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

net/publication/367254872

COUPLED INSTABILITY PROBLEM OF SOLIDS WITH PERIODIC MICRO-

STRUCTURE: SYMBOLIC COMPUTATION AND FINITE ELEMENT MODEL

Conference Paper · March 2021

CITATIONS READS

0 38

2 authors:

Mariana Figueiredo Miguel Matos Neves

Instituto Superior Técnico University of Lisbon
1 PUBLICATION   0 CITATIONS    93 PUBLICATIONS   859 CITATIONS   

SEE PROFILE SEE PROFILE

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

Energy-consuming Elements View project

Stability of Plates and Shells View project

All content following this page was uploaded by Miguel Matos Neves on 19 January 2023.

The user has requested enhancement of the downloaded file.

 SYMCOMP 2021
Évora, 25-26 March 2021
©ECCOMAS, Portugal

COUPLED INSTABILITY PROBLEM OF SOLIDS WITH

PERIODIC MICRO-STRUCTURE: SYMBOLIC
COMPUTATION AND FINITE ELEMENT MODEL
Mariana João Figueiredo1 and Miguel Matos Neves2∗
1: MEAer, DEM
Instituto Superior Técnico
Universidade de Lisboa
Av. Rovisco Pais 1049-001 Lisboa, Portugal
e-mail: mariana.a.figueiredo@tecnico.ulisboa.pt

2: IDMEC
Instituto Superior Técnico
Universidade de Lisboa
Lisboa
Av. Rovisco Pais 1049-001 Lisboa, Portugal
e-mail: miguel.matos.neves@tecnico.ulisboa.pt

Keywords: Multi-scale Buckling, Stability, Cellular Materials, Homogenization, Sym-

bolic computation, Finite Elements

Abstract. In this paper, the authors analyze the stability problem of structures built
from solids with periodic micro-structure, obtained from the minimization of the total po-
tential energy. From the obtained equilibrium equations (elastic stability problem at the
microscopic level, the one for coupled macroscopic and microscopic levels and the one for
macro-scale level), the authors focus this work on the second one, which establishes a con-
nection between macro-scale and micro-scale instabilities as a result of an initial stress field
at the micro-scale level resulting from a macroscopic strain field prior to bifurcation. The
implementation of the coupled instability problem uses a double discretization of the model
(one at the macro-scale and another at the micro-scale levels). In this study, a macro-
scopic element can be seen as a domain discretized in microscopic elements, meaning that
the global matrices are obtained from macro-scale elemental matrices, which are obtained
from the matrices of micro-scale elements. The objective is, then, to analyze the possibility
of implementation of such equation and check the meaning of the obtained results. A few
results are presented, which illustrate the convergence to macro-scale instability for micro-
structures of high material density and the increasing difference in buckling strength for
micro-structures of decreasing material density as well as the appearance of micro-scale
instabilities for micro-structures of low material density as higher order buckling loads are
analyzed.

149
 Mariana João Figueiredo and Miguel Matos Neves

1. INTRODUCTION
Buckling occurs when a structure is subjected to an increasing load that, after reaching
some critical value, results in a sudden change of its equilibrium configuration or shape
such as in a column under axial compression. In the case of gradually increased load, this
static equilibrium position modification causes the vanishing of the structure’s stiffness
and may occur before the structure starts to yield or fails. This means that when a
component of a given assembly buckles, the remaining ones may have to support the load
beyond this critical value (the one that caused the buckling). This makes the buckling
phenomena an important factor when assessing the design of a structure.
Many of the nowadays structural solutions employ the use of cellular materials, which
have found a wide range of applicability, including, for instance, infill structures for addi-
tive manufacturing. However, the strength capacity of these materials is also limited by
micro-structural instabilities which may occur when their slender structural members are
subjected to compressing loads ([1],[2]).
This means that structures built with periodic cellular materials can buckle on multi
spacial scale levels (macro-scale, micro-scale and mixed-scale levels).
The instability phenomena of structures built with these materials is here addressed
by means of a linearized stability theory based on a double-scale asymptotic technique
[3]. In this reference, equations describing buckling at macro-scale and micro-scale levels
and a new equation regarding mixed-scale buckling behaviour are obtained.
The stability problems at separated scale levels have already found some applicability,
for instance, in topology optimization techniques for improved buckling performance of
structures and micro-structures/materials [1].
In the context of linearized buckling strength of periodic materials, Neves et al. [4]
proposed a methodology for the maximization of the critical load for micro-structures
exhibiting micro-scale buckling modes with the same periodicity of the cellular material.
Later, this work was extend to include buckling modes of different wave-lengths by means
of the Bloch-wave theory [5], and recently a further extension which includes the complete
Bloch wave analysis is provided in [1] and recommended to read.
As for the macro-scale buckling strength of structures based on continuum models, a
methodology for the maximization of the linearized buckling load using topology opti-
mization is presented in [6].
However, while computational models as well as applications have been developed for
the linearized buckling behaviour at separated macro- and micro-scale levels (even though
not very extensively), no such implementations were found for the mixed-scale buckling
case.
This paper proposes, then, a numerical model of the couple-scale stability equation,
which is further applied to simple structural examples to compare the mixed-scale stability
behaviour to the limit-scale behaviours (macro-scale and micro-scales instabilities).

