0% found this document useful (0 votes)

84 views

Extended Multiscale Finite Element Method: Its Basis and Applications For Mechanical Analysis of Heterogeneous Materials

Uploaded by

TAWHID

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

84 views

Extended Multiscale Finite Element Method: Its Basis and Applications For Mechanical Analysis of Heterogeneous Materials

Uploaded by

TAWHID

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 27

Comput Mech

DOI 10.1007/s00466-013-0924-x

ORIGINAL PAPER

Extended multiscale finite element method: its basis

and applications for mechanical analysis of heterogeneous
materials
H. W. Zhang · Y. Liu · S. Zhang · J. Tao ·
J. K. Wu · B. S. Chen

Received: 18 June 2013 / Accepted: 10 September 2013

Abstract Extended multiscale finite element method 1 Introduction

(EMsFEM) has been proved to be an efficient method for the
mechanical analysis of heterogeneous materials. The key fac- Many kinds of materials are heterogeneous in microscop-
tor for efficiency and accuracy of EMsFEM is the numerical ical scale. For example, composite materials combine the
base functions (NBFs). The paper summarizes the general advantages of each phase and improve performance such as
method for constructing NBFs and proposes a generalized the strength and fracture toughness. The specified-functional
isoparametric interpolation based on the rigid displacement materials could be manufactured artificially by changing the
properties (RDPs) of NBFs. We prove that the NBFs con- microscopical structure of the materials. Numerical simula-
structed by linear, periodic and rotational angle boundary tions are preferred ways to analyze and design such kinds of
conditions satisfy the RDPs, which is independent with the materials, which are promising to be a substitution of costly
shape and material properties of unit cells. The properties field tests.
of NBFs for oversampling technique are also comprehen- However, highly efficient and reliable numerical methods
sively discussed. The algorithm complexity is discussed in remain a challenge. As pointed out in [1–5], an extremely
theoretical and numerical aspects, which concludes that the fine spatial discretization mesh is necessary for the direct
computation quantity of EMsFEM is much smaller than the numerical simulation of heterogeneous materials to capture
direct solutions. The algorithm is validated by linear analy- the effects of microscale heterogeneities, and the computa-
sis of the materials with random impurities and holes and the tion quantity exceeds the ability of a general computer, even
efficiency is improved further by parallel computing. for the supercomputer. As indicated by Zohdi [3], solutions
to a small elastic model with a few heterogeneities at the
Keywords Extended multiscale finite element method infinitesimal strains are still open questions. In short, com-
(EMsFEM) · Numerical base function · Heterogeneous plete solutions are virtually impossible. Multiscale model-
material · Algorithm complexity ing, with such features as high efficiency of macroscopical
models and considerable accuracy of microscopical models,
is a promising method to overcome the difficulty [5]. Other
advantages of multiscale modeling on science and engineer-
ing are summarized in the review paper by Horstemeyer [6].
Nowadays, many kinds of multiscale methods are devel-
oped, which can be, on the whole, classified by hierarchi-
H. W. Zhang (B) · Y. Liu · S. Zhang · J. Tao · J. K. Wu · B. S. Chen cal methods and concurrent methods [5–8]. For the hier-
State Key Laboratory of Structural Analysis for Industrial Equipment, archical multiscale methods, the mechanical information is
Department of Engineering Mechanics, Faculty of Vehicle Engineering
passed from microscopical to macroscopical scale. This kind
and Mechanics, Dalian University of Technology,
Dalian 116024, People’s Republic of China of methods include homogenization method [3], the sec-
e-mail: [email protected] ond order homogenization method [9], multilevel multiscale
Y. Liu method [10,11], mean field approaches [1,12], statistical
e-mail: [email protected] homogenization [13] and so on. Multiscale methods based

123
Comput Mech

on the homogenization technique have been extended to results. The assumptions that are central to the homogeniza-
multi-physical field problems (such as electroactive [4] and tion theory, such as periodicity and separation of length scales
thermal elasticity materials [14,15]) and all kinds of nonlin- [19], are no longer necessary for EMsFEM. Analogous to
ear problems (such as elasto-plastic problem [10,11], finite analytical base functions, the necessary conditions of NBFs
deformation [16,17] and material instability [18]). Readers are the rigid displacement properties (RDPs). Many kinds of
could refer to [19] for a comprehensive summary of the construction modes, such as linear or periodic boundary con-
homogenization method. Meanwhile, some drawbacks of the ditions and oversampling technique, have been investigated
homogenization methods arise in simulations. As indicated [28]. However, verifications of the RDPs of NBFs depend
by Unger and Eckardt [7], after the homogenization process only on the numerical tests and, especially, the rotational
of the fine scale model, the essential properties are averaged RDP has not been explained yet. It is no doubt that inves-
and the actual failure state is only captured in a homogeneous tigations on the NBFs in the theoretical aspect are of great
way. However, fracture or electric breakdown will depend on significance to avoid heavy labor of trial and error. Thus, the
specific detail in the microstructure and the averaging is not paper makes a detailed discussion on the RDPs of NBFs and
acceptable [20]. For the concurrent multiscale methods, the the principles for constructing NBFs for the first time, which
domain is decomposed into fine-scale and coarse-scale mod- aims to provide a rational construction scheme for NBFs. All
els, which could be considered as one of the DDMs [8,21]. kinds of boundary conditions are investigated, among which,
Different models are solved concurrently and the informa- a new one with rotational angle DOFs on the coarse mesh is
tions between the subdomains are passed on by strongly cou- proposed based on the proposed principles, which seems to
pling. Typical examples of the concurrent multiscale models be helpful for improving accuracy without increasing com-
include atomistic/continuum multiscale methods [22,23] and putational costs. As pointed out above, the aim of EMsFEM
quantum/atomistic/continuum multiscale methods [24,25]. is to overcome the shortages of computational resource in
A comprehensive summary on Atomistic/continuum multi- the numerical simulations of heterogeneous materials. In this
scale methods can be found in the review article by Miller context, the algorithm complexity of EMsFEM is analyzed
and Tadmore [26]. A general framework for designing mul- in both theoretical and numerical aspects and a theoretical
tiscale algorithms is reviewed in a recent paper by Abdulle formulation for determining the optimal size of coarse mesh
et al. [27] and the references therein. approximately is proposed to minimize the computational
Extended multiscale finite element method (EMsFEM) is time.
a new multiscale method proposed by Zhang et al. [28,29] in The paper is organized as follows. In Sect. 2, the basic
recent years, which aims to deal with the mechanical analysis principle of EMsFEM is reviewed simply. In Sect. 3, the con-
for heterogeneous materials. The idea of EMsFEM could be ditions ensuring the RDPs of NBFs are deduced in the two-
traced back to the work presented by Babuška and cowork- dimensional case, of which, the rotational RDP is newly pro-
ers [30,31], in which the concepts of NBF is proposed. posed. Moreover, the general construction modes of NBFs
Afterwards, a multiscale finite element method (MsFEM) are discussed and the properties of additional terms are also
is developed and push forward further by the work of Hou investigated. In Sect. 4, the construction methods and prop-
et al. [2,32], in which the convergence of the method is dis- erties of NBFs for all kinds of boundary conditions, such as
cussed comprehensively. A recently published monograph linear, periodic and rotational angle boundary conditions, are
by Efendiev and Hou [33] makes an extensive theoretical discussed and extended to more general geometries and mate-
discussion on the MsFEM with many applications, such as rial distributions. Oversampling technique for constructing
transport equations, Richards’ equations and fluid-structure oscillating NBFs is explored thoroughly in Sect. 5. Section 6
interactions. In comparison with the method developed by makes a discussion on algorithm complexity of EMsFEM and
Babuška and Hou, EMsFEM extends the elliptical problems corresponding numerical examples are given. As for Sect.
of scalar field to vector field. Moreover, the solution accu- 7, some numerical examples concerning the heterogeneous
racy is improved by introducing coupling terms to NBFs materials with random holes or impurities are considered.
and developing new boundary conditions for constructing Finally, some conclusions are presented in Sect. 8.
NBFs. The microscopical stress could be obtained by the
downscale calculation, which is essential for the nonlinear
mechanical analysis. The basic principles of EMsFEM are 2 Principle of EMsFEM
explained in details in [28] and the applications of the method
to mechanical problems, such static and dynamic analysis The direct discretization and solution of heterogeneous mate-
[28,34], elasto-plastic analysis [29,35], and strain localiza- rials which contain many micro structures at the mesoscopic
tion analysis [36], are also investigated. scale as shown in Fig. 1, lead to enormous DOFs of the fine
The core technique of EMsFEM is the construction of mesh model. The basic idea of EMsFEM is to convert the
NBFs, which shows great influence on the accuracy of the boundary value problems (BVPs) of the fine mesh model

123
Comput Mech

Fig. 1 The basic idea of EMsFEM

to the coarse mesh one by constructing the NBFs of UCs, where N = [N 1 , . . . , N n ], N i = N xi , N yi , û =
T T
which act as the bridges of different scales. The UC with a
û1T , û2T , . . . , ûnT , ûi = û xi , û yi . The equivalent stiff-
specific size contains some microscopical structures for het-
ness matrix and external force at the macroscopical level
erogeneous materials and the NBFs are obtained by solving
are computed using the NBFs. The UC is assumed to be
a group of BVPs of the UC. The displacement field of the UC
discretized by four-node elements (microscopical elements),
is expressed by linear combination of the NBFs whose coef-
whose energy functional and stiffness matrix are
ficients correspond to the DOFs of the coarse-mesh nodes,
as depicted in Fig. 1. Afterwards, the principle of EMsFEM
1 T
would be reviewed following the work in [28]. e = u K e ue , K e = BeT De Be de (3)
2 e
The microscopical displacement field of a two-dimension- e
al UC is defined by
where ue is the element displacement vector, Be and Dε De

n
n
are the strain matrix and constitutive matrix, respectively.
u= N xi ûxi + N yi û yi (1)
Subsequently, from Eq. (2), the equation which relates the
i=1 i=1
displacement of the microscopical element and that of the
where û xi and û xi denote the displacement of the coarse- UC can be defined by
mesh nodes, Âi , i = 1, . . . , n, in the x and y directions,
respectively. Here, n is the number of coarse-mesh nodes ue = N e û (4)
(macroscopical nodes). The displacement vector of the UC,
T where
u, could be expressed as u = u1T , u2T , . . . , uTN , where
T
T
uk = u xk , u yk are assumed to be the displacement vec-
N e = Re1
T
, . . . , Re4
T
,
tor of the microscopical node, Bk , k = 1, . . . , N . Here, N

is the total amount of the microscopical nodes within the N x x1 (i) N x y1 (i) . . . N x yn (i)
UC. The NBFs, N xi and N yi , could be obtained by applying Rei = (5)
N yx1 (i) N yy1 (i) . . . N yyn (i)
some specific boundary conditions on the UC and solving
the BVPs. The coupling terms are included in N xi and N yi The energy functional of the microscopical element and UC,
comparing with the traditional analytical base function. For expressed by the macroscopical variable, are as follows
example, the value of N xi on Bk is
T 1 T T
N xi (k) = N x xi (k) , N yxi (k) e = û N e K e N e û,
2 n

ne
1 T e
where the coupling terms N yxi (k) reflect the influence of the E = ei = û T
N ei K ei N ei û (6)
unit displacement of Âi in the x direction on the NBF of Bk 2
i=1 i=1
in the y direction.
The simplified form of Eq. (1) could be expressed as where n e is the number of the microscopical element in the
UC. Thus, we obain the equivalent stiffness matrix of the UC
u = N û (2) and the assembled global one at the macroscopical level

123
Comput Mech

Ne
NE
1, û yi = 0, i = 1, 2, . . . , n or û = 1̂x , displacement of
KE = N eT K e N e , K = Ai=1 K Ei (7) microscopical nodes in the UC should be u xk = 1, u yk =
e=1 0, k = 1, 2, . . . , N or u = 1x . Substitution of û and u into
where N E is the number of coarse mesh and A is the stiff- Eq. (2) yields
ness matrix assembly operator. Thus, the final problem is
1x = N 1̂x or N xi = 1x (10)
transformed into the macroscopical level equation
i
KU = Fext (8) which ensure the RDP of NBFs in the x direction. Equation
(10) could be expressed as a more specified form [28]
where Fext is the external force vector [29]. The displacement

at microscopical level can be obtained by downscale calcula- N x xi = 1, N yxi = 0 (11)
tion using Eq. (4) and the element strain and stress could be i i
achieved by the displacement results, similar to the traditional
In analogy to the RDP in the x direction, the following equa-
finite element analysis. One could refer to [28] for the details.
tion ensures the RDP in the y direction
It could be found that the informations between the macro-
scopical and microscopical level are passed on by the NBFs, 1 y = N 1̂ y or N yi = 1 y (12)
and the constitutive equation in the macroscopical level is i
not necessary. Thus, the scale separation assumption is of
no necessity to be taken into consideration for EMsFEM, 2. Rotational RDP
which differs significantly from the other multiscale meth-
ods based on the homogenization. One of superiorities of This property requires that, given a small rigid rotational dis-
NBFs is to implement the downscale calculation and obtain placement of the coarse-mesh node at the vicinity of initial
the strain and stress in the microscopical level, which provide configuration, the displacements of the microscopical nodes
convenience for nonlinear analyses. Meanwhile, as the whole in the UC should be the same rotational rigid displacement.
solution domain is divided into several UCs, the parallel com- As depicted in Fig. 2, after a small rigid rotation ϕ, the dis-
puting could be implemented without hard work, which is placement vector of the coarse-mesh node, Âi x̂i , ŷi , is
similar to the famous domain decomposition method (DDM) T
[21]. However, the NBFs is not introduced for DDM, with- ûi = ϕ r̂i t̂i = ϕr̂i {− sin θi , cos θi }T = ϕ − ŷi , x̂i
out considering the deformation features of heterogeneous = ϕr̂wi (13)
materials, and the equilibrium iterations are always needed 2
where t̂i is orthogonal to r̂i and r̂i = x̂i2 + ŷi2 . Similarly,
at the interfaces of subdomains, which may be negative for
the displacement of the microscopical node, Bk (xk , yk ), is
the computation efficiency.
uk = ϕ {−yk , xk }T = ϕrwk . Thus, the rotational rigid dis-
placement vector of macroscopical and microscopical nodes

3 Basic properties of numerical base function

3.1 Rigid displacement properties

Similar to the analytical base function, N x xi and N yyi should

be able to describe the rigid displacement mode of UC
and satisfy the Kronecker delta property at the coarse-mesh
node, Â j :

N x xi Â = δi j , N yyi Â = δi j (9)
j j

which is satisfied automatically after applying the boundary

conditions on a UC. The rigid displacements of NBFs include
two translational modes and a rotational one for the plane
problems. In this part, we will discuss the property in details.

