higher order thry
TRANSCRIPT
-
8/4/2019 Higher Order Thry
1/56
Higher-order theory for functionally graded materials
J. Aboudia, M.-J. Pinderab,*, S.M. Arnoldc
aTel-Aviv University, Ramat-Aviv 69978, IsraelbUniversity of Virginia, Charlottesville, VA 22903, USA
cNASA Glenn Research Center, Cleveland, OH 44135, USA
Received 18 June 1999; accepted 12 August 1999
Abstract
This paper presents the full generalization of the Cartesian coordinate-based higher-order theory for functionally graded materials
developed by the authors during the past several years. This theory circumvents the problematic use of the standard micromechanicalapproach, based on the concept of a representative volume element, commonly employed in the analysis of functionally graded composites
by explicitly coupling the local (microstructural) and global (macrostructural) responses. The theoretical framework is based on volumetric
averaging of the various field quantities, together with imposition of boundary and interfacial conditions in an average sense between the
subvolumes used to characterize the composites functionally graded microstructure. The generalization outlined herein involves extension
of the theoretical framework to enable the analysis of materials characterized by spatially variable microstructures in three directions.
Specialization of the generalized theoretical framework to previously published versions of the higher-order theory for materials functionally
graded in one and two directions is demonstrated. In the applications part of the paper we summarize the major findings obtained with the
one-directional and two-directional versions of the higher-order theory. The results illustrate both the fundamental issues related to the
influence of microstructure on microscopic and macroscopic quantities governing the response of composites and the technologically
important applications. A major issue addressed herein is the applicability of the classical homogenization schemes in the analysis of
functionally graded materials. The technologically important applications illustrate the utility of functionally graded microstructures in
tailoring the response of structural components in a variety of applications involving uniform and gradient thermomechanical loading.
1999 Elsevier Science Ltd. All rights reserved.
Keywords: Higher-order theory; Functionally graded materials; Microstructural tailoring/optimization
1. Introduction
Functionally graded materials (FGMs) are a new genera-
tion of engineered materials wherein the microstructural
details are spatially varied through non-uniform distribution
of the reinforcement phase(s), by using reinforcement with
different properties, sizes and shapes, as well as by inter-
changing the roles of reinforcement and matrix phases in a
continuous manner [1]. The result is a microstructure thatproduces continuously or discretely changing thermal and
mechanical properties at the macroscopic or continuum
scale. Examples illustrating different types of functionally
graded microstructures are presented in Fig. 1 (where the
terminology to describe the gradation type is used to reflect
the spatial variation of macroscopic properties). This new
concept of engineering the materials microstructure marks
the beginning of a revolution both in the materials science
and mechanics of materials areas as it allows one, for the
first time, to fully integrate the material and structural
considerations into the final design of structural compo-
nents. The rapidly growing interest in FGMs, originally
conceived in Japan, is evidenced by the large number of
recent conferences and special issues of technical journals
devoted to the analysis, design and fabrication of these
materials [27].
Functionally graded materials are ideal candidates forapplications involving severe thermal gradients, ranging
from thermal structures in advanced aircraft and aerospace
engines to computer circuit boards. In one such application,
a ceramic-rich region of a functionally graded composite is
exposed to hot temperature while a metallic-rich region is
exposed to cold temperature, with a gradual microstructural
transition in the direction of the temperature gradient,
Fig. 1(a). By adjusting the microstructural transition appro-
priately, optimum temperature, deformation and stress
distributions can be realized. This concept has been success-
fully employed to enhance the thermal fatigue resistance
Composites: Part B 30 (1999) 777832
1359-8368/99/$ - see front matter 1999 Elsevier Science Ltd. All rights reserved.
PII: S1359-8368(99) 00053-0
www.elsevier.com/locate/compositesb
* Corresponding author. Tel.: 1-804-924-1040; fax: 1-804-982-2951.
E-mail address: [email protected] (M.-J. Pindera)
-
8/4/2019 Higher Order Thry
2/56
and life of ceramic thermal barrier coatings [8,9]. Micro-
structural grading through non-uniform reinforcementspacing, Fig. 1(b), or through the use of different types of
reinforcement, Fig. 1(c), can also be effectively used to
reduce the mismatch in the thermomechanical properties
between differently oriented, adjacent plies in a laminated
plate. Thus, the reduction of thermally induced interlaminar
stresses at the free edge of a laminate (which result from the
large property mismatch between the adjacent plies) can be
realized by using the functional grading concept to smooth
out the transition between dissimilar plies. Along similar
lines, joining of dissimilar materials can be made more effi-
cient through the use of functionally graded joints [10 12].
Other benefits to be realized from the use of functionally
graded architectures include fracture toughness enhance-ment in ceramic matrix composites through tailored inter-
faces [13], and/or introduction of a second phase that creates
compressive stress fields in critical, crack-prone regions.
Many more applications of FGMs can be found in the afore-
mentioned conference proceedings and special issues of
technical journals, including the most recent focus on the
solar energy conversion devices [14], dental implants [15],
and naturally occurring biological FGMs [16,17].
Owing to the many variables that control the design of
functionally graded microstructures, full exploitation of the
FGMs potential requires the development of appropriate
J. Aboudi et al. / Composites: Part B 30 (1999) 777832778
Nomenclature
dpa ; h
qb ; l
rg dimensions of the subcell (ab g) in the (p,q,r)th generic cell
eabgijl;m;n coefficients of the total strain series expansion in the subcell (ab g)
kabgi coefficients of heat conductivity of the material in the subcell ( abg)
p; q; r indices used to identify the cell (p; q; r)q
abgi components of the heat flux vector in the subcell ( ab g)
vpqra;b;g volume of the subcell (abg) in the (p,q,r)th generic cell
uabgi displacement components in the subcell (abg)
xa1 ; x
b2 ; x
g3 local subcell coordinates
Cabgijkl elements of the stiffness tensor of the material in the subcell (abg)
1abgij local strain components in the subcell (abg)
1inabgij local inelastic strain components in the subcell (a bg)
Np, Nq, Nr number of cells in the x1, x2 and x3 directions, respectively
Qabgil;m;n
average values of the subcell heat flux component qabgi
when l
m
n
0; higher-order heat flux terms for
other values of l,m,n
Pn Legendre polynomial of order n
Rabgijl;m;n inelastic strain distribution components in the subcell (abg)
Sabgijl;m;n average values of the subcell (a bg) stress components s
abgij when l m n 0; higher-order stress compo-
nents for other values of l,m,n
T(abg) temperature field in the subcell (abg)
Tabgl;m;n temperature at the center of the subcell (abg) when l m n 0; coefficients associated with higher-order
terms in the temperature field expansion within the subcell ( abg) for other values of l,m,n
Wabgil;m;n displacement components at the center of the subcell (abg) when l m n 0; coefficients associated with
higher-order terms in the displacement field expansion within the subcell (abg) for other values of l,m,n
a ,b,g indices used to identify the subcell (abg)
m(abg) shear modulus of an isotropic material in the subcell (abg)
sabgij local stress components in the subcell (abg)
sTabgij local thermal stress components in the subcell (abg)
tabgijl;m;n coefficients of the stress series expansion in the subcell (ab g)
tTabgijl;m;n coefficients of the thermal stress series expansion in the subcell (abg)
Gabgij elements of the thermal tensor of the material in the subcell (abg)
-
8/4/2019 Higher Order Thry
3/56
modeling strategies for their response to combined thermo-
mechanical loads. Presently, most computational strategies
for the response of FGMs do not explicitly couple the mate-
rials heterogeneous microstructure with the structural
global analysis. Rather, local effective or macroscopic prop-
erties at a given point within the FGM are first obtained
through homogenization based on a chosen micromechanics
scheme, and subsequently used in a global thermomechani-
cal analysis, Fig. 2. In this approach, the local micromecha-nical analysis is carried out in an independent fashion from
the global macromechanical analysis, essentially decou-
pling the influence of the surrounding spatially inhomoge-
neous microstructure on the local response of the equivalent
(homogenized) continuum point. This is what is meant by
the absence of micro-macrostructural coupling in this type
of an analysis, as will be elaborated in the next section. The
exclusion of the possibility of coupling between local and
global effects often leads to potentially erroneous results in
the presence of macroscopically non-uniform material prop-
erties and large field variable gradients. This is particularly
true when the temperature gradient is large with respect to
the dimension of the inclusion phase, or the characteristic
dimension of the inclusion phase is large relative to the
global dimensions of the composite and the number of
uniformly or non-uniformly distributed inclusions is rela-
tively small. Perhaps the most important objection to
using the standard micromechanics approach based on the
concept of a representative volume element (RVE) in
the analysis of FGMs is the lack of a theoretical basis for
the definition of an RVE, which clearly cannot be unique in
the presence of continuously changing properties due to
non-uniform inclusion spacing [1820].
As a result of the limitation of the standard micromecha-
nics approaches in FGM applications, a new higher-order
micromechanical theory (HOTFGM), which explicitly
couples the local (microstructural) and global (macrostruc-
tural) effects, has been developed and applied to function-
ally graded composites. The theoretical framework is based
on volumetric averaging of the various field quantities,
satisfaction of the field equations in a volumetric sense,and imposition in an average sense of boundary and inter-
facial conditions between the subvolumes used to character-
ize the composites functionally graded microstructure. The
need for development of such a theory has been demon-
strated by comparison with results obtained using a standard
micromechanics approach that neglects the micro-macro-
structural coupling effects.