150
 Mariana João Figueiredo and Miguel Matos Neves

2. ELASTIC STABILITY THEORY FOR SOLIDS WITH PERIODIC MICRO-

STRUCTURE: RESULTS FROM SYMBOLIC COMPUTATION
The stability problem is here analyzed by means of a linearized theory based in [3], [7] and
the references therein, which formulation was fully verified in this work by the development
of symbolic computation scripts (using the Symbolic Toolbox from Matlab).
Let a linear-elastic body be defined by a domain Ωε and a boundary Γ, being quasi-
statically loaded with: prescribed displacements in Γu , surface tractions t in Γt and body
forces f in Ωε . The superscript ε indicates a dependency on the micro-structure, as it is
defined as the unit cell scale parameter, ε = Dd  1, where d is a characteristic dimension
of the unit cell and D a characteristic dimension of the structure.
The loading is gradually increased, only applied at the macro-scale level and also
considered proportional to some reference loading, t = λtref and f = λf ref .
Assuming the body has a uniform micro-structural shape, the solid can be represented
by a periodic repetition of a unit cell, defined by the domain Y =]0, Y1 [x]0, Y2 [x]0, Y3 [
and, here, built from a base material with holes in it. This base material is homogeneous,
linear-elastic and isotropic, meaning that the constitutive relations can be given by the
generalized Hooke’s law, σij (uε ) = Eijkm ekm (uε ), where σ and e are the stress and strain
tensors, respectively, and E is the constitutive matrix.
The equilibrium positions can be given by the minimization of the total potential
energy, given by
Π(uε ) = A(uε ) − R(uε ), (1)
where A(uε ) is the total elastic strain energy and R(uε ) is the applied forces potential
defined as Z
1
A(uε ) = Eijkm ekm (uε )eij (uε )dΩ, (2)
2 Ω
Z Z
R(uε ) = λ firef uεi dΩ + λ tref ε
i ui dΓ, (3)
Ω Γt

where    
ε 1 ∂uεi ∂uεj 1 ∂uεk ∂uεk
eij (u ) = + + . (4)
2 ∂xj ∂xi 2 ∂xi ∂xj
The displacement field, which is a function of both the macro-spatial, x, and micro-
spatial, y, variables, is represented using an infinitesimal real displacement parameter
α:
uε = u0ε + αu1ε , (5)
where u0ε is the displacement related with the unique primary equilibrium configuration
and u1ε is a relative displacement, which, when multiplied by α, represents the possible
”jump” to the secondary equilibrium position.

151
 Mariana João Figueiredo and Miguel Matos Neves

Based on the periodicity of the material, it is assumed that the displacement field terms
can be expressed by asymptotic expansions in terms of the scale parameter ε,
x
u0ε (x, y) = u00 (x, y) + εu01 (x, y) + ε2 u02 (x, y) + ..., y = , (6)
ε
x
u1ε (x, y) = u10 (x, y) + εu11 (x, y) + ε2 u12 (x, y) + ..., y =
, (7)
ε
where the functions uab with indices a=0,1 and b=0,1,2,... are here assumed Y-periodic
(i.e. have the same periodicity of the unit cell) although they can be generalized using a
Bloch-wave formulation (not used here).
The minimization of the total potential energy is expressed by

δΠ(uε ) = δA(uε ) − δR(uε ) = 0. (8)

Performing the differentiation of Y-periodic functions as dxd j F (x, y) = ∂F∂x (x,y) 1 ∂F (x,y)
j
+ ε ∂yj ,
considering only the first two terms of the asymptotic expansions (6) and (7) and intro-
ducing a perturbation δuε = α{v 10 (x, y) + εv 11 (x, y) + ...}, where v 10 and v 11 ∈ VΩ×Y =
{v(x, y) : v|Γu = 0 and v is Y-periodic}, equation (8) becomes (separated in terms of α
powers)
Z Z Z
α Eijkm {e0ij (uε )eIkm (v ε )}dΩ − αλ ti v εi dΓ − αλ f i v εi dΩ = 0 ∀ v ∈ VΩ×Y , (9)
Ωε Γt Ωε
Z
2
α Eijkm {e0ij (uε )eII ε 0 ε II ε I ε I ε I ε I ε
km (v ) + eij (v )ekm (u ) + eck (u )ecm (v ) + eck (v )ecm (u )}dΩ = 0
Ωε
(10)
∀ v ∈ VΩ×Y .
where e0ij (uε ), eIij (uε ) and eII ε 0 1 2
ij (u ) are the terms in α , α and α of the strain tensor (see
[3]).
The equations that characterize the small elastic deformation before bifurcation are
obtained from the analysis of equation (9). The analysis is performed by grouping the
terms in ε powers and setting each of them to zero.
From the term in ε−2 , it can be shown that

u00 (x, y) = u00 (x). (11)

i.e. the displacement u00 is only function of x.