1. Translational RDP

The property demands that, given a rigid translational dis- Fig. 2 A small rigid rotation of a UC at the vicinity of the initial con-
placement of the coarse-mesh node in the x direction: û xi = figuration

123
Comput Mech

could be expressed as, respectively, 1, q = mod (i + 1, 4) + 1, r = mod (i + 2, 4) + 1. The

symbol mod (·, ·) denotes the remainder after division. The
û = ϕr̂w , u = ϕrw (14)
DOF of the element doesn’t increase, while the accuracy
T T T
Here, r̂w = r̂w1 T T
, r̂w2 T
, . . . , r̂wn and rw = rw1 , rw2 , is promoted a lot by the coupling terms. The translational
T
4
. . . , rwN . Substituting Eq. (14) into Eq. (2), one obtains
T RDPs are verified easily by the equations: Niξ (ξ, η) =
i=1
rw = Nr̂w (15)
4
4
4
Niη (ξ, η) = 1 and Niξ η (ξ, η) = Niηξ (ξ, η) =
i=1 i=1 i=1
which could be written as a more detailed form 0. Due to the determinants

rw = − N xi ŷi + N yi x̂i (16)
x1 x2 x3 x4 y1 y2 y3 y4
i i
1 1 1 1
or det = 0 and det 1 1 1 1 = 0
x1 x2 x3 x4
x1 x2 x3 x4
y= N x xi ŷi − N x yi x̂i , x y1 y2 y3 y4 y1 y2 y3 y4
i i
(21)
= − N yxi ŷi + N yyi x̂i (17)
i i

4
4
where y and x are composed by the coordinates of microscop- we have βi ξi ηi xi = 0 and βi ξi ηi yi = 0, and then,
ical nodes in the UC. Supposing that N x yi = N yxi = 0 and i=1 i=1

N x xi = N yyi in Eq. (17), then we obtain the isoparametric

interpolations
4
4
−y = − Niξ (ξ, η) yi + Niξ η (ξ, η) xi ,

y= N yyi ŷi , x = N x xi x̂i (18) i=1 i=1
i i
4
4
x= Niη (ξ, η) xi − Niηξ (ξ, η) yi (22)
Therefore, the above isoparametric interpolation, degener-
i=1 i=1
ated from Eq. (17), fulfills the rotational RDP automatically.
Hence, the rotational RDP of NBFs could be considered as Consequently, the base functions of the generalized isopara-
an extension of the traditional isoparametric interpolations. metric element satisfy the rotational RDP, which are also
It should be noted that for scalar field problems, such as heat verified by the numerical tests in [34].
conduction and seepage analysis, no coupling terms appear Equations (10), (12) and (15) are the necessary and suffi-
in the NBFs, which means that the rotational RDP is not nec- cient conditions for the RDPs of NBFs in two-dimensional
essary and only the translational RDPs, shown in Eq. (10) cases, which could be used to illustrate feasibilities of the
or Eq. (12), is indispensable [2]. Thus, it is much easier to NBFs on the boundary or interior of UCs.
construct NBFs for scalar fields than that for vector fields.
The analytical base functions with coupling terms are also
constructed by Zhang et al. [37]. The proposed element is 3.2 Construction theory of NBFs
called the generalized isoparametric element, which adds
coupling terms to the traditional bilinear isoparametric ele- In this section, we will prove that if the NBFs on the boundary
ment. The displacement interpolations of the element are of a UC satisfy the RDPs, then so will the NBFs inside the
UC.

4
4
The BVPs for solving two-dimensional NBFs of a UC
u= Niξ (ξ, η) u i + Niξ η (ξ, η) νi ,
i=1 i=1
could be expressed as
4 4
ν= Niη (ξ, η) νi + Niηξ (ξ, η) u i (19) LN i = 0 in K
(23)
i=1 i=1 N i = N̂, i = 1, 2, . . . , n
where

1 where L is the elasticity operator satisfying Lu = div D : 21
Niξ = Niη = (1 + ξi ξ ) (1 + ηi η) ,
4 ∇u + (∇u)T and is the boundary of the domain K . The
5 (1 + μ)
above BVPs come down to the block equation
Niξ η = Niηξ βi ξi ηi 1 − ξ 2 1 − η2 (20)
32 (3 − μ)
Here, Niξ η and Niηξ are the coupling base functions, μ is K bb K bs Nb fb
2S
= (24)
the Poisson’s ratio and βi = pqr
S with p = mod (i, 4) + K sb K ss Ns 0

123
Comput Mech

where the subscripts, b and s, denote the terms related to the where the original NBFs are assumed to satisfy the RDPs.
boundary and interior of the UC, respectively. The interior The translational RDPs of N xi and N yi lead to the equations

NBFs could be solved from the second equation of Eq. (24): N xi = 1x and N yi = 1 y , which could deduce the
i i
N si = −K −1
ss K sb N bi (25) relations about the additional terms

Ñ xi = 0, Ñ yi = 0 (30)
There are three multiple zero eigenvalues of the coeffi-
i i
cient matrix in Eq. (24) for plane problems. The eigen-
values correspond to three orthogonal eigenvectors: N i = Then, substituting Eq. (29) into the condition about the rota-
T T T
tional RDPs of the new NBFs, Eq. (16), the additional terms
N bi N si , i = 1, 2, 3, which satisfy N si = −K −1
ss K sb N bi . are found to follow
The arbitrary rigid displacement G could be expressed as the
linear combination of N i : Ñ xi ŷi − Ñ yi x̂i = 0 (31)
i i

3
Thus, the additional terms meeting Eqs. (30) and (31) could
ai N i = G (26)
certify the RDPs of the final NBFs. One could calculate the
i=1
additional terms directly by Eqs. (30) and (31). However, the
T T equations are overdetermined in general, which is an unreal-
Here, G = GbT , GsT and ai = N i G. Provided the RDPs
of N bi , there are three coefficients ci which make the equation istic way to construct NBFs.
established Afterwards, two possible construction methods for the
NBFs with additional terms will be discussed: (1) applying

3
external force on the interior DOF of a UC; (2) introducing
ci N bi = Gb (27) multipoint constraints on the boundary DOF of a UC.
i=1

Subsequently, utilizing Eqs. (25)-(27), we have 3.3.2 Method I: applying external force

3
3
Introducing an external force f s on the interior DOF of a UC,
ci N si = −K −1 −1
ss K sb Gb = −K ss K sb ai N bi the second equation in Eq. (24) becomes
i=1 i=1

3 K sb N b + K ss N s = f s (32)
= ai N si = Gs (28)
Given N bi and f si , the displacement of the interior DOF could
i=1
be deduced as
which indicates that Gs could be composed by N si with the
N si = −K −1 −1
ss K sb N bi + K ss f si = N̂ si + Ñ si (33)
same coefficients ci . Equations (27) and (28) illustrate the
conclusion made in the start of this section, which implies the In the above equation, if N bi meets the RDPs, then so will
principle for constructing NBFs by solving the BVPs of a UC N̂ si , due to the conclusions in Sect. 3.2, and Ñ si could be
and for determining whether the NBFs satisfy RDPs or not. treated as the additional term, which should satisfy Eqs. (30)
Although the shapes and material distributions of a UC have and (31). Thus, the properties about the external force are
influence on a single NBF, the RDPs of NBFs just depend
on that of UC’s boundary. Moreover, the constructions of f sxi = 0, f syi = 0 (34)
i i
NBFs in two-dimensional domains are transformed into one-
dimensional UC’s edges and the difficulties for achieving f sxi ŷi − f syi x̂i = 0 (35)
reasonable NBFs are alleviated a lot. i i

Equations (34) and (35) guarantee the translational and rota-

3.3 Additional terms of NBFs and its properties tional RDPs, respectively.

3.3.1 Additional terms of NBFs 3.3.3 Method II: introducing multipoint constraints

The performance of NBFs, such as the one constructed by The new NBFs on the UC’s boundary could be expressed,
periodic boundary condition (PBC) in Sect. 4.2, could be similar to Eq. (29), as
improved by superposing the additional terms Ñ xi and Ñ yi
N bi = N̂ bi + Ñ bi (36)
to the original ones N̂ xi and N̂ yi :
where N̂ bi and Ñ bi denote the original NBF and its additional
N xi = N̂ xi + Ñ xi , N yi = N̂ yi + Ñ yi (29) term, respectively. It could be verified that, if appropriate

123
Comput Mech

multipoint constraints are applied on the boundary of the For the additional terms, Ñ byi , the equation could be similarly
UC, the RDPs of N bi could be guaranteed. The multipoint given by
constraints of N bi could be expressed as a general form
Ñ byi = T Ñ byi = 0 (44)

N bi = N̂ bi + Ñ bi = N̂ bi + T Ñ bi (37) i i

Subsequently, one could deduce the equation about rotational

where Ñ bi is the independent DOF after the constraints are rigid displacement
introduced, T is a constant matrix, the element of which is −1
related to the type of constraint. For example, applying fixed Ñ bxi ŷi − Ñ byi x̂i = K bb T T K bb
constraints on the pth DOF of Ñ bi , then, we have T ( p, :) = i i

0, which means that the elements at the pth row are zero.
If the pth DOF of Ñ bi is equal to that of the qth and the × N̂ bxi ŷi − N̂ byi x̂i =0 (45)

i i
qth DOF is retained in Ñ bi , then, T ( p, q) = 1 and other

elements at the pth row are zero accordingly.
which employs K bb N̂ bxi ŷi − N̂ byi x̂i = 0, on the
Introducing Eq. (37) into the energy functional, concerned
i
i
with the boundary DOF, by the Lagrange multiplier, it yields basis of the fact that N bxi ŷi − N byi x̂i represents the
i i
1 T T rotational rigid displacement mode on the UC’s boundary.
= N bi K bb N bi − N bi − N̂ bi − T Ñ bi λ (38) Thus, utilizing Eqs. (37) and (45) leads to the condition con-
2
cerning the additional terms
There are three independent variables in the above equation,

i.e., N bi , Ñ bi and λ. Utilizing the variational equation, δ =
Ñ bxi ŷi − Ñ byi x̂i = T Ñ bxi ŷi − Ñ byi x̂i
0, we get the equilibrium and constraint equations i i i i
⎧ =0 (46)
⎨ K bb N bi − λ = 0
⎪
N bi − N̂ bi − T Ñ bi = 0 (39) As a result, the additional terms, deduced from the multi-
⎪
⎩ T point constraints, satisfy Eqs. (43), (44) and (46), thus, the
T λ=0
RDPs of the new NBFs in Eq. (37) are guaranteed due to the
where λ is the constraint force and the third equation rep- conclusions in Sect. 3.3.1.
resents the relation of constraint forces induced by the con- It seems that the form of T have no influence on the
straints. From the above equations, the final equation with RDPs of the new NBFs, which means that the RDPs are not
respect to the independent DOF could be obtained by restricted by some specified form of multipoint constraints;
it is reasonable that only the singularity of K bb is eliminated.

