You are on page 1of 46

AN INTERFACE FINITE ELEMENT FOR THE

SIMULATION OF LOCALIZED
MEMBRANE-BENDING DEFORMATION IN
SHELLS
published in Computer Methods in Applied Mechanics and Engineering, Vol. 200, No. 29-32. (July
2011), pp. 2378-2396, doi:10.1016/j.cma.2011.04.009

A. Giampieri, U. Perego
Department of Structural Engineering
Politecnico di Milano
Piazza L. da Vinci, 32 - 20133 Milano (Italy)

corresponding author: umberto.perego@polimi.it

April 30, 2015

Abstract
The folding of thin shells around localized lines occurs in several situations of en-
gineering interest, such as in buckling induced deformation, as the consequence of
crash, in industrial processes like metal forming, in the deployment of folded mem-
branes, in packaging related processes. In this paper an interface element, allowing for
displacement and rotation discontinuities, to be placed between two adjacent 4-node
Mindlin-Reissner shell finite elements, is developed for a computationally effective
simulation of this type of localized deformation. The large displacement and rotation
kinematics of the interface element is discussed, arriving at a rigid-rotation free ele-
ment tangent stiffness matrix. An adaptive dynamic relaxation scheme is proposed for
the evolutionary analysis of the quasi-static structural response. To test the element
formulation, a simple coupled membrane-bending elastoplastic behavior is assumed.
The element is used for the simulation of few examples taken from the literature, where
the structural response is characterized by the formation of plastic hinges, exhibiting
good accuracy and computational effectiveness.
Keywords: shells; folding; cohesive interface; crease.

1 Introduction
The localized folding of shells occurs in several situations of engineering interest, such as
in buckling induced deformation, as the consequence of crash, in industrial processes like
in metal forming [1, 19, 21, 37], in the deployment of folded membranes or in packaging re-
lated processes. Large gossamer space structures such as space telescopes, space inflatable
antenna reflectors or solar sails, must be folded for packaging before launching. When they
are deployed, wrinkles are generated by compressive stresses. The folding and packaging

1
processes create some permanent creases which may interfere with the wrinkle pattern and
impact the intended performance of the deployed structure which has tight shape require-
ments [34, 36, 25, 35]. In industrial carton packaging processes, the final performance of
the package may depend substantially on the quality of the folding which has to produce
well defined edges and corners, without damaging the external surface of the container. To
facilitate the folding of the paperboard around the prescribed lines the paperboard blank is
“creased”, i.e. the folding lines are scored onto the paperboard by pressing it by a male die
with a rule onto a grooved female die. The creasing locally damages the paperboard reduc-
ing its bending stiffness and promoting the folding around the design lines [14, 22, 32, 6].
Nowadays, this type of problems can be attacked using nonlinear finite element shell
codes where a highly refined mesh is used to model the localized region where the fold-
ing occurs. Even though the fast growth of computing power is making this continuous
approaches more and more viable, the need to resolve the small size of the folding region
still leads in most cases to unaffordable computational costs. This is particularly evident
in the cases mentioned above, with large structures with highly localized crease lines. The
alternative approach, which is pursued in the present work, consists of developing ad hoc
finite elements incorporating the kinematics of discontinuous displacements and rotations.
The localized folding of shells has received considerable attention in the literature on
limit analysis of plates and shells, where the notion of cylindrical plastic hinge has been ex-
tensively used to predict the collapse mechanism in elastic-perfectly plastic structures (see
e.g. the classical books by Hodge [13] and Massonet and Save [20]). In a different context,
Rice and Levy [27] and Lee and Parks [15, 16] used continuously distributed line springs
in plates and shells to simulate the localized increase of bending compliance due to a part-
through crack. The basic idea was that the damage in the plate or shell could be accounted
for by localizing the bending stiffness reduction in a line of distributed membrane-bending
springs. A computational assessment on the use of this type of elements is presented in
[12]. Recently, Armero and Ehrlich [4] proposed a finite element method for the modeling
of softening hinge lines in Mindlin-Reissner plates. They formulated small displacements
triangular and quadrilateral finite elements incorporating embedded discontinuous rotations
across the hinge line according to the Strong Discontinuity Approach, where the finite el-
ements are enhanced with the singular strain fields associated with the discontinuities of
hinge lines.
Shell finite elements incorporating a discontinuous kinematics in the spirit of the Ex-
tended Finite Element Method (XFEM) have been proposed for the simulation of crack
propagation in shell structures by Areias and Belytschko [2] and Areias et al. [3], who for-
mulated a Mindlin-Reissner and a Kirchhoff type shell element, respectively, for fracture
analysis based on the Extended Finite Element Method. Song and Belytschko [30] devel-
oped an XFEM shell element, endowed with a nonlocal strain-based fracture criterion. The
method was applied to the simulation of crack propagation in metallic pipes.
In this paper an interface element, allowing for displacement and rotation discontinu-

2
ities, to be placed between two adjacent 4-node Mindlin-Reissner shell finite elements, is
developed for a computationally effective simulation of localized deformation due to fold-
ing. Interface shell elements have been proposed in the literature by several authors for
the simulation of crack propagation. Li and Siegmund [17], developed a cohesive interface
element for the simulation of crack growth in thin metal sheets to be used in shell element
meshes of the finite element code Abaqus. Cirak et al. [10] proposed an inter-element co-
hesive crack model based on Kirchhoff shell theory. Zavattieri [38] proposed an interface
cohesive element to be placed between four node Belytschko-Lin-Tsay [7] shell elements.
This latter formulation allows to consider the effects of through-the-thickness propagation
due to bending of part-through cracks by defining a nondimensional effective displacement
jump which includes the contribution of the rotation discontinuity.
The main features of the cohesive interface element presented in this paper are that it is
formulated in terms of generalized variables and that it allows for large displacements and
rotation jumps across the interface. The element generalized kinematic variables are defined
on the basis of a virtual work equivalence so that they are work conjugate to the element
generalized internal forces. Particular attention is devoted to the definition of compatibility
conditions between the element degrees of freedom and the generalized kinematic variables
that make the element internal work insensitive to rigid body rotations. The element kine-
matics is completely developed and the expression of the geometric tangent stiffness ma-
trix is provided. The adaptive dynamic relaxation scheme proposed by Oakley and Knight
[23, 24] has been used for the evolutionary analysis of quasi-static structural responses.
The element can be used in conjunction with any material constitutive model expressed
in terms of generalized variables. In particular, the interface element here proposed is suited
for the simulation of localized deformation along predefined lines, such as crease lines,
where the shell has been either intentionally or accidentally weakened. However, it can be
used also in situations where the localized deformation develops as a consequence of the
load application (buckling, plastic collapse) whenever the position of the localization lines
can be estimated a priori with reasonable accuracy. To test the element formulation, a sim-
ple coupled membrane-bending elastic-perfectly plastic behavior is assumed in this paper.
The element is used for the simulation of few examples taken from the literature, where
the structural response is characterized by the formation of plastic hinges, exhibiting good
accuracy and computational effectiveness. Other applications, requiring the formulation of
more complex constitutive models for the description of the material in the crease region,
will be discussed in a forthcoming paper.
In this paper, bold symbols will be used to define either vector and tensor quantities, or
matrix quantities, the proper nature of the variable being clear from the context. A reference
frame characterized by a basis e1 , e2 , e3 will be identified by the notation {ei }. If Ω is a
tensor, the notation [Ω]{ei } will be used to denote the matrix of the components of Ω with
respect to the reference frame {ei }.
Superposed symbols (bar, tilde and hat) will be used with the following meaning:

3
• superposed bar, ¯•, will be used to denote quantities belonging to the shell middle
surface;

• superposed tilde, ˜•, will be used to denote quantities belonging to the interface refer-
ence surface;

• superposed hat, ˆ•, will be used to denote quantities defined in the interface co-
rotational reference frame.

2 Kinematic description
2.1 Shell Geometry and deformation
Consider a shell body of constant thickness h immersed in a fixed reference frame {ii },
i = 1, 2, 3. Within the context of the inextensible director shell theory, the geometry of the
shell is described by the mapping

h h
X = Φ(ξ 1 , ξ 2 , ξ 3 ) = Φ̄(ξ 1 , ξ 2 ) + ξ 3 T(ξ 1 , ξ 2 ) − ≤ ξ3 ≤ (1)
2 2
where X is the position vector of a material point in the shell body identified by the con-
vective system of coordinates ξ 1 , ξ 2 , ξ 3 ; X̄ = Φ̄(ξ 1 , ξ 2 ) is the position vector of points
belonging to the shell middle surface M (ξ 3 = 0), with boundary ∂M, and the unit vector
T(ξ 1 , ξ 2 ) denotes the director field. In the present formulation, T is assumed to be normal
to the middle surface in the original configuration but, according to the Mindlin-Reissner
assumption, it is not forced to remain normal during the deformation.
A shell deformed configuration at time t ∈ [0, T ] is defined by the mapping x = χt (X)
which, in terms of the middle surface convective coordinates ξ 1 , ξ 2 , ξ 3 , can be expressed as

h h
x = φ(ξ 1 , ξ 2 , ξ 3 ) = φ̄(ξ 1 , ξ 2 ) + ξ 3 t(ξ 1 , ξ 2 ) − ≤ ξ3 ≤ (2)
2 2
with obvious meaning of symbols. The deformation (2) accounts for the inextensibility
assumption of the director field. The displacement u at a point X(ξ 1 , ξ 2 , ξ 3 ) follows from
this kinematic description as

u(ξ 1 , ξ 2 , ξ 3 ) = x(ξ 1 , ξ 2 , ξ 3 ) − X(ξ 1 , ξ 2 , ξ 3 ) = ū(ξ 1 , ξ 2 ) + ξ 3 [t(ξ 1 , ξ 2 ) − T(ξ 1 , ξ 2 )] (3)

In the fixed orthogonal reference system {ii }, the position and displacement fields are ex-
pressed as

X = X i ii , x = xi ii , T = T i ii , t = ti ii u = ui ii (4)

4
The deformation of the inextensible unit director t can be described as a rigid rotation of the
reference director T.

t = R(θ)T (5)

where R(θ) is an orthogonal rigid rotation tensor and θ is its axial rotation vector, such that
Rθ = θ. The rotation tensor can be conveniently expressed in terms of the axial rotation
vector using the exponential mapping defined by Rodrigues formula, which associates any
rotation tensor R to its corresponding skew-symmetric tensor Ω(θ) such that

(Ω(θ)T) · T = (θ × T) · T = 0. (6)

One has
sin θ 1 − cos θ 2
R(θ) = exp[Ω(θ)] = I + Ω(θ) + Ω (θ) (7)
θ θ2
where I is the identity tensor and θ = ∥θ∥ defines the amplitude of the rigid rotation. It has
been shown ([28]) that there is only one axial vector θ such that θ · T = 0 and t = R(θ)T.
This allows to unambiguously define the axial vector θ as a vector belonging to the plane
locally orthogonal to the director T, i.e. θ · T = 0 will be assumed henceforth. The
displacement field can also be expressed in terms of the axial rotation vector as

u(ξ 1 , ξ 2 , ξ 3 ) = ū(ξ 1 , ξ 2 ) + ξ 3 (R(θ) − I)T(ξ 1 , ξ 2 ) (8)

The present description of the shell deformation naturally leads to the definition of two
types of local reference frames in the initial and deformed configurations (Figure 1).