The original formulation of HOTFGM was developed for
the thermoelastic analysis of metal matrix composite plates
functionally graded in the thickness direction, constrained
from deforming in the plane of the plate, and subjected to a
through-thickness temperature gradient [2123]. Thisversion of the higher-order theory is known as HOTFGM-
1D as it reflects the one-directional grading character of the
materials microstructure. Extensive comparison between
the predictions of the one-directional version of the
higher-order theory and the results generated using the
finite-element analysis [24] and the boundary-element
analysis [25] established the theory as a viable tool for the
analysis of functionally graded composites. Subsequent
incorporation of inelastic constitutive models with tempera-
ture-dependent parameters for the response of the constitu-
ent phases into HOTFGM-1Ds theoretical framework
made possible the analysis of through-thickness function-
ally graded metal matrix composites over a wide tempera-ture range [26]. Relaxation of the constraint on the inplane
deformation through a partial homogenization procedure in
the non-functionally graded (periodic) directions extended
the theorys applicability to practical structural problems by
allowing simulation of generalized plane strain inplane
boundary conditions [27].
The analytical framework of HOTFGM, which results in
a closed-form system of equations for the unknown micro-
variables that govern the distribution of internal tempera-
ture, displacement and stress fields at each point within the
heterogeneous material, greatly facilitates the incorporation
J. Aboudi et al. / Composites: Part B 30 (1999) 777832 779
Fig. 1. Examples of different types of functionally graded microstructures.
-
8/4/2019 Higher Order Thry
4/56
of HOTFGM into an optimization algorithm as demon-
strated by Aboudi et al. [28] using HOTFGM-1D coupled
to the commercially available optimizer DOT [29]. In that
investigation, optimum through-thickness fiber distributions
were determined that minimized the inplane force and
moment resultants in unidirectionally and bidirectionally
reinforced MMC plates subjected to a through-thickness
thermal gradient.The recently developed two-directional version of the
higher-order theory, HOTFGM-2D, allows coupled micro-
macrostructural analysis of orthogonal composite laminates,
with finite dimensions in the plane containing arbitrarily
spaced inclusions or continuous fibers, subjected to
combined thermomechanical loading [30,31]. In particular,
the technologically important interlaminar stress fields in
laminated composites in the vicinity of the free edge were
analyzed and the advantages of functionally graded micro-
structures in reducing the free-edge stress concentrations
were demonstrated. In the most recent application of the
two-directional theory to functionally graded thermal
barrier coatings subjected to thermal gradient cyclic load-
ing, the creep induced stress redistribution was shown to
depend on the level of microstructural refinement, further
illustrating the shortcomings of the homogenization-based
analysis [32,33]. Summaries of selected developments of
the higher-order theory during the periods 1992 through
1995 and 1995 through 1997 presented in the above refer-ences have been provided by Pindera et al. [34,35].
As the development of the higher-order theory was driven
not only by our desire to address fundamental issues related
to microstructural effects in composites but also by various
technologically important applications, the notation
employed in constructing the theory was not always
uniform. In particular, the notation employed in construct-
ing the one-directional version was influenced by the nota-
tion used in developing the first-order homogenization
theory known in the literature as the generalized method
of cells [36,37]. In contrast, the extension of the theory
J. Aboudi et al. / Composites: Part B 30 (1999) 777832780
Fig. 2. Homogenization-based micromechanical analysis of functionally graded materials.
-
8/4/2019 Higher Order Thry
5/56
that enables the analysis of functionally graded microstruc-
tures in two directions necessitated the adaptation of a more
general notation. This most recent notation makes further
generalization of the higher-order theory to three dimen-
sions relatively straightforward.
One of the objectives of this review article, therefore, is to
present the full generalization of the higher-order theory to
three dimensions from which the two-directional and one-
directional versions are obtained as special cases using a
unified notation. In addition, the major results that illustrate
both the fundamental issues related to the influence ofmicrostructure on macroscopic and microscopic quantities
governing the response of composites and the technologi-
cally important applications are summarized. A major
theme that the authors wish to highlight is the issue of the
applicability of the classical homogenization schemes in the
analysis of functionally graded materials. The technologi-
cally important applications summarized herein, on the
other hand, illustrate the utility of functionally graded
microstructures in tailoring the response of structural
components under spatially uniform and non-uniform ther-
momechanical load histories.
2. Applicability of classical homogenization schemes
An alternative solution methodology for the thermal and
mechanical fields in functionally graded materials involves
a combination of the standard micromechanics and conti-
nuum approaches in which local (micromechanical) and
global (macromechanical) analyses are carried out inde-
pendently of each other, i.e. in a decoupled manner. In
this approach it is assumed that the heterogeneous micro-
structure of an FGM can be replaced by an equivalent
continuum with a set of macroscopic properties that vary
with spatial coordinates in a manner commensurate with
the materials heterogeneity, Fig. 2. The macroscopic prop-
erties at each point of the equivalent continuum are gener-
ated by assuming that it is possible to define an RVE in the
presence of spatially variable microstructure and then
carrying out an RVE-based micromechanical analysis
using a chosen micromechanics model. That is, the macro-
scopic properties are generated on the premise that no
coupling exists between local and global responses asthese properties are calculated without explicitly taking
into account the influence of the adjacent spatially variable
microstructural details. Subsequently, the continuum or
macrolevel results are obtained by solving the given
boundary-value problem of a homogenized medium with
spatially variable equivalent material properties calculated
in the manner described above. As stated previously, this
decoupled two-step procedure is the standard micro-macro-
mechanical approach currently employed by most
researchers working in the area of functionally graded
materials. With the knowledge of the macroscopic thermo-
mechanical fields, the stresses in the individual phases at acontinuum point may then be estimated by applying
average thermomechanical field quantities at the selected
point (or region), treating it as an RVE within the frame-
work of the chosen micromechanical model. A problem
arises, however, when the microstructural scale is large
relative to the thermomechanical field gradients so as to
invalidate the basic assumptions on which the RVE
concept is based. The situation becomes more complicated
when the microstructure changes continuously as in this
case, the micromechanical concept of an RVE, and thus
material property, is not well defined.
The various micromechanical approaches used to calcu-
late effective properties of composites include use of simpleReuss and Voigt hypotheses, self-consistent schemes and
their generalizations, differential schemes, the Mori
Tanaka method, concentric cylinder models, bounding tech-
niques and approximate (e.g. the generalized method of
cells) or numerical (e.g. finite-element or finite-difference)
analyses of periodic arrays of inclusions or fibers in the
surrounding matrix phase [38]. The central assumption in
applying these well-established techniques is the existence
of an RVE and the ability to apply homogeneous boundary
conditions to such an element, Fig. 3. These homogeneous
boundary conditions can be specified either in terms of
J. Aboudi et al. / Composites: Part B 30 (1999) 777832 781
Fig. 3. Calculation of macroscopic moduli based on the concept of a repre-
sentative volume element.
-
8/4/2019 Higher Order Thry
6/56
surface displacements
uiS 10ijxj 1
or in terms of prescribed surface tractions
Ti
S
s0ijnj
2
where ni is the unit outward normal vector on the boundary
surface S of the composite, xi are the Cartesian coordinates
of the surface, 10ij and s0ij are constants, and repeated index
implies summation. For a heterogeneous medium the
constants 10ij and s0ij are the volume averaged strains and
stresses under the prescribed boundary conditions given
by Eqs. (1) and (2), respectively. This result is a conse-
quence of the following relations:
1ij 1
V
V
1ijxk dV 1
V
S
1
2uinj ujni dS 3
sij 1
V
V
sijxk dV 1
V
S
1
2Tixj Tjxi dS 4
where Vis the volume enclosed by the surface S. The above
relations hold provided that: the displacements ui are contin-
uous; the tractions Ti are continuous at all interfaces of the
heterogeneous medium; and body forces vanish.
In practice, the average strains and stresses that result
from the application of homogeneous boundary conditions
are calculated for an RVE whose macroscopic behavior is
indistinguishable from the behavior of the composite-at-
large. By applying the homogeneous boundary conditions
to the bounding surface of the RVE, which are the same as
the boundary conditions applied to the entire composite, its
average behavior can be calculated. This average behavior,
in turn, defines the composites macroscopic properties. To
qualify as an RVE, the volume of the element used to calcu-
late average composite behavior must meet two criteria.
First, it must be sufficiently small with respect to the dimen-
sions of the composite-at-large in order to be considered a
material point in the equivalent homogeneous continuum
(i.e. hp H; see Fig. 4). Second, it must be sufficiently
large with respect to the inclusion phase (i.e. dp h) so
that to the first order the elastic strain energy induced by
both sets of homogeneous boundary conditions is the same,rendering the effective elastic properties independent of the
manner in which boundary conditions are applied [18]. In
this case, the microstructure of the RVE can be replaced by
a fictitious homogeneous material with effective or homo-
genized moduli or compliances, Cijkl or Sijkl, respectively,
that connect the average stress and strain quantities as
follows:
sij Cijkl 1kl 5
1ij Sijkl skl 6
where Sijkl C1ijkl ; Fig. 3. In the case of periodic fiber arrays,the repeating unit cell is interpreted as an RVE provided that
the homogeneous boundary conditions are replaced by
either symmetry conditions on the deformation of the unit
cell or periodic boundary conditions, depending on the type
of loading. It should be mentioned that the use of periodic
boundary conditions in conjunction with a multi-scale
asymptotic expansion of the displacement field in the
repeating unit cell forms the basis of the so-called homo-
genization methods for estimating the macroscopic proper-
ties of periodic composites [39,40].