From the term in ε−1 , acknowledging that v are independent variations that can assume
any value, ε is very small and using result (11), one obtains

∂u00
k (x)
u01 km
i (x, y) = −χi (y) . (12)
∂xm

152
 Mariana João Figueiredo and Miguel Matos Neves

To satisfy the previous equation, the characteristic displacements χkm

i must be the
solution of the km static problems at micro-scale level,
Z Z
∂χkm
i ∂vi ∂vi
Eijpq dY = Eijkm dY ∀ v ∈ VY = {v is Y-periodic}, (13)
Y ∂yq ∂yj Y ∂yj
where periodic displacements at the unit cell’s boundary are prescribed.
As for the term in ε0 , using previous results, the equation expressing macroscopic static
equilibrium is obtained ∀ v 10 ∈ VΩ ,
Z Z ! Z Z
1 ∂χkm
p ∂u00 ∂vi10
Eijkm − Eijpq dY k dΩ = λ tref
i vi
10
dΓ + λ firef vi10 dΩ,
Ωε |Y | Y ∂y q ∂x m ∂x j Γt Ωε

(14)
From (14), the homogenized elastic material properties can be defined as
Z !
km
H 1 ∂χ p
Eijkm = Eijkm − Eijpq dY. (15)
|Y | Y ∂yq

From the term in ε, it is possible to realize that

u01 (x, y) = u01 (y). (16)

i.e. the displacement u01 is only function of y.

To obtain the linearized elastic buckling response, equation (10) is analysed by setting
to zero each of the terms in different ε powers.
Starting with the term in ε−2 , considering u10 (x, y) = u10 (y) and taking into account
the periodicity of the displacements, the elastic stability problem at the micro-scale level
can be obtained:
Z Z
∂u10
i ∂vk
10 10
0 ∂uc ∂vc
10
Eijkm dY + σkm dY = 0, ∀ v 10 ∈ VΩ×#Y , (17)
#Y ∂y j ∂y m #Y ∂y k ∂y m

0 0
where σkm = σkm (x, y) is the initial stress at micro-scale level resulting from the macro-
scopic strain field prior to bifurcation and is defined as [1]
!
0
∂χkm
p ∂u00
i
σij = Eijkm − Eijpq . (18)
∂yq ∂xj

From the term in ε−1 , a new expression of the connection between the macroscopic and
microscopic instabilities is obtained:
Z  10 10  Z  10 10 
∂ui ∂vk 0 ∂uc ∂vc
Eijkm dΩ + σkm dΩ+
Ωε ∂yj ∂xm Ωε ∂ym ∂xk

153
 Mariana João Figueiredo and Miguel Matos Neves

Z   Z  
∂u10 10
i ∂vk 0 ∂u10 10
c ∂vc
+ Eijkm dΩ + σkm dΩ = 0∀ v 10 ∈ VΩ×Y . (19)
Ωε ∂xj ∂ym Ωε ∂xk ∂ym
As for the term in ε0 , assuming u01 = u01 (y), u10 = u10 (x) and that terms involving
the product of three displacement derivatives may be neglected, the macroscopic insta-
bility problem is obtained
Z Z
H ∂u10
i ∂vk
10 10
0H ∂uc ∂vc
10
Eijkm dΩ + σkm dΩ = 0 ∀ v 10 ∈ VΩ×Y , (20)
Ω ε ∂x j ∂x m Ωε ∂x k ∂x m

where
∂u00
i
H
σij0H = Eijkm . (21)
∂xj
From this theory the equations describing the linear-elastic pre-buckling response (the
elastostatic problems at separated micro-scale and macro-scale levels and the homoge-
nized elastic properties) and the equations describing the linearized buckling response
(the elastic stability problems at separated micro-scale and macro-scale levels and the
coupled-scale elastic stability problem) are obtained.

3. COUPLED-SCALE INSTABILITY EQUATION

From the aforementioned obtained equations, the authors focus this work on the equa-
tion describing the coupled-scale stability problem. This equation is first studied and
analysed through a finite element model which is after implemented numerically.

3.1. Study of equation and finite element model

From the macro-scale elastostatic problem, equation (14), it can be seen that u00
depends linearly on the load factor λ. Then, one can write u00 = λu, where u is the
displacement field associated with the reference loading that solves the the linear system
of static equilibrium equations [7]
Z Z Z
∂uk ∂vi ref
H
Eijkm dΩ = ti vi dΓ + firef vi dΩ, ∀ v ∈ VΩ . (22)
Ω ε ∂x m ∂x j Γt Ωε

Furthermore, because this is a linearized model, it is not possible to obtain the dis-
placement values; instead only the instability modes can be identified: u10 = cte φ and
v 10 = cte ψ, where φ denotes the instability modes and ψ the virtual instability modes
(test functions). By these considerations, the coupled-scale instability problem, equation
(19), can be rewritten as
Z     
∂φi ∂ψk ∂φi ∂ψk
Eijkm + Eijkm dΩ+
Ωε ∂yj ∂xm ∂xj ∂ym
Z     
0 ∂φc ∂ψc 0 ∂φc ∂ψc
+λxy σkm + σkm dΩ = 0, (23)
Ωε ∂ym ∂xk ∂xk ∂ym