1. the surface convected frames, denoted by {Ai } and {ai }, i = 1, 2, 3, where Aα =


Φ̄,α and aα = φ̄,α , α = 1, 2, define a covariant basis in the plane tangent to the
middle surface at ξ 1 , ξ 2 , with the associated metric tensors

A = Aαβ Aα ⊗ Aβ , Aαβ = Aα · Aβ (9)


a = aαβ a ⊗ a ,
α β
aαβ = aα · aβ (10)

and unit vectors normal to the middle surface

A1 × A2 a1 × a2
A3 = , a3 = (11)
∥A1 × A2 ∥ ∥a1 × a2 ∥

2. the director orthogonal frame, denoted by {ei }, i = 1, 2, 3, with e3 = T, e1 · T = 0


and e2 = T × e1 . According to this definition, the direction of e1 is arbitrary in a
plane orthogonal to T.

5
eˆ t ≃ tɶ
a3
a2
a1 eˆ c

Γ
t−
c eˆ n
− Γ c+
e 2

i3 e1− e1+ t+
i2 Γɶ c
i1 e 2+

Figure 1: Reference frames and interface reference surface.

The condition that the axial vector θ belongs to the plane perpendicular to T makes it
particularly convenient to express its components in the director reference frame {ei }, so
that only two independent parameters, θ̄1 and θ̄2 , govern the director rotation:

θ = θ1 i1 + θ2 i2 + θ3 i3 = θ̄1 e1 + θ̄2 e2 (12)

and
T = T 1 i1 + T 2 i2 + T 3 i3 = e3 (13)

The transformation between the director and the fixed reference frames is obtained in a
standard way using the orthogonal tensor Q defined below

ei = Qii = Qjh (ij ⊗ ih ) · ii = Qji ij , Qij = ii · ej , Q = [e1 e2 e3 ]{ii } (14)

where the notation [ ]{ii } means that the components of the bracketed quantity are expressed
in the {ii } reference frame. The components θ̄i of the axial rotation vector θ in the director
reference frame {ei } can be expressed in terms of its components in the fixed reference
frame {ii } as

θ̄i = θ · ei = Qji θ · ij = Qji θj (15)

which in matrix notation reads

[θ]{ei } = QT [θ]{ii } (16)

where [θ]{ei } and [θ]{ii } are the column arrays containing the components of θ in the
director and fixed reference frames, respectively.
The strain in the shell body can be measured by the Green-Lagrange strain tensor

1
E = (gi · gj − Gi · Gj )Gi ⊗ Gj = Eij Gi ⊗ Gj (17)
2

6
where Gi and Gi are the covariant and contravariant base vectors, respectively, in the ref-
erence configuration and gi is the covariant basis in the current configuration. These are
defined as

Gα = Φ,α = Aα + ξ 3 T,α (18)


G3 = Φ,3 = T = A3 (19)
gα = φ,α = aα + ξ 3 t,α (20)
g3 = φ,3 = t (21)

The non-zero components of E, up to linear terms in ξ 3 , are then given by

1
Eαβ = (ū,α Aβ + Aα ū,β +ū,α ū,β ) +
2
1 3
ξ [(t,α −T,α ) · Aβ + Aα · (t,β −T,β ) + ū,α ·t,β +t,α ·ū,β ] (22)
2
1
Eα3 = [ū,α ·t + Aα · (t − T)]
2

In Eαβ , as usual in thin shell formulations (see e.g. [9]), the quadratic term in ξ 3 has been
neglected under the assumption of small shell thickness.

2.2 Interface geometry and deformation


The interface region is modeled as a zero thickness interface cutting the shell through its
depth. In geometrical terms, in the initial configuration the interface is a surface defined by
its intersection Γc (henceforth referred to as reference interface line) with the shell middle
surface and the corresponding director field. Hence, points belonging to the interface are
defined by the mapping

Xc = Φ̄c (ξ 1 (ζ), ξ 2 (ζ)) + ξ 3 Tc (ξ 1 (ζ), ξ 2 (ζ)) (23)

where Φ̄c (ζ) is the parametric representation of Γc on the middle surface and ζ is a scalar
parameter.
It is convenient to define a couple of covariant basis vectors for the interface surface

∂ξ α ∂ξ α
Gc = (Φ̄c,α + ξ 3 Tc,α ) = Ac + ξ 3 Tc,α (24)
∂ζ ∂ζ
G3 = T(ξ 1 (ζ), ξ 2 (ζ)) = Tc (25)

∂ξ α
where Ac = ∂ζ Φ̄c,α is the tangent vector to the interface line Γc . The normal field to the
interface surface is defined by the unit vector

7
Gc × G3
Gn = = Gn (26)
∥Gc × G3 ∥
while

A c × Tc
An = = An (27)
∥Ac × Tc ∥
is the unit normal to the interface line on the middle surface.
The interface allows for displacement jumps

JuK = u+ − u− = JūK + ξ 3 JtK (28)



c ×[− 2 , 2 ] and Γc ×[− 2 , 2 ] of the interface, where the two curves,
h h h h
between the two sides Γ+
Γ+ −
c and Γc have the same parametric representation in terms of the convective coordinate
ζ. Following [10], in the deformed configuration it is convenient to define the interface
reference line as the average curve Γ̃c between Γ+ −
c and Γc (see Figure 1). The deformed
interface reference surface Γ̃c × [− h2 , h2 ] is then defined as the locus of points of coordinates

h h
˜ (ξ 1 (ζ), ξ 2 (ζ)) + ξ 3 t̃(ξ 1 (ζ), ξ 2 (ζ))
x̃ = φ̄ − ≤ ξ3 ≤ (29)
2 2
where

1 1
˜ = (φ̄+ + φ̄− ),
φ̄ t̃ = (t+ + t− ) (30)
2 2
Consistent with the notation in (29) and (30), a superposed tilde will be used henceforth to
denote quantities defined on the deformed interface reference surface. Note that, according
to the definition in (30)2 , t̃ is not a unit vector.
The covariant basis vectors for the reference deformed interface surface are

∂ξ α ∂ξ α
g̃c = ˜ ,α +ξ 3 t̃,α ) = ãc + ξ 3
(φ̄ t̃,α , g̃3 = t̃ (31)
∂ζ ∂ζ
˜ ,α ∂ξ α /∂ζ defines the tangent direction to the deformed reference interface
where ãc = φ̄
line Γ̃c while the unit normal vectors are given by

g̃c × g̃3 ãc × t̃


g̃n = = g̃n and ãn = = ãn (32)
∥g̃c × g̃3 ∥ ∥ãc × t̃∥

3 Weak form of equilibrium


Let PdS be the internal force, due to the interface deformation, acting on an elementary
area dS = ∥Gc × G3 ∥dξ 3 dζ of the undeformed interface reference surface Γc × [− h2 , h2 ].
With these definitions, the weak form of equilibrium is obtained for arbitrary compatible
variations of the configuration through the virtual work expression

8
δΠint − δΠext = 0 (33)

where

∫ ∫
δΠext = b · udV + f · udS (34)
V ∂V
∫ ∫ h ∫ ∫ h
2 2
δΠint = δΠint,V + δΠint,c = S : δE µdξ 3 dM + P · δw γdξ 3 dΓc
M\Γc −h
2
Γc −h
2
(35)

In equations (34) and (35), V is the volume occupied by the shell body in the reference
configuration, ∂V is the portion of the boundary where traction boundary conditions are
prescribed, b and f denote the assigned body and traction forces, respectively, dM = ∥A1 ×
A2 ∥dξ 1 dξ 2 is the middle surface infinitesimal element, S is the second Piola-Kirchhoff
stress tensor, µ and γ are defined as

∥G1 × G2 ∥ ∥Gc × G3 ∥
µ= , γ= √ (36)
∥A1 × A2 ∥ Ac · Ac
and account for the curvature of the middle surface and of the interface line, respectively,
while δE, δw are the strain variations for the shell and the interface, conjugate to S and P
and will be defined in the next section.

3.1 Shell and Interface strain variations


The variation δE of the Green-Lagrange strain tensor is defined as (see e.g. [9])

1 1
δEαβ = (δ ū,α ·aβ + δ ū,β ·aα ) + ξ 3 (δt,α ·aβ + δt,β ·aα + δ ū,α ·t,β +δ ū,β ·t,α )
2 2
1
δEα3 = (δt · aα + δ ū,α ·t) (37)
2

The variations in equations (37) require the definitions of the variations δ ū, δ ū,α , δt and
δt,α . Since the displacement ū of the shell middle surface is a primary variable for the
description of the shell deformation, its variations can be readily obtained from the adopted
displacement model. In contrast, the variations δT and δt of the director fields depend on
the variation of the axial rotation vector and have to be obtained through the variation of
the exponential mapping defined in (5), computed in the original and current configurations
[28]. In view of the inextensibility constraint on the directors, their admissible variations
have to belong to the tangent spaces of the unit sphere, i.e.

δT = Ω(δθ)T = δθ × T; δt = Ω(δϑ)t = δϑ × t (38)

9
δ ur + = δ ϑɶ × (u + ξ 3 t + )

δ ur − = δ ϑɶ × ξ 3 t −
ξ 3t − δ ϑɶ ξ 3t +
O− u O+

δ u = δ ϑɶ × (u + ξ 3 t + ) − δ ϑɶ × ξ 3 t − = δ ϑɶ × u + δ ϑɶ × ξ 3  t 


r

Figure 2: Interface virtual rigid body rotation starting from open configuration.

where δθ is a virtual rotation of the director T around an axis belonging to the orthogonal
plane to T while δϑ = R(θ)δθ denotes a virtual rotation of the director t in the current
configuration and R(θ) is the rotation tensor which transforms T into t, see equation (5).
According to (38), the variation δt is orthogonal to t, i.e. t · δt = 0. Taking into account
that t = R(θ)T, equation (38) can also be written as

δt = R(θ) (δθ × T) = R(θ)δT (39)

The variation δJuK of the displacement jump is defined as

δJuK = δJūK + ξ 3 δJtK (40)

and requires the definition of the virtual director jump δJtK:

δJtK = δt+ − δt− = δϑ+ × t+ − δϑ− × t− = δJϑK × t̃ + δ ϑ̃ × JtK (41)

where δ ϑ̃ = 12 (δϑ+ + δϑ− ) and t̃ is defined in equation (30)2 . Hence,

δJuK = δJūK + ξ 3 δJϑK × t̃ + ξ 3 δ ϑ̃ × JtK (42)

Note that the displacement discontinuity (42) does not describe pure straining modes of
the interface. Let us denote by δJuKt and δJuKr the jump variations due to virtual rigid
body translations and rotations, respectively. One has that a rigid translation corresponds
to δ ū+ = δ ū− ̸= 0 and δϑ+ = δϑ− = 0, and hence δJuKt = 0, i.e. a virtual rigid body
translation does not produce any jump variations. On the contrary, a rigid body rotation δ ϑ̃,
produces the jump variation (see Figure 2)