Clearly, the range of applicability of the aforementioned
micromechanical approaches is limited to composites rein-
forced by numerous fibers with very small diameters such
as graphite or carbon fibers. In such composites, a typical
RVE contains a sufficiently large number of fibers while
occupying a very small volume of the entire composite,
allowing one to disregard boundary-layer effects near the
bounding surfaces of the RVE upon application of either
type of homogeneous boundary conditions. As a result,even in the presence of highly inhomogeneous deformation
gradients within the composite-at-large, the field quantities
within the RVE will not vary significantly, thereby permit-
ting the definition of a material property at a point in the
equivalent homogeneous continuum. In contrast, in compo-
sites with few and relatively large-diameter fibers (with
respect to the thickness of a single ply), the variation of
the quantities of interest within the RVE (assuming that it
can be defined) invalidates the basic assumptions on which
the concept of effective properties is based. These local
variations of the field quantities within the RVE may
give rise to unexpected phenomena rooted in the local-global coupling which is neglected in the traditional micro-
mechanical homogenization schemes. For instance, differ-
ent thermal conductivities of the individual phases together
with their directional arrangement may produce thermal
gradients in the individual phases which are quite different
from the thermal gradients in the homogeneous composite
with equivalent effective properties subjected to identical
boundary conditions, Fig. 4. This, in turn, may alter the
local conductivity characteristics and produce unexpected
effects such as localized hot spots for instance. The size
of the RVE in relation to the thickness of the composite
and the temperature gradient obviously will play an impor-
tant role in the above scenario.The preceding discussion raises questions about the
applicability of the traditional micromechanical approach
based on the concept of an RVE in the presence of large
thermal gradients and coarse or spatially variable micro-
structure. In light of this discussion, the current practice of
decoupling the local response from the global response by
calculating pointwise effective thermoelastic properties of
functionally graded materials without regard to whether the
actual microstructure admits the presence of an RVE, and
subsequently using these properties in the global analysis of
the heterogeneous material, remains to be justified. These
J. Aboudi et al. / Composites: Part B 30 (1999) 777832782
-
8/4/2019 Higher Order Thry
7/56
issues were discussed qualitatively as early as 1974 by
Pagano [41] with regard to mechanical loading of macro-
scopically homogeneous composites. Most recently, discus-
sions regarding the admissibility of the classical RVEconcept in the homogenization analysis of random heteroge-
neous media can be found in the works of Huet [42], Ostoja-
Starzewski and Schulte [43], Ostoja-Starzewski et al. [44]
and Ostoja-Starzewski [45]. To account for the materials
random microstructure, these authors used the concept of a
material window containing different numbers of inhomo-
geneously distributed inclusions in order to study the effect
of the window size on the local macroscopic properties.
In order to resolve the issue of the applicability of the
classical (in the sense of Hill) RVE concept in the analy-
sis of FGMs, a model is required that explicitly couples
the microstructural and macrostructural analyses. The
higher-order micromechanical theory, developed by the
authors during the past several years and presented in
the following section, is a step in this direction for appli-cations involving, but not limited to: composites with
uniformly or non-uniformly spaced; large-diameter fibers
subjected to through-thickness thermal gradients; finite
cross-section metal matrix composite laminates subjected
to combined thermomechanical loading; and thermal
barrier coatings. The generalization of this higher-order
theory for materials functionally graded in three orthogo-
nal directions is outlined next. Finally, we mention that
the present approach is based on deterministic considera-
tions, in contrast with the statistical approaches mentioned
above.
J. Aboudi et al. / Composites: Part B 30 (1999) 777832 783
Fig. 4. Applicability of the representative volume element for heterogeneous materials in the presence of a thermal gradient.
-
8/4/2019 Higher Order Thry
8/56
3. Generalized three-directional higher-order theory
3.1. Model overview
The generalized higher-order theory is based on the
geometric model of a heterogeneous parallelepiped occupy-
ing the region 0 x1 D; 0 x2 H; 0 x3 L
(Fig. 5). The loading applied on the bounding surface of
the parallelepiped may involve an arbitrary temperature or
heat flux distribution and mechanical effects represented by
a combination of surface displacements and/or tractions
consistent with global equilibrium requirements. Thecomposite may be reinforced by arbitrarily spaced planar
arrays of continuous fibers (with arbitrary in-plane spacing)
oriented along any of the three axes x1, x2, x3, or finite-length
inclusions that are arranged arbitrarily in the three planes.
The microstructure of the heterogeneous composite is
discretized into Np, Nq and Nrcells in the intervals 0 x1
D; 0 x2 H; 0 x3 L; respectively. The generic cell
(p,q,r) used to construct the composite, highlighted in Fig. 5,
consists of eight subcells designated by the triplet (abg),
Fig. 6, where each index a ,b,g takes on the values 1 or 2
which indicate the relative position of the given subcell
along the x1, x2 and x3 axis, respectively. The indices p, q
and r, whose ranges are p 1; 2; ;Np; q 1; 2; ;Nq andr 1; 2; ;Nr; identify the generic cell in the three planes.The dimensions of the generic cell along the x1, x2 and x3axes, d
p1 ; d
p2 ; h
q1 ; h
q2 ; and l
r1 ; l
r2 ; can vary in an arbitrary
fashion such that
D Np
p1dp1 dp2 H
Nqq1
hq1 hq2
L Nrr1
lr1 lr2
Given the applied thermomechanical loading, an approx-
imate solution for the temperature and displacement fields is
constructed based on volumetric averaging of the field equa-
tions together with the imposition of boundary and continu-
ity conditions in an average sense between the subvolumes
used to characterize the materials microstructure. This is
accomplished by approximating the temperature and displa-
cement fields in each subcell of a generic cell using a quad-
ratic expansion in the local coordinates xa; xb; xg
centered at the subcells mid-point. A higher-order repre-
sentation of the temperature and displacement fields is
necessary in order to capture the local effects created bythe thermomechanical field gradients, the microstructure
of the composite and the finite dimensions in the function-
ally graded directions. This is in sharp contrast with
previous treatments involving fully periodic composite
media which employed linear expansions [38]. We note
that it is possible to specialize this temperature and displa-
cement field representation to model composites that are
functionally graded in one direction with the remaining
directions periodic, or composites that are functionally
graded in a plane with periodicity in the out-of-plane direc-
tion. In either case, the temperature and displacement fields
in the periodic directions are approximated using a linear
expansion in local coordinates. In the former case, the one-directional version of the theory is obtained, while the latter
case yields the two-directional theory. Both specialized
cases will be elaborated upon in Section 4.
The unknown coefficients associated with each term in
the temperature and displacement field expansions are
obtained by constructing systems of equations that satisfy
the requirements of a standard boundary-value problem for
the given field variable approximations. That is, the zeroth,
first, and second moments of the heat and equilibrium equa-
tions are satisfied in a volumetric sense. Similarly, the ther-
mal and heat flux, as well as the displacement and traction,
J. Aboudi et al. / Composites: Part B 30 (1999) 777832784
Fig. 5. Schematic of a composite functionally graded in three directions.
-
8/4/2019 Higher Order Thry
9/56
continuity conditions (the mechanical analogues) within agiven generic cell, and between a given generic cell and its
neighbors, are imposed in an average sense across the inter-
facial planes. The solution for these coefficients in the
generalized higher-order theory follows the general frame-
work for the solution of the corresponding two-directional
thermoelastic and thermoinelastic problems discussed
previously [30,31]. Therefore, only a summary of the
governing equations for the temperature and displacement
fields in the individual subcells within the rows and columns
of cells considered in solving the outlined boundary-value
problem is given in the following sections. Detailed deriva-
tion of these equations can be found in the above references.
3.2. Thermal analysis
Let the functionally graded parallelepiped be subjected to
steady-state temperature or heat flux distributions on its
bounding surfaces. Under steady-state heat conduction,
the heat flux field in the material occupying the subcell
(abg) of the (p,q,r)th generic cell, in the region xa1 12
dpa ; x
b2
12
hqb ; x
g3
12
lrg ; must satisfy:
21qabg1 22q
abg2 23q
abg3 0 7
where 21 2=2 xa1 ; 22 2=2 xb2 ; 23 2=2 xg3 : The
components qabg
i of the heat flux vector in this subcellare obtained from the Fouriers heat conduction law foranisotropic materials,
qabgi kabgij 2jTabg; i;j 1; 2; 3 8
where kabgij are the coefficients of heat conductivity of the
material in the subcell (abg), with kabgij kabgi dij (no sum
on i) for orthotropic materials, and no summation is implied
by repeated Greek letters in the above and henceforth.
The temperature distribution in the subcell (a bg) of the
(p,q,r)th generic cell, measured with respect to a reference
temperature Tref, is denoted by T(abg). We approximate this
temperature field by a second-order expansion in the local
coordinates xa1 ; xb2 ; x
g3 as follows:
Tabg Tabg000 xa1 Tabg100 xb2 Tabg010 xg3 Tabg001
1
23 xa21
dp2a
4
2 3T
abg200
1
23 x
b22
hq2b
4
2 3T
abg020
1
23 x
g23
lr2g
4
2 3T
abg002 9
J. Aboudi et al. / Composites: Part B 30 (1999) 777832 785
Fig. 6. Generic cell (p,q,r) of a composite functionally graded in three directions showing the dimensions and the designation convention of the internal subcells
(a b g).