K bb Ñ bi = −T T K bb N̂ bi (40) Thus, all kinds of multipoint constraints could be developed
to construct suitable NBFs. The conclusions in this section
where K bb = T T K bb T. Therefore, the independent variable will be used to discuss the RDPs of the NBFs for periodic
becomes boundary condition (NBFs-PBC).
−1
Ñ bi = − K bb T T K bb N̂ bi (41)
4 Some construction modes of NBFs
Then, Ñ bi and N bi are derived by substituting the above equa-
tion into Eq. (37). For the additional NBFs in the x direction, 4.1 Linear boundary condition

Ñ bxi , one could obtain
In this section, the construction of NBFs for linear bound-

−1
Ñ bxi = − K bb T
T K bb N̂ bxi = 0 (42) ary condition (NBFs-LBC) will be discussed in the theo-
i i
retical aspect on the basis of reference [28]. As the UC
depicted in Fig. 3, the edge is discretized by = 12 ∪
which could be deduced by summing Eq. (41) with respect to
23 ∪ · · · ∪ n1 , where i j denotes the directional edge
index i and utilizing the equation K bb N̂ bxi = K bb 1x = 0. with the macroscopical nodes Âi and Â j . i j is counter-
i
Due to Eqs. (37) and (42), the additional terms satisfy clockwise as j = mod (i, n) + 1, and clockwise as j =
n− mod (n − i + 1, n), where n is the number of the macro-

Ñ bxi = T Ñ bxi = 0 (43) scopical node and mod (·, ·) denotes the modulus operator
i i after division. Considering the BVPs corresponding to Âi ,

123
Comput Mech

2. The macroscopical DOFs, û xi and û yi , which correspond

to N xi and N yi , are meaningful in physics, so the method
is of convenience for the assembly of the global stiff-
ness matrix on the coarse mesh and applying boundary
conditions.
3. Only one decomposition process of K ss is needed for
solving all BVPs, which is helpful for improving compu-
tational efficiency. In the previous work for the MsFEM
or EMsFEM, the times for solving the linear equations
about the UC are almost equal to amount of NBFs. For
example, in the work of Hou et al. [2], three BVPs need
Fig. 3 LBC for the UC with an arbitrary shape to be solved in each rectangular UC to obtain three NBFs
for scalar-field problems (the fourth one is computed by
the RDPs). Thus, the simple technique would make a
the prescribed NBFs on the UC’s boundary are assumed to significant improvement on computation efficiency.
be N bi = N bxi , N byi , where N bxi and N byi correspond
4. High order boundary conditions could also be con-
to the DOFs of Âi , i.e., û xi and û yi , respectively. N bxi is
structed by introducing middle points on the edge of
set to be uxi j and uxik in the x direction and to be zero in
the UC, similar to the serendipity element, and this is
the y direction on i j and ik . Here, uxi j is distributed lin-
a promising way to improve the accuracy of NBFs.
early along i j , which is equal to one at Âi and zero at Â j
and uxik conforms to the similar distribution. On the other
boundaries, both displacements in the x and y directions are
4.2 Periodic boundary condition
set to be zero. In the case of Â2 , the NBFs on 21 and 23
are shown in Fig. 3, corresponding to û x2 or û y2 .
The construction of NBFs-PBC is explained detailedly in
The BVP corresponding to û xi could be defined by
[28]. The displacements at the macroscopical nodes are spec-
⎧
⎪ LN xi = 0 in K ified, while the NBFs on the edges need to be solved by the
⎪
⎨ multipoint constraints. The RDPs with respect to the NBFs
N xi = uxi j , N xi = uxik (47)
⎪ ij ik obtained by the multipoint constraints are discussed in Sect.
⎪
⎩ N xi
b \( i j ∪ ik )
= 0 3.3.3. Here, we just need to find a multipoint constraints
equation for PBC similar to Eq. (37). The boundary NBF is
The prescribed displacement on i j and ji satisfies uxi j + partitioned into the block form
ux ji = 1xi j due to the above discussions on LBC, where 1xi j T
denotes the rigid displacement of i j in the x direction. The N b = N Tp , N qT , N rT (51)
above BVP comes down to the equation as Eq. (24). There
are no coupling terms within the NBFs-LBC on the UC’s The node location corresponding to each block could be
boundary, thus, the corresponding RDPs could be verified as found in Fig. 4, in which N p is related to the macroscop-
follows

N bxi = 1bx (48)
i

N byi = 1by (49)
i

yb = N byi ŷi , xb = N bxi x̂i (50)
i i

Here, xb and yb are composed by the coordinates of micro-

scopical nodes on the UC’s boundary. It can be inferred from
the conclusions in Sect. 3.2 that, the NBFs-LBC satisfy the
RDPs. Some conclusions could be drawn about the NBFs-
LBC.

1. The RDPs have nothing to do with the shape and material

distribution of the UC. Fig. 4 PBC for the rectangular UC and the boundary DOF blocks

123
Comput Mech

ical nodes and highlighted with the solid dots and the nodes 4.3 Rotational angle boundary condition
corresponding to N q and N r are represented by the shadow
and blank dots, respectively. N p is known in advance as a The idea for constructing the NBFs of Rotational Angle
macroscopical node is constrained as shown in Fig. 4, while Boundary Condition (NBFs-RABC) could be traced down
N q and N r satisfy the multipoint constraint equations. to the membrane element with a drilling DOF. Introducing
Incorporating the features of PBC, the NBFs in Eq. (51) the drilling DOF to a membrane element is beneficial not only
could be decomposed as for improving the element’s accuracy, but also for construct-
⎧ ⎫ ⎧ ⎫ ⎧ ⎫ ing the plane shell element by combining with a suitable plate
⎨ N pi ⎬ ⎪ ⎨ N̂ pi ⎪
⎬ ⎨ Ñ pi ⎬ element. The first membrane element with the drilling DOF
N = N̂ qi + Ñ qi (52) is proposed by Allman [38]; soon afterwards, many famous
⎩ qi ⎭ ⎪ ⎩ N̂ ⎪ ⎭ ⎩ Ñ ⎭
N ri ri ri works on this topic appear, such as Hughes and Brezzi [39]
and Ibrahimbegovic et al. [40]. The study makes use of the
which could be simplified as N bi = N̂ bi + Ñ bi . The original membrane with the drilling DOF by Hughes and Brezzi [39],
NBFs, N̂ bi , with respect to the x and y in Fig. 4, distrib- where the independent rotational field is defined as

1
ute linearly along the two boundaries next to the node, Âi ,
thus, the RDPs of N̂ bi are guaranteed. The additional terms, ψ= ∇u − ∇uT (56)
− − 2
Ñ bi , corresponding to u B , u A etc. need to be calculated.
The multipoint constraints in Fig. 4 lead to Ñ pi = 0 and where u is the displacement field and ∇ = ∂∂x i + ∂∂y j repre-
Ñ qi = Ñ ri . Then, the additional terms of the NBFs could be sents the two-dimensional gradient operator. The above con-
expressed as straint equation is realized by introducing a penalty function
⎧ ⎫ ⎡ ⎤ into the original energy functional. On the basis of this theory,
⎨ Ñ pi ⎬ 0 Ibrahimbegovic et al. [40] presents the quadrilateral element
Ñ qi = ⎣ I ⎦ Ñ qi = T Ñ qi (53) with a drilling DOF. The rotational field of the element is
⎩ ⎭
Ñ ri I interpolated by a bilinear function; meanwhile, the displace-
ment along the element edge is linear and vertical to the edge
Thus, Eq. (52) could be rewritten as N bi = N̂ bi + T Ñ qi , quadratic. In the light of this theory, this section presents a
where Ñ qi is the independent unknown. It could be found that boundary condition with a rotational angle DOF on the coarse
the construction of NBFs-PBC coincides with the framework mesh.
in Sect. 3.3.3. Consequently, from Eqs. (41) and (53), one Each macroscopical node, Âi , contains three DOFs for
could obtain the solution of the independent variable T
RABC, i.e., ûi = û xi , û yi , ϕ̂i , in which û xi , û yi are the
−1 translational DOFs and ϕ̂i the rotational angle DOF. There are
Ñ bi = −T K̂ bb T T K̂ bb N̂ bi (54) 3n groups of BVPs to be solved to obtain all the NBFs, where
n denotes the number of coarse-mesh nodes. As depicted in
Fig. 5, there are six groups of BVPs related to the prescrip-
where K̂ bb = T T K̂ bb T and from Eqs. (43), (44) and (46), it tions of the nonzero displacement on i j , corresponding to
follows that the six DOFs of Âi and Â j . The parallel and vertical direc-
tions of i j are assumed to be si j and ni j , then, the displace-
Ñ bxi = 0, Ñ byi = 0,
ments of the two directions are supposed to be
i i

Ñ bxi yi − Ñ byi xi = 0 (55) u s = L1 u si + L2 u s j
i i (57)
u n = H1 u ni + H2 u n j + H3li j φi + H4 li j φ j
The above equatio2ns ensure the RDPs of the boundary
NBFs, so the NBFs inside the UC also meet the RDPs. Thus, or a more concise form
PBC could be treated as a combination of LBC and periodic
2
additional terms. ul = N ui ûli + N φ ûφ (58)
The above analysis indicates that, only one decomposition i=1
process of stiffness matrix is required for all BVPs, which is
beneficial for the efficiency and storage space of mechanical where ûli = {u si , u ni }T denotes the translational DOFs
analyses. In addition, PBC could be extended to the UCs with referred to the local coordinate of the edge, i.e., si j and ni j ,
T
even edges and arbitrary material distributions. We have to and ûφ = φi , φ j is the rotational DOF. Li and Hi belong
point out that the node number on a pair of opposite edges to the function collections of linear interpolation and Hermite
should be the same, so as to apply multipoint constraints. interpolation

123
Comput Mech

Fig. 5 RABC for the UC with an arbitrary shape and its boundary NBFs. a shows the sketch of RABC and the meaning of the drilling DOF on
Ân ; b and c are the linear and Hermitian NBFs prescribed on the UC’s edges, respectively

L = [1 − ξ, ξ ] (59) displacement vector between the ( local)and global coordinate

c −s
H = 1−3ξ 2 +2ξ 3 , 3ξ 2 −2ξ 3 , ξ − 2ξ 2 + ξ 3 , ξ 3 − ξ 2 systems. Substitution of R = into Eq. (62) yields
s c
(60) the new interpolation function matrices
( )
the distributions of which on the UC’s edge are shown in Li c2 + Hi s 2 Li sc − Hi sc
N ui = RN ui RT = ,
Fig. 5b, c. Here, ξ = l/li j ∈ [0, 1] and l is the distance from Li sc − Hi sc Li s 2 + Hi c2
a point on i j to Âi , as shown in Fig. 5a. The normal and i = 1, 2 (63)
length of i j could be represented by ( )
−sH −sH
% & % & N φ = RN φ = li j 3 4
(64)
sin αi j s cH3 cH4
ni j = = ,
− cos αi j −c The above equations are the NBFs on the UC’s boundary,
'
2 2 which infer that the coupling terms of the NBFs are not zero.
li j = xi − x j + yi − y j (61) Table 1 shows the NBFs on i j corresponding to the DOF of
It could be indicated that the boundary displacement is lin- Âi and Â j .
ear in si j , and Hermite cubic polynomial in ni j . The linear It should be noted that there are no rotational angle DOFs
displacement fields could be reconstructed in both directions in the microscopical level which only exist in the macroscop-
si j and ni j . ical scale. The rotational DOF represents the rigid rotation
It could be verified that, conforming to the process in of the corresponding node in physics. As shown in the mag-
Sect. 3.1, the displacement interpolations in Eq. (58) sat-
isfy the RDPs, i.e., Eqs. (10), (12) and (17). However, the
Table 1 The NBFs on i j corresponding to six groups of BVPs con-
DOF of the same node would be different on two neighbor-
cerning to Âi and Â j
ing edges, since the interpolations in Eq. (57) are referred to
the local coordinate system. The displacement interpolations Node DOF u x on i j u y on i j
could be transformed into the global coordinate system, xoy,
Âi û xi L1 c2 + H1 s 2 L1 sc − H1 sc
after applying the rotational transformation to Eq. (58):
û yi L1 sc − H1 sc L1 s 2 + H1 c2

2
2
ϕ̂i −s H3 li j −s H4 li j
ug = RN ui ûli + RN φ uφ = N ui ugi + N φ uφ (62) Â j û x j L2 c2 + H2 s2 L2 sc − H2 sc
i=1 i=1
û y j L2 sc − H2 sc L2 s 2 + H2 c2
where ug = Rul denotes the displacement in global coordi- ϕ̂ j c H3 l i j c H4 l i j
nate system, and ûli = RT ugi suggests the relation of node