δJuKr = δ ϑ̃ × JūK + ξ 3 δ ϑ̃ × JtK = δ ϑ̃ × JuK (43)

which is in general different from zero if JuK ̸= 0, i.e. when the variation occurs starting

10
δ ϑɶ δ ϑɶ
ξ 3t +
ξ 3t −
O+
O− u

δ ϑ+ = δ ϑ− = δ ϑɶ ; δ u = 0

Figure 3: Interface shear deformation starting from open configuration for virtual rotation
δ ϑ̃.

from an open interface configuration. Note that δ ū+ = δ ū− = 0, δϑ+ = δϑ− ̸= 0 does not
define a pure rigid body rotation, since it contains a deformation component of the interface
given by −δ ϑ̃ × JūK, which has to be included into the virtual internal work (see Figure 3).
Taking into account the expression (42) of δJuK, the interface pure straining mode w,
which will be used in the next section for the definition of the virtual internal work of the
interface, is therefore defined by

δw = δJuK − δJuKt − δJuKr = δ w̄ + ξ 3 δJϑK × t̃ − δ ϑ̃ × w̄ (44)

where the notation w̄ = JūK has been introduced. It is convenient to group in (44) the terms
which are constant through the thickness

δw = δη + ξ 3 δJϑK × t̃ (45)

where δη is defined as
δη = δ w̄ − δ ϑ̃ × w̄ (46)

Note that the term δ ϑ̃ × w̄ is a geometric contribution to the generalized strains which can
be neglected whenever the interface opening w̄ can be assumed to be infinitesimal (small
strain assumption). Three-dimensional views of a pure straining mode δw and of a rigid
body rotation are shown in Figures 4 and 5.

3.2 Weak form of equilibrium in generalized variables at interface


In view of the definition (44) of the interface virtual strain, using the permutation rule of the
triple box product, the internal virtual work on the cohesive interface is written as

11
u
a)

w = u

Γ c−
t Γ c+
e 2−
i3 e1− e1+ t+
i2
+
i1 e 2

b)
δ u = δ ϑɶ × u
r

δ ϑɶ
O
u

Γ c−
t− Γc+

e 2
e2−
i3 e 1− e1+ t+
i2 t−
+
i1 e 2

Figure 4: a) Interface open configuration. b) Rigid body virtual rotation around axis passing
through point O, with director rotations δ ϑ̃ = δϑ+ = δϑ− .

∫ ∫ h
2
δΠint,c = P · δw γdξ 3 dΓc
Γc −h
∫ (∫
2
h
) ∫ (∫ h
)
2 2
= Pγdξ 3 · δηdΓc + (t̃ × P) ξ 3 γdξ 3 · δJϑKdΓc
Γc −h
2
Γc −h
2

(47)

The internal work δΠint,c can also be conveniently expressed in terms of stress resultants. If
the vectors N of the internal forces per unit length and M of internal couples for unit length
are defined as
∫ ∫
( )
h h
2 2
N= Pγdξ ; 3
M= t̃ × P ξ 3 γdξ 3 (48)
−h
2
−h
2

δΠint,c can be rewritten as


∫ ∫
δΠint,c = N · δη dΓc + M · δJϑK dΓc (49)
Γc Γc

According to (49), δη and δJϑK turn out to be the generalized strains conjugate in the virtual

12
u(ζ ,ξ ) + δ w (ζ ,ξ )
3 3

u(ζ ,ξ ) 3

ζ
ξ3
Γc−
t− − w (ζ ) = u (ζ ) Γ c+
e 2

i3 e1− e1+ t+
i2
+
i1 e 2

Figure 5: Virtual pure straining mode δw(ζ, ξ 3 ) starting form open interface configuration
with director rotations δϑ+ = −δϑ− ⇒ δ ϑ̃ = 0.

work sense to the generalized internal forces N and M, respectively.


For the definition of the constitutive behavior of the interface, it is convenient to intro-
duce at points on Γ̃c a local co-rotational orthogonal reference frame {êc , êt , ên } defined as
(see Figure 1)

ãc
êc = , ên = ãn , êt = ên × êc (50)
∥ãc ∥
It should be noted that, according to this definition, in general êt × t̃ ̸= 0 so that the co-
rotational basis is aligned with the tangent ãc to Γ̃c and with the normal ãn to the interface
reference surface, but not with the average director t̃. However, in most cases, the bending
stiffness for rotations around the interface material axis (see Figure 6), here identified by
the vector ãc , tangent to Γ̃c , is smaller than around the other directions. Rotations around ãc
are then expected to be significantly larger than around the other covariant axes, realizing
between the “+”and “-” sides of the shell a type of joint similar to a cylindrical hinge in
plates. As a consequence, large rotations are expected only around ãc , while êt and t̃ are
expected to remain almost parallel throughout the interface deformation process.
The components of the generalized forces per unit length in the co-rotational basis are
given by (see Figure 6):

Vc = N · êc , N = N · ên , Vt = N · êt (51)


Mc = M · êc , Mn = M · ên , Mt = M · êt ≃ M · t̃ = 0 (52)

where N is the normal force to the reference interface surface, Vc and Vt are the components
of the shear force V = (I − ên ⊗ ên )N, Mc is the bending moment around the interface
axis, Mn is the torque around the normal axis and the drilling moment Mt is taken equal to
zero as discussed above.
The generalized strains δη and δJϑK are similarly projected onto the co-rotational axes

13
Vt
Vc
eˆ t ≃ tɶ
Mc
eˆ c Mn

eˆ n N
interface material axis Γ
ɶ
c

interface reference surface

Figure 6: Co-rotational components of internal forces and couples.

to obtain the kinematic quantities conjugated to the generalized forces and couples

δηn = δη · ên , δηc = δη · êc , δηt = δη · êt


δϑn = δJϑK · ên , δϑc = δJϑK · êc

With this notation, the internal work is written as



δΠint,c = [N δηn + Vc δηc + Vt δηt + Mn δϑn + Mc δϑc ] dΓc (53)
Γc

3.3 Matrix formulation of the internal work


For the purpose of the subsequent discretization of the balance equation (33), it is useful
to express the internal work (35) in matrix form. As in (16), a couple of square brackets
around a tensor or a vector quantity [v]{ii } will be used henceforth to denote the matrix (or
column vector) gathering the components of v in the reference frame ii . According to this
notation, let [S]{Ai } and δ[E]{Ai } be the column vectors collecting the five independent
components of the second Piola-Kirchhoff stress tensor and of the virtual variations of the
Green-Lagrange strain tensor, respectively, in the surface convected frame {Ai }
   
S 11 δE11
 22   
 S   δE22 
   
[S]{Ai } =
 S
12 
 δ[E]{Ai } =
 2δE12

 (54)
 13   
 S   2δE13 
S 23 {Ai }
2δE23 {Ai }

The bulk contribution to the internal work can then be expressed in matrix form as

14
∫ ∫ h ∫ ∫ h
2 2
δΠint,V = [S]T{Ai } δ[E]{Ai } dξ 3 µdM = [S]T{Ai } B δ[χ] dξ 3 µdM
M\Γc −2 h
M\Γc −h
2
(55)
where [ ]
δ[ū]
δ[χ] = (56)
δ[t] {ii }

defines the virtual variation of the configuration expressed in the fixed reference frame and
B is the compatibility matrix operator defined as

 ( )T 
[a1 ] + ξ 3 [t,ξ1 ] ∂ξ∂ 1 ξ 3 [a1 ]T ∂ξ∂ 1
 ( )T 
 [a2 ] + ξ 3 [t,ξ2 ] ∂ξ∂ 2 ξ 3 [a2 ]T ∂ξ∂ 2 
 ( )T ( )T ( ) 
 
B= [a2 ] + ξ 3 [t,ξ2 ] ∂ξ∂ 1 + [a1 ] + ξ 3 [t,ξ1 ] ∂
ξ3 [a2 ]T ∂ξ∂ 1 + [a1 ]T ∂ξ∂ 2 
 ∂ξ 2 
 [t]T ∂
[a1 ]T 
 ∂ξ 1 
[t]T ∂ξ∂ 2 [a2 ] T
{ii }
(57)
Since δϑ and t are orthogonal, from (38) one has that δϑ = t × δt. Hence, the gener-
alized strains in (49) can be expressed in terms of the virtual directors δt+ and δt− as

1[ ]
δη = δ w̄ + Ω(w̄)Ω(t+ )δt+ + Ω(w̄)Ω(t− )δt−
2
δJϑK = Ω(t+ )δt+ − Ω(t− )δt− (58)

where Ω(w̄), Ω(t+ ) and Ω(t− ) are skew-symmetric tensors with axial vectors w̄, t+ and
t− respectively, such that δϑ = t × δt = Ω(t)δt and w̄ × δϑ = Ω(w̄)Ω(t)δt.
With these definitions, the interface contribution to the internal work writes
∫ ( )
δΠint,c = [N]T{êi } B N + [M]T{êi } B M δJχKdΓc (59)
Γc

where [N]T{êi } = [N Vc Vt ] and [M]T{êi } = [Mn Mc 0] are the column vectors of the gen-
eralized internal forces in the co-rotational frame,
[ ]
δJχKT = δ[ū+ ] δ[ū− ] δ[t+ ] δ[t− ]
T T T T
(60)
{ii }

is the vector gathering the variables which define the interface current configuration, with
components expressed in the fixed reference frame {ii }, and B N , B M are the linearized
matrix compatibility operators of the cohesive interface, defining the generalized virtual

15
strains in the co-rotational frame so that
   −

δηn [ên ]T −[ên ]T 2 [ên · Ω(w̄)Ω(t )]
1 +
2 [ên · Ω(w̄)Ω(t )]
1
   − 
[δη]{êi } =  δηc  = B N δJχK; B N =  [êc ]T −[êc ]T 2 [êc · Ω(w̄)Ω(t )]
1 +
2 [êc · Ω(w̄)Ω(t )]
1

[êt ]T −[êt ]T −
2 [êt · Ω(w̄)Ω(t )] 2 [êt · Ω(w̄)Ω(t )]
1 + 1
δηt {ii }
   (61)
δϑn 0T 0T + −
[ên · Ω(t )] −[ên · Ω(t )]
   T 
δ [JϑK]{êi } =  δϑc  = B M δJχK; BM = 0 0T [êc · Ω(t+ )] −[êc · Ω(t− )] 
0 0T 0T 0T 0T {ii }
(62)
In (62) 0T is a column vector with zero entries, i.e. 0T = [0 0 0].
Even though, for discretization convenience (see section 5), the virtual work is ex-
pressed in terms of the middle surface displacement field and of the director field, the latter
is not taken as a primary kinematic variable, but rather it is expressed in terms of the two
non-zero components of the rotation vector in the local director orthogonal frame {ei }. As
discussed in [28], this allows for a formulation with only five degrees of freedom per node.
Recalling the definition θ = ∥θ∥, in matrix form one has

     sin θ

θ1 0 cos θ 0 θ θ2
     
[θ]{ei } =  θ2  ; [T]{ei } =  0  ; [R(θ)]{ei } =  0 cos θ − θ θ1
sin θ

0 1 − sinθ θ θ2 sin θ
θ θ1 cos θ
(63)
where in the expression of the rotation matrix [R(θ)]{ei } it has been taken into account that
Ω2 (θ)T/θ2 = −T in the Rodrigues map (7), because θ · T = 0.
With the notation in (63), one has

 sin θ

θ θ2
 
[t]{ei } = [R]{ei } [T]{ei } =  − θ θ1
sin θ
 and [t]{ii } = Q[t]{ei } = [e1 e2 e3 ]{ii } [t]{ei }
cos θ
(64)
Similarly, one has

