A modeling and resolution framework for wrinkling in hyperelastic sheets at finite membrane strain

Journal of the Mechanics and Physics of Solids 124 (2019) 446–470
Contents lists available at ScienceDirect
Journal of the Mechanics and Physics of Solids

journal homepage: www.elsevier.com/locate/jmps
A modeling and resolution framework for wrinkling in

hyperelastic sheets at finite membrane strain
C. Fu a, T. Wang a, F. Xu a,∗, Y. Huo a,∗, M. Potier-Ferry b
a
Institute of Mechanics and Computational Engineering, Department of Aeronautics and Astronautics, Fudan University, 220 Handan
Road, Shanghai 200433, P.R. China
b
Université de Lorraine, CNRS, Arts et Métiers ParisTech, LEM3, Metz F-57000, France
a r t i c l e i n f o a b s t r a c t
Article history: Wrinkles commonly occur in uniaxially stretched rectangular hyperelastic membranes with
Received 14 August 2018 clamped-clamped boundaries, and can vanish upon excess stretching. Here we develop a
Revised 5 October 2018
modeling and resolution framework to solve this complex instability problem with highly
Accepted 9 November 2018
geometric and material nonlinearities. We extend the nonlinear Föppl-von Kármán thin
Available online 10 November 2018
plate model to finite membrane strain regime for various compressible and incompress-
Keywords: ible hyperelastic materials. Under plane stress condition, 2D hyperelastic constitutive mod-
Hyperelastic membranes els can be systematically deduced based on general 3D strain energy potentials, e.g.,
Strain-stiffening materials Saint-Venant Kirchhoff, neo-Hookean, Mooney-Rivlin, Gent model and Gent-Gent model.
Instability Moreover, we establish a novel and efficient numerical resolution framework combining
Spectral method a path-following continuation technique by Asymptotic Numerical Method (ANM) and a
Asymptotic Numerical Method discretization by a spectral method. The main advantages of this framework include the
generality for both compressible and incompressible materials, ease of programming, high
precision and efficient continuation predictor. Based on the proposed approach, effect of
different incompressible constitutive models on the post-buckling response is investigated,
which shows that restabilization points and wrinkling amplitudes are quantitatively influ-
enced. However, for compressible materials, Poisson’s ratio plays a critical role in the wrin-
kling and restabilization behavior. We find that smaller Poisson’s ratio makes later onset of
wrinkling, lower amplitude and earlier disappearance of wrinkles. Besides, severe strain-
stiffening phenomena are explored by accounting for phenomenological models such as
Gent model and Gent-Gent model. Efficiency and accuracy of the proposed modeling and
resolution framework were examined by comparing with some benchmarks.
© 2018 Elsevier Ltd. All rights reserved.
1. Introduction
Thin-walled membrane structures are not only observed abundantly in nature and daily life (Dervaux and Ben Amar,
2008; Destrade et al., 2016; Sharon et al., 2004; Vandeparre et al., 2011; Yoon et al., 2012), but also widely applied in
light-weight deployable space structures and modern architectures such as solar sails, telescopes, space inflatable antennas
and suspended roofs of complex shape (Alioli et al., 2017; Elsabbagh, 2015; Fu et al., 2016; Krivoshapko, 2017; Nguyen
et al., 2015). Thin membranes, however, are prone to lose stability since they are hardly to accommodate any compression.
∗
Corresponding authors.
E-mail addresses: [email protected] (F. Xu), [email protected] (Y. Huo).
https://doi.org/10.1016/j.jmps.2018.11.005
0022-5096/© 2018 Elsevier Ltd. All rights reserved.
C. Fu, T. Wang and F. Xu et al. / Journal of the Mechanics and Physics of Solids 124 (2019) 446–470 447
Instead, imposed compressive stresses are relieved by buckling out of plane, and various wrinkling phenomena have raised
considerable attention over the past years (Abdelkhalek et al., 2015; Attipou et al., 2015; Cerda and Mahadevan, 2003; Cerda
et al., 2002; Destrade et al., 2016; Fischer et al., 20 0 0; Friedl et al., 20 0 0; Healey et al., 2013; Jacques and Potier-Ferry,
2005; Kim et al., 2012; Lecieux and Bouzidi, 2010; 2012; Li and Healey, 2016; Li, 2018; Luo et al., 2017; Nayyar et al., 2011;
2014; Plucinsky and Bhattacharya, 2017; Puntel et al., 2011; Rammerstorfer, 2018; Silvestre, 2016; Sipos and Fehér, 2016;
Taylor et al., 2015; Wong and Pellegrino, 20 06a; 20 06b; 20 06c; Yan et al., 2014; Yang et al., 2018; Zhao et al., 2017; Zheng,
2009; Zhu et al., 2018). Friedl et al. (20 0 0) revealed that a stretched rectangular sheet would induce transverse compression
somewhere which makes the sheet wrinkle therein. A diagram with critical buckling coefficients as a function of aspect
ratios was provided to determine the critical tensile stress. Later, Cerda et al. (2002) derived scaling laws for determining
the wrinkling wavelength and amplitude by minimizing an energy functional that incorporates both membrane and bending
contributions. Jacques and Potier-Ferry (2005) obtained an analytical solution of the critical buckling stress based on the
quadratic part of potential energy (van der Heijden, 2009). Puntel et al. (2011) and Kim et al. (2012) unveiled that the
applied critical strain depends on aspect ratios, and verified the scaling laws in Cerda et al. (2002) through both theoretical
