Karihaloo 2001 Computers & Structures

download Karihaloo 2001 Computers & Structures

of 16

Transcript of Karihaloo 2001 Computers & Structures

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    1/16

    Homogenization-based multivariable element method for

    pure torsion of composite shafts

    B.L. Karihaloo a,*, Q.Z. Xiao a,b, C.C. Wu b

    a Division of Civil Engineering, School of Engineering, Cardi University, Queen's Buildings, P.O. Box 686, Cardi CF24 3TB, UKb Department of Modern Mechanics, University of Science and Technology of China, Hefei 230026, China

    Received 13 March 2000; accepted 28 June 2001

    Abstract

    Application of the hybrid stress element to torsion of composite shafts is restricted, if the volume fraction of re-

    inforcement is very large. The homogenization method is the most suitable for such problems because it gives not only

    the equivalent material properties but also detailed information of local elds with less computational cost. In this

    paper, the extension of the homogenization method to the pure torsion of composite shafts with reinforcing bers

    aligned along its axis and the application of high performance multivariable elements are studied. The incompatible

    element based on a modied potential principle and the enhanced-strain element based on the HuWashizu principle

    are appropriate for the analysis of the representative unit cell, whereas the hybrid stress element is appropriate for the

    macro-homogenized problem. Numerical examples are provided and discussed. 2001 Elsevier Science Ltd. All rights

    reserved.

    Keywords: Composite shaft; Enhanced-strain element; Homogenization; Hybrid element; Incompatible element; Representative unitcell; Torsion

    1. Introduction

    In the mechanics of composite materials, it is of great

    importance to determine the eective properties of the

    composite from the distribution and basic properties of

    constituents and the detailed distribution of elds on the

    scale of microconstituents [14]. The hybrid stress ele-

    ment introduced by the authors [5] is of course a usefultool in the analysis of composite shafts because it gives

    accurate results for the warping displacement, the angle

    of twist per unit length, as well as the shear stress by

    using a relatively coarse mesh. However, if the volume

    fraction of reinforcement is very large, it is not realistic to

    obtain the microelds by this method because the degrees

    of freedom needed to model the entire macrodomain

    with a grid size comparable to that of the microscale

    features are too many. In this case, the mathematical

    homogenization method which has received considerable

    attention in recent years [616] seems to be the most

    suitable one. It is a kind of singular perturbation method

    suitable for problems with boundary layers [17] that exist

    at regions near the interfaces of dierent phases in a

    heterogeneous medium. With the help of multiple scale

    expansion, it gives not only the eective properties of thecomposite, but also detailed distribution of elds on the

    scale of microconstituents with acceptable cost. In con-

    trast to the most widely used methods in determining

    the macro properties, i.e., the Eshelby method, the self-

    consistent method, the MoriTanaka method, the dif-

    ferential scheme and the bound theories [14], the

    homogenization method takes into account the interac-

    tion between phases naturally and avoids assumptions

    other than the assumption of periodic distribution of

    constituents. On the other hand, it accounts for micro-

    structural eects on the macroscopic response without

    explicitly representing the details of the microstructure in

    the global analysis. The computational model at the

    Computers and Structures 79 (2001) 16451660

    www.elsevier.com/locate/compstruc

    * Corresponding author. Tel.: +44-29-2087-4934; fax: +44-

    29-2087-4597.E-mail address: [email protected] (B.L. Karihaloo).

    0045-7949/01/$ - see front matter 2001 Elsevier Science Ltd. All rights reserved.

    P II: S0 0 4 5 -7 9 4 9 (0 1 )0 0 0 9 1 -8

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    2/16

    lower scales is only needed if and when there is a necessity

    to do so. In recent years it has been employed for the

    solution of complex problems in conjunction with the

    nite element (FE) method [810,1214]. Since accuracy

    of the widely used isoparametric element method is not

    satisfactory, high performance multivariable elementsare necessary to be introduced in the homogenization

    method to improve the accuracy [14].

    In this study, we will employ the homogenization

    method for the solution of pure torsion of composite

    shafts with bers aligned along its axis and study the

    application of high performance multivariable elements.

    In accordance with the mathematical homogenization

    method, control dierential equations for the represen-

    tative unit cell (RUC) are obtained in Section 2. The

    corresponding variational principles are then deduced as

    the basis of the FE method in Section 3. The formula-

    tion and practical application of incompatible and en-hanced-strain elements to the analysis of RUC are

    discussed in detail in Sections 4 and 5. A 4-noded in-

    compatible element is also introduced. In Section 6, a

    penalty function method is discussed to enforce the pe-

    riodicity boundary condition of the RUC. In Section 7,

    composite shafts of square as well as rectangular cross-

    section reinforced with circular and elliptic bers are

    analyzed as illustrative examples. For a square shaft

    containing 16 bers, the problem is also solved directly

    by the hybrid stress element. The shear modulus from

    the homogenization method is compared with that ob-

    tained by the hybrid stress element and by the Voigt

    Reuss theory [4]. A comparison of the computed resultsshows common features of the local elds. Conclusions

    and discussion follow in Section 8.