      
0 0 δθ2 0 δθ2 0 −1 [ ]
       δθ1
δ[T]{ei } = [Ω(δθ)]{ei } [T]{ei } = 0 0 −δθ1   0  =  −δθ1  =  1 0 
δθ2
−δθ2 δθ1 0 1 0 0 0
(65)

16
4 Linearization of the internal work
The weak form of equilibrium (33) is a nonlinear function of the configuration and its
solution requires an iterative strategy. In most cases this is based on the definition of the
tangent stiffness operator of the system. If the mapping x = χ(X) is used to define a shell
deformed configuration, the tangent stiffness operator is obtained through the directional
derivative of the internal virtual work as follows

d
D δΠint (χ; δχ)[dχ] = δΠint (χ + εdχ; δχ) (66)
dε ε=0
where dχ is an arbitrary compatible perturbation of the configuration. To simplify the
discussion, the compact notation d(δΠint ) will be used hereafter to denote the directional
derivative instead of the longer notation in (66). Following the decomposition in (35), the
linearized internal work can be separated into the contributions of the bulk and of the cohe-
sive interface
∫ ∫ h ∫ ∫ h
2 2
3
d(δΠint,V ) = dS : δE µdξ dM + S : d(δE) µdξ 3 dM
M\Γc −h
2
M\Γc −h
2
∫ ∫ h ∫ ∫ h
2 2
d(δΠint,c ) = dP · δw γdξ dΓc +
3
P · d(δw) γdξ 3 dΓc
Γc −h
2
Γc −h
2
(67)

The linearization of the interface contribution is discussed below, while the linearization of
the contribution due to the shell bulk is obtained through a standard procedure (see e.g. [8])
and is not presented here.
Having in mind the expression (44) of δw and taking into account that d(δ w̄) =
d(δJϑK) = d(δ ϑ̃) = 0 since δ w̄ and δϑ are primary kinematic variables, the variation
d(δw) can be written as

d(δw) = ξ 3 δJϑK × dt̃ − δ ϑ̃ × dw̄ (68)

Defining the variations dN and dM of the generalized static quantities as


∫ h ∫ h
2 2
dN = dP γdξ ; 3
dM = (dt̃ × P + t̃ × dP) ξ 3 γdξ 3 (69)
−h
2
−h
2

one has that the variation d(δΠint,c ) of the interface virtual work takes the form

∫ ∫ ∫
d(δΠint,c ) = dN · δη dΓc + dM · δJϑKdΓc − N · (δ ϑ̃ × dw̄) dΓc (70)
Γc Γc Γc

The first two integrals on the r.h.s. of equations (70) represent the contributions of the

17
material tangent moduli, while the third integral represents the geometric contribution to
the tangent operator.
As far as the material contribution is concerned, since small deformations are expected
in directions other than around the interface axis, unlimited and decoupled elasticity is as-
sumed for the shear sliding and for the torque rotation, while an elastoplastic response is
postulated for the membrane-bending behavior around the interface axis. The matrix Dc of
the tangent moduli in the co-rotational frame is therefore expressed as
 
knep 0 0 ep
0 knϑ 0
 c

 kc 0 0 0 0 
  [ ]
 0 
 kt 0 0  DN DN M
Dc =  = (71)
 kϑn 0 
0  DTN M DM

 kϑepc 0 
 symm. 
0

where kc , kt are elastic constants while knep and kϑepc are the elastoplastic tangent stiffnesses
ep
for the membrane and bending behavior around the interface axis and knϑ c
accounts for the
membrane-bending elastoplastic interaction. The derivation of these elastoplastic moduli
will be discussed in section 6.
Expressing the components in the co-rotational frame {êi }, from equations (61), (62),
(70) and (71), in matrix form one has for the material contribution

∫ ∫ [ ]
B N
(δη · dN + δJϑK · dM) dΓc = δJχKT [B T
N B T ]
M Dc dJχK dΓc (72)
Γc Γc B M

where dJχK is defined as in (60). Similarly, the geometric contribution in (70) takes the
form ∫ ∫
T
δ ϑ̃ · (N × dw̄)dΓc = δJχKT B̃ [Ω(N)]{êi } B̄ dJχKdΓc (73)
Γc Γc

where in equation (73) use has been made of the permutation rule of the triple box product,
[Ω(N)]{êi } is the matrix of the co-rotational components of the skew symmetric tensor
having N as axial vector and B̃ and B̄ are such that

δ ϑ̃ = B̃ dJχK ; δ w̄ = B̄ dJχK (74)

i.e.
 
0T 0T [ên · Ω(t+ )] [ên · Ω(t− )]
1 T 
B̃ =  0 0T [êc · Ω(t+ )] [êc · Ω(t− )]  (75)
2
0T 0T [êt · Ω(t+ )] [êt · Ω(t− )] {i }
i

18
 
[ên ]T −[ên ]T 0T 0T
 
B̄ =  [êc ]T −[êc ]T 0T 0T  (76)
T
[êt ] −[êt ]T 0T 0T {i }
i

Summarizing, the linearized contribution of the interface to the internal virtual work is
obtained substituting the geometric contribution (73) and the material contributions (72) in
equation (67)

∫ [ ] [ ]
B N
d(δΠint,c ) = δJχK T
B T
N B T
M Dc dJχKdΓc
Γc B M

T
+ δJχKT B̃ [Ω(N)]{êi } B̄ dJχKdΓc (77)
Γc

5 Finite element discretization


5.1 Geometry discretization
5.1.1 Discretization of the shell

The shell middle surface M is discretized by four node quadrilateral elements according to
the standard isoparametric concept:


4
h
X̄ = N a (r1 , r2 )X̄a (78)
a=1

where the superscript h denotes the discretized fields and X̄a ∈ M are the element nodal
coordinates in the global reference frame {ii }, r1 , r2 are intrinsic coordinates defined in the
reference 2 × 2 finite element and

(r1 , r2 ) ∈ [−1, 1] × [−1, 1]


1
N a (r1 , r2 ) = (1 + r1 ra1 )(1 + r2 ra2 ) (ra1 , ra2 ) = (−1, −1); (1, −1); (1, 1); (−1, 1)
4
a = 1, 2, 3, 4
(79)
are the standard four node quadrilateral interpolation functions.
Since the middle surface displacement and director fields have been used for the de-
scription of the shell kinematics and for the formulation of the weak form of equilibrium,
these are the fields which are directly interpolated over the finite element. This choice has
the advantage that all the adopted nodal variables appear linearly in the expression of the
internal virtual work and of its linearized version. However, the nodal components of the
virtual director field are not used as primary variables in the finite element implementation.
As discussed in Section 3.3, through equation (65) the three nodal components of δTa are

19
expressed in terms of the two nodal virtual rotations (δθ1 )a ,(δθ2 )a which, together with the
nodal displacements δua , represent an array of 5 degrees of freedom for each node. To sim-
plify the notation, the matrix formulation which follows will be derived in terms of the six
component vector δ[χ]Ta = [δ ūTa δtTa ], being intended that the director increments will be
expressed in terms of the primary variables (δθ1 )a ,(δθ2 )a in the final implementation (see
section 7). A discussion of the implications of the different choices for the nodal variables
can be found in Simo et al. [28].
The same four node quadrilateral elements used for the geometry discretization are also
used for the interpolation of the kinematic fields. However, while the standard displace-
ment formulation does not exhibit significant membrane locking, it is known to be severely
affected by shear locking. To avoid this inconsistent response, the Assumed Natural Strain
enrichment proposed by Bathe and Dvorkin [5] has been implemented.
One can set


4 ∑
4
ūh (r1 , r2 ) = N a (r1 , r2 )ūa Th (r1 , r2 ) = N a (r1 , r2 )Ta
a=1 a=1

4
th (r1 , r2 ) = N a (r1 , r2 )ta (80)
a=1

where the nodal director Ta in the reference undeformed configuration is defined as the
average of the nodal normals to the middle surface of the elements converging in the same
node.
The discretization of the variations δ ū, δT and δt and of the spatial derivatives of ū, T,
t, δ ū, δT and of δt are obtained straightforwardly as

20

4 ∑
4
δ ūh (r1 , r2 ) = N a (r1 , r2 )δ ūa δTh (r1 , r2 ) = N a (r1 , r2 )δTa
a=1 a=1

4
δth (r1 , r2 ) = N a (r1 , r2 )δta
a=1

4 ∑4
ū,hα (r1 , r2 ) = N,aα (r1 , r2 )ūa T,hα (r1 , r2 ) = N,aα (r1 , r2 )Ta
a=1 a=1

4
t,hα (r1 , r2 ) = N,aα (r1 , r2 )ta
a=1

4 ∑4
δ ū,hα (r1 , r2 ) = N,aα (r1 , r2 )δ ūa δT,hα (r1 , r2 ) = N,aα (r1 , r2 )δTa
a=1 a=1

4
δt,hα (r1 , r2 ) = N,aα (r1 , r2 )δta
a=1

4
d(δt,hα (r1 , r2 )) = N,aα (r1 , r2 )d(δta ) =
a=1

4
=− N,aα (r1 , r2 )(dta · δta )ta
a=1
(81)

The linearization of the virtual work requires also the discretization of the increments
dū, dT and dt which are interpolated as their corresponding variations δ ū, δT and δt.

5.1.2 Discretization of the cohesive interface

The discretization of the displacement and director jumps, which define the deformation of
the interface element, come directly from the discretization of the corresponding fields of
the adjoint shell elements, specialized onto the “-”and “+” sides of the interface and do not
require an independent discretization, e.g.


2 ∑
2
1 s ∈ [−1, 1]
w̄h = b −
Ñ b (s)ū+ Ñ b (s)ū−
b Ñ b = (1 + s sb ), (82)
b=1 b=1
2 sb = 1; −1

where s is the intrinsic coordinate running on the interface element reference line.