123
Comput Mech

nification of Ân in Fig. 5a, the macroscopical node, Ân , is modulus and Poisson’s ratio. The same variables in Figs. 7
assumed to be shared by four UCs, C1 − C4 , and a rigid rota- and 8 will not be explained again. The NBFs with respect to
tion with an angle, ϕ0 , takes place at Ân . Then, each edge the macroscopical DOF, û x1 and û y1 , are shown in the figure,
next to Ân rotates ϕ0 . Consequently, the same rotation would in which the length and orientation of the red arrows represent
take place at the material point of Ân , due to the microscopi- the size and direction of displacement vector, respectively.
cal elements next to Ân . Since each microscopical element is For the NBFs of the rectangular UC, shown in the Fig. 6a,
interpolated by the analytical base function, which satisfies the coupling terms of NBF on the boundary are zero, while
the rotational RDP automatically. that inside the UC are not. The coupling terms inside the UC
We can show that the boundary NBFs in Eq. (62) sat- reflect the influence of the macroscopical DOF, û x1 , on the
isfy the RDPs. For example, substituting the rotational rigid NBFs in the y direction and such effects are not included
displacement of node Âi and Â j , i.e., ϕ − ŷi , x̂i , 1, − ŷ j , for the traditional analytical base function. In addition, the
T
x̂ j , 1 , into Eq. (62) and taking advantage of the geomet- NBFs for the other UCs, such as the irregular quadrilateral
rical relation li j = cxi j + syi j , the displacement of an and irregular hexagon, show the similar characteristic with
arbitrary point (x, y) on i j could be deduced as ϕ (−y, x). the rectangular one.
Thus, the rotational RDP is satisfied. For the translational The six error indicators in Eqs. (65) and (66) are listed on
RDPs, interested readers could verify that Eqs. (10) and (12) the right side of each UC. As one might expect, all the values
similarly. are small, thus, the NBFs-LBC satisfy the RDPs. Moreover,
In the numerical aspect, all BVPs result in the solution the RDPs have nothing to do with the geometries and material
of Eq. (24). The boundary NBFs, listed in Table 1, are pre- distributions of the UCs. Thus, the application scope for LBC
scribed in advance, and the interior NBFs could be solved has been extended a lot relative to the previous work [28].
by Eq. (25). Only one decomposition process of stiffness The displacement on the UC’s edge varies linearly, which
matrix is needed for RABC, similar to the solutions of the maybe a little rigid in some cases.
NBFs-LBC. Therefore, the computation quantity is of almost As shown in Fig. 7, the NBFs-PBC for all kinds of UCs,
no difference from LBC. RABC has no constrictions on the including the rectangular, regular hexagon, diamond and so
geometries and material distributions of the UCs, which is on, are depicted. The NBFs corresponding to the coarse-mesh
superior to PBC. One could refer to Sect. 4.4 for more details. DOF û x1 and û y1 are plotted. The coupling terms exist not
only inside the UC, but also on the edge, comparing with that
4.4 Numerical examples of LBC. Just as the NBFs of the rectangular UC shown in
Figs. 6a and 7a, the deformed edges for PBC become curved,
The indicators below are introduced to verify the RDPs of while that for LBC keep straight. The flexibility of boundary
NBFs: may decide the higher accuracy of PBC in most cases [28].
* * The error indicators are all small, which suggest the RDPs of
* *+
* *
ex = * N xi − 1x * 1x NBFs-PBC. Total amount of UC’s edge for PBC should be
* * even and the node number on the two opposite edges should
i
* *
* *+* * be the same, thus, the periodic multipoint constraints could
* *
ey = * N yi − 1 y * *1 y * be applied.
* *
i The NBFs-RABC are shown in Fig. 8, enumerating the
* *
* *+ UCs such as the rectangular, irregular quadrilateral and irreg-
* *
er ot = *rw + N xi ŷi − N yi x̂i * rw (65) ular hexagon. The vector diagram of the NBFs with respect
* *
i i to û x1 , û x1 and ϕ̂1 are depicted. Taking the rectangular UC
which denote the errors for the translational and rotational shown in Fig. 8a as an example, the displacement along
RDPs, respectively. · represents the second type norm of a the UC’s edge varies linearly and that perpendicular to the
matrix. Theoretically, three zero eigenvalues λi should exist edge Hermite cubic polynomial. As for the rotational angle,
for K E as shown in Eq. (7), thus, the following indicators are ϕ̂1 , the displacement vertical to the adjacent edges is stimu-
also considered: lated. It could be found that the NBFs-RABC are more flex-
ible than the NBFs-LBC concerning the deformation ver-
λ̂i = |λi | /λm , i = 1, 2, 3 (66)
tical to the UC’s edge. Thus, the NBFs-RABC could be
where λm is the smallest theoretical nonzero eigenvalue of more suitable to describe the bending deformation of the
KE. UC.
The NBFs-LBC for several types of UCs are depicted in The error indicators of Eqs. (65) and (66) are all small,
Fig. 6. The mesh and material information are referred to the which suggest the RDPs of the NBFs-RABC. The displace-
descriptions below the UC, in which N E and N N denote the ment on the UC’s edge is conforming, so no averaging oper-
number of the element and node, E and ν represent elastic ations are needed on the UC’s edge.

123
Comput Mech

Fig. 6 Distributions of the NBFs and the error indicators for LBC

123
Comput Mech

Fig. 7 Distributions of the NBFs and the error indicators for PBC

123
Comput Mech

Fig. 8 Distributions of the NBFs and the error indicators for RABC

123
Comput Mech

Some features of NBFs could be summarized based on the-

oretical analyses in Sect. 4.1–4.3 and above numerical exam-
ples: (1) the NBFs of vector fields contain coupling terms,
which take the mutual influence of different directions into
consideration and should be beneficial for the accuracy of the
results. (2) The NBFs are obtained by the solution of BVPs,
which represent the typical deformations of the UC. (3) The
NBFs are discretized inside the UC in nature, so that there
are no difficulties for NBFs to adapt to the UCs with intricate
geometries and material distributions. Even for the polygo-
nal UC, the construction of NBFs wouldn’t generate essential
difficulties. On the contrary, it may be hard for analytical base
functions to extend shape of element to arbitrary polygons.
(4) Each NBF corresponds to a macroscopical physical DOF
Fig. 9 Oversampling technique for the rectangular UC
on coarse mesh, which is convenient for assembling stiffness
matrix and applying boundary condition.
the x direction will not be affected due to the discussions in
Appendix 1.
The NBFs on are assumed to be φ x = φ x x1 , . . . , φ x x4 ,
5 Oversampling technique which are solved and extracted from the above BVPs. How-
ever, the Kronecker delta property of φ x xi shown in Eq. (9) is
Oversampling technique is first proposed by Hou et al. [2] not satisfied on Â j . Thus, it is not convenient to apply φ x xi
to overcome the difficulty due to scale resonance between directly on the boundary of the OrgUC. Consequently, the
the grid scale and the scales of continuous problem. The new boundary NBFs are introduced by a linear combination
technique is further developed by Zhang et al. [28] for the of φ x :
vector field analyses, with the aim to gain the oscillating
NBFs on the UC’s edge. In this section, the mechanism of
4
ψ x xi = C xi j φ x x j or ψ x = φ x C x (67)
the RDPs for the NBFs of oversampling technique will be
j=1
investigated in theory, which provides some illuminations for
constructing new NBFs with oversampling technique. The which is forced to satisfy the Kronecker delta property on Â j :
numerical examples on oversampling technique are not given
ψ x xi | Â j = δi j (68)
in this paper, and one could refer to [28,34].
Thus, the above NBFs could be exerted on directly. The
coefficient matrix, C x , in Eq. (67) can be solved by the fol-
5.1 Basic properties of oversampling technique lowing equation deduced from the Kronecker delta property

In this section, general properties about the oversampling T x Cx = I (69)

technique are discussed. To construct the oscillating NBFs where T x is a 4 × 4 matrix constructed by the value of φ x
on the boundary of the Original UC (OrgUC), K , an over- at Âi and I is a diagonal unit matrix. The NBFs, ψ y , could
sampling unit cell (OsUC) overlapping the OrgUC, K , is also be constructed similarly. The new NBFs calculated by
introduced, as shown in Fig. 9. For the most common rectan- Eq. (67) keep some properties of φ x below.
gular UC, as shown in Fig. 9, the NBF, φ x x1 , corresponding

to the DOF, û 1 , of Âi , is prescribed on . φ x x1 is equal to Property 5.1 Suppose that φ x satisfies the translational

one at Â1 and varies linearly along the two boundaries con- RDPs, so will ψ x .

nected by Â1 . A displacement constraint in the y direction is
Proof Due to the translational RDPs of φ x , one could obtain
applied on Â3 to avoid rigid displacements as shown in Fig.
9, which will be the same for different prescribed bound-
4

ary displacements, i.e., φ x xi , i = 1, . . . , 4, so that only one φ x xi = 1 or φ x 1 = 1 (70)

decomposition of the OsUC’s stiffness matrix is involved for i=1

solving four NBFs in the x direction. The coupling NBFs where 1 denotes a vector whose element are all one and its
on the boundary of the OsUC, φ yxi , i = 1, . . . , 4, should dimensionality is determined as appropriate. The dimension-
be solved by Eq. (97). Such BVPs lead to the Eq. (97) with ality of left vector 1 is four and the right one is the same as
the boundary condition similar to Eq. (100) and the NBFs in that of φ̂ x . The provision will not be emphasized hereinafter.

123
Comput Mech

Since T x is composed by the value of φ x at Â j , one could Then, combining Properties A.1 and A.3, an equation about
obtain the unit decomposition of T x as follows the coupling terms could be obtained

4

4
T xi = 1 or T x 1 = 1 (71) φ yxi = α1 (78)
i=1 i=1
and then
where α is determined by the constraint value ṽm in the y
C x 1 = C x T x 1 = I1 = 1 (72) direction (see Appendix 1), which we need not to care about.
In conclusion, the boundary NBFs obtained by the oversam-
Equation (72) indicates that the coefficient matrix, C x , also
pling technique satisfy the equation
satisfies the unit decomposition. Thus, utilizing Eqs. (67),
% &
(70) and (72), we obtain the the translational RDPs of ψ x 4
φ x xi 1
= (79)
ψ x 1 = φx Cx 1 = φx 1 = 1 (73) φ yxi α1
i=1

Property 5.2 Suppose that φ x satisfies the rotational RDP, which denotes a translational rigid displacement. Subse-
T
so will ψ x .
quently, the NBFs on , φ Txxi , φ Tyxi , could be obtained
Proof From Eq. (17), the rotational RDP of φ x degenerates by Eq. (24), and φ x xi should satisfy the unit decomposition
into due to the discussions in Sect. 3.2:

ŷ = T x ŷ , y = φ x ŷ (74)
4
φ x xi = 1 (80)
since the coupling terms of φ x are zero. Here, ŷ = ŷi 4×1

i=1
and ŷ = ŷi denote the y coordinate vector of Âi and Therefore, ψ x xi deduced from Eq. (67) satisfies the transla-
4×1

Âi , respectively, y is composed by y coordinates of micro- tional RDPs as a result of the Property 5.1. The same process
scopic nodes on . Due to Eqs. (67), (69) and (74), the equa- could be adopted to prove the translational RDPs of ψ y .
tion, implying the rotational RDPs of ψ x , could be obtained The rotational RDP for the oversampling technique in Eq.
(17) degenerates into
y = φ x ŷ = ψ x T x ŷ = ψ x ŷ (75)

4
4
The similar method could be employed to deduce ψ y , which x= ψ yyi x̂i , y = ψ x xi ŷi (81)
also satisfies the properties similar to Properties 5.1 and 5.2. i=1 i=1

where x and yare the coordinate of microscopical nodes on

5.2 Rigid displacement properties and x̂i , ŷi is the coordinate of Âi . Due to the Property
5.2, the second equation in Eq. (81) is equivalent to Eq. (74).
T Thus, to verify Eq. (74), one could firstly obtain the boundary
The NBFs on are assumed to be φ T T
x xi , φ yxi , in which
equation, by multiplying Eq. (76) by ŷi , and summing over
φ x xi are the prescribed NBFs in the x direction and φ yxi are the index i:
solved from Eq. (97). Thus, one could obtain an equation ⎧ ⎫ ⎧ ⎫
⎪ y ⎪
similar to Eq. (101) K̂ x x K̂ x y ⎨ ⎬ ⎨ 4 ⎬
f̂ ŷ
% & 4 = xi i (82)
K̂ x x K̂ x y φ x xi f̂ xi K̂ yx K̂ yy ⎩ ⎪ ⎪ ⎩ i=1
φ yxi ŷi ⎭ ⎭
= (76) i=1
0
K̂ yx K̂ yy φ yxi 0

4
K̂ x x K̂ x y The relation φ x xi ŷi = y , on account of φ x xi = φ yyi ,
where, K̂ = denotes the stiffness matrix corre- i=1
K̂ yx K̂ yy is made use of to deduce the above equation. Due to the
sponding to the DOF on and f̂ xi is the constraint reaction Property A.2 and Eq. (98), the property about the coupling
in the x direction. It could be inferred from the last section terms is obtained by
4
that φ x xi = 1. Thus, summing Eq. (76) over index i leads
4

i=1 φ yxi ŷi = −x + γ 1 (83)
to
⎧ ⎫ i=1
⎪1 ⎪ 4 % &
K̂ x x K̂ x y ⎨ ⎬ where γ is determined by the constraint value ṽm (see Appen-
f̂ xi
’
4 = (77)
K̂ yx K̂ yy ⎪ ⎩ φ yxi ⎪
⎭ i=1 0 dix 1), which we do not need to care about. Thus, the terms
i=1 on the left side of Eq. (82) could be written as

123
Comput Mech

4
φ x xi y and Zhang et al. [29], which may have some distinctions
ŷi = (84) comparing with the actual computations.
i=1 φ yxi −x + γ 1
Suppose that K is large enough and the structure is com-
The above equation represents the rigid displacement composed by Q coarse meshes, then, we have K ≈ Q · M,
posed by the rotational and translational components. Based where M is the number of microscopical nodes in each coarse
T
on the conclusions in Sect. 3.2, φ Txxi , φ Tyxi will also meet mesh on average. The macroscopical nodes with respect
the similar equation to the coarse mesh become P ≈ Q, due to the conclu-
sions in Appendix 2. The computation quantity for the con-
4
φ x xi struction of NBFs for each