    2. Mathematical homogenization

    Consider a uniform composite shaft of arbitrary

    cross-section twisted by couples applied at the ends.

    Without loss of generality, the origin of coordinates is

    taken at the left end cross-section, with the x1- and x2-

    axes as the principal axes of inertia, and the x3-axisalong the axis of the shaft and pointing to its other end.

    According to St. Venant's theory of torsion, the dis-

    placement components are [18]

    u1 hx3x2

    u2 hx3x1

    u3 wx1;x2

    1

    where h represents the angle of twist per unit length

    (clockwise about the x3-axis).

    Assume the microstructure of the cross-section Xe to

    be locally periodic with a period dened by a statistically

    homogeneous volume element, denoted by the repre-

    sentative volume element or unit cell Y, as shown in Fig.

    1. In other words, the composite material is formed by a

    spatial repetition of the unit cell. The shaft has two

    length scales; a global length scale D that is of the order

    of the section size, and a local length scale dthat is of the

    order of the unit cell and proportional to the wavelength

    of the variation of the microstructure. The size of the

    unit cell is much larger than that of the constituents but

    much smaller than that of the section. The relation be-tween the global coordinate system xi for the section and

    the local system yi for the minimum repeated unit cell

    can then be written as

    yi xi

    ei 1; 2 2

    where e is a very small positive number representing the

    scaling factor between the two length scales. The local

    coordinate vector yi is regarded as a stretched coordinate

    vector in the microscopic domain. For an actual hetero-

    geneous body subjected to external forces, eld quan-

    tities such as displacements, strains and stresses are

    assumed to have slow variations from point to point

    Fig. 1. Illustration of a problem with two length scales.

    1646 B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    3/16

    with macroscopic (global) coordinate x as well as fast

    variations with local microscopic coordinate y within a

    small neighborhood of size e of a given point x. That is,

    displacements, strains and stresses have two explicit

    dependencies: one on the macroscopic level with coor-

    dinates xi, and the other on the level of microconstitu-ents with coordinates yi

    ue3 ue3x;y

    ce3j ce3jx;y

    se3j se3jx;y

    3

    where j 1, 2. Due to the periodicity of the micro-structure, functions ue3, c

    e3j and s

    e3j are assumed to be Y-

    periodic, i.e., ue3x;y ue3x;y kY, c

    e3jx;y c

    e3jx;y

    kY and se3jx;y se3jx;y kY, where Yyi is the size

    of the unit cell, or the basic period of the stretched co-

    ordinate system y and k is a non-zero integer.The unknown displacement ue3, the angle of twist per

    unit length h, and the non-zero strain ce3j and stress se3j

    can be solved from the following equations

    Equilibrium :ose3j

    oxj 0 in Xe 4

    Kinematical :ce31ce32

    o

    ox1x2

    o

    ox2x1

    " #ue3h

    in Xe

    5

    Constitutive : se3i Ceijc

    e3j in X

    e 6

    and

    Mt

    ZXe

    s32x1 s31x2 dX 7

    together with the traction free condition on the surface

    of the shaft, and the traction and displacement condi-

    tions at the interfaces between the microconstituents.

    For the sake of simplicity and clarity, we assume that the

    elds are continuous across the interfaces. The material

    property tensor Ceij

    is symmetric with respect to indices

    i;j. Mt is the torque applied at the ends. The super-script e denotes Y-periodicity of the corresponding

    function. The convention of summation over the re-

    peated indices is used.

    The displacement ue3x;y is expanded in powers ofthe small number e [614]

    ue3x;y u03 x;y eu

    13 x;y e

    2u23 x;y 8

    where u03 , u

    13 , u

    23 ,. . ., are Y-periodic functions with

    respect to y. Substituting Eq. (8) into Eq. (5) gives the

    expansion of the strain ce3j:

    ce3jx;y e1c13j c

    03j ec

    13j 9

    where

    c13j

    ou03

    oyj

    c03j c

    0x3j c

    0y3j

    c0

    x31 ou

    03

    ox1 hx2

    c0

    x32 ou

    03

    ox2 hx1

    c0

    y3j ou

    13

    oyj

    c13j

    ou13

    oxjou

    23

    oyj

    10

    Substituting Eq. (9) into the constitutive relation (6)

    gives the expansion of the stress se3j

    se3jx;y e1s

    13j s

    03j es

    13j 11

    where

    s13i Cijc

    13j 12

    s03i Cijc

    03j 13

    s13i Cijc

    13j 14

    Inserting the asymptotic expansion for the stress eld

    (11) into the equilibrium equation (4) and collecting theterms of like powers in e gives equations

    Oe2 :os

    13j

    oyj 0 15

    Oe1 :os

    13j

    oxjos

    03j

    oyj 0 16

    Oe :os

    03j

    oxjos

    13j

    oyj 0 17

    We rst consider the Oe2

    equilibrium equation (15) inY. Premultiplying it by u

    03 , integrating over Y, followed

    by integration by parts, yieldsZY

    u03

    os13j

    oyjdY

    IoY

    u03 s

    13j nj dC

    ZY

    ou03

    oyjCji

    ou03

    oyidY 0 18

    where oY denotes the boundary ofY. The boundary in-

    tegral term in Eq. (18) vanishes due to the periodicity of

    the boundary conditions in Y, because u03 and s

    13j are

    identical on the opposite sides of the unit cell, while the

    corresponding normals nj are in opposite directions.

