COMPOSITES SCIENCE AND TECHNOLOGY

Composites Science and Technology 66 (2006) 713722 www.elsevier.com/locate/compscitech

Computational modelling of delamination

Rene de Borst
a

a,b,*

, Joris J.C. Remmers

Koiter Institute Delft, Faculty of Aerospace Engineering, Delft University of Technology, P.O. Box 5058, NL-2600 GB Delft, The Netherlands b LaMCoS CNRS UMR 5514, INSA de Lyon, F-69621 Villeurbanne, France Available online 19 January 2005

Abstract Delamination in composite structures is best modelled at a mesoscopic level. In this approach, the plies are modelled as continua, which can either be assumed to behave linearly elastically or to degrade according to a damage law. Delamination in the interfaces between them is modelled using a discrete relation between interface tractions and relative displacements. Key in this type of modelling is the presence of a work of separation or fracture energy, which governs the growth of the delamination. This cohesive-surface approach has traditionally been implemented numerically using special interface elements which connect the continuum elements that are used to model the plies. Exploiting the partition-of-unity property of nite element shape functions to model the interface separation process oers some advantages, since interfaces can be inserted at the onset of delamination and not a priori, as in the conventional approach. As a consequence, elastic compliance of the interface prior to onset of delamination, spurious traction oscillations ahead of the delamination front and spurious wave reections because of the presence of a high stiness value are avoided. Moreover, unlike the conventional approach, unstructured meshes can be employed. 2005 Elsevier Ltd. All rights reserved.
Keywords: C. Delamination; C. Finite element analysis; Interfaces

1. Introduction Composite structures can be modelled at various scales. At the macroscopic or structural level the laminae are normally modelled via a layered shell approach, where the dierent directions of the bres in the layers are taken into account through an anisotropic elasticity model. If this (anisotropic) elasticity model is augmented by a damage or plasticity model, degradation phenomena like matrix cracking, bre pull-out or bre breakage can also be taken into account, albeit in a smeared manner. The proper modelling of delamination is more dicult, at least at this level. Delamination the debonding of dierent layers is a discrete process which is not very amenable to smeared models, which
Corresponding author. Tel.: +31 15 278 5464; fax: +31 15 261 1465. E-mail address: R.deBorst@LR.TUDelft.nl (R. de Borst). 0266-3538/$ - see front matter 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.compscitech.2004.12.025
*

are, from a practical point of view, the only feasible modelling approach at the structural level. Nevertheless, delamination is one of the major failure modes which endanger the reliability of composite structures. Its modelling can be done accurately at a mesoscopic level. Then, the laminae are considered as continua and are discretized using standard nite elements while the delamination is modelled in a discrete manner using interface elements [15]. While in [2,3] possible matrix degradation eects like matrix cracking, bre pull-out and bre breakage have not been taken into account, Allix ` and Ladeveze [1] have supplemented their interface fracture model by a damage model to account for the deterioration processes in the laminae quoted above. ` The Allix/Ladeveze damage model [1] is a phenomenological approach, which models the underlying microscopic degradation processes like matrix cracking, bre pull-out and bre breakage in a smeared manner at the meso level. A signicant amount of research is

714

R. de Borst, J.J.C. Remmers / Composites Science and Technology 66 (2006) 713722

damage model

contact model

Fig. 1. Finite element model of a laminated composite. The individual layers are modelled with generalized plane-strain or shell elements. Interface elements equipped with either a damage/plasticity or a contact model are applied between the layers.

currently being invested in deriving bridges between the micro level where pull-out and breakage of individual bres are modelled in a discrete manner as well as the modelling of matrix cracks and the meso level in an attempt to derive the damage parameters postulated at the meso level from micromechanical considerations and to substantiate the assumptions made when developing the meso level damage model for the plies [6]. Ultimately, such eorts should also result in the incorporation of an internal length scale in the mesoscopic damage model, since its present absence makes it possible that the boundary value problem becomes ill-posed for certain congurations and loading conditions. A severe and unphysical mesh sensitivity can result [7]. In the above approaches, the numerical modelling at the meso level is achieved by stacking continuum elements for the plies and the interface nite elements, which can capture the delaminations that arise between the plies, Fig. 1. More specically, generalized planestrain elements are often used to model free-edge delamination as in [2,3], while stacks of shell elements and interface elements are applicable to cases of delamination near holes or other cases where a three-dimensional modelling is necessary [8,9]. This contribution will start by reviewing this approach and will show some typical examples of mixed-mode free-edge delamination and delaminationbuckling near an initial circular delamination. Next, some drawbacks of this way of modelling are discussed and a modern way of numerically representing discontinuities, including delaminations, will be outlined. Illustrative examples will accompany the discussion.