5.2 Discretization of the interface contribution to the virtual internal work


The finite element solution of the equilibrium problem (33) requires the discretization of
the unknown kinematic fields leading to the discrete version of the nonlinear equilibrium

21
equations

δΠhint − δΠhext = 0 (83)

The discretization of the compatibility operators B N and B M in equations (61) and (62)
are obtained straightforwardly as

 
Ñ b [ên ]T −Ñ b [ên ]T Ñ b 12 [ên · Ω(w̄)Ω(t+ )] Ñ b 12 [ên · Ω(w̄)Ω(t− )]
 
BbN =  Ñ b [êc ]T −Ñ b [êc ]T Ñ b 12 [êc · Ω(w̄)Ω(t+ )] Ñ b 12 [êc · Ω(w̄)Ω(t− )] 
Ñ b [êt ]T −Ñ b [êt ]T Ñ b 21 [êt · Ω(w̄)Ω(t+ )] Ñ b 12 [êt · Ω(w̄)Ω(t− )] {i }
i
(84)
 
0T 0T [ên · Ω(t+ )]Ñ b −[ên · Ω(t− )]Ñ b
 
BbM =  0T 0T [êc · Ω(t+ )]Ñ b −[êc · Ω(t− )]Ñ b  (85)
0T 0T 0T 0T {i } i

The contribution of the interface element e to the internal work is then written as
( )e ∫
δΠhint,c = (δη · N + δJϑK · M) dΓc = δJχKTe Feint,c (86)
Γc

with

2 ∫ [ ]
Feint,c = A
b=1 Γe
(BbN )T [N]{êi } + (BbM )T [M]{êi } dΓc (87)
(24×1) c (12×3) (3×1) (12×3) (3×1)

2
where Ab=1 defines the assembly operator.

5.3 Tangent stiffness matrix of the interface element


The expression of the tangent stiffness matrix is obtained from the discretization of the
linearized internal virtual work, equation (66), so that

Ns Nc
d(δΠint ) = δ[χ] Kd[χ] = T
A
e=1
δ[χ]Te Kes d[χ]e + A δJχKTe KecdJχKe
e=1
(88)

where Ns and Nc are the total number of shell and interface elements in the finite element
model, K is the assembled tangent stiffness matrix and Kes and Kec are the tangent stiffness
matrices of the shell and interface elements. The expression of the former is given e.g. in
[8], while the latter is defined below.
The expression of the interface element tangent stiffness matrix Kec in (88) comes from
the discretization of the integrals on the r.h.s. of equation (77)

22
2
( )
Kec = A
a,b=1
kab ab
M c + kG c (89)
(24×24) (12×12) (12×12)

The material contribution kab


M c is given by

 b
∫ [ ]a BN
BTN BTM  (3×12) 
kab
Mc = Dc   dΓc (90)
(12×12) Γc (12×3) (12×3) (6×6) BM
(3×12)

where BaN , BaM have been defined in (84), (85) and Dc in (71).
The geometric contribution kab
G c is given by

kab
Gc = B̃aT [Ω(N)]{êi } B̄b dΓc (91)
(12×12) Γc (12×3) (3×12)

where B̃ and B̄ are defined as


 
0T 0T [ên · Ω(t+ )]Na [ên · Ω(t− )]Na
1 T 
B̃a =  0 0T [êc · Ω(t+ )]Na [êc · Ω(t− )]Na  (92)
2
0T 0T [êt · Ω(t+ )]Na [êt · Ω(t− )]Na {i }
i

 
Ñ b [ên ]T −Ñ b [ên ]T 0T 0T
 
B̄b =  Ñ b [êc ]T −Ñ b [êc ]T 0T 0T  (93)
Ñ b [êt ]T −Ñ b [êt ]T 0T 0T {i }
i

6 Interface elastoplastic behavior


The interface element has the purpose to facilitate the localization of membrane and bending
deformation in the shell along pre-determined lines, coinciding with the edges of shell finite
elements. Even though the most natural application is to problems where the shell exhibits
weakened properties across the interface, as in the case of crease lines, the interface element
can be used also for the limit analysis of plates or shells, using an approach similar to the
yield-line approach of Johansen (see e.g. [18]) and assuming that the pattern of the yield-
lines is known in advance and coinciding with the position of the interface elements. These
type of applications are particularly convenient for validation of the interface kinematics,
since the numerical solution can be easily interpreted and, in some cases, even checked
against analytical or experimental solutions. An application of the interface element in
conjunction with a constitutive model apt to simulate the folding behavior of paperboard
crease lines can be found in [11].
A simplified elastoplastic membrane-bending interaction model is developed in this sec-
tion, intended for application to this type of analyses. In the spirit of Johansen approach,

23
m

1 f1 = 0

−r 1 n

f2 = 0
−1

Figure 7: Interface material elastic domain in m-n plane.

purely elastic shear and torsional behaviors are assumed across the interface, while drilling
rotations are not allowed. On the other hand, since the interface element kinematics allows
for large membrane displacement discontinuities, membrane-bending interaction is taken
into account in the definition of the yield criterion for the cylindrical plastic hinge. The
elastoplastic transition, with stress redistribution across the thickness, from the purely elas-
tic sectional response to the fully plastic one is also neglected and an elastic-perfectly plastic
behavior is assumed, with additive decomposition between elastic and plastic strains.
Let σ0+ and σ0− denote the material yield limits in tension and in compression, respec-
tively. Defining the limit membrane forces N0+ and N0− , their ratio r and the limit bending
moment M0 as

h N0+ N0−
N0+ = σ0+ h, N0− = σ0− h, r = −N0− /N0+ , M0 = − (94)
2 N0+ − N0−
and defining the nondimensional variables

N Mc
n= , m= (95)
N0+ M0
the interface limit plastic domain turns out to be defined in the m−n plane by the conditions
(see Figure 7)

n2 1 − r
f1 (n, m) = − n+m−1≤0 (96)
r r
n2 1 − r
f2 (n, m) = − n−m−1≤0 (97)
r r

The plastic dissipation rate per unit interface length is given by

D = N η̇np + Mc ϑ̇pc (98)

Assuming an associative flow rule, the generalized plastic strain rates η̇np and ϑ̇pc are defined

24
as
( )
∂f1 ∂f2 1 ∂f1 ∂f2
η̇np = λ̇1 + λ̇2 = + λ̇1 + λ̇2 (99)
∂N ∂N N0 ∂n ∂n
( )
∂f1 ∂f2 1 ∂f1 ∂f2
ϑ̇pc = λ̇1 + λ̇2 = λ̇1 + λ̇2 (100)
∂Mc ∂Mc M0 ∂m ∂m

where λ̇1 and λ̇2 are nonnegative plastic multipliers, defined by the Kuhn-Tucker loading-
unloading conditions

f1 ≤ 0, f2 ≤ 0, λ̇1 ≥ 0, λ̇2 ≥ 0, f1 λ̇1 = f2 λ̇2 = 0 (101)

The computation of the cohesive interface internal work (53) and of its linearized coun-
terpart (86) requires the computation of the finite-step stress increments and of the material
consistent tangent matrix at the element Gauss points. According to the standard displace-
ment approach, finite increments of the generalized strains are assumed to be assigned at
the interface element Gauss points. The corresponding generalized stress and plastic strain
increments are computed through a backward-difference integration of the constitutive law,
generalized to the case of multisurface plasticity. The time-integration and the derivation
of the consistent tangent moduli follow a standard approach [26, 29] and are here only
sketchily described.
Denoting by ∆ηn and ∆ϑc the computed interface total strain increments at a Gauss
point at a certain time-step, a purely elastic trial stress is first computed

∆N tr = kn ∆ηn , ∆Mctr = kϑc ∆ϑc (102)

where kn and kϑc are the interface normal and bending elastic stiffnesses. If both f1 (n, m) ≤
0 and f2 (n, m) ≤ 0, the increment is purely elastic and the generalized stresses are updated
using the trial elastic increments in (102). If either one of the conditions is not satisfied, the
plastic correction is activated. Four plastic loading conditions need be distinguished in this
latter case, depending on the plastic mode which is activated in the step:

1. f1 is active, f2 is not active;

2. f2 is active, f1 is not active;

3. the vertex at the intersection between f1 and f2 on the tension side is active;

4. the vertex at the intersection between f1 and f2 on the compression side is active;

The algorithm used for the identification of the active mode is detailed in box 1.
In the most general case (a corner where both f1 and f2 are active), the plastic multipli-
ers are computed by defining the generalized stress increments

25
1: evaluate f1 (ntr , mtr ) and f2 (ntr , mtr )
2: if f1 (ntr , mtr ) ≤ 0 and f2 (ntr , mtr ) ≤ 0 then
3: activation mode: elastic
4: else if mtr > 0 then
5: activation mode: case 1
6: solve for n, m
7: evaluate f2 (n, m)
8: if f2 (n, m) ≤ 0 then
9: the correct yielding mode has been chosen
10: exit
11: else if ntr > 0 then
12: new hypothesis on the yielding mode
13: activation mode: case 3
14: else if ntr < 0 then
15: new hypothesis on the yielding mode
16: activation mode: case 4
17: end if
18: else if mtr < 0 then
19: activation mode: case 2
20: solve for n, m
21: evaluate f1 (n, m)
22: if f1 (n, m) ≤ 0 then
23: the correct yielding mode has been chosen
24: exit
25: else if ntr > 0 then
26: new hypothesis on the yielding mode
27: activation mode: case 3
28: else if ntr < 0 then
29: new hypothesis on the yielding mode
30: activation mode: case 4
31: end if
32: else
33: if ntr > 0 then
34: activation mode: case 3
35: else if ntr < 0 then
36: activation mode: case 4
37: end if
38: end if
Algorithm 1: Algorithm chart for the selection of the active yielding mode

26
( )
kn ∂f1 ∂f2
∆n = ∆n − +2 tr
∆λ1 + ∆λ2
N0 ∂n ∂n
( )
kϑc ∂f1 ∂f2
∆m = ∆m − 2
tr
∆λ1 + ∆λ2 (103)
M0 ∂m ∂m

and substituting in the corresponding plastic consistency conditions f1 = 0 and f2 = 0. The


resulting nonlinear system of equations is then solved for ∆λ1 and ∆λ2 using a Newton-
Raphson iterative scheme. It should be remarked that, as it is standard in finite-step plas-
ticity models, the finite increment material law in (103) is holonomic, in the sense that the
non-reversibility constraint ∆λi ≥ 0 is enforced only on finite increments of plastic mul-
tipliers, and not on their rates λ̇i . This implies that at each Gauss point the increment is
either elastic or plastic. No unloading can take place at an intermediate instant within the
time step.
The computation of the linearized internal work in (77) requires the definition of the
algorithmic consistent material tangent matrix Dc . The tangent moduli kc , kt and kn in
Dc are constant, since pure elastic behavior has been assumed for torsion and shear. For
the computation of the remaining tangent moduli, as usual it is assumed that the plastic
modes activated in the corrector phase will remain active also in the tangent projection.
This implies that, if mode 3 or mode 4 (i.e. corner modes) where found active at the end
of the corrector phase, the stress point has to remain in the same corner after the tangent
projection and the corresponding stiffnesses vanish. Hence

∂∆N ∂∆Mc ∂∆N ∂∆Mc


knep = = 0, kϑepc = = 0, ep
knϑ = = kϑepc n = = 0 (104)
∂∆ηn ∂∆ϑc c ∂∆ϑc ∂∆ηn

In the other cases, when mode i (i = 1 or 2) were found active, the tangent moduli are such
that
[ ]
ep 1 ∂ 2 fi 1 ∂fi
δ(∆N ) = knep δ(∆ηn )
+ knϑ = kn δ(∆ηn ) − +2
δ(∆ϑc ) δ(∆N )∆λi − + δ(∆λi )
c
N0 ∂n2 N0 ∂n
[ ]
ep ep 1 ∂fi
δ(∆Mc ) = kϑc n δ(∆ηn ) + kϑc δ(∆ϑc ) = kϑc δ(∆ϑc ) − δ(∆λi ) (105)
M0 ∂m

where δ(∆·) denotes a perturbation of the incremental quantity computed through the back-
ward difference integration. The perturbed plastic multipliers are obtained from the corre-
sponding perturbed consistency condition

[ ] [ ]
∂fi kn 1 ∂fi ∂fi kϑc 1 ∂fi
δfi = δ(∆ηn ) − + δ(∆λi ) + δ(∆ϑc ) − δ(∆λi ) = 0
∂n N0+ N0 ∂n ∂m M0 M0 ∂m
(106)
The tangent moduli are then obtained by substituting back in (105) the value of δ∆λi com-

27
puted from (106).