ŷi =
y
(85) coarse mesh, as shown in Sect.
φ yxi −x + γ 1 4.1–4.3, is about O M 3 ; thus, the computation quantity
i=1
(just for solving the linear algebraic equation, which will
Thus, Eq. (74) has been proved. A similar way could be not beemphasized again)
for all coarse meshes
becomes

adopted to deduce the rotational RDPs of φ yyi , i.e., x = N · O M 3 ≈ K · O M 2 ≈ O K 3 /Q 2 . About O Q 3
4 flops are needed on the macroscopical level, and the total
φ yyi x̂i’ . Therefore, ψ x and ψ y in Eq. (67) also meet the
i=1 computation quantity could be obtained by
requirement of the rotational RDP.
In conclusion, the NBFs on fulfills the translational and S = O K 3 /Q 2 + O Q 3 = O K 3 /Q 2 + N 3 (86)
rotational RDPs, so will the NBFs on K . The NBFs of the
oversampling technique generally oscillate on the OrgUC’s Afterwards, two extreme situations about the above equation
edge for heterogeneous materials, which suggest the hetero- are investigated. (1) The amount of the macroscopical node
geneity of material. Four stiffness matrix decompositions are is almost equal to that of the fine mesh, i.e., Q ≈ K , as the
needed to calculate the NBFs for oversampling technique for amount of coarse mesh becomes
small enough.
Thus, Eq.
the quadrilateral UC, which leads to a little larger compu- (86) degenerates into O K + K 3 ≈ O K 3 . (2) On the
tation quantity comparing with the boundary conditions in contrary, as the size of coarse mesh becomes large enough, its
Sect. 4.1–4.3. One could refer to [28,34] for the discussions number in the whole structure
approximates to one, then Eq.
on numerical performance of the oversampling technique. (86) also tends to be O K 3 . The analyses state clearly that
too many or too few coarse meshes are both disadvantageous
on the computational efficiency. In other words, there exists a
6 Size effects suitable coarse-mesh number, which makes the computation
quantity minimum. Let the derivation of Eq. (86) be zeros,
For the homogenization method, different sizes of UCs have the optimal coarse-mesh number could be obtained by
considerable influence on the computational accuracy, which
2 3 0.2
is related to the material properties and analysis type. For N= K = 0.922K 0.6 (87)
3
example, the constitutive equations of periodic composite
materials show some dependence on the size of representative and the corresponding minimum computation quantity
volume element [41], and the recent work on this problem can becomes
refer to [9,42]. The UC’s size for EMsFEM also has effects
on both computational accuracy and efficiency, which have Ŝ = O 1.960K 1.8 (88)
been discussed in the paper of Hou et al. [30] and Zhang et al.
which indicates that the computation quantity for EMsFEM is
[29]. On the basis of their work, more exhaustive discussions
proportional to K 1.8 in ideal cases. The computation quantity
will be presented in this section.
relative to the direct solution is as follows
, , , 3
6.1 Size effects on algorithm complexity S = S O K 3 = 1 Q2 + Q K (89)

6.1.1 A special case of O K 3 the logarithmic curves of which versus the coarse-mesh num-
ber are shown in Fig. 10, corresponding to different number
Firstly, the flops for the solution of a linear algebraic equation of microscopical nodes, i.e., 1×105 , 1×106 , 1×107 , 1×108 .

are assumed to be proportional to K 3 , denoted as O K 3 , The minimum relative computation quantity
without thinking of the bandwidth and sparse storage, see the +
Appendix 2. Here, K is the total node number of a structure S min = O 1.960K 1.8 O K 3 = 1.960K −1.2
and O (·) represents a linear time operator in some software = 1.666Q −2 (90)
and hardware environments. The total computation quantity
is assumed to be about O (K ) in the work by Hou et al. [30] is depicted by the dotted line in the same figure.

123
Comput Mech

1
p − 1 2 p−1 2 p−1p
Q= K ,
p
⎛ ⎞
1− p p
p − 1 2 p−1 p − 1 2 p−1 p(1−p)
S min =⎝ + ⎠K 2 p−1 (92)
p p

It can be verified that the above equations degenerate into

Eqs. (89) and (90) as p = 3. The logarithmic curves of S
versus N are depicted in Fig. 11, with respect to different
values of p, i.e., 1, 1.5, 2, 2.5 and 3. Figure 13a, b are related
to K = 108 and K = 109 , respectively.
Several conclusions could be drawn from the figure. (1)
The trend of S versus to Q is the same as that shown in
Fig. 10, so the conclusions deduced from Fig. 10 are also
applicative to that of Fig. 11. (2) S diminishes with p for a
Fig. 10 The relative computation quantity versus the coarse-mesh fixed Q; for example, S is around 10−2 as p is equal to 1.5 as
number corresponding to different K shown in Fig. 11a, while decreases down to about 10−8 as p
is equal to three, which indicates that advantages on the com-
putation efficiency of EMsFEM will be more apparent as the
Some messages could be inferred from the figure. (1) The direct LDLT decomposition is employed to solve the equa-
values of S are almost less than one, which indicate that tion. Thus, the conclusions for p = 3 in the last subsection
less computation quantities are consumed for EMsFEM com- are the most optimistic for the computation quantity saving.
paring with the direct solutions. (2) Indeed, for each curve (3) One could find that S is equal to one as p approximates
there exists a minimum relative computation quantity, which to one. In this case, the computation quantity of EMsFEM is
diminishes with Q or K , as shown in the magnification of proportional with the scale of the problem.
the figure; at a fixed value of Q, S decreases with K , which The minimums of S min are marked by full circles as shown
means that the more the DOF of a FE model is, the more com- in Fig. 11, from which, it could be found that Q p , correspond-
putational resource will be saved for EMsFEM; (3) S tends ing to S min , increases with p. In general, for a fixed amount
to be stable at the vicinity of S min with respect to very large of microscopical nodes, the more the coarse-mesh number
K , such as the curve in case of K = 1×108 . In this situation, is, the higher accuracy will be obtained for EMsFEM.
more coarse meshes are probable to be helpful for the accu- The above analysis is based on the series computing. In
racy promotion without increasing too much computational fact, it is of considerable convenience for the parallel comput-
time. ing, as the UC is independent with each other. Readers could
refer to Sect. 7.2 for the details. Meanwhile, the NBFs are
6.1.2 General cases for O (K p ) just calculated once for some material nonlinear problems,
such as strain localization and elasto-plastic problems [29],
In this section, more general assumptions on the solu- and the equilibrium iteration is implemented on the coarse
tion of the linear equation, i.e., O (K p ), are adopted. mesh; in such cases, more computational resources could be
The case p = 3 represents the standard LDLT decom- saved.
position in the last section. The exponent p should be
approximately between one and three, considering the band-
6.2 Numerical example
width or sparsity of the stiffness matrix, see Appendix 2.
The analysis flow in the last section is also useful for
In this section, the homogeneous beam with sizes of 6 × 18
O (K p ), thus, the computation quantity and its relative value
is considered as shown in Fig. 12. The elastic modulus and
become
Poisson’s ratio of the materials are 2.0×1011 and 0.3, respec-
, p tively. The left side of the beam is fixed and a 1,000 concen-
S = O K p Q 1− p + Q p , S = Q 1− p + Q K
trated force is applied at the bottom right corner. The size of
(91) fine mesh is set to be 0.02 × 0.02, which leads to 270,000
fine elements and 271,210 nodes. The configuration of mesh
The optimal number of coarse mesh and the corresponding size for this example aims to be convenient for comparing
minimum computation quantity are expressed as, by means the computation quantity of each model. Such fine mesh is
of the limit analysis of above equations: surely not necessary in practice.

123
Comput Mech

Fig. 11 The relative computation quantity versus the coarse-mesh number for different p, the total microscopical nodes in a and b are 108 and
109 , respectively

where TF1 and TF2 denote the time corresponding to the

direct solutions. The relative time versus the coarse-mesh
number are depicted in Fig. 13a, b, respectively. The relative
time of LBC, PBC and RABC are given, from which, one
can find that the time decreases at first and then increases
with the coarse-mesh number; the trend is the same as that
shown in Figs. 10 and 11.
The minimum of T E x1 takes place at about Q 1 = 1, 200,
while the theoretical optimal value is about Q 0 = 1, 670
Fig. 12 The testing model for the comparison of algorithm complexity as Eq. (87). However, there are almost no affections on
the computation quantity, as N distributes in the interval of
(500, 3000), so the theoretical optimal value still show some
The following ten groups of the coarse-mesh size are cho- reasonability. It also could be found that the minimum of
sen T E x1 is less than 0.1, which means the efficiency of EMsFEM
for the matrix decompositions is ten times higher than that of
ac ∈ 3 2 1.2 1 0.6 0.5 0.4 0.3 0.2 0.1 (93)
the direct solutions. The similar trend of T E x2 versus to Q
which leads to 108/ac2 coarse meshes. The coarse meshes could be detected from Fig. 13b; about Q 2 = 500 minimizes
corresponding to ac ∈ {3, 2, 1.2, 1} are depicted at the left T E x2 within the interval between 0.1 and 0.2, which indi-
bottom of the beam with different rectangulars as shown in cates that efficiency of EMsFEM is about an order higher than
Fig. 12. that of the direct simulations. T E x1 and T E x2 for different
The computational program is implemented by MATLAB boundary conditions are almost the same for all coarse-mesh
with version R2010b and carried out on a 64 bits Windows number.
Server 2007 system with a 2.66 GHz CPU and 36 GBytes of The errors of the displacements and Von Mises Stress are
memory. The preprocessing is implemented by the commer- shown in Fig. 13c, d, respectively. It could be revealed that the
cial software ANSYS. errors decrease rapidly at the low number of coarse mesh, and
The variables TE x1 and TE x2 , which represents the time then tend to be stable with the higher coarse-mesh number,
for the LDLT decomposition and that from reading the pre- which indicates the convergence of EMsFEM. The errors of
processing files to obtaining the displacement results, are maximum displacements become stable at about 300 coarse
collected. The nondimensionalized time indicators are intro- meshes, while that of stress at about 1,200 coarse meshes;
duced, as a matter of convenience for comparisons thus, the necessary amount of coarse meshes should be great
, , than 1,000 and more coarse meshes are helpful for improving
T E x1 = TE x1 TF1 , T E x2 = TE x2 TF2 (94) the accuracy.

123
Comput Mech

Fig. 13 Relative computational time and accuracy versus the number of coarse mesh for the cantilever beam. a and b represent T E x1 and T E x2 ,
respectively. c and d are the errors of displacement and Von Mises stress, respectively

To know more influence of the coarse-mesh number on the advantages on efficiency for EMsFEM become more obvious
computation efficiency, the fine-mesh size of the structure in than the direct simulations, which exhibit a good perspective
Fig. 12 is altered to the denser one, i.e., 0.01 × 0.01, which on the large scale scientific computing. It could be found eas-
results in 1.0824 × 106 nodes. Twelve groups of coarse mesh ily that the inequality Q 0 > Q 1 > Q 2 is the same as that of
are chosen with sizes the last model. Thus, considering the efficiency and accuracy
simultaneously, theoretical coarse-mesh number in Eq. (87)
ac ∈ 3 2 1.2 1 0.6 0.5 0.4 0.3 0.2 0.1 0.06 0.05 (95) is an acceptable choice.

The relative time T E x1 and T E x2 for the new model are

depicted in Fig. 14a, b, the trends of which are absolutely 7 Numerical examples: applications for heterogeneous
the same as that in Fig. 13a, b. As the coarse-mesh num- materials
ber becomes Q 1 = 2, 700, T E x1 descends to the minimum
value, while the theoretical optimal coarse-mesh number in Many kinds of engineering materials appear heterogeneous
Eq. (87) is Q 0 = 3, 850. However, the computation quantity in the microscopical level, such as composites [16], on
differs a little at the range Q ∈ [2700, 10800], so the theo- the section of which, there are many elliptical impurities
retical optimal coarse-mesh number also show some reason- approximately, and concrete [7], containing many disorderly-
ability. Comparing with the relative time in Figs. 13 and 14, arranged aggregates with different sizes embedded in the
the minimum of relative time of the former is larger than the matrix. Modeling and simulating the mechanical behavior
later one. In simple words, as the node number increases, the of such kinds of materials are of significance.