-
8/4/2019 Higher Order Thry
10/56
where Tabg000 which is the temperature at the center of the
subcell, and Tabglmn l; m; n 0; 1; or 2 with l m n 2
are unknown coefficients which are determined from condi-
tions that will be outlined subsequently. It should be noted
that no terms of the form xi x
j i j appear in the tempera-ture field representation, as well as in the displacement field
representation given in the sequel, due to the averaging
procedure employed in our higher-order theory.
Given the seven unknown quantities associated with each
subcell (i.e. Tabg000 ; ; T
abg002 ) and eight subcells within each
generic cell, 56NpNqNr unknown quantities must be deter-
mined for a composite with Np, Nq, and Nrrows and columns
of cells containing arbitrarily specified materials. These
quantities are determined by first satisfying the heat conduc-
tion equation, as well as the first and second moment of this
equation, in each subcell in a volumetric sense in view of the
temperature field approximation given by Eq. (9). Subse-
quently, continuity of heat flux and temperature is imposed
in an average sense at the interfaces separating adjacent
subcells, as well as neighboring cells. Fulfillment of thesefield equations and continuity conditions, together with the
imposed thermal boundary conditions on the bounding
surfaces of the composite, provides the necessary 56NpNqNrequations for the 56NpNqNr unknown coefficients in the
temperature field expansion.
We begin the outline of steps to generate the required
number of equations by first considering an arbitrary
(p,q,r)th generic cell in the interior of the composite (i.e.
p 2; ;Np1; q 2; ;Nq1; and r 2; ;Nr1: Thisproduces 56Np 2Nq 2Nr 2 equations. The addi-tional equations are obtained by considering the boundary
cells (i.e. p 1;Np
;q 1
;Nq and r 1
;Nr). For thesecells, most of the relations obtained from considering the
interior cells will also hold, with the exception of some of
the interfacial continuity conditions between adjacent
interior cells which will be replaced by the specified
boundary conditions.
In the course of satisfying the steady-state heat equation
in a volumetric sense, it is convenient to define the follow-
ing flux quantities:
Qabgil;m;n
1
npqrabg
dpa=2d
pa=2
hqb=2h
qb=2
lrg=2l
rg=2
xa1 l xb2 m
xg3 nqabgi d xa1 d xb2 d xg3 10
where l; m; n 0; 1; or 2 with l m n 2; and vpqrabg d
pa h
qb l
rg is the volume of the subcell (a bg) in the (p,q,r)th
generic cell. For l; m; n 0; Qabgi0;0;0 is the volume averagevalue of the heat flux component q
abgi in the subcell,
whereas for other values of (l,m,n) Eq. (10) defines
higher-order heat fluxes. These flux quantities can be eval-
uated explicitly in terms of the coefficients Tabglmn by
performing the required volume integration using Eqs. (8)
and (9) in Eq. (10). This yields the following non-vanishing
zeroth-order and first-order heat fluxes in terms of the
unknown coefficients in the temperature field expansion:
Qabg10;0;0 kabg1 Tabg100 11
Qabg11;0;0 kabg1
dp2a
4T
abg200 12
Qabg
20;0;0 kabg
2 Tabg
010 13
Qabg20;1;0 kabg2
hq2b
4T
abg020 14
Qabg30;0;0 kabg3 Tabg001 15
Qabg30;0;1 kabg3
lr2g
4T
abg002 16
Subsequently, satisfaction of the zeroth, first and second
moment of the steady-state heat equation results in the
following eight relationships among the first-order heat
fluxes Qabgil;m;n
in the different subcells (abg) of the
(p,q,r)th generic cell, after some involved algebraic manip-
ulations (see [21,22]):
Qabg11;0;0=d2a Qabg20;1;0=h2bQabg30;0;1=l2gp;q;r 0 17where the triplet (abg) assumes all permutations of the
integers 1 and 2.
The continuity of heat fluxes at the subcell interfaces and
between individual generic cells, imposed in an average
sense, is ensured by:
12Q1bg11;0;0=d1 Q2bg10;0;0 6Q2bg11;0;0=d2p;q;r
Q2bg
10;0;0 6Q2bg
11;0;0=d2p1;q;r
0 18
Q1bg10;0;0 12 Q2bg10;0;0 3Q
2bg11;0;0=d2p;q;r
12Q2bg10;0;0 6Q2bg11;0;0=d2p1;q;r 0 19
12Qa1g20;1;0=h1 Qa2g20;0;0 6Qa2g20;1;0=h2p;q;r
Qa2g20;0;0 6Qa2g20;1;0=h2p;q1;r 0 20
Qa1g20;0;0 12 Qa2g20;0;0 3Q
a2g20;1;0=h2p;q;r
12 Qa2g20;0;0 6Qa2g20;1;0=h2p;q1;r 0 21
12Qab130;0;1=l1 Qab230;0;0 6Qab230;0;1=l2p;q;r
Qab230;0;0 6Qab230;0;1=l2p;q;r1 0 22
Qab130;0;0 12 Qab230;0;0 3Q
ab230;0;1=l2p;q;r
12Qab230;0;0 6Qab230;0;1=l2p;q;r1 0 23
Eqs. (18)(23) provide us with 24 additional relations
among the zeroth-order and first-order heat fluxes. These
J. Aboudi et al. / Composites: Part B 30 (1999) 777832786
-
8/4/2019 Higher Order Thry
11/56
relations, together with Eq. (17), can be expressed in terms
of the unknown coefficients Tabglmn by making use of
Eqs. (11)(16), providing a total of 32 of the required 56
equations necessary for the determination of these
coefficients in the interior (p,q,r)th generic cell.
An additional set of 24 equations necessary to deter-
mine the unknown coefficients in the temperature field
expansion is subsequently generated by the thermal
continuity conditions imposed on an average basis at
each subcell and cell interface. Imposing the interfacial
thermal continuity conditions we obtain the following
relations:
T1bg000 12 d1T1bg100
14
d21 T
1bg200 p;q;r
T2bg000 12 d2T2bg100
14
d22T
2bg200 p;q;r 24
T2bg000 12 d2T2bg100
14
d22 T
2bg200 p;q;r
T1bg
000 1
2 d1T1bg
100 1
4 d
2
1T1bg
200 p1;q;r
25Ta1g000 12 h1T
a1g010
14
h21T
a1g020 p;q;r
Ta2g000 12 h2Ta2g010
14
h22T
a2g020 p;q;r 26
Ta2g000 12 h2Ta2g010
14
h22T
a2g020 p;q;r
Ta1g000 12 h1Ta1g010
14
h21T
a1g020 p;q1;r 27
Tab1000 12 l1Tab1001
14
l21T
ab1002 p;q;r
Tab2
000 12 l2T
ab2
001 14 l
2
2Tab2
002 p;q;r
28
Tab2000 12 l2Tab2001
14
l22T
ab2002 p;q;r
Tab1000 12 l1Tab1001
14
l21T
ab1002 p;q;r1 29
which comprise the required additional 24 relations.
The steady-state heat equations, Eq. (17), together with
the heat flux and thermal continuity equations, Eqs. (18)
(23) and Eqs. (24)(29), respectively, form 56 linear
algebraic equations which govern the 56 field variables
Tabglmn in the eight subcells (abg) of an interior generic
cell
p; q; r
; p
2; ;Np1; q
2; ;Nq1; r
2; ;Nr1: For the boundary cells p 1;Np; q 1;Nqand r 1;Nr a different treatment must be applied. For
p 1; for instance, the above relations are operative, butrelations (18) and (19), on the other hand, which follow
from the continuity of heat flux between a given generic
cell and the preceding one are not applicable. They are
replaced by the condition that the heat flux at the interface
between subcell (1bg) and (2bg) of the cell (1,q,r) is
continuous, as well as the applied temperature relation at
the bottom surface x1 0 (see Fig. 5). For the cell p Np;the previous equations are applicable except Eq. (25) which
are obviously not operative. These equations are replaced by
the specific temperature applied at the top surface x1 D:The boundary conditions at these surfaces are
T1bg1;q;r Tbottomx2;x3; x11 12 d
11 30
T2bgNp ;q;r Ttopx2;x3; x21 12 d
Np2 31
where q 1; ;Nq; r 1; ;Nr:Similar reasoning holds for the cells q 1 and q Nq:
That is, the temperature in the cells (p,1,r) and (p,Nq,r) at the
x2 0 and x2 H(i.e. left and right) surfaces, respectively,must equal the applied temperature,
Ta1gp;1;r Tleftx1;x3; x12 12 h
11 32
Ta2gp;Nq ;r Trightx1;x3; x22 12 h
Nq2 33
where p 1; ;Np; r 1; ;Nr:Finally, the temperature in the cells (p,q,1) and (p,q,Nr) at
the x3 0 and x3 L (i.e. front and back) surfaces, respec-tively, must equal the applied temperature,
Tab1p;q;1 Tfrontx1;x2; x13 12 l
11 34
Tab2p;q;Nr Tbackx1;x2; x23 12 l
Nr2 35
where p 1; ;Np; q 1; ;Nq: Alternatively, it ispossible to impose mixed-boundary conditions involving
temperature and heat flux at different portions of the
boundary.
The governing equations for the interior and boundary
cells form a system of 56NpNqNr algebraic equations in the
unknown coefficients Tabglmn : Their solution determines the
temperature distribution within the functionally gradedcomposite subjected to the boundary conditions (30)(35).
The final form of this system of equations is symbolically
represented below
kT t 36where the structural thermal conductivity matrix k contains
information on the geometry and thermal conductivities of
the individual subcells (abg) in the NpNqNr cells spanning
the x1, x2 and x3 directions, the thermal coefficient vector T
contains the unknown coefficients that describe the tempera-
ture field in each subcell, i.e.