7 Finite-step incremental procedure


The discrete form (83) of the weak statement of equilibrium consists of a nonlinear system
of equations. Its solution requires: a subdivision of the analysis total duration in finite time-
steps, where finite increments are assigned to the external loads; an iterative scheme for
the step solution of the nonlinear system of equations under the assigned load increment;
a scheme for the geometry and material internal variables updating after convergence has
been reached in the current time-step.
In view of an effective parallelization of the iterative procedure, and to avoid con-
vergence difficulties in problems with contact, the Adaptive Dynamic Relaxation method
(ADR) in the form proposed by Oakley and Knight [23, 24] has been adopted. The ADR
method is briefly summarized with emphasis on its specialization to the present shell model
with interfaces. In what follows, the attention will be focussed on an individual time-step
∆t = tn+1 − tn , with the assumption that the equilibrium state of the shell structure at time
tn is completely known.
With dynamic relaxation, the static solution of the finite-step problem at time tn+1 is
obtained as the final steady-state response to a pseudo-transient dynamic analysis of the
system. The term “pseudo” alludes to the fact that fictitious mass and damping matrices are
used to accelerate the achievement of the steady-state response, implicitly defining in this
way a “pseudo-time ” τ for the pseudo-dynamic analysis, which is unrelated to the physical
time-interval ∆t.
The pseudo-motion of the dynamic system is described by the semi-discretized finite
element equation of motion (the symbol M used here to denote the mass matrix should not
be confused with the vector M of internal couples defined in equation (48)2 ):

Mk d̈k + Ck ḋk + Fint (dk ) = Fn+1


ext (107)

where the superscript k denotes quantities evaluated at time τ k , M is the mass matrix, C
the damping matrix, Fint the vector of nodal equivalent internal forces, Fn+1
ext the vector of
nodal equivalent external forces prescribed at the end of the physical time tn+1 , d the vector
of nodal unknowns (three displacements and two rotations per node in this case, where the
rotations are expressed in the reference {ei }k defined at τ k ). ḋ = dd/dτ and d̈ = d2 d/dτ 2
are the pseudo velocities and accelerations of the nodal variables.
Time integration at time τ k+1 = τ k + ∆τ is carried out by means of a half-station
central difference method, where

1 1 1 1 1
ḋk+ 2 = (dk+1 − dk ); d̈k = (ḋk+ 2 − ḋk− 2 ) (108)
∆τ ∆τ

28
1 1
Note that in using ḋk− 2 in (108), the rotation d.o.f.s computed at τ k− 2 need be expressed
in the current frame {ei }k .
The velocity at time τ k , needed in the damping term in equation (107), is obtained as

1 1 1
ḋk = (ḋk+ 2 + ḋk− 2 ) (109)
2
In equation (107), M is assumed to be diagonal and a mass proportional damping C = cM
is assumed, c being the damping coefficient. Substituting (108)b and (109) in (107), the
1
velocity vector at τ k+ 2 is immediately obtained as
( ) ( )
2 − c∆τ 2∆τ
ḋ k+ 12
= k− 12
ḋ + M−1 (Fn+1
ext − Fint )
k
(110)
2 + c∆τ 2 + c∆τ
where M−1 is computed straightforwardly since M is diagonal. This also allows to compute
explicitly the increment of the solution vector at τ k+1 as

1
∆d = dk+1 − dk = ∆τ ḋk+ 2 (111)

Since the integration method is only conditionally stable, the time-step size cannot be
arbitrarily chosen. However, in order to reach in the shortest possible time the steady-state,
∆τ is assigned a priori and kept fixed throughout the motion (it is usually taken equal to
one) while, to fulfill the stability requirement in the ADR, the diagonal entries Mik of Mk
are computed from an estimate of the maximum eigenvalue of the system [23]

∆τ 2 ∑ k
4
Mik = |K̄ij | (112)
4
j=1

where K̄k is the stiffness tangent matrix obtained from (88) substituting the director in-
crements with their expression (65) in terms of the nodal rotation increments dθk . The
damping coefficient c is chosen so as to approximate the critical damping

c = 2ω0 (113)

where ω0 is the minimum structural eigenfrequency which is estimated at each time-step as


illustrated in [23].
The time-marching procedure for the shell and interface elements described in the pre-
vious sections can be summarized as follows

• At the beginning τ k of a new pseudo-time-increment, a new orthogonal frame {ei }k ,


aligned with the current position of the director (Ta )k at node a, is defined and the
tangent matrix is assembled. The current director (Ta )k is taken as coincident with
the rotated director (ta )k{ei }k−1 at the end of the previous time-step, but its compo-
nents are expressed in the updated local frame {ei }k . The rotations at each node are
set to zero.

29
• The increments of all nodal variables are computed from (111).

• The position of the director at each node is updated according to (64)

[Ta ]k+1 k a a k
{ii } = Q [R(∆θ )]{ei } [T ]{ei } (114)

Note that, apart from the initial configuration, [Ta ]0 , [Ta ]k is in general not normal
to the middle surface in the configuration at τ k .

• In the shell elements, Green-Lagrange strains Eαβ with respect to the initial configu-
ration X0 are computed at each integration point according to (22), where T = T0 .
Increments of displacements and rotations jumps along the interface element e are
computed in an incremental way, to account for the moving co-rotational reference
frame, using the linearized compatibility matrices BN and BM defined in (84) and
(85), which directly provide co-rotational components of the interface strains:
 
ηnk+1
[ k+1 ]  k+1  [ k] ∑
η k =  ηc
{êi }e  = η {ê }k + 2a=1 (BaN )ke ∆ [[χ]]a
i e
ηtk+1 k
 k {êi }e (115)
[ ] ϑn [ ]
k+1  k  k ∑
[[ϑ]] k
=  ϑ c  = [[ϑ]] k
+ 2a=1 (BaM )ke ∆ [[χ]]a
{êi }e {êi }e
0 {ê }k
i e

The finite increments ∆ηn and ∆ϑc to be used in (102) are computed from (115) as

∆ηn (sg ) = ηnk+1 (sg ) − ηn0 (sg )


(116)
∆ϑc (sg ) = ϑk+1
c (sg ) − ϑ0c (sg )

where sg is the intrinsic coordinate of the interface Gauss point and ηn0 and ϑ0c are the
converged generalized strains at the end of the previous step, i.e. at the physical time
tn .

• Stresses at the integration points are computed at time τ k+1 through implicit inte-
gration of the elastoplastic constitutive law (for the interface elements, see section 6)
and using the matrix of elastic moduli of the shell element [9]. The definition of the
finite strain increments in (116) leads to a material model which is reversible in the
step, as discussed in section 6. In the case of non-reversible material behavior, small
time-steps ∆t are required to avoid accuracy loss.

• The interface equivalent internal nodal forces Fk+1


int,c are computed according to equa-
tion (87).

• Convergence is checked. According to the ADR, convergence is reached when the

30
static response is achieved. This is identified with the condition of vanishing inertia
forces, i.e. convergence is reached when

∥Fk+1
int − Fext ∥
k+1
≤ ϵtol (117)
∥Fk+1
int ∥

ϵtol being a preassigned tolerance.

• If the check (117) is satisfied, a new load increment is assigned and the ADR proce-
dure is started again. If (117) is not satisfied, the time index k is incremented and the
integration of a new step of the pseudo-motion is carried out.

8 Numerical Examples
8.1 Cantilever plate
The analysis of a cantilever plate subject to an imposed transversal displacement at its free
end has been considered by Armero and Ehrlich [4]. A rectangular plate of length L =
3 mm, width b = 1 mm and thickness t = 0.1 mm is considered. It is clamped at one of its
short edges, while a transversal displacement v is imposed to the opposite side, as shown
in Figure 8. The corresponding reaction force R is recorded. Since a straight hinge line is
expected to develop at the clamped edge, it is possible to embed interface elements along
this edge of the mesh since the beginning of the analysis and observe the development
of the hinge under the applied displacement v. The material model used in [4] for this

Figure 8: Cantilever plate. Test case geometry: clamped at one side, under imposed
transversal displacement at the other.

test case is characterized by a linear elastic-plastic, isotropic response with the following
material properties: Young’s modulus E = 3.6 ∗ 104 N/mm2 , Poisson’s coefficient ν = 0
(a transverse shear coefficient s = 5/6 is used, according to the proposed shell model) and
yield stress σ0 = 280 N/mm2 . Furthermore, in [4] a linear softening behavior was assumed
beyond the yield limit, while, in this paper, a perfectly plastic hinge is used.

31
Figure 9: Cantilever plate unstructured mesh: clamped edge (left); moving edge (right).

At the clamped edge, transverse displacements and director vectors are constrained only
at the external side of the interface element, letting both displacement jumps and angular
discontinuities develop. Under small strains assumption, the plastic hinge forms when the
limit moment Mc = M0 = 1/4 σ0 h2 = 0.7 N is reached for the first time. An unstruc-
tured mesh of 602 quadrilateral shell elements is used, as shown in Figure 9. This mesh is
comparable to the unstructured mesh of 528 quadrilateral elements used in [4].
The parameters for the interface material model are reported in Table 1. The simulations
of a rigid plastic behavior would lead to the condition kθc → ∞; the stiffness parameters
have been defined so as to approximate this condition avoiding numerical problems. The
yield limits are computed as the resultants of the normal stresses under axial full yielding
condition N0 = σ0 ∗ h.

Table 1: Summary of the parameters adopted in the cantilever plate case.


σ0− [N/mm2 ] σ0+ [N/mm2 ] kθc [N/deg] kθn [N/deg]
−280.0 280.0 2.6 1.0
h [mm] kn [N/mm2 ] kt [N/mm2 ] kc [N/mm2 ]
0.1 3.6 ∗ 104 1.8 ∗ 103 1.8 ∗ 103

A comparison between the numerical results obtained with the proposed interface model
and the softening hinge model presented in [4] is shown in Figure 10a, where the reaction
force R is plotted against the imposed displacement v, while the final displaced configura-
tion is shown in Figure 11b. The present interface element solution captures well the peak
in the reaction force but, for increasing imposed displacements, deviates from the reference
softening solution in [4], which was obtained under the small displacements assumption. In
the present analysis, the geometric effect produces a progressive reduction of the lever of
the reaction force. This requires a progressive increase of the reaction force to maintain the
limit value of the bending moment in the plate near the clamp, which explains the positive
slope after the peak of the R − v dashed curve in Figure 10a. When the imposed vertical
displacement becomes of the same order of magnitude of the plate length, towards the end

32
Figure 10: Cantilever plate: a) reaction force vs. transversal imposed displacement; b) load
path in Mc − N plane at interface Gauss point.

a) b) interface reference surface


original plate undeformed configuration

3 mm

3 mm

final plate displaced configuration

Figure 11: Cantilever plate. Lateral view of: a) initial configuration; b) final displaced
configuration.