    B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660 1647

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    4/16

    Taking into account the positive deniteness of the

    symmetric constitutive tensor Cij, we have

    ou03

    oyj 0 A u03 u

    03 x 19

    and

    c13j x;y 0 s

    13j x;y 0 20

    Next, we proceed to the Oe1 equilibrium equation(16). From Eqs. (10) and (13) and taking into account

    Eq. (20), it follows that

    o

    oyjCjic

    0y3iu

    13

    oCji

    oyjc

    0x3i u

    03 21

    Based on the form of the right-hand side of Eq. (21)

    which permits a separation of variables, u13 may be

    expressed as

    u13 x;y v

    3j3 yc

    0x3ju

    03 22

    where v3j3 y is a Y-periodic function dened in the unit

    cell Y. Substituting Eq. (22) into Eq. (21), and taking

    into account the arbitrariness of the macroscopic strain

    eld, cx3ju03 within a unit cell, we have

    o

    oyjCjic

    0y3iv

    3k3 y

    oCjk

    oyj23

    We now consider the Oe equilibrium equation (17).Substituting Eq. (22) into Eq. (10), followed by the latter

    into Eq. (13), and nally the result into Eq. (17) yields

    o

    oxjCjidikh

    v3k3;yi c0

    x3ku03 i

    os

    13j

    oyj 0 24

    where dik is the Kronecker Delta. Integrating Eq. (24)

    over the unit cell domain Y and taking into account the

    periodicity of s13j yields

    o

    oxjCHjkc

    0x3ku

    03

    h i 0 25

    This is an equilibrium equation for a homogeneous

    medium (cf. Eq. (4)) with constant material propertiesCHjk, which are usually termed as the homogenized or

    eective material properties and are given by

    CHjk 1

    Y

    ZY

    Cjidik v3k3;yi

    dY 26

    where Y is the area of the unit cell.

    3. Variational principles and nite elements

    To solve the torsion of composite shafts by the ho-

    mogenization method, together with numerical meth-

    ods, e.g. the FE method adopted here, we will rst solve

    for v3j3 y from Eq. (23) assuming it to be a Y-periodic

    function dened in Y. The eective material properties

    CHjk are given by Eq. (26). We then solve the homoge-

    neous St Venant torsion problem by the hybrid stress

    element [5] and obtain the macroscopic elds: warping

    displacement u03 , the angle of twist per unit length h,strains c

    0x3j and stresses (given by C

    Hji c

    0x3i ). If the distri-

    bution of the microscopic elds in the neighborhood of

    point x is of interest, we use Eq. (22) to calculate the

    higher order displacement term, and then use Eqs. (10)

    and (13) to calculate the higher order strain and stress

    terms.

    The key problem here is to develop powerful nite

    element methods to solve Eq. (23).

    Corresponding to the equilibrium equation (23), the

    virtual work principle states that

    ZYdv

    3k3

    o

    oyj Cji

    ov3k3oyi dY ZY dv3k3 oCjkoyj dY 0 27a

    where dv3k3 are arbitrary Y-periodic functions dened in

    the unit cell Y. Integration of Eq. (27a) by parts yieldsIoY

    dv3k3 Cjiov3k3oyi

    nj ds

    IoY

    dv3k3 Cjknj ds

    ZY

    odv3k3oyj

    Cji

    ov3k3oyi

    dY

    ZY

    odv3k3oyj

    Cjk dY 0

    The boundary integral terms in the above equation

    vanish due to the Y-periodicity ofv3k3 and dv3k3 . Thus, we

    haveZY

    odv3k3oyj

    Cjiov3k3oyi

    dY

    ZY

    odv3k3oyj

    Cjk dY 0 27b

    Based on Eq. (27b), displacement elements can be es-

    tablished in a standard manner.

