part B 30(6):1667–1669
38(4):469–480 Yan J, Hu WB, Wang ZH, Duan ZY (2014) Size effect of lattice
Sigmund O, Torquato S (1997) Design of materials with extreme thermal material and minimum weight design. Acta Mech Sinica 30(2):
expansion using a three-phase topology optimization method. J 191–197
Mech Phys Solids 45(6):1037–1067 Yan J, Yang SX, Duan ZY, Yang CQ (2015a) Minimum compliance
Sørensen SN, Lund E (2013) Topology and thickness optimization of optimization of a thermoelastic lattice structure with size-coupled
laminated composites including manufacturing constraints. Struct effects. J Therm Stresses 38(3):338–357
Multidiscip Optim 48(2):249–265 Yan J, Hu WB, Duan ZY (2015b) Structure/material concurrent
Sørensen SN, Sørensen R, Lund E (2014) DMTO–a method for discrete optimization of lattice materials based on extended multiscale
material and thickness optimization of laminated composite struc- finite element method. Int J Multiscale Comput Eng 13(1):
tures. Struct Multidiscip Optim 50(1):25–47 73–90
Stegmann J, Lund E (2005) Discrete material optimization of general Zhang Y (1998) Solving large-scale linear programs by interior-point
composite shell structures. Int J Numer Methods Eng 62(14): methods under the Matlab environment. Optim Methods Softw
2009–2027 10(1):1–31

View publication stats