can either be assumed to behave linearly elastically or can be degraded according to a damage law. In the approach suggested by Schellekens and de Borst [2,3], elastic anisotropy and curing of the composite are taken into account by including possible thermal and hygral eects, but eventual damage which can evolve in the plies is lumped into the interface. This approach is reasonable as long as the energy dissipation due to processes like matrix cracking is small compared to the energy needed for delamination growth, as for mode-I delaminations and for mixed-mode delaminations where the bres are (almost) parallel to the intralaminar cracks. If this condition is not met, the interface delamination model must be supplemented by a damage model for the ply, as has been suggested in [1]. A drawback of existing damage approaches for modelling intralaminar cracks, bre breakage and debonding is that no localization limiter is incorporated, which renders the governing equations ill-posed at a generic stage in the loading process and can result in a severe dependence of the results on the spatial discretization [7]. Delamination in the interfaces between the plies is modelled using a discrete relation between the interface tractions td and the relative displacements v td td v; j 1 with j a history parameter. After linearization, necessary to use a tangential stiness matrix in an incremental-iterative solution procedure, one obtains _ td T_ v 2

with T the material tangent stiness matrix of the discrete traction-separation law T otd otd oj : ov oj ov 3

Whichever type of modelling is used, a key element is the presence of a work of separation or fracture energy, Gc , which governs delamination growth and enters the interface constitutive relation (1) in addition to the tensile strength ft [10,11]. It is dened as the work needed to create a unit area of fully developed crack: Z 1 Gc r du 4
u0

ft

ft

2. Meso-level approach In the meso-level approach to numerically model delamination the plies are modelled as continua and

Fig. 2. Stressdisplacement curves for ductile separation (left) and quasi-brittle separation (right).

R. de Borst, J.J.C. Remmers / Composites Science and Technology 66 (2006) 713722

715

with r and u the stress and the displacement across the fracture process zone. It thus equals the area under the decohesion curves as shown in Fig. 2. Evidently, cohesive-surface models as dened above are equipped with an internal length scale, since the quotient Gc =E, with E a stiness measure for the plies, has the dimension of length.

3. Examples of the standard nite element approach As discussed in Section 1, the standard nite element approach to delamination consists of stacking continuum elements to model the plies and interface elements equipped with a decohesion relation for the interfaces. The decohesion relation in the interface elements can be based on a plasticity formalism, as in [2,3], or on a damage formalism, as proposed in [1]. For monotonic loadings the dierences in results between both approaches are minor, but from a numerical point of view, damage-based formats are to be preferred, since they tend to exhibit better convergence characteristics [9]. This is because, in contrast to plasticity, elasticity-based damage models yield a straightforward computation of the stress for a given strain history. The rst example which we will discuss, mixed-mode free-edge delamination in a composite strip loaded in uniaxial tension, employs a plasticity formalism. In the second example, delamination-buckling of a uniaxially compressed panel with an initial circular delamination, a damage formalism is used. For the example of mixed-mode free-edge delamination, a composite strip loaded in uniaxial tension has been considered, as shown in Fig. 3. The ply material was a Fibredux 6376C/35/135/HTA graphite epoxy. To introduce the initial stresses that are caused by the manufacturing process, a temperature drop DT = 150 C has been imposed prior to incrementing the axial elongation. Since the cross-sectional dimensions of the strip are small compared to its length, a generalized plane-strain modelling has been applied, which is reasonable away from the grips. Due to mismatch of