    It is easy to prove that Eqs. (27a) and (27b) is the rst

    order variation of the following potential functional

    PPv3k3

    ZY

    1

    2

    ov3k3oyj

    Cjiov3k3oyi

    dY

    ZY

    ov3k3oyj

    Cjk dY 28

    If we dene the strain

    ~ck3i ov3k3oyi

    and the stress

    ~sk3j Cji ~ck3i

    so that

    ~ck3i C1ij

    ~sk3j

    which are Y-periodic functions in the unit cell, we have a

    2-eld HellingerReissner functional

    1648 B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    5/16

    PHRv3k3 ; ~s

    k3j

    ZY

    1

    2~sk3iC

    1ij ~s

    k3j ~s

    k3i

    ov3k3oyi

    oCjk

    oyjv3k3

    dY

    29a

    or equivalently

    PHRv3k3 ; ~s

    k3j

    ZY

    1

    2~sk3iC

    1ij

    ~sk3j ~sk3i

    ov3k3oyi

    Cjkov3k3oyj

    dY

    29b

    By making use of the Lagrange multiplier method and

    relaxing the compatibility condition in the potential

    principle (28), or by employing Legendre transformation

    on the HellingerReissner principle (29b), one arrives at

    the 3-eld HuWashizu functional

    PHWv3k3 ; ~c

    k3j; ~s

    k3j ZY

    1

    2~ck3iCij ~c

    k3j ~s

    k3i

    ~ck3i ov3k3oyi

    Cjkov3k3oyj

    dY 30

    Based on the functionals (29) and (30), multivariable

    nite elements can be established.

    Although hybrid elements based on the Hellinger

    Reissner principle or the HuWashizu principle can in

    general improve the accuracy of the approximate dis-

    placement and stress solutions, they will not be used

    here as it is dicult to meet the Y-periodicity condition

    of the stress on the boundary of the unit cell. The general

    isoparametric elements are also not satisfactory because

    of the gradients of v3k3 that appear in Eq. (26) in the

    evaluation of the homogenized material properties. For

    these reasons, in this paper we will instead introduce

    displacement-incompatible elements based on the po-

    tential (28) and enhanced-strain elements based on Eq.

    (30).

    4. Displacement-incompatible elements

    Subdivide the unit cell domain Y into nite element

    subdomains Ye, such that Ye Y, Ya Yb Y andoYa oYb Sab (a, b are arbitrary elements).

    In each element, v3k3 is divided into a compatible part

    v3k3q and an incompatible part v3k3k, so that the functional

    (28) can be rewritten as

    PPv3k3 v

    3k3q v

    3k3k

    X

    e

    ZYe

    1

    2

    ov3k3oyj

    Cjiov3k3oyi

    dY

    ZYe

    ov3k3q

    oyjCjk dY

    31

    Taking the variation of the above functional, integrating

    by parts and making use of the periodicity condition on

    the outer boundary of the unit cell, yields

    dPPv3k3

    Xe

    ZYe

    dv3k3qo

    oyjCji

    ov3k3oyi

    oCjk

    oyj

    dY

    Xa;b ZSabdv3k3q Cji

    ov3k3oyi

    ( Cjk

    nj

    a

    Cjiov3k3oyi

    Cjknj

    b)ds

    X

    e

    ZYe

    odv3k3koyj

    Cjiov3k3oyi

    dY

    The stationary condition of the functional (31) gives the

    equilibrium equation (23) and the equilibrium of trac-

    tion between the elements if the following condition is

    met a priori

    XeZYe odv

    3k3k

    oyj Cjiov3k3oyi dY 0

    A convenient way to meet this condition is to satisfy the

    following strong form (i.e. the sucient but not the

    necessary condition) in each elementZYe

    odv3k3koyj

    Cjiov3k3oyi

    dY 0

    Since a constant stress state is recovered in each element

    as its size is reduced to zero and since dv3k3k is arbitrary,

    the above constraint reduces to the general patch test

    condition (PTC) [19,20]ZYe

    ov3k3koyj

    dY 0 or equivalentlyIoYe

    v3k3knj ds 0

    32

    The incompatible functions meeting the PTC can now

    be easily formulated.

    If we refer to the 4-noded isoparametric element

    shown in Fig. 2, the compatible displacement v3k3q is re-

    lated to the nodal values eqk via the bilinear interpola-

    tion functions

    v3k3q Neqk 33

    where

    N N1 N2 N3 N4

    and

    Ni 141 nin1 gig

    n; g represent the isoparametric coordinates, ni; gi arethe isoparametric coordinates of point i with the global

    coordinates xi;yi, i 1, 2, 3, 4.The incompatible term v3k3k is related to the element

    inner parameters ekk via the shape functions Nk

    v3k3k Nkekk 34

    B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660 1649

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    6/16

    Here, two incompatible terms are employed in each ele-

    ment as derived in Refs. [19,20]

    Nk1 n2 D Nk2 g

    2 D 35

    D 2

    3

    J1

    J0n

    J2

    J0g

    where J0, J1 and J2 are related to the element Jacobian asfollows

    jJj J0 J1n J2g

    a1b3 a3b1 a1b2 a2b1n a2b3 a3b2g

    36

    and coecients ai and bi i 1; 2; 3 are dependent onthe element nodal coordinates

    a1 b1

    a2 b2

    a3 b3

    264

    375

    1

    4

    1 1 1 11 1 1 1

    1 1 1 1

    24

    35

    x1 y1x2 y2x3 y3x4 y4

    2664

    3775 37

    With the above assumed displacements (33) and (34), we

    have

    ov3k3

    oy1ov3k

    3

    oy2

    8

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    7/16

    The stationary condition of the functional (41) gives the

    equilibrium equation (23), the stressstrain relations and

    the equilibrium of traction between the elements if the

    following condition is met a priori

    XeZ

    Yed~sk3i ~ckk3idY 0

    Following the procedure employed in Section 4, the

    above constraint can be simplied to the PTC [20,23]ZYe

    ~ckk3idY 0 42

    It is evident that Eq. (42) is an alternative formulation of

    the PTC (32), if the enhanced strain ~ckk3i corresponds to

    the incompatible displacement v3k3k.