33
of the plot, the plate response switches from prevailing bending to prevailing membrane
response, with a much higher stiffness, which accounts for the sudden increase of the ver-
tical reaction. The bending-membrane transition is evidenced also by the plot in Figure
10b, where the load path inside the limit domain at the interface gauss point is shown. As
the unconstrained directors at the plate side of the interface are free to rotate, the interface
reference surface Γ̃c × [−h/2, h/2] also rotates (Figure 11b). Since the generalized inter-
nal forces N and M are referred to the co-rotational frame {êi } defined on the reference
interface surface, the vertical reaction force R turns out to have an increasing component in
the direction ên , normal to the interface surface, giving rise to growing membrane internal
force.
The membrane N and shear Vt internal forces vs. imposed transverse displacement plot
is shown in Figure 12a. The cantilever rotates under prevailing bending conditions until the
imposed displacement approaches the plate length of 3 mm (Figure 11b). At this point, N
and Vt grow abruptly as the membrane and shear elastic stiffnesses are mobilized. The plot
of the total and plastic rotation jumps in Figure 12 shows that the plastic hinge undergoes
a rotation of about 80◦ , which implies a rotation of about 40◦ of the interface reference
surface Γ̃c .

8.2 Circular plate


This example is intended to assess the capabilities of the interface model in the case of
curved hinge lines. The classical case of a circular plate under uniformly applied load is
considered. Two boundary conditions for the plate are treated: the simply supported case
and built-in case, respectively b and c in Figure 13. In both cases, an unstructured mesh
consisting of 1873 quadrilateral elements is employed, and a line of interface elements is
placed around the plate boundary. The interface elements are fully constrained on the outer
side. The simply supported case is then obtained adopting a very small bending stiffness. In
this way, the simply supported case allows to verify that the interface kinematics does not
exhibit spurious internal rotational constraints.
The plate is characterized by the following dimensions: R = 50 mm, h = 2 mm. The
isotropic, linear elastic, perfectly plastic plate is characterized by the following material
properties: Young’s modulus E = 106 N/mm2 , Poisson’s coefficient ν = 0.5 and yield
stress σ0 = 250 N/mm2 . An extremely stiff material has been used to remain within the hy-
pothesis of small displacements, so that membrane effects are negligible and a meaningful
comparison with the analytical solutions is possible.
The analytical solution reported in [33] in terms of pressure p vs. deflection v, up to the
yield limit Mc = M0 , occurring at the center of the plate, gives the results:

33 Eh3 32 M0
p= v, p0 = (118)
64 R4 7 R2

34
Table 2: Summary of the parameters adopted in the supported plate case.
σ0− [N/mm2 ] σ0+ [N/mm2 ] kθc [N/deg] kθn [N/deg]
−250.0 250.0 10−3 1.0
h [mm] kn [N/mm2 ] kt [N/mm2 ] kc [N/mm2 ]
2.0 106 3.3 ∗ 105 3.3 ∗ 105

Table 3: Summary of the parameters adopted in the clamped plate case.


σ0− [N/mm2 ] σ0+ [N/mm2 ] kθc [N/deg] kθn [N/deg]
−250.0 250.0 105 1.0
h [mm] kn [N/mm2 ] kt [N/mm2 ] kc [N/mm2 ]
2.0 106 3.3 ∗ 105 3.3 ∗ 105

The interface parameters for the hinge line at the transversally constrained edge are reported
in Table 2. The comparison with the analytical solution is shown in Figure 14.
The solution for the clamped case, including the study of the response of the plate be-
yond the elastic range, is reported in [31]. The solution in terms of pressure p vs. deflection
v of the plate center point is characterized by two significant load values: the first yielding,
which occurs at the clamped edge when the yield limit Mc = M0 is reached; the second
yielding, which occurs at the center of the plate under the same condition. In the elastic
range the p − v curve is governed by the equation:

64 Eh3
p= v (119)
9 R4

while the two yielding loads are defined as:

M0 64 M0
p1 = 8 , p2 = (120)
R2 7 R2

For higher pressure values, plasticity starts spreading towards the plate central zone and the
response cannot be reproduced anymore by a plastic hinge model. In the clamped case, the
circular edge is characterized by infinite elastic rotational stiffness around the hinge axis,
which is again approximated using a high value of kθc . The interface parameters for the
hinge line at the constrained edge are reported in Table 3.
The comparison between the analytical and the numerical solutions is shown in Figure
14. Good agreement can be observed in both cases. The interface element is able to capture
well the elastic behavior by a proper calibration of the rotational stiffness around the circular
hinge axis. In the clamped case, both the first yielding and the post-yielding behavior up to
the second yield point are well captured. Since the increase of the central deflection, after
first yielding, occurs under plastic flow at the plastic hinge lines at the edge, the same slope

35
as the supported case is obtained. These examples confirm that the element kinematics
allows rotations around curved lines without introducing spurious locking effects.

8.3 Doubly-curved shell forming


The last example deals with the simulation of the stamping of a rectangular plate between
a die and a matching punch to obtain a doubly-curved shell (Figure 15). Yu et al. [37]
presented the results of experimental tests on shell forming, together with the results of a
rigid-plastic analytical model with plastic hinges, used to investigate the deformation modes
involved in the stamping process.
During the forming process, the reacting force P under the die is recorded and plotted
against the displacement v of the punch. The process is simulated introducing interface
elements in the positions defined by [37] in their rigid plastic approximations.
Consider a rectangular plane plate of dimensions L1 = 300 mm, L2 = 250 mm and
thickness t = 1.6 mm. The case of curvature ratio ω = 2/3, for the punch and the die,
is considered; the geometry of the die is characterized by two radii: R1d = 360 mm and
R2d = 240 mm. The radii of curvature of the punch are related to those of the die by the
thickness of the plate, that is Rip = Rid − h, where (i = 1, 2) (see Figure 15).
The plate is made of mild steel, which is modeled as an isotropic linear elastic-perfectly
plastic material. It is characterized by the following properties: Young’s modulus E =
2.1 ∗ 105 N/mm2 , Poisson’s coefficient ν = 0.3 and yield stress σ0 = 275 N/mm2 .
In Figure 16, the experimental die-reaction vs. punch-travel curve obtained by Yu et
al. [37] is reported. From the experimental curve (solid line) it is possible to discern the
different dominant deformation modes developing during the process (Figure 17). In the
first stage, the punch contacts the center of the plate, which is supported at its four corners,
pushing it until the yield limit is reached at the middle of the longer span. The ensuing defor-
mation is well described by a model with a cylindrical plastic hinge with its axis parallel to
the shortest side L2 , as sketched in Figure 17a. Under increasing punch-pressure, the mid-
point at the center of the long side contacts the die, causing a rapid increase in the reaction
force (sharp hardening branch in Figure 16), since the plate is now supported on six points
rather than on four. After primary bottoming, depending on the side length ratio L1 /L2 ,
the plate deformation can be modeled either as in (b) or in (c). In the multiple cylindrical
model (b), more hinges are created parallel to the central one, while in the four-segment
model (c) a cruciform localized hinge develops, involving membrane-bending coupling.
When a deformation as in mode (c) is taking place (which is the case for the L1 /L2 ratio
here considered), a six-segment deformation mode of the type shown in Figure 17d can be
triggered for increasing load, characterized by 8 support points (4 at the corners and 2 for
each plate long side) and by the presence of flattened regions at midspan of the long sides.
As the load increases, these regions are easily subject to buckling phenomena, with large
plastic flow, resulting in the softening branch visible in Figure 16. This type of deforma-

36
tion develops until secondary bottoming occurs along the short sides and the reaction force
increases again, since the plate becomes supported on ten points.
Two numerical simulations have been performed, with different patterns of embedded
interface elements. In the first test, an interface pattern in agreement with the four-segment
model (Figure 17c) is used, and the rectangular plate is modeled by means of a structured
mesh. In the second test, the six-segment model (Figure 17d) is employed, with the flattened
region length of 60 mm, as experimentally measured by [37], the plate being discretized
with an unstructured mesh of quadrilateral elements. In both cases, the length of the average
element side is approximatively le = 3 mm, for a total of 8296 elements in the first mesh
and 9656 in the second one, as shown in Figure 18. A regular mesh is used also for both
dies, which are treated as rigid bodies. The fineness of the plate mesh is dictated by the
need to have an accurate resolution of the contact regions.
The load is imposed controlling the vertical displacement of the punch; master-slave
surface-to-surface interactions with penalty approach without friction are employed to en-
force contact conditions between die and plate, and between punch and plate.
Also in this case, the interface parameters, reported in Table 4, are defined in order
to numerically approximate the rigid plastic behavior. A linear elastic behavior has been
assumed for the shell elements.

Table 4: Summary of the parameters adopted in the metal forming case.


σ0− [N/mm2 ] σ0+ [N/mm2 ] kθc [N/deg] kθn [N/deg]
−275.0 275.0 782.0 100.0
h [mm] kn [N/mm2 ] kt [N/mm2 ] kc [N/mm2 ]
1.6 1.05 ∗ 105 4.2 ∗ 104 4.2 ∗ 104

The numerical results are plotted in Figure 16 (dashed curves), where they are compared
to the experimental results reported in [37].
It can be observed that the bottoming points and the global behavior are well captured.
The slopes of the numerical and experimental curves are already different in the elastic
range. This difference is due to the lack of friction in the numerical model, which makes the
overall response more compliant, since no lubrication between plate and dies was applied in
the experiments. The two numerical models are in perfect agreement up to the increase in
reaction after primary bottoming, where the transition to the six-segment deformation mode
is activated; the six-segment interface pattern clearly better captures the experimental trend
in the peak-and-softening branch up to secondary bottoming, after which the two numerical
responses overlap again.
In Figure 19, three points are highlighted on the experimental curve, marking three
important stages in the plate forming process: primary bottoming, early softening and sec-
ondary bottoming. In Figure 20, the deformed configurations of the plate at these three

37
stages of the six-segment analysis are compared with the experimentally observed defor-
mation states at the same values of imposed punch displacement. In Figure 20, the pictures
labeled as A show views from the direction normal to the long side (L1 ), while those labeled
as B are views from the direction normal to the short side (L2 ). Good qualitative agreement
can be observed in all the three stages, particularly with reference to: the almost-cylindrical
deformation mode up to primary bottoming (state 4); the development of lateral flattened
regions at the center of the long sides, promoted by local buckling (state 5); the development
of a sharp angle of deflection along the short sides at secondary bottoming (state 6).
In Figure 21, the evolution of the numerical contact imprints is shown, at the starting
elastic stage and at the three notable points previously defined. In the early elastic stage,
the punch contacts the plate at its center, which is supported at four corners (state elastic in
Figure 21). Then the contact region extends along the longer center-line of the plate up to
the splitting into two different zones (state 4). The supporting points, after the peak in the
punch-force response, spread along the long sides, at the borders of the flattened regions
(state 5). The typical imprints developing along the bisectors of the four inner angles can
be recognized under higher loads (state 6), after secondary bottoming has occurred. The
development of the contact regions between plate and dies is also reported in [37], as shown
in Figure 22, but only for a load value of 13.4 kN, which is outside the range of values
considered in Figure 16.