T T111
111 ; ; T222
NpNqNrwhere
Tabgpqr T000; T100; T010; T001; T200; T020; T002abgpqrand the thermal force vector t contains information on the
boundary conditions.
3.3. Mechanical analysis
Given the temperature field generated by the applied
surface temperatures and/or heat fluxes obtained in the
preceding section, we proceed to determine the resulting
J. Aboudi et al. / Composites: Part B 30 (1999) 777832 787
-
8/4/2019 Higher Order Thry
12/56
displacement and stress fields. This is carried out for arbi-
trary mechanical loading, consistent with global equili-
brium requirements, applied to the surfaces of the
composite.
The stress field in the subcell (a bg) of the (p,q,r)th
generic cell generated by the given temperature field must
satisfy the equilibrium equations
21sabg1j 22s
abg2j 23s
abg3j 0; j 1; 2; 3 37
where the operator 2ii 1; 2; 3 has been definedpreviously. The components of the stress tensor, assuming
that the material occupying the subcell (abg) of the
(p,q,r)th generic cell is orthotropic, are related to the
strain components through the familiar generalized
Hookes law:
sabgij Cijkl1abgkl 1inabgkl Gabgij Tabg 38
where Cijkl are the elements of the stiffness tensor, 1in
abg
klare the inelastic strain components, and the elements
Gabgij of the so-called thermal tensor are the products
of the stiffness tensor and the thermal expansion coeffi-
cients. In this paper, we consider either elastic ortho-
tropic materials or inelastic materials which are
isotropic in both elastic and inelastic domains. Hence,
Eq. (38) reduces to
sabgij Cijkl 1abgkl 2mabg1inabgij sTabgij 39
where m (abg) is the elastic shear modulus of the mate-
rial filling the given subcell (a bg), and the term
sTabgij ; henceforth referred to as thermal stress, standsfor the thermal contribution G
abgij T
abg: The compo-
nents of the strain tensor in the individual subcells
are, in turn, obtained from the strain-displacement rela-
tions
1abgij 12 2iu
abgj 2ju
abgi ; i;j 1; 2; 3 40
The displacement field in the subcell (abg) of the
(p,q,r)th generic cell is approximated by a second-order
expansion in the local coordinates xa1 ; xb2 ; and x
g3 as
follows:
uabg1 Wabg1000 xa1 Wabg1100 xb2 Wabg1010 xg3 Wabg1001
1
23 xa21
dp2a
4
2 3W
abg1200
1
23 x
b22
hq2b
4
2 3W
abg1020
1
23 x
g23
lr2g
4
2 3W
abg1002 41
uabg2 Wabg2000 xa1 Wabg2100 xb2 Wabg2010 xg3 Wabg2001
1
23 xa21
dp2a
4
2 3W
abg2200
12
3 xb22 h
q2b4
2 3Wabg2020
1
23 x
g23
lr2g
4
2 3W
abg2002 42
uabg3 Wabg3000 xa1 Wabg3100 xb2 Wabg3010 xg3 Wabg3001
1
23 xa21
dp2a4
2 3W
abg3200
1
23 x
b22
hq2b4
2 3W
abg3020
1
23 x
g23
lr2g
4
2 3W
abg3002 43
where Wabgi000 ; which are the displacements at the center of
the subcell, and the higher-order terms Wabgilmn i 1; 2; 3
must be determined from conditions similar to those
employed in the thermal problem. In this case, there are
168 unknown quantities in a generic cell (p,q,r). The deter-
mination of these quantities parallels that of the thermalproblem. Here, the heat conduction equation is replaced
by the three equilibrium equations, and the continuity of
tractions and displacements at the various interfaces
replaces the continuity of heat fluxes and temperature.
Finally, the boundary conditions involve the appropriate
mechanical quantities. As in the thermal problem, we start
with the internal cells and subsequently modify the govern-
ing equations to accommodate the boundary cells p 1;Np; q 1;Nq; and r 1;Nr:
In the perfectly elastic case, the quadratic displacement
expansion, Eqs. (41) (43), produces linear variations in
strains and stresses at each point within a given subcell. In
the presence of inelastic effects, however, a linear strainfield generated by Eqs. (41)(43) does not imply the linear-
ity of the stress field due to the path-dependent deformation.
Thus the displacement field microvariables must depend
implicitly on the inelastic strain distributions, giving rise
to a higher-order stress field than the linear strain field
generated from the assumed displacement field representa-
tion. In the presence of inelastic effects, this higher-order
stress field is represented by a higher-order Legendre poly-
nomial expansion in the local coordinates. Therefore, the
strain field generated from the assumed displacement
field, and the resulting mechanical and thermal stress fields,
J. Aboudi et al. / Composites: Part B 30 (1999) 777832788
-
8/4/2019 Higher Order Thry
13/56
must also be expressed in terms of Legendre polynomials:
1abgij
l0
m0
n0
1 2l1 2m1 2n
p
eabgijl;m;nPlza1 Pmzb2 Pnzg3 44
sabgij
l0
m0
n0
1 2l1 2m1 2n
p
tabgijl;m;nPlza1 Pmzb2 Pnzg3 45
sTabgij
l0
m0
n0
1 2l1 2m1 2n
p
tTabgijl;m;nPlza1 Pmzb2 Pnzg3 46
where the non-dimensionalized variables z
i s, defined in theinterval 1 zi 1; are given in terms of the local
subcell coordinates as za1 xa1 =dpa =2; zb2 xb2 =hqb =2; and zg3 xg3 =lrg =2: For the given displace-
ment field representation, Eqs. (41)(43), the upper limits
on the summations in Eq. (44) become 1, while for the given
temperature distribution, Eq. (9), the upper limits on the
summations in Eq. (46) become 2. Alternatively, the
upper limits on the summations in Eq. (45) are chosen so
that an accurate representation of the stress field (which
depends on the amount of inelastic flow) is obtained within
each subcell. This will be discussed in Section 3.5 and
Appendix A where the solution of the resulting equations
for the unknown coefficients appearing in the displacementexpansions is outlined. The coefficients e
abgijl;m;n; t
abgijl;m;n;
tTabgijl;m;n in the above expansions are determined as described
below.
The strain coefficients eabgijl;m;n are explicitly determined in
terms of the displacement field microvariables of Eqs. (41)
(43), using orthogonal properties of Legendre polynomials.
Similarly, the thermal stress coefficients eTabgijl;m;n can be
expressed in terms of the temperature field microvariables
Tabglmn of Eq. (9). For example,
eabg110;0;0 Wabg1100
and
tTabg110;0;0 Gabg11 Tabg000
The complete sets of non-zero strain and thermal stress
coefficients, eabgijl;m;n and t
Tabgijl;m;n; respectively, are given in
Appendix A.
The stress coefficients tabgijl;m;n are expressed in terms of
the strain coefficients, the thermal stress coefficients and the
unknown inelastic strain distributions, by first substituting
the Legendre polynomial representations for 1(abg), s(abg),
and sT(abg) into the constitutive equations, Eq. (39), and
then utilizing the orthogonality of Legendre polynomials:
tabgijl;m;n Cijkleabgkll;m;n tTabgijl;m;n Rabgijl;m;n 47
The Rabgijl;m;n terms represent inelastic strain distributions
calculated in the following manner,
Rabgijl;m;n mabgLlmn 1 1
1 1
1 1
1inabgij Plza1
Pmzb2 Pnzg3 dza1 dzb2 dzg3 48
where
Llmn 141 2l1 2m1 2n
pIt should be noted that the choice of the Legendre poly-
nomials in the above expansions is motivated by the simpli-
city of their orthogonal properties (i.e. the weight function is
equal to one). It is possible, of course, to use any other
orthogonal set of polynomials.In the course of satisfying the equilibrium equations in a
volumetric sense, it is convenient to define the following
stress quantities:
Sabgijl;m;n
1
npqrabg
dpa =2d
pa =2
hqb
=2
hqb
=2
lrg =2 l
rg =2
xa1 l xb2 m
xg3 nsabgij d xa1 d xb2 d xg3 49
For l m n 0; Eq. (49) provides volume average stres-ses in the subcell, whereas for other values of ( l,m,n) higher-
order stresses are obtained that are needed to describe the
governing field equations of the higher-order continuum.