123
Comput Mech

Fig. 14 Relative computational time versus coarse-mesh number, a and b represent T E x1 and T E x2 , respectively

Fig. 15 The geometry of ellipse impurities or holes

Two main concepts to generate heterogeneous materials

can be distinguished, i.e., processing the sequential section-
ing image of natural materials and the artificial generation Fig. 16 Unit cells with impurities or holes corresponding to different
parameters. For a, a, b and ϕ are fixed and the center coordinate (xc , yc )
[7]. The second type material generator has been investi- is randomly distributed. In b, a is set to be equal to b and ϕ is out-of-
gated by many researches, such as generator by Clément et operation, meanwhile, a and (xc , yc ) are generated randomly. For c, ϕ
al. [16], based on the statistical theory and Galli et al. [43], and (xc , yc ) are controlled periodically. In d, ϕ conforms to the random
on the basis of Delaunay tetrahedron. In the paper, only the distribution and the other parameters are fixed
two-dimensional cases are considered and the impurities or
holes included in the heterogeneous materials are supposed
to be ellipse, shown in Fig. 15, which follows the parametric
length of the long or short axis. The angle ϕ could be used
equation
for controlling the orientation of the ellipse.
As shown in Fig. 16, all kinds of UCs with elliptical impu-
x = xc + a cos θ cos ϕ − b sin θ sin ϕ rities generated by Eq. (96) are plotted. Thus, it can be seen
(96)
y = yc + a cos θ sin ϕ + b sin θ cos ϕ that various micro-structures could be designed by control-
ling the parameters in Eq. (96). One could refer to [7] for
Here, θ is the variable parameter, ranging from 0 to 2π ; more exhaustive descriptions for the three-dimensional gen-
(xc , yc ) denotes the central coordinate and a or b are the erators of heterogeneous materials.

123
Comput Mech

7.1 Cantilever beam with random holes

The cantilever beam with four large circular holes, as shown

in Fig. 17, is composed by the materials with random ellip-
tical holes in the microscopical view. The size of the beam
is 10 × 40 and the radius of the hole is 2. The left side of
the beam is fixed, while each coarse-mesh node on the right
side are applied a force of 100. At the microscopical level, the
location, size and orientation of the elliptical hole conform to
random distributions, as shown in the magnified zone in Fig.
17. The elastic modulus and Poisson’s ratio of the materials
are 2.0 × 1011 and 0.3, respectively. In the macroscopical
level, there are 453 nodes, and 374 irregular-shape quadrilat-
eral UCs with feature size of one. There are 65,592 quadrilat-
eral elements with feature size of 0.1 and 71,941 nodes at the
fine scale. Each UC contains 175 fine elements on average.
It should be noted that, the drilling DOF for RABC on the Fig. 18 Transversal displacements for different models
fixed boundary is set to be zero during simulations, which
will not be mentioned repeatedly in the next example.
assumption of periodical distribution of the reinforcing phase
Comparisons of transversal displacement at y = 0 are
[44], while the UC shown in Fig. 17 is not periodic. There-
depicted in Fig. 18, which includes the reference solution
fore, it is reasonable to extend PBC to non-periodic structure
of the fine mesh, EMsFEM for LBC, PBC and RABC. As a
in the framework of EMsFEM. This example is not large
whole, the error of EMsFEM models relative to the fine-mesh
enough for comparing the efficiency, readers could refer to
solution is tiny, even the maximum error does not exceed 1%.
the next example.
It could be found that, from the magnification of the tip, the
result of LBC seems to be a little stiffer than the other solu-
tions, which means that PBC and RABC are more recom- 7.2 Heterogeneous structure with random impurities
mended for the mechanical analysis of such kind of materi-
als. The Von Mises stress for different models are depicted in The structure shown in Fig. 20 is a curve beam composed by
Fig. 19, from which, one could found that the results of EMs- the materials with random elliptical impurities. The bottom
FEM are almost the same as the reference solution, and stress of the beam is fixed and each coarse-mesh node on the right
concentration is also captured at the bottom and top of the side of the top is applied a downward force of 100. The het-
fixed side. The results indicate that the UC’s shape for EMs- erogeneous materials have two phases, i.e., matrix and impu-
FEM could extend to arbitrary quadrilateral, which expands rities, with the elastic modulus 2.0 × 1010 and 2.0 × 1011 .
application scope of EMsFEM to geometrically complicated The Poisson’s ratio of both phase are 0.3. There are 1500
domain. In addition, PBC is originally deduced based on the rectangular coarse meshes with feature size 0.1, correspond-
ing to 1661 nodes in the macroscopical level. Some ellipti-
cal impurities with the feature size 0.01 are included in the
each coarse mesh, as shown in the magnified zone A and
B in Fig. 20. The heterogeneity at the microscopical level
leads to 1,112,493 fine elements with the feature size 0.001
and 1,112,042 nodes, which is amount to about 2.2 million
DOFs.
The comparison of displacement for different models is
depicted in Fig. 21, which is located at the line P1 P2 P3 as
shown in Fig. 20. The displacement results of EMsFEM for
LBC, PBC and RABC are almost the same as that of the fine
mesh. The Von Mises stress, shown in Fig. 22, draws the
similar conclusions. Especially, the stress concentration at
the corner is also captured well, which may be beneficial for
the nonlinear analyses. In conclusion, although EMsFEM
Fig. 17 The coarse-mesh model of the cantilever beam with random solve the final problem at the macroscopical level, almost
holes identical accuracy as the direct solution could be obtained,

123
Comput Mech

Fig. 19 Results of Von Mises stress for different models. a denotes the reference solution, b–d are the EMsFEM solutions for LBC, PBC, RABC

Fig. 21 Displacements of different models

Fig. 20 The coarse mesh of the curved beam with random impurities

which is mainly because the fine properties of NBFs. The consider the time for linear boundary condition. The relative
example also indicates the feasibility of PBC for the non- time tn /t0 is depicted in Fig. 23a. It could be found that the
periodic materials. relative time decreases with the number of processors. As
Parallel computing is implemented for this example to n = 1, then tn /t0 ≈ 0.2, and tn /t0 approximate to 0.02, as
compare the computational efficiency with the direct method. n = 10, which indicates that the efficiency of EMsFEM is 50
We utilize the parallel computational tool box in Matlab. times higher than the direct method. The scale-up, denoted
The time from assembling the stiffness matrix to achieving by t1 /tn , is shown in Fig. 23b. It could be inferred that the
the displacements is collected for both direct solutions and scale-up is linear with the number of processors approxi-
EMsFEM, which are assumed to be t0 and tn , respectively. mately, which means that a problem may be solved in half
Here, tn is the time corresponding to n processors. We have time when doubling the number of processors. The discrep-
known that, in Sect. 6.2, the time of EMsFEM for differ- ancy between the EMsFEM and the optimum may be induced
ent boundary conditions are almost the same. Thus, we just by communications between processors [45].

123
Comput Mech

Fig. 22 Results of Von Mises

stress for different models. a
denotes the reference solution,
b–d are the EMsFEM solutions
for LBC, PBC, RABC

Fig. 23 Time analysis for parallel computing of EMsFEM-LBC a The relative time of EMsFEM to the direct solution. b Scale-up for parallel
computing

8 Conclusions proposed in analytical form for the first time, which could
be considered as an extension of the traditional isoparamet-
The key of EMsFEM is the construction of NBFs. In ric interpolation and differs from that for the the scalar field.
the paper, the necessary conditions of NBFs for the two- We prove theoretically that if the NBFs on the boundary of
dimensional vector field, that is the RDPs, are investigated a UC satisfy the RDPs, then so will the NBFs inside the
comprehensively in theory. Therein, the rotational RDP is UC. The proposed conclusion has nothing to do with the

123
Comput Mech

geometries and material distributions of the UCs and could Appendix 1: Properties of BVPs for oversampling tech-
be considered as a basic principle to construct NBFs. The nique
principle transforms the construction of NBFs on the two-
dimensional domain into a line and alleviates the difficulties Before the displacement boundary conditions are applied, the
for constructing NBFs a lot. In addition, we find that only two-dimensional BVPs are assumed to be
one decomposition of the stiffness matrix is necessary for all % & % &
Kxx Kxy u fx
NBFs of each UC, which improves the efficiency a lot relative = (97)
K yx K yy v fy
to the former works. The RDPs of LBC and PBC are demon-
strated on the basis of the proposed theory. We find that the where u and v are the displacement vectors in the x and y
NBFs for PBC could be considered as a combination of LBC directions, respectively. Several properties about K are not
and periodic additional terms. A new rotational angle bound- so hard to be deduced.
ary condition is proposed, based on the membrane element T
with a drilling DOF. The new boundary condition is conduc- Property A.1 Given the rigid displacement 1T 0T or
T T T
tive to improve the computation accuracy relative to LBC and 0 1 in Eq. (97), then
alleviate the modeling complexity comparing with PBC. The
Kx x 1 = Kx y 1 = K yx 1 = K yy 1 = 0 (98)
oscillating NBFs constructed by oversampling technique are
also discussed in the theoretical aspect, which is helpful for Property A.2 Given the rotational rigid displacement
us to understand the mechanism for describing oscillating T T T
−y x in Eq. (97), then
displacement of heterogeneous materials.
The algorithm complexity of EMsFEM is analyzed from − Kx x y + Kx y x = −K yx y + K yy x = 0 (99)
both theoretical and numerical aspects. The results indicate
an interesting conclusion that there exists a suitable UC size, Property A.3 zero eigenvalue multiplicity of Kx x and K yy
which makes the total computation quantity minimum. Thus, is one, and the corresponding eigenvector is αI, α, α is an
we have a criterion to decide which size of UC approxi- arbitrary non-zero real number.
mates to the best efficiency. Meanwhile, as the problem’s The Properties A.1 and A.2 could be obtained directly by
scale increases, more computation resources could be saved substituting the corresponding rigid displacement into Eq.
relative to the direct solutions and the advantages of EMs- (97). For the Property A.3, assuming that multiplicity of zero
FEM will be more remarkable. In the paper, applications of eigenvalue of K x x and K yy is equal or great than two, then
EMsFEM to the heterogeneous materials with random holes the multiplicity of zero eigenvalue for K in Eq. (97) should be
or impurities are implemented and the parallel computing is equal or great than four, which is contradict with three zero
also realized. The results state the accuracy and efficiency of eigenvalues. Thus, zero eigenvalue for K x x and K yy should
the method for large scale problems. be less than two. In addition, from the Property A.1, there is
The present EMsFEM has also been applied to inelastic at least one zero eigenvalue of K x x and K yy . Therefore, the
problems, such as elasto-plastic analysis of heterogeneous Property A.3 follows immediately.
materials [29,35] and strain localization of lattice truss mate- Supposing that multiple displacement constraints are
rials [36]. Numerical examples for elasto-plastic problems applied in the x direction and only one in the y direction,
show excellent precision of the stress in the microscopical the final BVP comes down to Eq. (97), with the boundary
scale, even for the path-dependent cases with loading and condition
unloading. The constructions and the properties of NBFs for
inelastic problems are almost the same as presented in the u i = ũ i , on x and vk = ṽk , on y (100)
paper and the final problems are also solved in the macro- where x and y denote the node set applied the constraints in
scopical scale. Detailed implementations of the method for the x and y directions. Then, substituting the unique solution
inelastic problems are discussed in the above mentioned ref- T
of above BVP, uT vT , into Eq. (97), we have
erences. The computation practices in the present and pre- % & % &
vious publications demonstrate that EMsFEM could handle Kxx Kxy u fx
the large scale problems efficiently and provide a new tool = (101)
K yx K yy v 0
for large-scale scientific computation.
where f x represents the constraint reaction in the x direction.
Acknowledgments The supports of the National Natural Science It should be noted that constraint reaction corresponding to v
Foundation (11232003, 11072051, 51021140004), the 111 Project is 0, since only one displacement constraint is applied in the
( B08014) and the National Key Basic Research Special Founda-
y direction. Considering theProperty A.1, the new displace-
tion of China (2010CB832704) are gratefully acknowledged. The T
authors also thank H. Liu, J. Lü, D.S. Yang and L. Zhang for useful ment superposed by uT vT and the rigid displacement
discussions. in the y direction, i.e.,

123
Comput Mech

% & % & % &

U u 0 Some conclusions on the computation quantity are review-
= +γ (102)
V v 1 ed for the direct solution methods of the above equation.
Assuming K is a symmetric dense matrix with a size of n ×n,
is also the solution of Eq. (101), which indicates the following
(1) the flops are about 13 n 3 +2n 2 if the standard LDLT decom-
properties.
position is employed. In case n is large enough, 2n 2 could
be ignored, which lead the approximate total flops, 13 n 3 . (2)
1. If the constraint on Ak in the boundary condition (100) is Considering the bandwidth m of K and m << n, about 4m 2 n
replaced by that on Al with prescribed
% & displacement
% & %value
& flops are needed. (3) The case for a sparse matrix K becomes
Ul u 0
ṽl , the new solution becomes = + γl , much more complicated, in which non-zero elements of K,
Vl v 1
the sparse storage mode and some related special algorithms
which could be obtained by V l = vl + γl = ṽl , namely,
are all related to the computational time. In general, the flops
γl = ṽl − vl . Thus, the solutions for the different con-
are proportional to n in ideal case. The above conclusions
straints in the y direction could be deduced by adjusting
are referred to the monograph by Boyd and Lieven [46] and
γ.
Golub and Charles [47].
2. The solutions for different constraints in the yy direction
For the imporous plane FE model composed by quadrilat-
are assumed to be
eral elements, the following equation could be deduced as
% & % & % & % & % & % &
U1 u 0 U2 u 0 Nb Nb
= +γ1 , = +γ2 E d = Nn + − 1 = Nd − −1 (107)
V1 v 1 V2 v 1 2 2
(103) where E d , Nn and Nb are the number of elements, total
nodes and boundary nodes, respectively. In case all the ele-
the relative value of the two solutions is expressed by ments are uniform in sizes, the approximate relations could
be obtained by
% & % & % & % &
U 12 U2 U1 0 ac2 ac
= − = (104) Ed ≈ , Nb ≈
V 12 V2 V1 (γ2 − γ1 ) 1 (108)
ae2 ae