9 Conclusions
Folding of thin shells is a technological problem of interest in many engineering applica-
tions. The simulation of the folding process may be extremely useful for the correct design
of the folding procedure. In view of the highly localized deformations, standard shell finite
element models would require extremely fine meshes for a proper resolution of the folding
region.
In this paper we present a two–noded, cohesive, interface finite element, to be placed
between adjacent four–noded Mindlin-Reissner finite elements, for the simulation of fold-
ing processes. The element is formulated in terms of work conjugate static and kinematic
generalized variables. Since the element is designed to allow for large displacement and
rotation jumps across the interface, particular attention has been devoted to the derivation
of the interface pure (rigid rotation free) straining modes. A complete derivation of the
element matrices, including the tangent stiffness matrix, has been provided.
The element can be used in conjunction with any interface material model. For the
numerical examples, an elastic perfectly plastic behavior has been assumed. The element
has been tested in several problems for which either analytical or experimental results were
available, exhibiting good performances. Applications of the element to other folding prob-
lems requiring more sophisticated material models, such as those with pre-existing crease
lines, are currently in progress.

38
Acknowledgments
The financial support by Tetra Pak Packaging Solutions is gratefully acknowledged. The
second Author also acknowledge the support by MIUR-PRIN 2009AJ3PC9 004.

References
[1] W. Abramowicz. Thin-walled structures as impact energy absorbers. Thin-Walled
Structures, 41(2-3):91–107, February 2003.

[2] P. M. A. Areias and T. Belytschko. Non-linear analysis of shells with arbitrary evolv-
ing cracks using xfem. International Journal for Numerical Methods in Engineering,
62(3):384–415, 2005.

[3] P. M. A. Areias, J. H. Song, and T. Belytschko. Analysis of fracture in thin shells by


overlapping paired elements. Computer Methods in Applied Mechanics and Engineer-
ing, 195(41-43):5343–5360, 08/15 2006.

[4] F. Armero and D. Ehrlich. Finite element methods for the multi-scale modeling of
softening hinge lines in plates at failure. Computer Methods in Applied Mechanics
and Engineering, 195(13-16):1283–1324, February 2006.

[5] K. J. Bathe and E. N. Dvorkin. A four-node plate bending element based on


mindlin/reissner plate theory and a mixed interpolation. International Journal for
Numerical Methods in Engineering, 21(2):367–383, 1985.

[6] L. A. A. Beex and R. H. J. Peerlings. An experimental and computational study of


laminated paperboard creasing and folding. International Journal of Solids and Struc-
tures, 46(24):4192–4207, August 2009.

[7] T. Belytschko, J. I. Lin, and C. S. Tsay. Explicit algorithms for the nonlinear dynamics
of shells. Computer Methods in Applied Mechanics and Engineering, 42(2):225–251,
February 1984.

[8] B. Brank, F. B. Damjanić, and D. Perić. On implementation of a nonlinear four node


shell finite element for thin multilayered elastic shells. Computational Mechanics,
16(5):341–359, 1995.

[9] B. Brank, D. Peric, and F. B. Damjanić. On large deformations of thin elasto-plastic


shells: Implementation of a finite rotation model for quadrilateral shell element. In-
ternational Journal for Numerical Methods in Engineering, 40(4):689–726, 1997.

[10] F. Cirak, M. Ortiz, and A. Pandolfi. A cohesive approach to thin-shell fracture and
fragmentation. Computer Methods in Applied Mechanics and Engineering, 194(21-
24):2604–2618, June 2005.

[11] A. Giampieri, U. Perego, and R. Borsari. A constitutive model for the mechanical
response of the folding of creased paperboard. International Journal of Solids and
Structures, accepted for publication, 2011.

39
[12] J. Goncalves. Application of the line spring model to some complex geometries, and
comparison with three-dimensional results. International Journal of Pressure Vessels
and Piping, 76(8):551–560, August 1999.

[13] P. G. Hodge. Limit analysis of rotationally symmetric plates and shells. Prentice-Hall,
1963.

[14] P. Isaksson and R. Hagglund. A mechanical model of damage and delamination in cor-
rugated board during folding. Engineering Fracture Mechanics, 72(15):2299–2315,
October 2005.

[15] H. Lee and D. Parks. Line-spring finite element for fully plastic crack growth–i. for-
mulation and one-dimensional results. International Journal of Solids and Structures,
35(36):5115–5138, December 1998.

[16] H. Lee and D. Parks. Line-spring finite element for fully plastic crack growth-ii.
surface-cracked plates and pipes. International Journal of Solids and Structures,
35(36):5139–5158, December 1998.

[17] W. Li and T. Siegmund. An analysis of crack growth in thin-sheet metal via a cohesive
zone model. Engineering Fracture Mechanics, 69(18):2073–2093, December 2002.

[18] J. Lubliner. Plasticity Theory. Macmillan Publishing Company, 1990.

[19] J. Marsolek and H. Reimerdes. Energy absorption of metallic cylindrical shells with
induced non-axisymmetric folding patterns. International Journal of Impact Engi-
neering, 30(8-9):1209–1223, October 2004.

[20] C. E. Massonet and M. A. Save. Plastic analysis and design of plates, shells and disks.
North-Holland, 1972.

[21] D. Mohr and M. Doyoyo. Deformation-induced folding systems in thin-walled mono-


lithic hexagonal metallic honeycomb. International Journal of Solids and Structures,
41(11-12):3353–3377, June 2004.

[22] M. Nygårds, N. Hallbäck, M. Just, and J. Tryding. A finite element model for sim-
ulations of creasing and folding of paperboard. In 2005 Abaqus User’s Conference,
pages 373–387, May 2005.

[23] D. Oakley and N. F. Knight. Adaptive dynamic relaxation algorithm for non-linear
hyperelastic structures part i. formulation. Computer Methods in Applied Mechanics
and Engineering, 126(1-2):67–89, September 1995.

[24] D. Oakley and N. F. Knight. Adaptive dynamic relaxation algorithm for non-linear
hyperelastic structures part ii. single-processor implementation. Computer Methods
in Applied Mechanics and Engineering, 126(1-2):91–109, September 1995.

[25] A. Papa and S. Pellegrino. Systematically creased thin-film membrane structures.


Journal of Spacecraft and Rockets, 45(1):10–18, 2008.

[26] U. Perego. Explicit backward difference operators and consistent predictors for lin-
ear hardening elastic-plastic constitutive laws. Solid Mechanics Archives, 13:65–102,
1988.

40
[27] J. R. Rice and N. Levy. The part-through surface crack in an elastic plate. Journal of
Applied Mechanics, 39(1):185–194, 1972.

[28] J. C. Simo and D. D. Fox. On a stress resultant geometrically exact shell model. part i:
Formulation and optimal parametrization. Computer Methods in Applied Mechanics
and Engineering, 72(3):267–304, March 1989.

[29] J. C. Simo, J. G. Kennedy, and S. Govindjee. Non-smooth multisurface plasticity and


viscoplasticity. loading/unloading conditions and numerical algorithms. International
Journal for Numerical Methods in Engineering, 26:2161–2185, 1988.

[30] J.H. Song and T. Belytschko. Dynamic fracture of shells subjected to impulsive loads.
Journal of Applied Mechanics, 76(5):051301 1–051301 9, 2009.

[31] B. Tekinalp. Elastic-plastic bending of a built-in circular plate under a uniformly


distributed load. Journal of the Mechanics and Physics of Solids, 5(2):135 – 142,
1957.

[32] B. K. Thakkar, L. G. J. Gooren, R. H. J. Peerlings, and M. G. D. Geers. Experimen-


tal and numerical investigation of creasing in corrugated paperboard. Philosophical
Magazine, 88(28):3299–3310, 2008.

[33] S. P. Timoshenko and S. Woinowsky-Krieger. Theory of Plates and Shells. McGraw-


Hill, 1959.

[34] C. Wang, X. Du, and Z. Wan. Numerical simulation of wrinkles in space inflatable
membrane structures. Journal of Spacecraft and Rockets, 43(5):1147–1149, Septem-
ber 2006.

[35] C. G. Wang, X. W. Du, H. F. Tan, and X. D. He. A new computational method for
wrinkling analysis of gossamer space structures. International Journal of Solids and
Structures, 46(6):1516–1526, March 2009.

[36] K. Woo, K. Nandurkar, and C. H. Jenkins. Effective modulus of creased thin mem-
branes. Journal of Spacecraft and Rockets, 45(1):19–26, 2008.

[37] T. X. Yu, W. Johnson, and W. J. Stronge. Stamping rectangular plates into doubly-
curved dies. Proceedings of the Institution of Mechanical Engineers - Part C: Journal
of Mechanical Engineering Science, 198C(8):109–125, 1984.

[38] P. D. Zavattieri. Modeling of crack propagation in thin-walled structures using a co-


hesive model for shell elements. Journal of Applied Mechanics, 73(6):948–958, 2006.

41
Figure 12: Cantilever plate, evolution with imposed transverse displacement: a) interface
shear and membrane forces; b) plastic and total interface rotation jumps.

Figure 13: Circular plate: a) unstructured mesh, b) supported case, final deformed con-
figuration (100 times amplified), c) clamped case, final deformed configuration (100 times
amplified).

42
10
9.14

8
pR /M 0
2

4.57

Clamped (sol)
Supported (sol)
Clamped (num)
Supported (num)
0
0 1.13 1.75 2.36 3
3 2
Eh v/M0R

Figure 14: Circular plate: comparison between numerical (num) and analytical (sol) solu-
tions.

Figure 15: Doubly-curved shell: geometry of the punch and the die.

43
3.5
experimental
numerical 4sm
3 numerical 6sm

2.5

2
P [kN]

1.5

0.5

0
0 5 10 15 20 25 30 35 40 45
v [mm]

Figure 16: Shell forming: force-deflection curves; comparison between experimental (solid)
and numerical (dashed) solutions.

Figure 17: Deflection modes: a) single cylindrical bending, b) multiple cylindrical bending,
c) four-segment deformation, d) six-segment deformation.

44
Figure 18: Plate mesh: four-segment model (left), six-segment model (right).

Figure 19: Notable reference points in the experimental curve.

Figure 20: Deformed configurations of the plate: experimental (left), numerical (right).
Label A stands for views from side L1 , label B stands for views from side L2 . Images are
taken from [37].

45
Figure 21: Contact region patterns at different stages of analysis (upper and lower plate
sides are superposed).

Figure 22: Contact regions: experimental upper side (left) and experimental lower side
(right). Images taken from [37].

46

You might also like