    The nite element based on the stationary condition

    of functional (41) requires an independent approxima-

    tion of three elds: v3k3 , ~ckk3j and ~s

    k3j. In the enhanced-

    strain element, however, the independent stress eld is

    eliminated by selecting it to be orthogonal to the en-

    hanced strain eld, i.e.ZYe

    ~sk3i ~ckk3idY 0 43

    Thus, the two independent elds for the enhanced-strain

    formulation are the displacement v3k3 and the enhanced

    assumed strains ~ckk3j. The formulation here is the same as

    in Section 4 above, provided ~ckk3j are interpolated from

    the element inner parameters as follows

    ~ckk31~ckk32

    ( ) Bkk

    k 44

    Moreover, if the assumed strains ~ckk3j in Eq. (44) corre-

    spond to the incompatible displacement v3k3k in Eq. (34),

    the enhanced-strain formulation will be equivalent to

    the incompatible-displacement formulation discussed in

    Section 4. Therefore, only NQ6 introduced in Section 4

    will be employed in the computations to follow. Note

    however that the stress in the enhanced-strain formula-

    tion can be recovered with the help of the orthogonal-

    ization condition (43), as suggested in Ref. [21].

    6. Enforcing the periodicity boundary condition in the

    analysis of the RUC

    Assembling the discretized equations of equilibrium

    of all elements, yields the following system of equilib-

    rium equations

    Kqk fk 45

    Two dierent loading cases need to be analyzed in order

    to determine the characteristic deformations of the unit

    cell. The periodicity condition of the boundary dis-

    placement can conveniently be enforced by a penalty

    function technique [24]. Eq. (45) is the EulerLagrange

    equation of the following functional

    Pqk 12

    qkTKqk qkTfk 46

    The periodicity condition yields the following constraint

    Rqk 0

    If a couple of nodes, i and j, on the boundary have the

    same displacement because of the periodicity condition,

    i.e.

    qki qk

    j

    the above condition is equivalent to

    Ri; i 1 Ri;j 1 Ri; l T i;j 0

    In order to satisfy the above periodicity constraint by a

    penalty function technique the functional (46) is modi-

    ed as

    ePqk 12

    qkT

    Kqk qkT

    fk a

    2qk

    T

    RTRqk 47

    where a is a large positive number and taken to be 104 in

    our computations. Thus, instead of Eq. (45), we will

    solve the following equations

    K aRTRqk fk 48

    7. Numerical examples

    As the performance of the element using incompati-

    ble functions dened in Eq. (35) and of the hybrid stress

    element to be used here for torsion of shafts has been

    extensively studied in Refs. [5,19,20], no standard per-

    formance tests for both elements are included in thispaper. However, to illustrate the method described

    above, we solve the torsion of a composite shaft with

    square cross-section (length of side 80), as shown inFig. 3(a). Assume that the microstructure of the cross-

    section is locally periodic with a period dened by a

    RUC shown in Fig. 3(b), i.e. it consists of an isotro-

    pic circular ber of diameter 2a embedded in an iso-

    tropic square matrix with side 4a. a 5 is adapted inthis study. The problem is solved in two stages. First,

    we solve the RUC by using the incompatible element

    NQ6 introduced in Section 4, with the periodicity

    boundary condition enforced by the penalty function

    approach discussed in Section 6. We obtain the eld v3k3

    B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660 1651

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    8/16

    and its derivatives o=oyjv3k3 and calculate the homog-enized moduli from Eq. (26). Second, we solve the tor-

    sion of the square shaft shown in Fig. 3(a) with the

    homogenized moduli obtained at step one above, by

    using the hybrid stress element introduced in Ref. [5]. In

    this way, we calculate the warping displacement, tor-

    sional rigidity and the angle of twist per unit length, aswell as the shear stresses and strains. With the results so

    obtained, we can calculate the rst order warping dis-

    placement from Eq. (22) and the local strain and stress

    elds from Eqs. (10) and (13), respectively. For the

    present illustrative purpose, we choose e 0:25. Thecomplete shaft section from which the RUC has been

    extracted is shown in Fig. 3(c). In the gures to follow,

    lled triangles represent computed data. In all the gures

    that illustrate the stress distribution, a line segment

    represents the distribution within an element. In Figs. 5,

    6(b) and (c), the solid line represents the polynomial t

    of the corresponding computed data that is not satis-

    factorily smooth.

    The RUC shown in Fig. 3(b) is discretized into 896

    quadrilateral elements and 929 nodes, as shown in Fig.