These stress quantities can be evaluated explicitly in terms
of the unknown coefficients Wabgilmn by performing the
required volume integration upon substituting Eqs. (39),
(40) and (41)(43) in Eq. (49). This yields the following
non-vanishing zeroth-order and first-order stress compo-
nents in terms of the unknown coefficients in the displace-
ment field expansion:
Sabg110;0;0 Cabg11 Wabg1100 Cabg12 Wabg2010 Cabg13 Wabg3001
Gabg1 T
abg000 R
abg110;0;0 (50)
Sabg111;0;0
1
4d
p2a C
abg11 W
abg1200
1
12d
p2a G
abg11 T
abg100
1
2
3p dpa Rabg111;0;0 51
Sabg110;1;0
1
4h
q2b C
abg12 W
abg2020
1
12h
q2b G
abg11 T
abg010
1
23p hqb Rabg110;1;0 52
J. Aboudi et al. / Composites: Part B 30 (1999) 777832 789
-
8/4/2019 Higher Order Thry
14/56
Sabg110;0;1
1
4lr2g C
abg13 W
abg3002
1
12lr2g G
abg11 T
abg001
1
2
3p lrg Rabg110;0;1 53
with similar expressions for Sabg220;0;0; S
abg221;0;0; S
abg220;1;0;
Sabg
220;0;1; and Sabg
330;0;0; Sabg
331;0;0; Sabg
330;1;0; Sabg
330;0;1 and
Sabg120;0;0 Cabg66 Wabg1010 Wabg2100Rabg120;0;0 54
Sabg121;0;0
1
4d
p2a C
abg66 W
abg2200
1
2
3p dpa Rabg121;0;0 55
Sabg120;1;0
1
4h
q2b C
abg66 W
abg1020
1
2
3p hqb Rabg120;1;0 56
Sabg130;0;0 Cabg55 Wabg1001 Wabg3100Rabg130;0;0 57
Sabg131;0;0 14 dp2a C
abg55 W
abg3200 1
2
3p dpa Rabg131;0;0 58
Sabg130;0;1
1
4lr2g C
abg55 W
abg1002
1
2
3p lrg Rabg130;0;1 59
Sabg230;0;0 Cabg44 Wabg2001 Wabg3010Rabg230;0;0 60
Sabg230;1;0
1
4h
q2b C
abg44 W
abg3020
1
2
3p hqb Rabg230;1;0 61
Sabg23
0;0;1
1
4
lr2g C
abg44 W
abg2002
1
23p lrg R
abg23
0;0;1
62
where contracted notation has been employed for the stiff-
ness elements Cabgijkl :
Subsequently, satisfaction of the zeroth, first, and second
moments of the equilibrium equations results in the follow-
ing 24 relations among the volume-averaged first-order
stresses Sabgijl;m;n in the different subcells (abg) of the
(p,q,r)th generic cell, after lengthy algebraic manipulations
(see Refs. [21,22]):
Sabg1j1;0;0=d2a Sabg2j0;1;0=h2b Sabg3j0;0;1=l2gp;q;r 0 63where j
1; 2; 3 and, as in the case of Eq. (17), the triplet
(abg) assumes all permutations of the integers 1 and 2.The continuity of tractions at the subcell interfaces and
between adjacent cells, imposed in an average sense, is
ensured by the following relations:
12S1bg1j1;0;0=d1 S2bg1j0;0;0 6S2bg1j1;0;0=d2p;q;r
S2bg1j0;0;0 6S2bg1j1;0;0=d2p1;q;r 0 64
S1bg1j0;0;0 12 S2bg1j0;0;0 3S
2bg1j1;0;0=d2p;q;r
12S2bg1j0;0;0 6S2bg1j1;0;0=d2p1;q;r 0 65
12Sa1g2j0;1;0=h1 Sa2g2j0;0;0 6Sa2g2j0;1;0=h2p;q;r
Sa2g2j0;0;0 6Sa2g2j0;1;0=h2p;q1;r 0 66
Sa1g2j0;0;0 12 Sa2g2j0;0;0 3S
a2g2j0;1;0=h2p;q;r
12 S
a2g2j0;0;0 6S
a2g2j0;1;0=h2
p;q1;r 0 67
12Sab13j0;0;1=l1 Sab23j0;0;0 6Sab23j0;0;1=l2p;q;r
Sab23j0;0;0 6Sab23j0;0;1=l2p;q;r1 0 68
Sab13j0;0;0 12 Sab23j0;0;0 3S
ab23j0;0;1=l2p;q;r
12Sab23j0;0;0 6Sab23j0;0;1=l2p;q;r1 0 69
where j 1; 2, and 3.Eqs. (64)(69) provide us with 72 additional relations
among the zeroth-order and first-order stresses. These rela-tions, together with Eq. (63), can be expressed in terms of
the unknown coefficients Wabgilmn by making use of Eqs.
(50)(62), providing a total of 96 of the required 168 equa-
tions necessary for the determination of these coefficients in
the (p,q,r)th generic cell.
The additional 72 relations necessary to determine the
unknown coefficients in the displacement field expansion
are subsequently obtained by imposing displacement conti-
nuity conditions on an average basis at each subcell and cell
interface. This produces,
W1bgj000 12 d1W1bg
j100 14
d21 W
1bgj200p;q;r
W2bgj000 12 d2W2bg
j100 14
d22 W
2bgj200p;q;r 70
W2bgj000 12 d2W2bg
j100 14
d22 W
2bgj200p;q;r
W1bgj000 12 d1W1bg
j100 14
d21 W
1bgj200p1;q;r 71
Wa1gj000 12 h1Wa1g
j010 14
h21W
a1gj020p;q;r
Wa2gj000 12 h2Wa2g
j010 14
h22W
a2gj020p;q;r 72
W
a2gj
000
12
h2Wa2g
j
010
14
h22W
a2gj
020
p;q;r
Wa1gj000 12 h1Wa1g
j010 14
h21W
a1gj020p;q1;r 73
Wab1j000 12 l1Wab1
j001 14
l21W
ab1j002p;q;r
Wab2j000 12 l2Wab2
j001 14
l22W
ab2j002p;q;r 74
Wab2j000 12 l2Wab2
j001 14
l22W
ab2j002p;q;r
Wab1j000 12 l1Wab1
j001 14
l21W
ab1j002p;q;r1 75
which comprise the required additional 72 relations.
J. Aboudi et al. / Composites: Part B 30 (1999) 777832790
-
8/4/2019 Higher Order Thry
15/56
The equilibrium equations, Eq. (63), together with the
traction and displacement continuity equations Eqs. (64)
(69) and Eqs. (70)(75), respectively, form 168 equations in
the 168 Wabgilmn unknowns which govern the equilibrium of a
subcell (abg) within the (p,q,r)th generic cell in the inter-
ior. As in the thermal problem, a different treatment must be
adopted for the boundary cells (1,q,r), (Np,q,r), and (p,1,r),
(p,Nq,r), and (p,q,1), (p,q,Nr). For (1,q,r), the above relations
are operative, except Eqs. (64) and (65), which follow from
the continuity of tractions between a given generic cell and
the preceding one. These 12 equations must be replaced by
the conditions of continuity of tractions at the interior inter-
faces of the cell (1,q,r) and by the applied tractions at x1 0: For the cell (Np,q,r), the previously derived governing
equations are operative except for the 12 relations given
by Eqs. (71), which are obviously not applicable. These
are replaced by the imposed traction conditions at the
surface x1 D: Similar arguments hold for boundary cells(p,1,r), (p,Nq,r), and (p,q,1), (p,q,Nr).
The traction vector in the cells (1,q,r) and (Np,q,r) at thebounding surfaces must equal the applied surface loads,
s1bg1i
1;q;r tbottomx2;x3; x11 12 d11 76
s2bg1i
Np ;q;r ttopx2;x3; x21 12 dNp2 77
where q 1; ;Nq; r 1; ;Nr; and tbottom and ttop describethe spatial variation of these loads at the bottom and top
surfaces. Similar conditions hold for the traction vector in
the cells (p,1,r) and (p,Nq,r) at the left and right surfaces,
respectively,
sa1g
2i p;1;r
tleftx1;x3; x12 12 h
11 78
sa2g2i
p;Nq ;r trightx1;x3; x22 12 hNq2 79
where p 1; ;Np; r 1; ;Nr; and tleft and tright describethe spatial variation of these loads at the left and right
surfaces. Similar traction boundary conditions can be writ-
ten for the front and back faces. Alternatively, if the front or
back surfaces are rigidly clamped (say), then
uab1i
p;q;1 0; x13 12 l11 80
uab2i
p;q;Nr
0; x2
3
1
2lNr2
81
where p 1; ;Np; q 1; ;Nq: For other types of bound-ary conditions, Eqs. (76)(81) should be modified accord-
ingly.
Consequently, the governing equations for the interior
and boundary cells form a system of 168NpNqNr algebraic
equations in the unknown coefficients Wabgilmn : The final
form of this system of equations is symbolically represented
by
KU f g 82where the structural stiffness matrix K contains information
on the geometry and thermomechanical properties of the
individual subcells (abg) within the cells comprising the
functionally graded composite, the displacement coefficient
vector U contains the unknown coefficients that describe the
displacement field in each subcell, i.e.
U
U
111111 ; ; U
222NpNqNr
where
Uabg
pqr Wi000; Wi100; Wi010; Wi001;
Wi200; Wi020; Wi002abgpqr i 1; 2; 3and the mechanical force vector f contains information on
the boundary conditions and the thermal loading effects
generated by the applied temperature. In addition, the
inelastic force vector g appearing on the right-hand side of
Eq. (82) contains the inelastic effects given in terms of the
integrals of the inelastic strain distributions that are repre-
sented by the coefficients Rabgijl;m;n
: These integrals depend
implicitly on the elements of the displacement coefficient
vector U, requiring an incremental solution of Eq. (82) at
each point along the loading path.
3.4. Higher-order theory vs finite-element analysis
Despite some similarities between the finite-element
method and the outlined coupled higher-order theory, the
higher-order theory contains many unique features not
found in the traditional displacement-based finite-element
analyses. The common features include discretization of the
microstructure into subvolumes wherein the displacement
field is approximated by certain functions (in this casesecond-order expansions in the local subcell coordinates).