the Poisson eect between the layers, as caused by the dierent orientation of the bres, interlaminar stresses will develop between the plies at the free edges. At a generic stage in the loading process, these edge stresses will lead to delamination. A cross-section of the example that we will consider is shown in Fig. 4. For this specic lay-up, the delamination jumps from one of the 35/90 interfaces to that which is located at the other side of the symmetry line and back (mixed-mode delamination). The energy that is dissipated in the crack that runs through the plies is negligible, because the bres in this layer are parallel to the crack. Fig. 5 shows the results in terms of the axial strain at onset of delamination for three dierent layups. A clear size (thickness) eect is predicted because of the presence of the work of separation Gc in the constitutive model for the interface. This computed size effect was conrmed by experiments carried out at the Catholic University of Leuven after completion of the computations, Fig. 5 [2].
0 +35 35 90 90 35 +35 0
Fig. 4. Mixed-mode delamination in a uniaxially loaded laminated strip: jumping of a delamination front between two 35/90 interfaces.

% u 1.0

0.75

+
0.5

r3

11

0.25 n=1 0.0 8 n=2 n=3

r2 r1 h
b

16 24 number of plys

11
Fig. 3. Uniaxially loaded laminate strip.

Fig. 5. Computed and experimentally determined values for the ultimate strain u as a function of the number of plies [2]. Results are shown for laminates consisting of 8 plies (n = 1), 16 plies (n = 2) and 24 plies (n = 3). The triangles, which denote the numerical results, are well within the band of experimental results. The dashed line represents the inverse dependence of the ultimate strain on the laminate thickness.

716

R. de Borst, J.J.C. Remmers / Composites Science and Technology 66 (2006) 713722

For three-dimensional analyses of laminated composites, a range of shell elements is available. Most of them share the property that the strains in the thickness direction are discretized and, therefore, are two-dimensional elements in a three-dimensional space. Related to this is the fact that these elements are equipped with rotational degrees-of-freedom in addition to the usual translational degrees-of-freedom. This is inconvenient when stacking these elements and placing between them interface elements, which only have translation degrees-of-freedom. To preserve a proper connectivity, dependence relations have to be supplied, which is usually a costly manual process. For this reason, solid-like or 3D-like shell elements are to be preferred when modelling delamination in a three-dimensional setting. They have the added benet that they can be formulated such that Poisson locking is avoided for a large range of aspect ratios, which makes them applicable for very thin applications, and that the normal strain in the thickness direction varies linearly instead of being constant, which allows fully three-dimensional stressstrain relations to be used. Within this class of elements, we have employed the solid-like shell element of Parisch [8,12]. This solid-like shell element possesses eight or sixteen external nodes, each with three translational degrees-of-freedom, Fig. 6. Four internal degrees of freedom are used to add a quadratic term to

the displacement eld in the thickness direction (the socalled internal stretch of the element). Hence, the strain varies linearly over the thickness instead of being constant and Poisson locking is avoided. This implies that the element can also be used in thin applications, in practice up to an aspect ratio of 1:1000. The position of a material point in the shell in the undeformed conguration can be written as a function of the three curvilinear coordinates [n, g, f], see Fig. 7 Xn; g; f X0 n; g fDn; g; 5 where X0(n, g) is the projection of the point on the midsurface of the shell and D(n, g) is the thickness director in this point 1 X0 n; g Xt n; g Xb n; g; 2 6

1 7 Dn; g Xt n; g Xb n; g: 2 The subscripts ()t and ()b denote the projections of the variable onto the top and bottom surface, respectively. The position of the material point in the deformed conguration x(n, g, f) is related to X(n, g, f) via the displacement eld /(n, g, f) according to xn; g; f Xn; g; f /n; g; f; where /n; g; f u0 n; g fu1 n; g 1 f2 u2 n; g: 9 8

In this relation, u0 and u1 are the displacements of X0 on the shell mid-surface, and the thickness director D, respectively 1 u0 n; g ut n; g ub n; g; 2 1 u1 n; g ut n; g ub n; g 2 10 11

Fig. 6. Geometry of the eight node solid-like shell element. Each geometrical node I contains three degrees of freedom: ^x ; ^y ; ^z I . Each u u u ^ internal node J has one degree of freedom: wJ .

and u2(n, g) denotes the internal stretching of the element, which is colinear with the thickness director in the deformed conguration and is a function of an additional stretch parameter w u2 n; g wn; gD u1 n; g: 12

Fig. 7. Kinematic relations of the solid-like shell element.