    4(a). According to the denition of the RUC, its size

    should be enlarged four times as e 0:25. However,numerical results show that the results are unaected by

    whether or not the RUC size is enlarged, allowing us to

    use the original RUC size. Care must be taken in en-forcing the periodicity boundary condition at corner

    nodes. For the four corner nodes, i, j, k and l, shown in

    Fig. 3(b), the periodicity condition yields

    qki qk

    j qkk q

    kl

    The above condition can be rewritten as

    qki qk

    j

    qkj qkk

    qkk qkl

    Fig. 3. Geometry of a composite shaft of square prole: (a) square prole, (b) RUC, (c) square shaft with 16 bres.

    1652 B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    9/16

    and treated conveniently by the procedure discussed in

    Section 6. The ber and the matrix are considered to be

    isotropic with the shear moduli, Gf 10 and Gm 1,respectively. The computed homogenized shear moduli

    are

    C11 C12Sym C22

    1:38271 0:00138

    Sym 1:38467 49

    Thus the macroscopic behavior of the composite shaft is

    also isotropic. The numerical results for the character-

    istic displacements v3k3 and their derivatives o=oyjv3k3

    are saved for later use.

    The isotropic shaft of square cross-section shown in

    Fig. 3(a) is now analyzed with the homogenized shear

    moduli (49) obtained above. Only a quarter of the cross-

    section, the shaded part shown in Fig. 3(a), is dicretized

    because of symmetry. The warping displacements are

    xed on the axes of symmetry. The employed FE mesh

    with 400 quadrilateral elements and 441 nodes is shown

    in Fig. 4(b). One unit of torque is applied on the quarter

    section with its units being consistent with those of the

    shear moduli. The computed result for the torsional ri-

    gidity 4 1:9927 106 is very close to the accurate value7:9856 106 obtained from the formula [16]

    Torsional rigidity 0:141G2b4 50

    where the shear modulus G 1:38271, and the length ofside of the square cross-section 2b 80 in the presentexample. The numerical results for the local elds near

    or along the interface between the ber and the matrix

    adjacent to the point with global co-ordinates x1 30,x2 30 are shown in Figs. 5 and 6. Fig. 5(a)(c) showthe results along the line 36y16 7, y2 0 near the pointP in Fig. 3(b). Fig. 5(a) shows the distribution of

    warping displacement. Fig. 5(b) shows the polynomial

    tting of the computed shear stress sxz, on the scale of

    the gure results given by the upper and lower ele-

    ments adjacent to the line cannot be distinguished. Fig.

    5(c) shows the computed shear stress syz, data linked by

    solid and broken lines represent respectively the results

    Fig. 4. Discretised meshes used in the computation: (a) mesh of the RUC shown in Fig. 3(b), (b) mesh of a quarter of the cross-section

    shown in Fig. 3(a), (c) mesh of a quarter of the cross-section shown in Fig. 3(c).

    B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660 1653

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    10/16

    obtained from the upper and lower elements adjacent to

    the line in question. From the results it is seen that the

    gradient of the warping displacement changes rapidly

    across the interface y1 5 and that the distribution of

    sxz but not of syz is continuous across the interface. The

    distribution of warping displacement, and of normal

    and tangential shear stresses along the interface, which

    are given by

    Fig. 5. Numerical results on the line 36y1 6 7, y2 0, from the homogenisation method: (a) distribution of warping displacement, (b)distribution of sxz, (c) distribution of syz.

    1654 B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    11/16

    sn sxz cosu syz sinu

    st sxz sinu syz cosu51

    where u is the angle from the axis y1 as shown in Fig.

    3(b), are plotted in Fig. 6 (a)(c). In Fig. 6(b) and (c),

    data linked by broken lines represent the results ob-

    tained from the matrix side, the continuous solid line

    represents the polynomial t of the results obtained

    from the ber side of the interface. These results show

    that the warping displacement and normal shear stress

    Fig. 6. Numerical results along the interface from the homogenisation method: (a) distribution of warping displacement, (b) distri-

    bution of the normal shear stress sn, (c) distribution of the tangential shear stress st.

    B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660 1655

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    12/16

    sn vary continuously across the interface, whereas the

    tangential shear stress st has a signicant discontinuity.

    Although it will not be possible to compare the re-

    sults with those obtained by the present homogenization

    method, we will still solve directly the torsion of the

    composite shaft shown in Fig. 3(c) by the hybrid stresselement introduced in Ref. [5] to illustrate some typical

    features of local elds adjacent to the interface. Again,

    only a quarter of the cross-section is needed to be

    dicretized because of symmetry. The warping displace-

    ments are xed on the axes of symmetry. The FE mesh

    with 3584 quadrilateral elements and 3649 nodes is

    shown in Fig. 4(c). One unit of torque is applied on the

    quarter section with its units being consistent with those

    of the shear modulus. The computed result for torsional

    rigidity is 4 1:9456356 106, which according to theformula (50) corresponds to an isotropic shaft with

    shear modulus 1.34754. The result is reasonably close tothat obtained by the homogenization method (49). The

    latter predicts larger values of moduli because the em-

    ployment of the periodic boundary condition makes the

    system stier. The result given by the homogenization

    method is also within the lower bound 1.215 and the

    upper bound 2.767 as per the VoigtReuss theory [4].