The above equation indicates that there is no influence where ac and ae denotes the characteristic size of structure
on the displacement in the x direction, U , even if the and element. If ac >> ae or ac /ae >> 1, then we have the
constraint in the y direction is applied on different nodes. inequality E d >> Nb from the above relations. Therefore,
3. The relative displacement for the arbitrary different nodes one can conclude that if the element size is small enough
Am and An is related to the characteristic size of the structure discretized
by the quadrilateral elements, the total amount of elements
and nodes are approximately equal, due to Eq. (108), i.e.,
V 1,mn = V 1,m − V 1,n = v m − v n = V 2,mn (105)
E d ≈ Nd .
which suggests that the relative displacement value of two
arbitrary nodes is the same for different constraints in the
References
y direction. Furthermore, the invariance of the relative
displacement field also indicates that the strain and stress 1. Wu L, Noels L, Adam L, Doghri I (2012) A multiscale mean-
are invariable for different constraints. field homogenization method for fiber-reinforced composites with
gradient-enhanced damage models. Comput Methods Appl Mech
Eng 233–236:164–179
The similar conclusions could be drawn by applying the mul- 2. Hou TY, Wu XH (1997) A multiscale finite element method for
tiple constraints in the y direction and only one constraint in elliptic problems in composite materials and porous media. J Com-
the x direction. Such features of BVPs are shared with that put Phys 134:169–189
of oversampling technique. 3. Zohdi TI (2004) Homogenization methods and multiscale model-
ing. in encyclopedia of computational mechanics. Wiley, Hoboken
4. Kuznetsov S, Fish J (2012) Mathematical homogenization theory
for electroactive continuum. Int J Numer Methods Eng 91:1199–
Appendix 2: Some conclusions on algorithm complexity 1226
and FE model 5. Fish J (2006) Bridging the scales in nano engineering and science.
J Nanopart Res 8:577–594
6. Horstemeyer MF (2010) Multiscale modeling: a review. In:
The most time-consuming segment of structure analysis lies Leszczynski J, Shukla MK (eds) Practical aspects of computational
in solving the following linear equation chemistry. Springer, Berlin, pp 87–135
7. Unger JF, Eckardt S (2011) Multiscale modeling of concrete: from
Ku = f (106) mesoscale to macroscale. Arch Comput Method E 18:341–393

123
Comput Mech

8. Belytschko T, Song JH (2010) Coarse-graining of multiscale crack 27. Abdulle A, Weinan E, Engquist B, Vanden-Eijnden E (2012) The
propagation. Int J Numer Methods Eng 81:537–563 heterogeneous multiscale method. Acta Numer 21:1–87
9. Kouznetsova V, Geers MGD, Brekelmans WAM (2002) Multi- 28. Zhang HW, Wu JK, Lü J, Fu ZD (2010) Extended multiscale finite
scale constitutive modelling of heterogeneous materials with a element method for mechanical analysis of heterogeneous materi-
gradient-enhanced computational homogenization scheme. Int J als. Acta Mech Sin 26:899–920
Numer Methods Eng 54:1235–1260 29. Zhang HW, Wu JK, Fu ZD (2010) Extended multiscale finite ele-
10. Feyel F (2003) A multilevel finite element method (FE2) to describe ment method for elasto-plastic analysis of 2D periodic lattice truss
the response of highly non-linear structures using generalized con- materials. Comput Mech 45:623–635
tinua. Comput Methods Appl Mech Eng 192:3233–3244 30. Babuška I, Osborn JE (1983) Generalized finite element methods:
11. Feyel F, Chaboche JL (2000) FE2 multiscale approach for mod- their performance and their relation to mixed methods. SIAM J
elling the elastoviscoplastic behaviour of long fibre SiC/Ti com- Numer Anal 20:510–536
posite materials. Comput Methods Appl Mech Eng 183:309– 31. Babuška I, Caloz G, Osborn JE (1994) Special finite element meth-
330 ods for a class of second order elliptic problems with rough coef-
12. Eshelby JD (1957) The determination of the elastic field of an ficients. SIAM J Numer Anal 31:945–981
ellipsoidal inclusion and related problems. Proc R Soc Lond A 32. Hou TY, Wu XH, Cai Z (1999) Convergence of a multiscale finite
Mat 241:376–396 element method for elliptic problems with rapidly oscillating coef-
13. Huang J, Krabbenhoft K, Lyamin A (2012) Statistical homogeniza- ficients. Math Comput 68:913–943
tion of elastic properties of cement paste based on X-ray microto- 33. Efendiev Y, Hou TY (2009) Multiscale finite element methods:
mography images. Int J Solids Struct 50:699–709 theory and applications. Springer, Berlin
14. Sengupta A, Papadopoulos P, Taylor RL (2012) A multiscale finite 34. Zhang HW, Liu H, Wu JK (2013) A uniform multiscale method for
element method for modeling fully coupled thermomechanical 2D static and dynamic analyses of heterogeneous materials. Int J
problems in solids. Int J Numer Methods Eng 91:1386–1405 Numer Methods Eng 93:714–746
15. Temizer I, Wriggers P (2011) Homogenization in finite thermoelas- 35. Zhang HW, Wu JK, Lv J (2012) A new multiscale computa-
ticity. J Mech Phys Solids 59:344–372 tional method for elasto-plastic analysis of heterogeneous mate-
16. Clément A, Soize C, Yvonnet J (2012) Computational nonlin- rials. Comp Mech 49:149–169
ear stochastic homogenization using a nonconcurrent multiscale 36. Zhang HW, Wu JK, Zheng YG (2012) An adaptive multiscale
approach for hyperelastic heterogeneous microstructures analysis. method for strain localization analysis of 2D periodic lattice truss
Int J Numer Methods Eng 91:799–824 materials. Philos Mag 92:3723–3752
17. Moraleda J, Segurado J, LLorca J (2009) Finite deformation 37. Zhang HW, Liu H, Wu JK, Zheng YG (2010) Plane 4 node gen-
of incompressible fiber-reinforced elastomers: a computational eralized isoparametric element. Chin J Comput Mech 27:397–402
micromechanics approach. J Mech Phys Solids 57:1596–1613 (in Chinese)
18. Nezamabadi S, Yvonnet J, Zahrouni H, Potier-Ferry M (2009) 38. Allman DJ (1984) A compatible triangular element including ver-
A multilevel computational strategy for handling microscopic tex rotations for plane elasticity analysis. Comput Struct 19:1–
and macroscopic instabilities. Comput Methods Appl Mech Eng 8
198:2099–2110 39. Hughes TJR, Brezzi F (1989) On drilling degrees of freedom. Com-
19. Kanouté P, Boso DP, Chaboche JL, Schrefler BA (2009) Multiscale put Methods Appl Mech Eng 72:105–121
methods for composites: a review. Arch Comput Method E 16:31– 40. Ibrahimbegovic A, Taylor RL, Wilson EL (1990) A robust quadri-
75 lateral membrane finite element with drilling degrees of freedom.
20. Bai YL, Wang HY, Xia MF, Ke FJ (2005) Statistical mesomechan- Int J Numer Methods Eng 30:445–457
ics of solid, linking coupled multiple space and time scales. Appl 41. Pecullan S, Gibiansky LV, Torquato S (1999) Scale effects on the
Mech Rev 58:372–388 elastic behavior of periodic andhierarchical two-dimensional com-
21. Lloberas-Valls O, Rixen DJ, Simone A, Sluys LJ (2011) Mul- posites. J Mech Phys Solids 47:1509–1542
tiscale domain decomposition analysis of quasi-brittle heteroge- 42. Sebsadji SK, Chouicha K (2012) Determining periodic represen-
neous materials. Int J Numer Methods Eng 89:1337–1366 tative volumes of concrete mixtures based on the fractal analysis.
22. Xiao SP, Belytschko T (2004) A bridging domain method for cou- Int J Solids Struct 49:2941–2950
pling continua with molecular dynamics. Comput Methods Appl 43. Galli M, Botsis J, Janczak-Rusch J (2008) An elastoplastic three-
Mech Eng 193:1645–1669 dimensional homogenization model for particle reinforced com-
23. Wagner GJ, Liu WK (2003) Coupling of atomistic and continuum posites. Comput Mater Sci 41:312–321
simulations using a bridging scale decomposition. J Comput Phys 44. Xia Z, Zhou C, Yong Q, Wang X (2006) On selection of repeated
190:249–274 unit cell model and application of unified periodic boundary con-
24. Abraham FF, Broughton JQ, Bernstein N, Kaxiras E (2007) Span- ditions in micro-mechanical analysis of composites. Int J Solids
ning the continuum to quantum length scales in a dynamic simula- Struct 43:266–278
tion of brittle fracture. Europhys Lett 44:783–787 45. Wriggers P (2008) Nonlinear finite element methods. Springer,
25. Broughton JQ, Abraham FF, Bernstein N, Kaxiras E (1999) Con- Berlin
current coupling of length scales: methodology and application. 46. Boyd S, Vandenberghe L (2004) Convex optimization. Cambridge
Phys Rev B 60:2391–2403 university press, Cambridge
26. Miller RE, Tadmor EB (2009) A unified framework and perfor- 47. Golub GH, Van Loan CF (1996) Matrix computations. Johns Hop-
mance benchmark of fourteen multiscale atomistic/continuum cou- kins University Press, Baltimore
pling methods. Model Simul Mater Sc 17:053001