154
 Mariana João Figueiredo and Miguel Matos Neves

where !
∂χkm
p ∂ui
σij0 = Eijkm − Eijpq (24)
∂yq ∂xj
and λxy denotes the coupled instability load factor.
Given a double discretization of the model (one at the macro-scale and another at
the micro-scale levels) and since in equation (23) the integrations are performed in the
macroscopic domain, the resultant global matrices can be obtained by the assembly of
macro-elemental corresponding ones. Furthermore, a macroscopic element ”can be seen”
as a domain discretized by microscopic elements. Following this line of thought, the
global matrices can be obtained by individual assemblies of microscopic matrices cor-
responding to micro-elements within a macro-element, followed by the assembly of the
obtained macroscopic matrices. Mathematically, for a given macro-scale element E, the
integrals in equation (23) become, by the same order they appear in equation (23),
Z n Z
X   Z   
∂φi ∂ψk ∂φi ∂ψk
Eijkm dY + Eijkm dY dΩ and (25)
ΩE e=1 Ye ∂yj ∂xm Ye ∂xj ∂ym
Z X n Z   Z   
0 ∂φc ∂ψc 0 ∂φc ∂ψc
σkm dY + σkm dY dΩ, (26)
ΩE e=1 Ye ∂ym ∂xk Ye ∂xk ∂ym
where n refers to the number of microscopic elements present in a macroscopic element.
The summation has the meaning of a finite element assembly operation (to obtain a
macro-elemental matrix).
If elements of the same type (including same number of nodes and shape) are used for
macro- and micro-scales and if the discretization is the same at both levels, then the nodes
of macro- and micro-elements are also the same and the assembly operations within each
macro-scale element drop out, but the double integrations (in both domains) remain.
From this double and equal discretization of both domains of the model, the instability
modes are then given by the interpolation of their nodal values:
NN
X nn
X NN
X nn
X
φ1 (x1 , y1 ) ≈ φI1 NxI + φi1 Nyi , φ2 (x2 , y2 ) ≈ φJ2 NxJ + φj2 Nyj , (27)
I=1 i=1 J=1 j=1

NN
X nn
X NN
X nn
X
ψ1 (x1 , y1 ) ≈ ψ1I NxI + ψ1i Nyi , ψ2 (x2 , y2 ) ≈ ψ2J NxJ + ψ2j Nyj , (28)
I=1 i=1 J=1 j=1

where N N denotes the number of nodal points per macroscopic element, nn the number
of nodes per microscopic element and Nx and Ny the shape functions for a macro-element
and for a micro-element, respectively. φI1 and φJ2 are the nodal modes of a macro-scale
element in directions 1 and 2, respectively, and φi1 and φj2 the nodal modes of a micro-scale
element in directions 1 and 2.

155
 Mariana João Figueiredo and Miguel Matos Neves

It should be noted that, in the given approximations, the exact ratio between each
sum of the product of nodal modes and shape functions is irrelevant, since all terms in
equation (23) involve the product between two derivatives, one of each scale.
Now, expanding the terms inside these double integrations and writing them in matrix
notation yields
   ∂φ1 
  h i E1111 E1122 E1112 ∂y
∂φi ∂ψk ∂ψ2    ∂φ12 
Eijkm = ∂ψ 1
∂x1
∂ψ2
∂x2
∂ψ1
∂x2
+ ∂x1
E2211 E2222 E2212  ∂y2  , (29)
∂yj ∂xm ∂φ ∂φ
E1211 E1222 E1212 ∂y2
1
+ ∂y1 2

   ∂φ1 
  h i E1111 E1122 E1112 ∂x1
∂φi ∂ψk ∂ψ1 ∂ψ2 ∂ψ1 E2211 E2222 E2212   
Eijkm = ∂y1 ∂y2 ∂y2
+ ∂ψ
∂y1
2

∂φ2
∂x2 , (30)
∂xj ∂ym ∂φ ∂φ
E1211 E1222 E1212 ∂x2
1
+ ∂x1 2

  ∂φ1 

  σ11 σ12 0 0 ∂y1
h i  ∂φ1 
0 ∂φc ∂ψc ∂ψ1 ∂ψ1 ∂ψ2 ∂ψ2 σ12 σ22 0 0 
  ∂y2 

σkm = ∂x2  0 , (31)
∂ym ∂xk ∂x1 ∂x2 ∂x1 0 σ11 σ12   ∂φ ∂y1 
2

0 0 σ12 σ22 ∂φ2

∂y2
   ∂φ1 
  σ11 σ12 0 0 ∂x
h i  ∂φ11 
0 ∂φc ∂ψc ∂ψ1 ∂ψ1 ∂ψ2 ∂ψ2 σ12 σ22 0 0 
  ∂x2 

σkm = ∂y2  0 . (32)
∂xk ∂ym ∂y1 ∂y2 ∂y1 0 σ11 σ12   ∂φ ∂x1 
2

0 0 σ12 σ22 ∂φ2

∂x2

By the application of the given instability modes’ approximations, equations (27) and
(28),
 
   ∂N 1 ∂N NN  φ11
∂φ1 x
0 ... ∂xx1 0  φ12 
 ∂φ ∂x1
 
∂x1
1 N N   
 ∂x22  =  0
∂Nx
... 0 ∂Nx
  ...  = Bx φx , (33)
∂x2 ∂x2  NN 
∂φ1
+ ∂φ2 ∂Nx1 ∂Nx1 ∂NxN N ∂NxN N φ1 
∂x2 ∂x1 ∂x2 ∂x1
... ∂x2 ∂x1
φN
2
N