    Zhao and Weng [25] have derived the nine eective

    elastic constants of an orthotropic composite reinforced

    with monotonically aligned, uniformly dispersed elliptic

    cylinders using the EshelbyMoriTanaka method. The

    problem studied above is the special case that the rein-

    forcements are bers with circular cross-section. The

    two shear moduli relevant to torsion given by Zhao andWeng [25] are

    C11

    Gm 1

    cfcma1a

    GmGfGm

    ;C22

    Gm 1

    cfcm

    1a Gm

    GfGm

    52

    where cf and cm are volume fractions of ber and matrix,

    respectively, and a is the cross-sectional aspect ratio of

    the reinforced ber. In our case, cf p=4, cm 1 p=4and a b=a 1, and hence the eective shear moduli

    C11 4:595947 C22 given by Eq. (52) are unreason-ably higher than the results by the direct FE analysis, as

    well as the results (49) by the homogenization method

    mentioned above. They are also above the upper bound

    of the VoigtReuss theory. The EshelbyMoriTanaka

    method cannot give good results, especially for high

    volume fraction of reinforcements, because Eshelby's

    tensor is based on the inclusion in an innite matrix,

    which takes into account of the interaction between re-

    inforcements in a very weak sense. On the other hand, it

    is evident that the homogenization method has the ad-

    vantage of taking the interaction between phases into

    account naturally and of not having to make assump-

    tions such as isotropy of material.

    The distribution of warping displacement and shear

    stresses along the line corresponding to Fig. 5 and the

    interface corresponding to Fig. 6 are plotted in Figs. 7

    and 8. (51) has been used to obtain the normal and

    tangential shear stresses in Fig. 8(b) and (c). A com-

    parison of Figs. 5 and 6 with Figs. 7 and 8, respectively,shows the obvious dierences of the results obtained by

    the homogenization method and the direct hybrid stress

    element. The dierences are to be expected in view of the

    limited number of bers that can be economically han-

    dled by the hybrid stress element. The homogenization

    method is suitable for problems involving a large num-

    ber of periodically distributed reinforcements so that the

    RUC occupies only a ``point'' in the physical domain

    [12]. The computed stress elds by the hybrid stress

    element are smoother than those obtained by the ho-

    mogenization method and smoothing techniques are

    unnecessary for the former since dierentiations areavoided in the computations. Notwithstanding these

    dierences, the results by the two methods reveal the

    common features of the local elds: a signicant dis-

    continuity exists in the tangential shear stress, while

    other elds are continuous adjacent to the interface.

    Having gained condence in the accuracy of the in-

    compatible element NQ6 developed from the homoge-

    nization theory in predicting the eective shear moduli,

    we study below the eect of the cross-sectional shape of

    the reinforcing bers. The RUC used is illustrated in

    Fig. 9(a), i.e. an elliptic cylindrical ber is embedded in

    the matrix of rectangular shape; the pattern of discreti-

    zation is similar to Fig. 4(b). The material properties ofthe ber as well as the matrix and their volume fractions

    are selected as above. The variation of the computed C11and C22 with the aspect ratio of the ber is plotted in

    Fig. 9(b). Again, the results from Eq. (52), i.e., the

    EshelbyMoriTanaka method, are unreasonably high-

    er. However, the predicted trend is the same with an

    increase in b=a, C11 decreases, and C22 increases.

    8. Conclusions and discussion

    The homogenization method is most suitable for

    problems involving a large number of periodically dis-

    tributed reinforcements so that the RUC can be re-

    garded as a ``point'' in the physical domain. It gives not

    only the equivalent material properties but also detailed

    information of local elds with much lower computa-

    tional cost. Such detailed information of the elds on

    the scale of microconstituents is almost impossible to

    obtain by using the hybrid stress element, because of the

    enormous degrees of freedom needed to model the en-

    tire macrodomain with a grid size comparable to that of

    the microscale features. When the number of the rein-

    forcement is not very large, numerical results by the

    1656 B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    13/16

    homogenization method without the terms of order

    higher than one are usually quantitatively dierent from

    those obtained by the direct hybrid stress element. The

    inclusion of higher terms may improve numerical ac-

    curacy, but it inevitably complicates the procedure. In

    the determination of the equivalent properties and in the

    evaluation of the local elds, the derivatives of the

    characteristic displacement v3k3 are needed. Therefore,

    the widely used isoparametric elements are not suitable

    for the analysis of the RUC. The hybrid stress element is

    also limited because it is dicult to enforce the period-

    icity condition on the assumed stresses. From a practical

    point of view, the incompatible element based on the

    modied potential principle and the enhanced-strain

    Fig. 7. Numerical results on the line 36y16 7, y2 0, from the hybrid stress element method: (a) distribution of warping displace-ment, (b) distribution of sxz, (c) distribution of syz.