R. de Borst, J.J.C. Remmers / Composites Science and Technology 66 (2006) 713722

717

In the remainder, we will consider the displacement eld / as a function of two kinds of variables; the ordinary displacement eld u, which will be split in a displacement of the top and bottom surfaces ut and ub, respectively, and the internal stretch parameter w / /ut ; ub ; w: 13 The derivation of the strains and the nite element formulation are given in [8,12]. Using the solid-like shell element, the behaviour of a Glare panel with a circular initial delamination and a sinusoidally shaped out-of-plane imperfection (with an amplitude of 0.003 mm) subject to a compressive load is examined. The failure mechanism is slightly complicated, since the direction in which the delaminated zone will propagate is not evident. Similar studies with carbon bre-reinforced epoxies [14] show that the delamination grows in a direction perpendicular to the main loading direction. As a result, the delaminated area transforms from a circular area into an ellipsoidal one. Consequently, the buckling mode will change as well, and some parts of the top layer will tend to move inwards. For this reason, the possibility of self-contact is included and a contact algorithm is activated. The specimen of Fig. 8 consists of an aluminium layer with thickness h1 = 0.2 mm and a Glare3 0/90 prepreg layer with a thickness h2 = 0.25 mm, [13]. An initially circular delamination area with radius 8 mm is assumed. The layers are attached to a thick backing plate in order to prevent global buckling. A uniaxial compressive loading in x-direction is considered (rx = r0, ry = 0.0). The nite element mesh is shown in Fig. 9. The material parameters for the Glare3 layer are taken from [15],

Fig. 9. Mesh used for the simulation of delamination growth in the Glare panel. The initial delamination is located at the darker elements. Note that just one quarter of the panel (x > 0, y > 0) is modelled.

see Table 1. The ultimate traction in normal direction in tension and compression are assumed to be t 50:0 MPa and c 150:0 MPa, respectively, and tn tn the ultimate traction in the two transverse directions equals s1 s2 25:0 MPa. The work of separation is t t Gc 1:1 N=mm. An initial stiness of the interface elements of dn = 50 000 N/mm3 has been assumed. The analytical estimation for the local buckling load of a clamped unidirectional panel with thickness h1 subjected to an axial compressive load r0 was derived in [16]. For this conguration, the lowest critical buckling load is equal to r0 = 113.2 MPa. For the contact algorithm the penalty stiness has been set equal to the initial stiness of the interface elements with the delamination model: dpen = 50 000 MPa.
Table 1 Material parameters for 0/90 Glare3 E11 E22 E33 33 170 MPa 33 170 MPa 9400 MPa G12 G23 G13 5500 MPa 5500 MPa 5500 MPa m12 m23 m13 0.195 0.032 0.06

Fig. 8. Geometry of Glare panel with a circular initial delamination.

Fig. 10. Out-of-plane displacement of top layer versus applied axial compressive load r0. The dashed line corresponds to the critical buckling load obtained by an eigenvalue analysis [13].

718

R. de Borst, J.J.C. Remmers / Composites Science and Technology 66 (2006) 713722

P=1000 N

100 mm 20 mm

250 mm
Fig. 11. Final deformation of the Glare laminate under uniaxial loading [13]. Fig. 12. Geometry of symmetric, notched three-point bending beam.

The out-of-plane displacement of the centre point of the panel is shown in Fig. 10. The local buckling load is in agreement with an eigenvalue analysis [13]. Initial delamination growth does not start until a load level r0 = 300 MPa, while progressive delamination begins at an external load level r0 % 950 MPa. As this value is far beyond normal stress levels, the analysis suggests that delamination buckling is of little concern in uniaxially compressed Glare panels. As expected, the delamination extends in a direction perpendicular to the loading direction, Fig. 11.