 1  
 ∂φ1  ∂Ny ∂Nynn
φ11
0 ... 0
∂y1  ∂N
∂y1 ∂y1  
 ∂φ1   y 1 ∂N nn
0  φ12 
 ∂y2   ∂y2 0 ... ∂yy2   

 ∂φ2  =  ∂Ny1 ∂Nynn   ...  = gy φy , (34)
 ∂y1   0 ... 0   nn 
 ∂y1 ∂y1  φ 1
∂φ2 ∂Ny1 ∂Nynn
∂y2 0 ∂y2
... 0 ∂y2
φnn
2

and the same goes for the other derivative vectors in equations (29), (30), (31) and (32).

156
 Mariana João Figueiredo and Miguel Matos Neves

Since the discretization is the same at both scale-levels, then Nx = Ny , Bx = By ,

gx = gy , φx = φy and ψx = ψy , which, as suggested by [3], results in the following
system of equations
(Kxy + λxy Gxy ) φ = 0. (35)
The global coupled stiffness matrix, Kxy , is obtained by the assembly of the corre-
sponding elemental ones given by
Z Z
 T 
E
Kxy = B E E e B e + B Te E e B E dY dΩ (36)
ΩE Ye

and the global coupled geometric stiffness matrix, Gxy , by the assembly of
Z Z
 T 0 
GExy = g E σ e g e + g Te σ 0 e g E dY dΩ, (37)
ΩE Ye

where e refers to a micro-scale element, E is the constitutive matrix and σ 0 is the matrix
form presented in equations (31) and (32) of the initial stress at micro-scale level resulting
from the macroscopic strain field prior to bifurcation.
Applying numerical integration to evaluate the integrals by means of the Gauss quadra-
ture rule yields
"
P G2 npg1
#
NXP G1 NX X npg2
X

KxyE
= B TE E e B e + B Te E e B E detJ y wd wc detJ x wb wa and (38)
a=1 b=1 c=1 d=1

NX
P G1 NX
P G2
"npg1 npg2 #
XX

GE
xy = g TE σ 0 e g e + g Te σ 0 e g E detJ y wd wc detJ x wb wa , (39)
a=1 b=1 c=1 d=1

where N P G1 and N P G2 refers to the number of Gauss points in both directions for the
macroscopic elements, npg1 and npg2 to the number of Gauss points for the microscopic
elements, wi the weight factor for the respective Gauss point and detJi the determinant
of the Jacobian matrix related with the element mapping with a master element.
It should be, again, noted that B E and B e are in this exploratory work assumed
the same matrices, since the shape functions for both scales are the same as a result of
choosing the same discretization for both domains.
Moreover, for the construction of the geometric stiffness matrix, the calculation of
the initial stress components is required. For it, the macroscopic elastostatic equations
must be solved for the fundamental equilibrium position for which, the homogenized
material properties must be obtained. To perform the homogenization of the unit cell,
the characteristic displacements, solution of the elastostatic problems at the micro-scale
level, must be found. This means that, before addressing the coupled instability problem,
it is necessary to, by this order:
• Solve the elastostatic problems at the micro-scale level;

157
 Mariana João Figueiredo and Miguel Matos Neves

• Calculate the homogenized material properties;

• Solve the elastostatic problem at the macro-scale level.

3.2. Finite element models: homogenized properties and elastostatic prob-

lems at separated scale levels
As outlined in the previous section, to address the coupled-scale instability problem,
one needs to solve the static problems at separated scales and to calculate the homogenized
material properties. Since the focus of this work is on the instabilities resulting from the
coupling of the two scales of the model, only the discretized equations of the referred
problems will be here given - the reader is referred to, for instance, [1] or [8] for more
details on obtaining these equations. Furthermore, the Voigt’s notation is now used, so
that the equations may be presented in a more compact way.
The elastostatic problems at the micro-scale level in equation (13), which are to be
solved for the characteristic displacements χj , take the linear form

Ky χj = F jy , (40)

after the discretization of the model. In this equation, Ky is the stiffness matrix related
with the unit cell and Fy j the characteristic loading at micro-scale level. This equation
represents 3 systems of equations [j = 1, 2, 3 or (k, m) = (1, 1), (2, 2), (1, 2), respectively]
in two-dimensional situations and they are solved considering periodic displacement bound-
ary conditions. The stiffness matrix and the characteristic loading are given by, respec-
tively,
XN Z XN Z
Ky = T j
B e E e B e dY, Fy = B Te E e ε̃j dY, (41)
e=1 Ye e=1 Ye

where the sum has the meaning of a finite element assembly operation, N is the total num-
ber of elements, E e is the constitutive matrix of element e, B is the strain-displacement
matrix and ε̃j = δij are three independent unit macroscopic strain fields.
From the solution of the previous problem, the homogenized constitutive matrix can
be calculated by
N Z
1 X
T

EH
ij = ε̃i − B e χie E e ε̃j − B e χje dY, (42)
|Y | e=1 Ye

where |Y | denotes the geometrical volume of the unit cell.