    B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660 1657

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    14/16

    element based on the 3-eld HuWashizu principle are

    the most appropriate for the analysis of RUC, whereas

    the hybrid stress element is appropriate for the macro-

    homogenized problem.

    For torsion of ber reinforced composite shafts, a

    signicant discontinuity exists in the tangential shear

    stress, while other elds are continuous along the in-

    terface between the ber and the matrix.

    Fig. 8. Numerical results along the interface from the hybrid stress element method: (a) distribution of warping displacement,

    (b) distribution of the normal shear stress sn, (c) distribution of the tangential shear stress st.

    1658 B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    15/16

    Acknowledgements

    CC Wu and QZ Xiao acknowledge the nancial

    support from the National Natural Science Foundation

    of China under grant number 19772051.

    References

    [1] Hashin Z. Analysis of composite materials a survey.

    J Appl Mech 1983;50:481505.

    [2] Christensen RM. A critical evaluation for a class of

    micromechanics models. J Mech Phys Solids 1990;38:379

    404.

    [3] Tolonen H, Sjolind SG. Eect of mineral bers on

    properties of composite matrix materials. Mech Compos

    Mater 1995;31:43545.

    [4] Du SY, Wang B. Micromechanics of composite materials.

    Beijing: Science Press; 1998.

    [5] Xiao QZ, Karihaloo BL, Li ZR, Williams FW. An

    improved hybrid-stress element approach to torsion of

    shafts. Comput Struct 1999;71:53563.

    [6] Bensoussan A, Lions JL, Pananicolaou G. Asymptotic

    analysis for periodic structures. New York: North-Hol-

    land; 1978.[7] Bakhvalov A, Panassenko GP. Homogenization: averaging

    process in periodic media. Dordrecht: Kluwer Academic

    Publisher; 1989.

    [8] Lene F, Leguillon D. Homogenized constitutive law for a

    partially cohesive composite material. Int J Solids Struct

    1982;18:44358.

    [9] Guedes JM, Kikuchi N. Preprocessing and postprocessing

    for materials based on the homogenization method with

    adaptive nite element methods. Comp Meth Appl Mech

    Engng 1990;83:14398.

    [10] Jansson S. Homogenized nonlinear constitutive properties

    and local stress concentrations for composites with peri-

    odic internal structure. Int J Solids Struct 1992;29:2181

    200.

    Fig. 9. Eect of the cross-sectional shape of the ber on the eective shear moduli: (a) a RUC of rectangular shape with an embedded

    elliptic cylindrical ber, (b) variation of C11 and C22 with the aspect ratio of the ber.

    B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660 1659

  • 8/8/2019 Karihaloo 2001 Computers & Structures

    16/16

    [11] Kalamkarov AL. Composite and reinforced elements of

    construction. New York: Wiley; 1992.

    [12] Fish J, Nayak M, Holmes MH. Microscale reduction error

    indicators and estimators for a periodic heterogeneous

    medium. Comput Mech 1994;14:32338.

    [13] Lukkassen D, Persson LE, Wall P. Some engineering and

    mathematical aspects on the homogenization method.

    Compos Eng 1995;5:51931.

    [14] Sun HY, Di SL, Zhang N, Wu CC. Micromechanics of

    composite materials using multivariable nite element

    method and homogenization theory. Int J Solids Struct

    2001;38:300720.

    [15] Mascarenhas ML, Trabucho L. Homogenised behavior of

    a beam with multicellular cross section. Applicable Anal

    1990;38:97119.

    [16] Mascarenhas ML, Polisevski D. The warping, the torsion

    and the Neumann problems in a quasi-periodically perfo-

    rated domain. Math Model Numer Anal 1994;28:3757.

    [17] Lin CC, Segel LA. Mathematics applied to deterministic

    problems in the natural sciences. New York: Macmillan;1974.

    [18] Timoshenko SP, Goodier JN. Theory of elasticity, 3rd ed.

    New York: McGraw-Hill; 1970.

    [19] Wu CC, Huang MG, Pian THH. Consistency condition

    and convergence criteria of incompatible elements, general

    formulation of incompatible functions and its application.

    Comput Struct 1987;27:63944.

    [20] Wu CC, Pian THH. Incompatible numerical analysis and

    hybrid element method. Beijing: Science Press; 1997.

    [21] Simo JC, Hughes TRJ. On the variational founda-

    tions of assumed strain methods. J Appl Mech 1986;

    53:514.

    [22] Simo JC, Rifai MS. A class of mixed assumed strain

    methods and the method of incompatible modes. Int

    J Numer Meth Eng 1990;29:1595638.

    [23] Wu CC, Buer H. Multivariable nite elements: consis-

    tency and optimization. Sci China (A) 1991;34:284

    99.

    [24] Bathe KJ. Finite element procedures in engineering ana-

    lysis. Englewood Clis, NJ: Prentice-Hall; 1982.

    [25] Zhao YH, Weng GJ. Eective elastic moduli of ribbon-reinforced composites. J Appl Mech 1990;57:15867.

    1660 B.L. Karihaloo et al. / Computers and Structures 79 (2001) 16451660