4. Shortcomings of standard FE approach In the above analyses, interface elements equipped with a cohesive-zone model have been inserted a priori in the nite element mesh. Before the tensile strength in the interface element is exceeded, which is monitored in the integration points of the interface element, no deformations should occur, since (elastic) deformations before the onset of fracture only take place in the bulk. Nevertheless, the formulation of conventional interface elements requires a nite stiness prior to the onset of cracking, thus giving rise to deformations in the interface before crack initiation. Prior to onset of delamination the stiness matrix in the interface reads 2 3 dn 0 0 6 7 T 4 0 ds 0 5 14 0 0 dt with dn the stiness normal to the interface and ds and dt the tangential stinesses. The undesired elastic deformations can be largely suppressed by choosing a high value for the stiness dn. Depending on the chosen spatial integration scheme, this high stiness value can lead to spurious traction oscillations in the pre-cracking phase, which may cause erroneous crack patterns [17,18]. An example of an oscillatory traction pattern ahead of a notch is given in Fig. 13 for the notched three-point bending beam of Fig. 12. When analysing dynamic delaminations, spurious wave reections can occur as a result of the introduction of such articially high stiness

Fig. 13. Traction proles ahead of the notch using linear interface elements with Gauss integration. Results are shown for dierent values of the stiness D = dn in the pre-cracking phase [17,18].

values prior to the onset of delamination. Moreover, the necessity to align the mesh with the potential planes of delamination, restricts the modelling capabilities, in particular for thin composite structures.

5. Partition-of-unity based interface modelling As an outgrowth of the research into meshless methods [19] recently a method has emerged in which a discontinuity in the displacement eld can be captured exactly. It has the added benet that it can be used advantageously at dierent scales, from microscopic to macroscopic analyses. The method makes use of the partition-of-unity property of nite element shape functions [2022]. A collection of functions /,P associated with nodes , form a partition of unity if n / x 1 with n the number 1 of discrete nodal points. For a set of functions / that satisfy this property, a eld u can be interpolated as follows: ! n m X X ux / x  a w| x~| a 15
1 |1

with  the regular nodal degrees-of-freedom, w|(x) the a enhanced basis terms, and ~| the additional degrees-ofa freedom at node which represent the amplitudes of the

R. de Borst, J.J.C. Remmers / Composites Science and Technology 66 (2006) 713722

719

|th enhanced basis term w|(x). In conventional nite element notation, the displacement eld of an element that contains a single discontinuity can be represented as a a a u u N HCd ~ N HCd N~  HCd ~ a u 16

with the Heaviside function HCd separating the continuous displacement elds  and ~. The matrix N contains u u the standard shape functions, and  and ~ collect the a a conventional and the additional nodal degrees-of-freedom, respectively. The interpolation (16) has a structure identical to u  HCd ~ u u 17

u Distinction between nC and nC is possible because ~ is d d not spatially constant. It is this property that makes it possible to capture buckling at the interface. Furthermore, it is emphasized that the kinematics are completely general: continuity is the only requirement on  and ~. u u In a standard manner, we set up the weak form of the equilibrium equations in the reference conguration: Z Z rX g : P dX0 g t0 dC0 ; 22
X0 C0

which describes a displacement eld that is crossed by a single discontinuity, but is otherwise continuous. Accordingly, the partition-of-unity property of nite element shape functions can be used in a straightforward fashion to incorporate discontinuities in a manner that preserves their discontinuous character. This approach for capturing discontinuities can be generalized to large displacement gradients in a straightforward and consistent manner [23]. To this end, one rst rewrites eq. (17) as x X  HC0 ~ u u
d

where a subscript X denotes dierentiation with respect to the reference conguration, P denotes the nominal stress tensor and t0 is the nominal traction at the external boundary. In a BubnovGalerkin method the test functions g for a single discontinuity are given by g g  HC 0 ~ : g
d

23

18

with x and X representing the current and reference conguration, respectively, and HC0 the Heaviside function d at the interface in the reference conguration, C0 , see d Fig. 14(a). The deformation gradient follows by dierentiation  ~ F F HC0 F dC0 ~  nC0 u
d d d

Pushing forward to the current conguration and requiring that the result holds for arbitrary  and ~ g g yields: Z Z  t dC; g g 24a rx  : r dX
X C