With the homogenized properties, the elastostatic problem at the macro-scale level
(equation (14)) can be solved for the displacement field u by means of the linear system
of equations
Kx u = Fx , (43)

158
 Mariana João Figueiredo and Miguel Matos Neves

where Kx is the macro-scale stiffness matrix and Fx is the applied loads vector at macro-
scopic level given by, respectively,
N Z
X N Z
X Z
Kx = B Te E H
e Be dΩ, Fx = N Te f ref dΩ + N Te tref dΓ. (44)
e=1 Ωe e=1 Ωe Γt

In the previous equation, N represents the shape functions matrix.

The displacement field u is used to compute the macro-scale strain field, which is used
to compute the initial stress field at micro-scale level:

ε0e = B e ue , σ 0e = E e (I − B e χe )ε0e , (45)

where I is the identity matrix and χe is the elemental characteristic displacement fields
matrix: χe = [χ1e χ2e χ3e ].

3.3. Numerical implementation

A flowchart of the computational model can be visualized in figure 1. This compu-
tational model was built by the development of a Matlab function (named Coupleds),
where four node isoparametric plane elasticity elements and plane stress relations were
used. Both the structure and micro-structure are assigned regular meshes (all elements
are equal) and all the required matrices are computed numerically using Gauss point
integration.
The homogenization problem implementation (i.e. elastic-static problems at the micro-
scale level and calculation of the homogenized material properties) was based on the
homogenize Matlab function presented in [8], which was modified for the case of micro-
structures built from a base material with holes in it.
This implementation has seven user specified inputs: the lengths in x− and y− direc-
tions of the structure, the lengths in x− and y− directions of the unit cell, the two Lamé
parameters (each of them has two entries: one for the base material and one for void;
when referred to void these parameters should be set to zero) and the material indicator
matrix x. This last parameter, x, works in the same way as in the homogenize function:
it allows the user to choose which elements of the unit cell refer to the base material and
which ones refer to void. As in the case of the homogenization computational model, x is
used to discretize the micro-structure and, by means of the ratios between the lengths in
each direction of the unit cell and the structure, the macroscopic discretization is deduced,
so that both discretizations are the same in both domains.

159
 Mariana João Figueiredo and Miguel Matos Neves

Figure 1: Coupled instability problem - a flowchart of its computational implementation.

160
 Mariana João Figueiredo and Miguel Matos Neves

4. APPLICATION EXAMPLES
Here, to illustrate the application of the above described numerical implementation, two
examples are given, which explore the behaviour of the coupled-scale stability formulation
against the considered two limit instabilities: macro-scale instabilities and micro-scale
instabilities.

4.1. Macro-scale instabilities versus Coupled-scale instabilities

Here, the variation of the critical load with the unit cell’s material density for the
coupled instability problem is compared with the corresponding variation observed for
the macroscopic buckling case.
For that purpose, a beam, constrained in both directions on the left side, free on the
other side and subjected to an axial compressing load is considered. This beam has a
length L = 30m, a height h = 2m, a thickness b = 1m and it is built from a periodic
material characterized by a square micro-structure of side measuring 0.5m. The unit cell
is built from a base material with Young’s modulus E = 2.1MPa and a Poisson’s ratio
ν = 0.3. The variation in density for the structure is achieved by varying all unit cell’s
density in the same way: starting from a uniform unit cell (density=1), the density is
decreased by introducing a hole in the micro-structure and by progressively increasing
the hole’s size.
The macro-scale buckling response is also obtained by the development of a Matlab
function, named Macro ms, which was also built using a finite element formulation of the
equations obtained by the linearized stability theory presented in this paper. In both
cases (macro-scale and couple-scale), the structure is discretized by the same number of
finite elements.
For the macro-scale problem, to obtain the effective material properties, the unit cell
is discretized in 10x10 finite elements. Taking this into account, and because the given
implementation of the coupled case relies on an equal discretization of both the structures
domains, the beam is discretized in 24000 finite elements. The density was varied from
an initial value of 0.36 up to a final value of 1 and the results of both functions are shown
if figure 2.
It can be observed in figure 2 that the decrease of the both critical loads with decreasing
densities is in agreement with the expected decrease in buckling strength as the material
density is reduced.
The case were density is equal to one, corresponds to the case of a homogeneous beam.
This justifies the fact that the results from both functions are exactly the same - there
is no representative micro-structure to account for. Furthermore, since the used mesh is
sufficiently refined, the correspondent buckling load should be, with a small error, the
2
theoretical one given by the Euler’s Column Formula Pcr = π4LEI 2 . This formula gives a

theoretical value of 3838.179N, whereas Coupleds and Macro ms give 3800N, resulting in
an error of less than 1%.

161
 Mariana João Figueiredo and Miguel Matos Neves

Figure 2: Critical load versus material density – results from Coupleds and Macro ms.