However, the present approach is based on the volumetric
averaging of the field equations (heat conduction and equi-
librium) in the individual subcells which, in turn, involves
the satisfaction of the zeroth, first, and second moments of
these equations in a volumetric sense. In contrast, the tradi-
tional finite-element analysis involves the minimization of
either an energy expression or a weak form of the field
equations taken over a subvolume (but not their higher
moments). The satisfaction of the higher moments of the
field equations in the present approach naturally gives rise to
the higher moments of the heat flux and stress quantities
given in Eqs. (10) and (49). The higher-order quantitiesappear in the higher-order moments of the field equations
that must be satisfied. The volume-averaged zeroth and
first-order quantities are similar to the force and moment
resultants introduced in the standard plate theories except
here we average throughout each subcells volume rather
than through the entire plates thickness. Such higher-order
quantities are not employed in the traditional finite-
element analysis based on the second-order displacement
expansion.
Additional differences between the present approach and
the traditional finite-element analysis involve the manner in
J. Aboudi et al. / Composites: Part B 30 (1999) 777832 791
-
8/4/2019 Higher Order Thry
16/56
which the continuity of displacements and tractions are
satisfied between the various subvolumes. In the present
formulation, both displacement and traction continuity
between the individual subcells is satisfied in an integral
(average) sense, in contrast with the traditional finite-
element analysis based on a second-order displacement
expansion where only the displacement continuity is speci-
fied between adjacent elements. Moreover, in the present
approach, exact displacement field continuity between
adjacent domains is not required, whereas finite-element
analysis requires satisfaction of exact displacement con-
tinuity at the nodes (even in the case of non-conformable
elements).
Finally, in the present theory, the governing field
equations and the displacement and traction continuity
conditions are always satisfied in a volumetric sense
irrespective of the subvolume discretization. Naturally,
more detailed (accurate) results are obtained upon further
discretization, but further refinement is not necessary to
ensure that the traction continuity is satisfied in an integralsense. This is in sharp contrast with the traditional dis-
placement-based finite-element analysis where refinement
typically leads to better satisfaction of the traction continu-
ity in a point-wise fashion, and thus to more accurate point-
wise results. The present theory makes possible consistent
analysis of different microstructures at different levels of
volume discretization, and the fully analytical nature of
the approach allows us to do this efficiently in a volumetric
sense, although not as accurately as the point-wise finite-
element analysis.
3.5. Solution of the governing equations
The choice of an appropriate technique for the solution of
Eq. (82) depends on the inelastic constitutive model
employed to calculate the inelastic strain distributions in
each subcell from which the coefficients Rabgijl;m;n can be
generated. For instance, if the classical incremental plasti-
city (PrandtlReuss) equations are employed to model the
inelastic response of the matrix phase, then Mendelsons
iterative method of successive elastic solutions is an appro-
priate technique for the determination of the plastic strains
needed in the solution of Eq. (82) at each increment of the
applied load [46]. This method has been employed by
Pindera et al. [47,48], in investigating the thermoplasticresponse of unidirectional metal matrix composites
subjected to axisymmetric loading for those situations
where rate effects can be neglected. An advantage of this
solution technique is its efficiency and relative quick
convergence even for relatively large load increments
[49]. If, on the contrary, a unified viscoplastic constitutive
theory is employed to model the inelastic response of the
matrix phase, then either an implicit or an explicit technique
can be employed to integrate the viscoplastic rate equations
at each increment of the applied load. The integration of
viscoplastic constitutive equations, however, may require
a substantial computational effort due to the potentially
stiff behavior of this class of equations.
In the examples presented in Section 7, three constitutive
models are employed to describe the inelastic response of
the matrix phase, namely the classical incremental plasticity
theory, the BodnerPartom unified viscoplasticity theory,
and the power-law creep model, briefly outlined in
Appendix A. The plasticity theory is employed to efficiently
model the inelastic constitutive response of the matrix phase
when rate effects can be neglected, whereas the more
computationally intensive BodnerPartom theory and the
power-law creep model are employed for those situations
where rate-dependent deformation must be taken into
account. It should be noted that the present formulation is
sufficiently general to accommodate other types of unified
viscoplastic theories.
4. Specialization of the higher-order theory
The generalized (i.e. three-directional) higher-order
theory can easily be specialized for materials functionally
graded in a given plane or along a specific direction. In the
former case we obtain the two-directional version of the
higher-order theory referred to in previous communications
as HOTFGM-2D, and in the latter case we obtain the one-
directional version referred to as HOTFGM-1D. Herein, we
briefly show how the generalized higher-order theory
presented in the preceding section can be specialized to
the above two cases.
4.1. Two-directional higher-order theory
To maintain consistency with our previous communica-
tions, we consider a composite functionally graded in the
x2x3 plane (see Fig. 7). Such a composite has a finite thick-
ness H and finite length L in the functionally graded direc-
tions and extends to infinity in the periodic x1 direction. In
the x2x3 plane, the composite is reinforced by an arbitrary
distribution of infinitely long fibers oriented along the x1axis or finite-length inclusions that have periodic spacing
in the direction of the x1 axis. Both aligned and arbitrary
fiber or inclusion architectures in the x2x3 plane, Fig. 7(a)
and (b), respectively, are admissible. The microstructure of
the heterogeneous composite is discretized into Nq and Nr
cells in the intervals 0 x2 H; 0 x3 L in the x2x3plane. The indices q and r of the generic cell p; q; r thatidentify its location in the x2x3 plane have the same ranges
as in the generalized higher-order theory, while the range of
the index p is infinite due to the microstructures periodicity
in the x1 direction. This index can therefore be either
suppressed or retained with the understanding that its
value is indeterminate. The dimensions of the generic cell
along the x1 axes, d1, d2, are fixed for a given configuration
whereas the dimensions along the x2 and x3 axes, hq1 ; h
q2 ;
and lr1 ; lr2 ; can vary in an arbitrary fashion as previously
discussed.
J. Aboudi et al. / Composites: Part B 30 (1999) 777832792
-
8/4/2019 Higher Order Thry
17/56
The temperature distribution T(abg)
in the subcell (ab g)of the (p,q,r)th cell is approximated as follows:
Tabg Tabg000 xb2 Tabg010 xg3 Tabg001
1
23 xa21
d2a
4
2 3T
abg200
1
23 x
b22
hq2b
4
2 3T
abg020
1
23 x
g23
lr2g
4
2 3T
abg002 83
In contrast with the generalized theory, the above represen-tation does not contain a linear term in the local coordinates
xa1 ; i.e.
Tabg100 0 84
(see Eq. (9)). This follows directly from the assumed peri-
odicity in the x1 direction and symmetry with respect to the
lines xa1 0 for a 1 and 2. In the two-directional theory,therefore, the temperature field within the subcell (abg) is
described by six unknown coefficients instead of seven.
Thus for a composite with Nq rows and Nr columns,
48NqNr unknown thermal coefficients must be determined
J. Aboudi et al. / Composites: Part B 30 (1999) 777832 793
Fig. 7. Composite with non-periodic fiber distributions in the x2x3 plane: (a) aligned inclusion architecture, (b) arbitrary inclusion architecture.
-
8/4/2019 Higher Order Thry
18/56
in order to be able to describe the temperature field in every
subcell (abg) of every generic cell p; q; r:The determination of these coefficients follows the same
methodology as outlined previously. The system of
equations for the unknown coefficients within the interior
subcells can be obtained from the generalized solution by
using Eq. (84) and suppressing those equations which
involve the variations associated with the index p appearing
in the superscripts, i.e. p 1; q; r; p; q; r and p; q; r;p 1; q; r: Thus the satisfaction of the zeroth, first, andsecond moment of the steady-state heat equation, Eq. (7),
results in eight relations among the first-order heat fluxes
Qabgil;m;n in the different subcells (abg) of the (p,q,r)th
generic cell which are the same as Eq. (17). Twenty addi-
tional relations are obtained among the zeroth and first-order
heat fluxes upon satisfaction of the continuity of heat fluxes
at the subcell interfaces and individual cells, whereas the
same number of additional equations is obtained upon
imposing thermal continuity at the subcell interfaces and
individual cells. The boundary subcells are dealt with inthe same manner as discussed previously. The above
relations provide the required 48NqNr equations for the
48NqNr unknown coefficients Tabg000 ; T
abg010 ; T
abg001 ; T
abg200 ;
Tabg020 ; T
abg002 ; and can be expressed in the same form as
Eq. (36).
Next, the displacement field in the subcell (abg) of the
(p,q,r)th generic cell is approximated by a second-order
expansion in the local coordinates xa1 ; xb2 ; and x
g3 as
follows, taking into account symmetry considerations:
uabg1 Wabg1000 xa1 Wabg1100 85
uabg2 Wabg2000 xb2 Wabg2010 xg3 Wabg2001
1
23 xa21
d2a
4
2 3W
abg2200
1
23 x
b22
hq2b
4
2 3W
abg2020
1
23 x
g23
lr2g
4
2 3W
abg2002 86
uabg3 Wabg3000 xb2 Wabg3010 xg3 Wabg3001
1
23 xa21
d2a
4
2 3W
abg3200
1
23 x
b22
hq2b
4
2 3W
abg3020
1
23 x
g23
lr2g
4
2 3W
abg3002 87
The first equation does not contain the linear terms in the
local coordinates xb2 and x
g3 ; i.e.
Wabg1010 Wabg1001 0 88
This follows from the assumed periodicity in the x1 direction
and symmetry with respect to xa1 0 a 1; 2: Further,the presence of the constant term W
abg
1000 in the firstequation, that represents subcell center x1 displacements,
produces uniform composite strain 111 upon application of
a partial homogenization scheme described in the next
section. This partial homogenization, which couples the
present higher-order theory and an RVE-based theory,
leads to an overall behavior of a composite, functionally
graded in the x2 and x3 directions, that can be described as
a generalized plane strain in the periodic x1 direction. Plane
strain behavior in the periodic direction is obtained by
setting the constant term to zero [27]. The absence of the
higher-order terms in the first equation, i.e.