Z
X

g rx ~ : r dX

Z
Cd

~ td dC g

Z
C

g HCd ~ t dC

24b

19

 ~ with F I o=oX; F o=oX and dC0 the Dirac u u d function at the interface in the reference conguration. With aid of Nansons relation for normals n to a surface C dC0 n det FFT n0 ; 20 dC the expressions for the normals at the side and at the + side of the interface can be derived (see Fig. 14(b)) dC0 d   T nC det FF nC0 21a ; d d dC d  ~  nC detF FF
d

with the subscript x signifying dierentiation with respect to the current conguration, r the Cauchy stress and td nC r the traction at the discontinuity in the d current conguration. With a standard interpolation  N ; g w ~ N~ ; g w 25 where N contains the interpolation polynomials and  ~ w and w contain the discrete values for the test functions, the discrete format of Eqs. (24a)(24b) reads Z Z BT r dX NT t dC; 26a
X C

Z
X

BT r dX

Z
Cd

NT td dC

Z
C

HCd NT t dC:

26b

dC0 d ~ FT nC0 d dC d

21b

In the cohesive-zone approach, interface tractions td are transmitted between C and C with dierent normals d d nC and nC . In a heuristic assumption in [23] it has been d d assumed that   T dC0 d  1~  1~ nC det F F F F nC0 27 : d d dC 2 2 d After substitution of the constitutive relations for the plies and that for the interface, a non-linear set of algebraic equations results, which can be solved in a standard manner using an incremental-iterative procedure.

Fig. 14. (a) Body X crossed by a discontinuity Cd. (b) Normals nC and nC at both sides of the discontinuity. d d

720

R. de Borst, J.J.C. Remmers / Composites Science and Technology 66 (2006) 713722

If a NewtonRaphson procedure is used, these equations have to be linearized in order to derive the structural tangential stiness matrix, see [23] for details. To exemplify the possibilities of this approach to model the combined failure mode of delamination growth and local buckling we consider the double cantilever beam of Fig. 15 with an initial delamination length a = 10 mm. Both layers are made of the same material with Youngs modulus E = 135 000 N/mm2 and Poissons ratio m = 0.18. Due to symmetry in the geometry of the model and the applied loading, delamination propagation can be modelled with an exponential mode-I decohesion law   tult tn tult exp vn ; 28 dis Gc dis where tn and vn are the normal traction and displacedis dis ment jump, respectively. The ultimate traction tult is equal to 50 N/mm2, the work of separation is Gc 0:8 N=mm. This case, in which failure is a consequence of a combination of delamination growth and structural instability, has been analysed using conventional interface elements in [4]. The beam is subjected to an axial compressive force 2P, while two small perturbing forces P0 are applied to trigger the buckling mode. Two nite element discretizations have been employed, a ne mesh with three elements over the thickness and 250 elements along the length of the beam, and a coarse mesh with only one (!) element over the thickness and 100 elements along the length. Fig. 16 shows that the calculation with the coarse mesh approaches the results for the ne mesh closely. For instance, the numerically calculated buckling load is in good agreement with the analytical solution. Steady-state delamination growth starts around a lateral displacement u = 4 mm. From this point onwards, delamination growth interacts with geometrical instability. Fig. 17 presents the deformed beam for the coarse mesh at a tip displacement u = 6 mm. Note that the displacements are plotted at true scale, but that the dierence in displacement between the upper and lower

4 3 2 1 0

Perfect bond Debonding (dense mesh) Debonding (coarse mesh)

4 u (mm)

Fig. 16. Loaddisplacement curves for delamination-buckling test [24].

Fig. 17. Deformation of coarse mesh after buckling and delamination growth (true scale) [24].

parts of the beam is for the major part due to the delamination and that the strains remain small. The excellent results obtained in this example for the coarse discretization have motivated the development of a layered plate/shell element in which delaminations can occur inside the element between each of the layers [25]. Because of the absence-of-rotational degrees of freedom, the solid-like shell element was taken as a point of departure. The shell of Fig. 18 is crossed by a disconti-

geometrical node internal node

Fig. 15. Double cantilever beam with initial delamination under compression. Fig. 18. Enhanced nodes (black) whose support contains a discontinuity (grey surface). The nodes on the edge of the discontinuity are not enhanced in order to ensure a zero delamination opening at the tip.