As the density decreases and for ”high” values of density, the buckling response when
considering both scales still captures the macro-scale behaviour, which is justified since the
model still resembles to a homogeneous macro-scale one. In this test, that only happens
for density values of 1 and 0.95. For values lower than that, the Coupleds function gives a
buckling load lower than the one given by Macro ms. This difference in the results from
both models is due to the fact that the elastic properties for elements corresponding to
void are null in Coupleds, whereas in Marco ms these are homogenized ones.
For all the tested density values, the obtained first and second instability modes given
by both functions are represented in figure 3, which, as it can be seen, resemble the ones
obtained from a macro-scale analysis.

(a) First instability mode (b) Second instability mode

Figure 3: Instability modes’ shapes versus undeformed shape - results from Coupleds.

162
 Mariana João Figueiredo and Miguel Matos Neves

4.2. Micro-scale instabilities versus Coupled scale instabilities

Here, a test case is developed in order to verify if the carried-out implementation of
the coupled instability problem can capture micro-scale buckling modes.
Instability phenomena is more likely to happen for cellular materials with lower relative
density. As a ”low material density” micro-structure, the hexagonal honeycomb is here
considered (see figure 4).

Figure 4: Unit cell, or RVE, of the honeycomb micro-structure.

To obtain the micro-scale instability behaviour of the honeycomb, another Matlab

function, named micro buckles was developed, which was also based on the bifurcation
theory here presented. This implementation is limited to the calculation of instability
modes having the same periodicity of the unit cell. The coupled-scale instability behaviour
is obtained from Coupleds, where a beam constrained in both directions on the left hand
side, free on right hand side and subjected to a compressing load is again considered.
To address the Y-periodic micro-scale instability behaviour, the honeycomb is dis-
cretized in 21x12 finite elements and its dimensions are assigned so that its area is one [9].
The base material of this unit cell has a Young’s modulus E = 2.1MPa and a Poisson’s
ratio ν =0.3 and the void regions are assigned null material properties. The unit cell is
subjected to a macroscopic strain field of ε0 = {−12.3417 9.8942 0}. This strain field was
obtained from a macro-scale elastostatic analysis of the the referred beam, which has the
same material properties as the unit cell, dimensions L =10.53m, h =3.04m and b =1m
and is subjected to an uniform stress distribution of 1MPa in the x−direction at the right
end.
The honeycomb micro-structure and its critical and second buckling modes given by
micro buckles can be visualized in figures 5a and 5b, respectively. The results from the
micro buckles function showed a critical buckling load factor of 0.0094 (and the corre-
sponding critical load Pcr = 7142.6560N).
Performing the coupled-scale stability analysis of the structure, and with a discretiza-
tion of the beam into 168x48 finite elements, the two first buckling modes/loads are
first considered. The obtained critical load values are Pcr = P1 = 1215.74N and P2 =
3951.15N, which are much lower than the micro-scale one. This indicates that the ob-
tained structure’s behaviour at these bifurcation points presents lower critical loads than
the Y-periodic ones, which is somehow to expect (but can be investigated with a Bloch

163
 Mariana João Figueiredo and Miguel Matos Neves

(a) First instability mode for the (b) Second instability mode for the
honeycomb micro-structure honeycomb micro-structure

Figure 5: Honeycomb – micro-scale buckling analysis using micro buckles.

wave formulation extension not considered in this work). The corresponding mode shapes
can be seen in figures 6 and 7.

Figure 6: Undeformed shape versus first instability mode shape - beam with honeycomb micro-structure.

The ”general” shapes of the two modes are very identical to the ones obtained in a
macro-scale analysis concerning the buckling of beams.
In the critical mode shape, one cannot identify any buckled unit cells, although it can
clearly be seen, specially in the vicinity of the fixed left side, the compressing of the lower
cells and the stretching of the upper ones.
For the second instability mode, the previous considerations also apply. However, on
the right hand side, where the loading is applied, some unit cells start to buckle (see detail
on figure 7). While the majority of them resembles the first honeycomb instability mode,
the lowest one seems to be on its second one.
Since the so far obtained bifurcation points cause the instability of non or very few
unit cells and since the corresponding buckling loads are low when compared to the one

164
 Mariana João Figueiredo and Miguel Matos Neves

Figure 7: Undeformed shape versus second instability mode shape - beam with honeycomb micro-
structure.

obtained from a micro-scale analysis of the honeycomb, it can be expected (or at least
worthy of verification), that more cells will become unstable for higher buckling modes.
With that in mind, the calculations from Coupleds were extended to obtain the first
thirty instability load factors and shapes.
From one bifurcation point to the next, the increase in buckling strength was approx-
imately 300N and the structure exhibited some repeated buckling loads. From this set of
buckling loads, four of them exceeded the micro-scale instability load of the honeycomb,
but non of them exhibited an approximate value to the critical load of the RVE. How-
ever, an intermediate load, P13 = P14 = P15 = 6382, 62N, and two of the loads exceeding
the critical load of the unit cell, P28 = P29 = 7902.291N and P30 = 8206.23N, are next
analysed.
From the thirteenth instability mode shape (see figure 8) , one can see a significant
increase in the number of buckled unit cells. These seem to be arranged in columns and
subsequent unstable column groups seem to be present in the structure in a mirror-like
position, resulting in a deformed but not unstable shape for the remaining cells.