Wabg
1200 Wabg
1020 Wabg
1002 0 89is a direct consequence of the periodicity in the out-of-plane
direction. The last two equations, on the other hand, do not
contain linear terms in the local coordinate xa1 ; i.e.
Wabg2100 Wabg3100 0 90
in order to ensure symmetry of the deformation field in the
out-of-plane direction. In the two-directional theory, there-
fore, the displacement field within the subcell (abg) is
described by 13 unknown coefficients instead of 21, noting
that the constant term Wabg1000 in the first equation will be
replaced by the applied strain111 through homogenization.Thus for a composite with Nq rows and Nr columns, 104NqNr
unknown coefficients must be determined in order to be able
to describe the displacement field in every subcell (abg) of
every generic cell (p,q,r).
As in the thermal case, the determination of these coef-
ficients follows the same methodology as outlined
previously. The system of equations for the unknown
coefficients within the interior subcells can be obtained
from the generalized solution by using Eqs. (88) (90),
and suppressing those equations which involve the varia-
tions associated with the index p appearing in the super-
scripts. Thus the satisfaction of the zeroth, first, and
second moment of the equilibrium equations, Eq. (37),results in 16 relations among the volume averaged first-
order stresses Sabgijl;m;n in the different subcells (abg) of
the (p,q,r)th generic cell which are the same as Eq. (63).
Forty-four additional relations are obtained among the
zeroth and first-order stresses upon satisfaction of the
continuity of tractions at the subcell interfaces and indi-
vidual cells. The remaining 44 relations necessary to
determine the unknown coefficients in the displacement
field expansion are obtained by imposing a partial homo-
genization procedure in the periodic x1 direction together
with displacement continuity conditions applied on the
J. Aboudi et al. / Composites: Part B 30 (1999) 777832794
-
8/4/2019 Higher Order Thry
19/56
average basis at each subcell and cell interface in the
functionally graded x2 and x3 directions.
The equations resulting from the partial homogenization
procedure are:
d1W1bg1100 d2W2bg1100p;q;r d1 d2 111 91
where 111 is the unknown uniform far-field strain in the
x2 x3 plane that is either specified or determined. In the
latter case, the additional equation for the unknown 111 is
obtained from the imposed s11: These equations replace
Eq. (70) in the generalized higher-order theory solution
for j 1: On the contrary, the remaining equations obtainedfrom the imposition of the continuity of displacements at
each subcell interface of the (p,q,r)th generic cell, as well as
between adjacent cells, in the functionally graded directions
are obtained by specializing Eqs. (71)(75) in the manner
described previously.
Finally, the boundary subcells are also dealt with in the
same manner as discussed previously. The above relations
provide the required 104NqNr equations for the 104NqNrunknown coefficients
Wabg1100; W
abg2000; W
abg2010; W
abg2001; W
abg2200; W
abg2020; W
abg2002
and
Wabg3000; W
abg3010; W
abg3001; W
abg3200; W
abg3020; W
abg3002
and can be expressed in the same form as Eq. (82).
4.2. One-directional higher-order theory
To maintain consistency with our original communica-
tions, we consider a functionally graded composite in the x1direction. Such a composite has a finite depth D in the
functionally graded direction and extends to infinity in
the x2x3 plane, Fig. 8. The composite is reinforced by
periodic arrays of fibers in the direction of the x2 axis or
the x3 axis, or both. In the direction of the x1 axis, the fiber
spacing between adjacent arrays may vary. The reinforcing
fibers can be either continuous or finite-length, Fig. 8(a) or
(b), respectively.The microstructure of the heterogeneous composite is
discretized into Np cells in the interval 0 x1 D: The
index p of the generic cell (p,q,r) that identifies its location
along the x1 axis has the same range as in the generalized
higher-order theory, while the range of the indices q and ris
infinite due to the microstructures periodicity in the x2 and
x3 directions. These indices can therefore be either
suppressed or retained with the understanding that their
values are indeterminate. The dimensions of the generic
cell along the x2 and x3 axes, h1, h2, and l1, l2, are fixed for
a given configuration whereas the dimensions along the x1
axis, dp1 ; d
p2 ; can vary in an arbitrary fashion as previously
discussed.
The temperature distribution T(abg) in the subcell
(abg) of the (p,q,r)th generic cell is approximated as
follows:
Tabg
Tabg
000 xa
1 Tabg
100 1
2 3 xa
2
1
dp2a
42 3Tabg200
1
23 x
b22
h2b
4
2 3T
abg020
1
23 x
g23
l2g
4
2 3T
abg002 (92)
The above representation does not contain linear terms in
the local coordinates xb2 and x
g3 ; i.e.
Tabg
010
T
abg
001
0
93
because of the assumed periodicity in these directions. In
the one-directional theory, therefore, the temperature field
within the subcell (abg) is described by five unknown
coefficients instead of seven. Thus for a composite with
Np planar slices in the functionally graded direction
containing inclusions with periodic spacing in the x2
x3 plane, 40Np unknown thermal coefficients must be
determined in order to be able to describe the tempera-
ture field in every subcell (abg) of every generic cell
(p,q,r).
The system of equations for the unknown coefficients
within the interior subcells can be obtained from thegeneralized solution by using Eq. (93) and suppressing
those equations which involve the variations associated
with the indices q and r appearing in the superscripts,
i.e. (p,q 1,r), (p,q,r) and (p,q,r), (p,q 1,r) for the q
variation, and p; q; r 1; (p,q,r) a nd (p,q,r), p; q; r1 for the r variation. Thus the satisfaction of the zeroth,first, and second moment of the steady-state heat equation,
Eq. (7), results in eight relations among the first-order
heat fluxes Qabgil;m;n in the different subcells (abg) of the
(p,q,r)th generic cell which are the same as Eq. (17).
Sixteen additional relations are obtained among the zeroth
and first-order heat fluxes upon satisfaction of the conti-
nuity of heat fluxes at the subcell interfaces and individualcells, whereas the same number of additional equations is
obtained upon imposing thermal continuity at the subcell
interfaces and individual cells. The boundary subcells are
dealt with in the same manner as discussed previously.
The above relations provide the required 40Np equations
for the 40Np unknown coefficients Tabg000 ; T
abg100 ; T
abg200 ;
Tabg020 ; T
abg002 ; and can be expressed in the same form as
Eq. (36).
Next, the displacement field in the subcell (a bg) of the
(p,q,r)th generic cell is approximated by a second-order
expansion in the local coordinates xa1 ; xb2 ; and x
g3 as
J. Aboudi et al. / Composites: Part B 30 (1999) 777832 795
-
8/4/2019 Higher Order Thry
20/56
follows, taking into account symmetry considerations:
uabg1
Wabg1000
xa1
Wabg1100
1
23 xa2
1
dp2a
42 3Wabg1200
1
23 x
b21
h2b
4
2 3W
abg1020
1
23 x
g23
l2g
4
2 3W
abg1002 (94)
uabg2 Wabg2000 xb2 Wabg2010 95
uabg3 Wabg3000 xg3 Wabg3001 96
The first equation does not contain linear terms in the local
coordinates xb2 and x
g3 ; i.e.
Wabg
1010 Wabg
1001 0 97in order to ensure symmetry of the deformation field in the
periodic directions. The second equation does not contain
linear terms in the local coordinates xa1 and xg3 ; i.e.
Wabg2100 Wabg2001 0 98
while the third equation does not contain linear terms in the
local coordinates xa1 and xb2 ; i.e.
Wabg3100 Wabg3010 0 99
This follows from the assumed periodicity in the x2 and x3
J. Aboudi et al. / Composites: Part B 30 (1999) 777832796
Fig. 8. Composite with non-periodic fiber distributions in the x1 direction: (a) unidirectionally reinforced material, (b) particulate inclusion reinforced material.
RCS denotes the representative cross section.
-
8/4/2019 Higher Order Thry
21/56
directions and symmetry with respect to xb2 0 b 1; 2
and xg3 0 g 1; 2: The constant terms Wabg2000 and
Wabg3000 in these equations, that represent subcell center
displacements in the x2 and x3 directions, respectively,
produce uniform composite strains 122 and 133 upon appli-
cation of a partial homogenization scheme. This partial
homogenization leads to an overall behavior of a composite
that can be described as a generalized plane strain in the
periodic directions. Plane strain behavior in the periodic
directions is obtained by setting the constant terms to zero
[21]. The absence of the higher-order terms in the last two
equations, i.e.
Wabg2200 Wabg2020 Wabg2002 0 100
and
Wabg3200 Wabg3020 Wabg3002 0 101
is a direct consequence of the periodicity in the x2 and x3
directions. In the one-directional theory, therefore, thedisplacement field within the subcell (abg) is described
by seven unknown coefficients instead of 21, noting that
the constant terms Wabg2000 and W
abg3000 in the last two equa-
tions will be replaced by the applied strains 122 and 122through homogenization. Thus for a composite with Npplanar slices in the functionally graded direction, 56Npunknown coefficients must be determined in order to be
able to describe the displacement field in every subcell
(abg) of every generic cell (p,q,r).
The system of equations for the unknown coefficients
within the interior subcells is obtained from the generalized
solution by using Eqs. (97)(101) and suppressing those
equations which involve the variations associated with the
indices q and rappearing in the superscripts. Thus the satis-
faction of the zeroth, first, a