R. de Borst, J.J.C. Remmers / Composites Science and Technology 66 (2006) 713722

721

Fig. 19. Loaddisplacement curve and deformations of shell model after buckling and delamination growth (true scale) [25].

nuity surface C0 which is assumed to be parallel to the d mid-surface of the thick shell. The displacement eld /(n, g, f) can now be regarded as a continuous regular  ~ eld / with an additional continuous eld / that determines the magnitude of the displacement jump. The position of a material point in the deformed conguration can then be written as  ~ x X / HC0 /:
d

29

Since the displacement eld / is a function of the variables ut, ub and w, we need to decompose these three terms as u ut t HC0 ~t ; u
d

u ub b HC0 ~b ; u
d

30

 ~ w w HC0 w:
d

Inserting Eq. (30) into Eqs. (10)(12) gives u0 0 HC0 ~0 ; u u

d

u1 1 HC0 ~1 ; u u
d

31

u2 2 HC0 ~2 ; u u
d

where 0 1 t b ; u u u 2 1  u1 2 t b ; u u   u u2 wD 1 ; ~0 1 ~t ~b ; u u u 2 1 ~ u1 2 ~t ~b ; u u ~ ~ u u2 wD 1 ~1 w~1 : u u 32

It is noted that the enhanced part of the internal stretch parameter u2, i.e., ~2 , contains regular variables as well u as additional variables. The elaboration of the strains, the equilibrium equations and the linearization follows standard lines [25]. The magnitude of the displacement jump at the discontinuity is governed by an additional set of degreesof-freedom which are added to the existing nodes of the model. Fig. 18 shows the activation of these additional sets of degrees of freedom for a given (static) delamination surface in the model. Both the geometrical and the internal nodes are enhanced when the corresponding element is crossed by the discontinuity. This

implies that each geometrical node now contains three additional degrees-of-freedom next to the three regular ones, giving six degrees-of-freedom in total. Each internal node has one extra degree-of-freedom added to the single regular degree-of-freedom. The discontinuity always stretches through an entire element. This avoids the need for complicated algorithms to describe the stress state in the vicinity of a delamination front within an element. As a consequence, the discontinuity touches the boundary of an element. The geometrical and internal nodes that support this boundary are not enhanced in order to assure a zero crack tip condition. At variance with conventional interface elements, a criterion is needed for the placement of the discontinuity upon propagation. This criterion is based on the stress state at the delamination front, which can be monitored by adding temporary sample points. When the criterion exceeds a threshold value, the discontinuity is extended into the new element. The corresponding nodes of this element are enhanced with an additional set of degrees-of-freedom. The example of Fig. 15 has been reanalysed with a mesh composed of eight node enhanced solid-like shell elements [25]. Again, only one element in thickness direction has been used. In order to capture delamination growth correctly, the mesh has been rened locally. Fig. 19 shows the lateral displacement u of the beam as a function of the external force P. The loaddisplacement response for a specimen with a perfect bond (no delamination growth) is given as a reference. The numerically calculated buckling load is in agreement with the analytical solution. Steady delamination growth starts around a lateral displacement u % 4 mm, which is in agreement with the previous simulations [4,24].

6. Concluding remarks Numerical models with separate nite elements for interfaces and plies are a powerful tool to analyse delaminations in composite structures, but have some restrictions. Because the interface elements have to be