165
 Mariana João Figueiredo and Miguel Matos Neves

Figure 8: Undeformed shape versus thirteenth instability mode shape - beam with honeycomb micro-
structure.

As for the twenty-eightieth buckling configuration in figure 9, one can see the appear-
ance of more unit cells exhibiting micro-scale second instability mode shapes, along with
some micro-scale first instability ones as well as with completely stable ones.

Figure 9: Undeformed shape versus twenty-eightieth instability mode shape - beam with honeycomb
micro-structure.

Regarding the thirtieth mode (figure 10), it can be observed that no stable cells are
present. The majority of them are on their second instability configuration.

166
 Mariana João Figueiredo and Miguel Matos Neves

Figure 10: Undeformed shape vs. thirtieth instability mode - beam with honeycomb micro-structure.

As a final note, one can see that while micro-scale instabilities arise as higher buckling
loads are analysed, the model also captures macro-scale instability modes, which are
characteristic of this type of structure (see, for instance, the critical and second instability
modes, which are very well documented for this type of beams). Furthermore, for the
analysed structure, the unit cells become unstable, but not in a periodic way and the
resultant buckling load do not reaches the value of the micro-scale analysis. This last
fact is justified, since there is imposition of periodicity in the boundary conditions for the
micro-scale analysis, while the results clearly indicate that the micro-scale modes are non-
periodic. For a more reliable comparison, a Bloch-wave formulation shall be considered
in the micro-scale analysis, as described in ([1],[5]).

5. CONCLUSIONS AND FURTHER COMMENTS

The presented computational model for the coupled-scale linearized elastic stability prob-
lem (Euler Buckling) enables the exploration of the mixed-scale instability equation.
The results illustrate its capability for capturing macro-scale as well as micro-scale
behaviours, at least for the presented examples. Authors consider positive the results
obtained as they are one more step in the verification of the coupled-scale linearized
elastic stability equation. Anyway, this is the first attempt to implement and analyze this
equation, as far as the authors knowledge.
Several questions arise from the presented computational model, mainly: the discretiza-
tion adopted, the integration in the two scales with the shape functions used, and con-
sequent numerical inefficiency. Even if accuracy achieved in the examples is clear rela-
tively to the macroscopic critical load, for the microscopic critical load the absence of a
Bloch-wave extension is evident because of the non-periodic nature of the modes. The
implementation as proposed here is not considered efficient, mainly because it takes prac-
tically the same computational effort than a macroscale analysis with very fine mesh to
deal with the details at its microstructure.
Furthermore, this coupled-scale model may eventually be used in topology optimization

167
 Mariana João Figueiredo and Miguel Matos Neves

problems addressing the coupled-scale critical load of structures. Actually, the models
presented in the literature works in separated responses, i.e. for the macroscale only or
for the microscale only. But, the coupled problem of coupled scale modes discussed here,
still be an open problem for which the authors hope this work may call attention - as well
as contribute for a better understanding.

ACKNOWLEDGMENTS
This work was supported by FCT, through IDMEC, under LAETA, project UIDB/50022/2020.

REFERENCES
[1] Thomsen, C.R., Wang, F., Sigmund, ”Buckling strength topology optimization of 2D
periodic materials based on linearized bifurcation analysis”, Comput. Methods Appl.
Mech. Engrg. ,Vol. 339, pp.115-136 2018.
[2] Wu, J., Sigmund, O., Groen, J.P., ”Topology optimization of multi-scale structures:
a review”, Struct Multidisc Optim, 2021.
[3] Neves, M.M., ”Symbolic Computation to Derive a Linear-Elastic Buckling Theory
for Solids with Periodic Microstructure”, International Journal for Computational
Methods in Engineering Science and Mechanics , Vol.20(6), pp. 523-539, 2019.
[4] Neves, M M., Sigmund, O., Bendsøe, M.P., ”Topology optimization of periodic mi-
crostructures with a penalization of highly localized buckling modes”, International
Journal for Numerical Methods in Engineering , Vol. 54, pp. 809-834, 2019.
[5] Neves, M.M., Sigmund, O., Bendsøe, M.P., ”Topology optimization of periodic mi-
crostructures with a buckling criteria”, In Proceedings of Fifth World Congress on
Computational Mechanics , Vienna University of Technology, Vienna, Austria, 2002.
[6] Neves, M.M., Rodrigues, H.C., Guedes, J.M., ”Generalized topology design of struc-
tures with a buckling load criterion”, Structural Optimization, Vol.10, pp.71-78, 1995.
[7] Neves, M.M., ”Otimização da Topologia de Estruturas com Constrangimentos de
Estabilidade” (Mechanical Engineering Master’s Thesis in portuguese, Topology op-
timization of structures subject to buckling load constraint) , Instituto Superior
Técnico, Universidade Técnica de Lisboa, Lisboa, Portugal, 1994.
[8] Andreassen, E., Andreasen, C.S., ”How to determine composite material properties
using numerical homogenization”, Computational Materials Science , Vol. 83, pp.
488-495, 2014.
[9] Sigmund, O. ”Materials with prescribed constitutive parameters: an inverse homoge-
nization problem”, International Journal of Solids and Structures , 31, pp. 2313-2329,
1994.

168

View publication stats