123

!!! - Finite Element Method For Solids and Structures - A Concise Approach (2021, Cambridge University Press) - Libgen - Li
No ratings yet
!!! - Finite Element Method For Solids and Structures - A Concise Approach (2021, Cambridge University Press) - Libgen - Li
370 pages
Instruction 1115 Fuel Limit Adjustment
100% (1)
Instruction 1115 Fuel Limit Adjustment
9 pages
Installation Manual: Trailer Edition Precedent™ C-600, C-600M, S-600, S-700, S-600M and S-610M/DE
100% (2)
Installation Manual: Trailer Edition Precedent™ C-600, C-600M, S-600, S-700, S-600M and S-610M/DE
80 pages
Splines Standards and Calculation - Rudolf Och
50% (10)
Splines Standards and Calculation - Rudolf Och
10 pages
ICS R490说明书（ENGLISH） PDF
No ratings yet
ICS R490说明书（ENGLISH） PDF
14 pages
Multiscale Modeling of Composite Materials and Structures With ANSYS
100% (1)
Multiscale Modeling of Composite Materials and Structures With ANSYS
12 pages
Accelerated Stability Testing
No ratings yet
Accelerated Stability Testing
5 pages
Preprint - A Multiscale Virtual Element Method For The Analysis of Heterogeneous Media
No ratings yet
Preprint - A Multiscale Virtual Element Method For The Analysis of Heterogeneous Media
40 pages
Coda Et Al. - 2022 - Alternative Multiscale Material and Structures Mod
No ratings yet
Coda Et Al. - 2022 - Alternative Multiscale Material and Structures Mod
19 pages
A multiscale eigenelement method and its application to periodical composite structures
No ratings yet
A multiscale eigenelement method and its application to periodical composite structures
11 pages
A Three-Dimensional Computational Multiscale Micromorphic Analysis of Porous Materials in Linear Elasticity
No ratings yet
A Three-Dimensional Computational Multiscale Micromorphic Analysis of Porous Materials in Linear Elasticity
22 pages
Feyel 2003
No ratings yet
Feyel 2003
12 pages
10.1016 J.cma.2016.11.029 An Extended Cohesive Damage Model For Simulating Arbitrary Damage Propagation in Engineering Materials
No ratings yet
10.1016 J.cma.2016.11.029 An Extended Cohesive Damage Model For Simulating Arbitrary Damage Propagation in Engineering Materials
34 pages
Multiscale extended finite element method for deformable fractured porous media
No ratings yet
Multiscale extended finite element method for deformable fractured porous media
19 pages
The Reduced Model Multiscale Method (R3M) For The Non-Linear Homogenization of Hyperelastic Media at Finite Strains
No ratings yet
The Reduced Model Multiscale Method (R3M) For The Non-Linear Homogenization of Hyperelastic Media at Finite Strains
28 pages
GGGG
No ratings yet
GGGG
4 pages
FINITE ELEMENT METHOD L-01_250217_142231
No ratings yet
FINITE ELEMENT METHOD L-01_250217_142231
16 pages
Composites: Part B: Adjovi Tchalla, Salim Belouettar, Ahmed Makradi, Hamid Zahrouni
No ratings yet
Composites: Part B: Adjovi Tchalla, Salim Belouettar, Ahmed Makradi, Hamid Zahrouni
11 pages
10 1177@1464420719886887
No ratings yet
10 1177@1464420719886887
20 pages
Reduced Multiscale Method 1
No ratings yet
Reduced Multiscale Method 1
34 pages
Finite Element Method 1
No ratings yet
Finite Element Method 1
10 pages
ST3573 IMT7A GARCIA MACIAS JIMENA CaseStudyP2
No ratings yet
ST3573 IMT7A GARCIA MACIAS JIMENA CaseStudyP2
13 pages
Nano Mechanics and Materials:: Theory, Multiscale Methods and Applications
No ratings yet
Nano Mechanics and Materials:: Theory, Multiscale Methods and Applications
19 pages
Bordas Phu
No ratings yet
Bordas Phu
42 pages
XFEM Library
No ratings yet
XFEM Library
42 pages
Finite Element Methods
No ratings yet
Finite Element Methods
20 pages
1 s2.0 S0045782598002278 Main PDF
No ratings yet
1 s2.0 S0045782598002278 Main PDF
35 pages
Homogeneization Periodic Guy Bonnet IJSS 2016
No ratings yet
Homogeneization Periodic Guy Bonnet IJSS 2016
47 pages
A. A. Zhuravlev Et Al., 2020. Multiscale Discrete Element Modeling
No ratings yet
A. A. Zhuravlev Et Al., 2020. Multiscale Discrete Element Modeling
11 pages
21-Multiscale Modelling For Material Failure
No ratings yet
21-Multiscale Modelling For Material Failure
103 pages
Unit1 Introduction
No ratings yet
Unit1 Introduction
32 pages
Multiscale Modeling in Solid Mechanics - Computational Approaches (Computational and Experimental Methods in Structures) (PDFDrive) PDF
100% (1)
Multiscale Modeling in Solid Mechanics - Computational Approaches (Computational and Experimental Methods in Structures) (PDFDrive) PDF
347 pages
Generalized Multiscale Finite-Element Method (Gmsfem) For Elastic Wave Propagation in Heterogeneous, Anisotropic Media
No ratings yet
Generalized Multiscale Finite-Element Method (Gmsfem) For Elastic Wave Propagation in Heterogeneous, Anisotropic Media
45 pages
Review On Finite Element Method Erhunmwun, Id Ikponmwosa, Ub
No ratings yet
Review On Finite Element Method Erhunmwun, Id Ikponmwosa, Ub
4 pages
Finite Element Method: From Wikipedia, The Free Encyclopedia
No ratings yet
Finite Element Method: From Wikipedia, The Free Encyclopedia
13 pages
Department of Civil Engineering
0% (1)
Department of Civil Engineering
13 pages
Chap 0 Contents
No ratings yet
Chap 0 Contents
8 pages
Varying Shape, Boundary Conditions and Loads Solutions Obtained Are Approximate
No ratings yet
Varying Shape, Boundary Conditions and Loads Solutions Obtained Are Approximate
17 pages
FEM - Cont. & Chapt-1 - Slide Presentation
No ratings yet
FEM - Cont. & Chapt-1 - Slide Presentation
10 pages
December 24, 2009 17:17 WSPC/IJCM-j050 00195
No ratings yet
December 24, 2009 17:17 WSPC/IJCM-j050 00195
34 pages
Newrobustandefficient Global Iterations For Large Deformation Finite Element Analysis of Beams and Shells With Material Nonlinearity
No ratings yet
Newrobustandefficient Global Iterations For Large Deformation Finite Element Analysis of Beams and Shells With Material Nonlinearity
28 pages
10.1016 J.proeng.2016.12.135 Application of EFGM and XFEM For Fatigue Crack Growth Analysis of Functionally Graded Materials
No ratings yet
10.1016 J.proeng.2016.12.135 Application of EFGM and XFEM For Fatigue Crack Growth Analysis of Functionally Graded Materials
8 pages
Ifem051 w1
No ratings yet
Ifem051 w1
18 pages
Modelling and Effective Properties Prediction of Metal Foams
No ratings yet
Modelling and Effective Properties Prediction of Metal Foams
7 pages
Meshless Methods1
No ratings yet
Meshless Methods1
4 pages
Simulation of Lamb Waves Using The Spectral Cell Method: A B B A A B
No ratings yet
Simulation of Lamb Waves Using The Spectral Cell Method: A B B A A B
11 pages
STUDY UNIT 1-INTRODUCTION TO THE STIFFNESS METHOD
No ratings yet
STUDY UNIT 1-INTRODUCTION TO THE STIFFNESS METHOD
18 pages
1 Introduction
No ratings yet
1 Introduction
23 pages
The Atomic-Scale Finite Element Method: B. Liu, Y. Huang, H. Jiang, S. Qu, K.C. Hwang
No ratings yet
The Atomic-Scale Finite Element Method: B. Liu, Y. Huang, H. Jiang, S. Qu, K.C. Hwang
16 pages
Atomic Scale Fem
No ratings yet
Atomic Scale Fem
16 pages
1st Assignment Theory Answers
No ratings yet
1st Assignment Theory Answers
7 pages
Multi Scale Crack
No ratings yet
Multi Scale Crack
17 pages
A Strain Smoothing Method in Finite Elements For Structural Analysis
No ratings yet
A Strain Smoothing Method in Finite Elements For Structural Analysis
220 pages
1 s2.0 S0168874X23000203 Main
No ratings yet
1 s2.0 S0168874X23000203 Main
16 pages
Lectures in Finite Element Method Ph. D. Students Course/ Structures Engineering 2hours/week
No ratings yet
Lectures in Finite Element Method Ph. D. Students Course/ Structures Engineering 2hours/week
66 pages
ComputationalParticleMechanics AnisotropicMaterialPaper Watermarked PDF
No ratings yet
ComputationalParticleMechanics AnisotropicMaterialPaper Watermarked PDF
22 pages
A Review of The Extended Finite Element Method On Macrocrack and Microcrack Growth Simulations
100% (1)
A Review of The Extended Finite Element Method On Macrocrack and Microcrack Growth Simulations
39 pages
Finite Cell Method: H-And P-Extension For Embedded Domain Problems in Solid Mechanics
No ratings yet
Finite Cell Method: H-And P-Extension For Embedded Domain Problems in Solid Mechanics
13 pages
Finite Element Methods For Materials Modelling
No ratings yet
Finite Element Methods For Materials Modelling
9 pages
CH3 Fundamentals For Finite Element Method - v2
100% (1)
CH3 Fundamentals For Finite Element Method - v2
8 pages
Introduction
No ratings yet
Introduction
3 pages
Mechanical Properties of Nanostructured Materials: Quantum Mechanics and Molecular Dynamics Insights
From Everand
Mechanical Properties of Nanostructured Materials: Quantum Mechanics and Molecular Dynamics Insights
Abdolhossein Fereidoon
No ratings yet
Finite Element Method
From Everand
Finite Element Method
Gouri Dhatt
1/5 (1)
Applied Iterative Methods
From Everand
Applied Iterative Methods
Louis A. Hageman
No ratings yet
Graph Layout Support for Model-Driven Engineering
From Everand
Graph Layout Support for Model-Driven Engineering
Miro Spönemann
No ratings yet
WCTE 2018 Full-Paper Citra-PURBA
No ratings yet
WCTE 2018 Full-Paper Citra-PURBA
8 pages
Determiner, Number and Gender Change: Topic
No ratings yet
Determiner, Number and Gender Change: Topic
82 pages
International Bureau
No ratings yet
International Bureau
15 pages
Topic: Neoclassical Period in English Literature: WWW - Uttoron.academy
100% (1)
Topic: Neoclassical Period in English Literature: WWW - Uttoron.academy
56 pages
Sultan Mahmud: Road To BB Ad 2020
No ratings yet
Sultan Mahmud: Road To BB Ad 2020
4 pages
Math Model Test # 03 Problems & Solutions By:: Road To BB Ad 2020
No ratings yet
Math Model Test # 03 Problems & Solutions By:: Road To BB Ad 2020
4 pages
Sultan Mahmud: Road To BB Ad 2020
No ratings yet
Sultan Mahmud: Road To BB Ad 2020
4 pages
Sciencedirect: Incorporation of Material Variability in The Johnson Cook Model
No ratings yet
Sciencedirect: Incorporation of Material Variability in The Johnson Cook Model
8 pages
03... 1000 Difficult Gre Words
No ratings yet
03... 1000 Difficult Gre Words
16 pages
Student Guide - Module 2 Machine Learning
No ratings yet
Student Guide - Module 2 Machine Learning
50 pages
Mercedes Uputstvo (English)
No ratings yet
Mercedes Uputstvo (English)
2 pages
College of Home Economics, Azimpur, Dhaka-1205.: Seat Plan, Group:-Science
No ratings yet
College of Home Economics, Azimpur, Dhaka-1205.: Seat Plan, Group:-Science
3 pages
Q4 Intro To Philo M4
No ratings yet
Q4 Intro To Philo M4
9 pages
Magnetism Science Unit Plan
No ratings yet
Magnetism Science Unit Plan
7 pages
Amazing Math Race
No ratings yet
Amazing Math Race
2 pages
2018 Regional 6th Grade Reading
No ratings yet
2018 Regional 6th Grade Reading
4 pages
Wire Ropes Used in Mines and Their Installation PDF
No ratings yet
Wire Ropes Used in Mines and Their Installation PDF
18 pages
PVS980 - 5MW Pre Commissioning Instruction
No ratings yet
PVS980 - 5MW Pre Commissioning Instruction
6 pages
QC Test Multiple Choice
No ratings yet
QC Test Multiple Choice
7 pages
DLL - Science 6 - Q3 - W2
No ratings yet
DLL - Science 6 - Q3 - W2
8 pages
Tire Choices Chariot Racing
No ratings yet
Tire Choices Chariot Racing
5 pages
Catalogo Flowtek
No ratings yet
Catalogo Flowtek
8 pages
2010 Ford Pickup SVT Raptor
100% (1)
2010 Ford Pickup SVT Raptor
115 pages
Analysis of Rolexs Marketing Strategy in Quartz C
No ratings yet
Analysis of Rolexs Marketing Strategy in Quartz C
5 pages
HIGH-DENSITY-PLANTING-IN-GUAVA-1
No ratings yet
HIGH-DENSITY-PLANTING-IN-GUAVA-1
13 pages
Qualitative Methods of Data Collection: The Case-Study Method Has Four Steps
No ratings yet
Qualitative Methods of Data Collection: The Case-Study Method Has Four Steps
4 pages
Ashrae Meteo (Islamabad Climate Data)
100% (1)
Ashrae Meteo (Islamabad Climate Data)
1 page
Killick 2014
No ratings yet
Killick 2014
17 pages
End of Year English Revision Booklet 2024
No ratings yet
End of Year English Revision Booklet 2024
15 pages
A Comprehensive Joint Learning System To Detect Skin Cancer
No ratings yet
A Comprehensive Joint Learning System To Detect Skin Cancer
43 pages
Thesis Sa Filipino Chapter 3
100% (2)
Thesis Sa Filipino Chapter 3
5 pages
Formula in Spider-Man (2002) and Spider-Man: Homecoming (2017) - An
No ratings yet
Formula in Spider-Man (2002) and Spider-Man: Homecoming (2017) - An
2 pages
Ecoesport 2013 Duratec Sigma
No ratings yet
Ecoesport 2013 Duratec Sigma
78 pages

Extended Multiscale Finite Element Method: Its Basis and Applications For Mechanical Analysis of Heterogeneous Materials

Uploaded by

Extended Multiscale Finite Element Method: Its Basis and Applications For Mechanical Analysis of Heterogeneous Materials

Uploaded by

Comput Mech

Extended multiscale finite element method: its basis

Received: 18 June 2013 / Accepted: 10 September 2013

Abstract Extended multiscale finite element method 1 Introduction

Fig. 1 The basic idea of EMsFEM

3 Basic properties of numerical base function

3.1 Rigid displacement properties

Similar to the analytical base function, N x xi and N yyi should

which is satisfied automatically after applying the boundary

could be expressed as, respectively, 1, q = mod (i + 1, 4) + 1, r = mod (i + 2, 4) + 1. The

N x xi = N yyi in Eq. (17), then we obtain the isoparametric

Equations (34) and (35) guarantee the translational and rota-

Subsequently, one could deduce the equation about rotational

2. The macroscopical DOFs, û xi and û yi , which correspond

Here, xb and yb are composed by the coordinates of micro-

1. The RDPs have nothing to do with the shape and material

L = [1 − ξ, ξ ] (59) displacement vector between the ( local)and global coordinate

Some features of NBFs could be summarized based on the-

In this section, general properties about the oversampling T x Cx = I (69)

ary displacements, i.e., φ x xi , i = 1, . . . , 4, so that only one φ x xi = 1 or φ x 1 = 1 (70)

where x and yare the coordinate of microscopical nodes on

It can be verified that the above equations degenerate into

where TF1 and TF2 denote the time corresponding to the

The relative time T E x1 and T E x2 for the new model are

Fig. 15 The geometry of ellipse impurities or holes

Two main concepts to generate heterogeneous materials

7.1 Cantilever beam with random holes

The cantilever beam with four large circular holes, as shown

Fig. 21 Displacements of different models

Fig. 22 Results of Von Mises

% & % & % &

You might also like

where x and yare the coordinate of microscopical nodes on