722

R. de Borst, J.J.C. Remmers / Composites Science and Technology 66 (2006) 713722 [8] Hashagen F, Schellekens JCJ, de Borst R, Parisch H. Finite element procedure for modelling bremetal laminates. Compos Struct 1995;32:25564. [9] de Borst R, Schipperen JHA. Computational methods for delamination and fracture in composites. In: Allix O, Hild F, editors. Continuum damage mechanics of materials and structures. Amsterdam (Oxford): Elsevier; 2002. p. 32552. [10] Barenblatt GI. The mathematical theory of equilibrium cracks in brittle fracture. Adv Appl Mech 1962;7:55129. [11] Dugdale DS. Yielding of steel sheets containing slits. J Mech Phys Solids 1960;8:1008. [12] Parisch H. A continuum-based shell theory for non-linear applications. Int J Numer Meth Eng 1995;38:185583. [13] Remmers JJC, de Borst R. Delamination buckling of bremetal laminates under compressive and shear loadings. In: 43rd AIAA/ ASME/ASCE/AHS/ASC structures, structural dynamics and materials conference, Denver, Colorado, USA 2002; CD-ROM. [14] Rinderknecht S. Delamination in Faserverbundplatten, ein vereinfachtes Berechnungsmodell. Dissertation, University of Stuttgart, Stuttgart, Germany; 1994. [15] Hashagen F, de Borst R. Numerical assessment of delamination in bre metal laminates. Comp Meth Appl Mech Eng 2000;185:14159. [16] Shivakumar K, Whitcomb J. Buckling of a sublaminate in a quasi-isotropic composite laminate. J Compos Mater 1985;19:218. [17] Schellekens JCJ, de Borst R. On the numerical integration of interface elements. Int J Numer Meth Eng 1993;36:4366. [18] Remmers JJC, Wells GN, de Borst R. Analysis of delamination growth with discontinuous nite elements. In: Pamin J, editor. Solids, structures and coupled problems, Proceedings of the second european conference on computational mechanics. Cracow (Poland): Cracow University of Technology; 2001. CDROM. [19] Belytschko T, Lu YY, Gu L. Element-free Galerkin methods. Int J Numer Meth Eng 1994;37:22956. [20] Babuska T, Melenk JM. The partition of unity method. Int J Numer Meth Eng 1997;40:72758. [21] Belytschko T, Black T. Elastic crack growth in nite elements with minimal remeshing. Int J Numer Meth Eng 1999;45:60120. [22] Moes N, Dolbow J, Belytschko T. A nite element method for crack growth without remeshing. Int J Numer Meth Eng 1999;46:13150. [23] Wells GN, de Borst R, Sluys LJ. A consistent geometrically nonlinear approach for delamination. Int J Numer Meth Eng 2002;54:133355. [24] Wells GN, Remmers JJC, de Borst R, Sluys LJ.A large strain discontinuous nite element approach to laminated composites. In: Miehe C, editor. IUTAM symposium on computational mechanics of solid materials at large strains, Dordrecht: Kluwer; 2003. p. 35564. [25] Remmers JJC, Wells GN, de Borst R. A solid-like shell element allowing for arbitrary delaminations. Int J Numer Meth Eng 2003;58:201340.

inserted a priori, spurious elastic deformations will occur prior to delamination onset. These undesired deformations can be partially suppressed by assigning a high value to the normal stiness modulus in the elastic range, but this can result in traction oscillations ahead of the crack tip and in erroneous wave reections when dynamic delaminations are analysed. Furthermore, this methodology restricts the freedom of the discretization in the sense that element boundaries have to be aligned with surfaces of potential delamination. Exploiting the partition-of-unity property of nite element shape functions enables placement of cohesive interfaces at arbitrary positions at the onset of delamination. Since interfaces are created at the moment when they are needed, the necessity of assigning an articially high stiness in the elastic regime no longer exists and traction oscillations or spurious wave reections are no longer an issue. The fact that alignment of the discretization with potential planes of delamination is no longer necessary makes possible that unstructured meshes can be used. The versatility of the method is further enhanced by a consistent extension to large strains and by the incorporation in a solid-like shell element. It is the latter extension which enables large-scale computations of composite structures taking into account delaminations.

References
` [1] Allix O, Ladeveze P. Interlaminar interface modelling for the prediction of delamination. Compos Struct 1992;22:23542. [2] Schellekens JCJ, de Borst R. Free edge delamination in carbonepoxy laminates: a novel numerical/experimental approach. Compos Struct 1994;28:35773. [3] Schellekens JCJ, de Borst R. A non-linear nite element approach for the analysis of mode-I free edge delamination in composites. Int J Solids Struct 1993;30:123953. [4] Allix O, Corigliano A. Geometrical and interfacial non-linearities in the analysis of delamination in composites. Int J Solids Struct 1999;36:2189216. [5] Alfano G, Criseld MA. Finite element interface models for the delamination analysis of laminated composites: mechanical and computational issues. Int J Numer Meth Eng 2001;50:170136. ` [6] Ladeveze P, Lubineau G. An enhanced mesomodel for laminates based on micromechanics. Compos Sci Technol 2002;62:53341. [7] de Borst R. Some recent issues in computational failure mechanics. Int J Numer Meth Eng 2001;52:6395.