and numerical analyses. All these theoretical investigations were constrained to the critical buckling conditions based on the
geometrically nonlinear Föppl-von Kármán plate theory, while the post-buckling analysis was not referred.
The post-buckling analyses of highly stretched thin sheets mainly resort to computational approaches due to the reso-
lution difficulty of strong nonlinearity, especially via the finite element method (Nayyar et al., 2011; Wong and Pellegrino,
20 06c; Zheng, 20 09). Zheng (20 09) first revealed a restabilization phenomenon in a highly stretched sheet, showing that
the wrinkle amplitude first increases, then decreases and finally is smoothed out with the continuous increase of stretching,
based on the linear elastic constitutive law. Later, Nayyar et al. (2011) explored the restabilization phenomenon of hypere-
lastic nHk (neo-Hookean) sheets. The effects of transverse compressive stress distribution and magnitude on wrinkles were
examined through a two-dimensional stress analysis. In a notable recent work by Healey et al. (2013), a thorough bifurca-
tion analysis was carried out for a SVK (Saint-Venant Kirchhoff) finite membrane strain sheet using conformal finite element
discretization and Euler-Newton arc-length continuation technique. They pointed out that the nonlinear in-plane terms in
geometric equations play a crucial role in the disappearance of wrinkles. Li and Healey (2016) further extended the consti-
tutive models to hyperelastic rubber-like materials, i.e., nHk and MR (Mooney-Rivlin) models, and mentioned that nHk and
MR constitutive laws are more accurate than linear elastic ones. The plate model of Li and Healey (2016) is extended here
to various compressible hyperelastic materials. Koiter’s nonlinear plate theory was applied by Taylor et al. (2014) to model
the nonlinear problem, and a dynamic relaxation method with finite difference discretization was implemented to solve
the nonlinear equations numerically. Despite the bending strain tensor involving exact geometric curvatures is accounted
in their plate model, the material law for in-plane membrane strain remains linear elastic. Note that most of the previous
studies assumed the materials of the sheet to be incompressible. The Poisson effect, the origin of transverse compression
during the longitudinal stretching, can affect the nonlinear buckling response, which has not yet been pursued. It is there-
fore in need of providing a general and systematic modeling approach for a large variety of nonlinear constitutive laws, by
taking material compressibility into account.
Here we aim to establish a comprehensive and systematic modeling framework for both compressible and incompressible
materials, which can be used to explore Poisson effect on wrinkling and restabilization behavior. Furthermore, we look into
the effect of strain-stiffening phenomena on wrinkles, based on phenomenological hyperelastic laws such as Gent model
(Gent, 1996). The strain-stiffening characteristic is widely observed in biomaterials and polymers. The stretchability of soft
tissues varies with age, pathology, humidity and the type of tissues. Tendons and ligaments can be uniaxially stretched up to
a large strain ∼ 15%, and even higher for aorta ( ∼ 100%) and cartilage ( ∼ 120%) (Holzapfel, 2001). For polymers, the limiting
√
stretch can be estimated as n, with n being the number of monomers between two crosslinkers. The value of n can range
from several to thousands, and the limiting chain extensibility parameter Jm can be correspondingly tuned by changing the
crosslink density or mixing polymers of different kinds.
To solve large deformations of hyperelastic plate models and to trace the nonlinear morphological evolution of wrinkles,
we develop a novel, efficient and robust numerical algorithm through the combination of ANM (Cochelin et al., 1994; 2007;
Damil and Potier-Ferry, 1990) and a spectral method (Boyd, 2001; Trefethen, 20 0 0; Weideman and Reddy, 20 0 0). The char-
acteristic of an effective numerical procedure to solve large scale instability problems lies in its ability to detect bifurcation
points. Indeed, these problems usually involve large matrix systems and thus reduced-order methods appear to be attractive,
which inspires us to incorporate a spectral method for spatial discretization with the ANM as a path-following continuation
technique to predict bifurcations. The ANM is based on expanding the variables of the nonlinear problems in the form of
Taylor series which are truncated at high orders. The nonlinear partial differential equations are then transformed into a
set of recurrent, well-posed linear problems which hold the same tangent operator. The solution is described continuously
along “steps” that are naturally adaptive and correspond to the range of loading where the power series converge. The main
interest of this method here lies in its ability to trace the post-buckling evolution on the equilibrium path, and to predict
secondary bifurcations without any special tools. The ANM gives interactive access to semi-analytical equilibrium branches,
which provides considerable advantage of reliability compared with classical iterative algorithms. Its efficiency has been
validated for a wide range of nonlinear smooth problems (Nezamabadi et al., 2011; Xu et al., 2015a; Xu and Potier-Ferry,
2016). The use of ANM, however, has been mainly popularized associated with finite element or finite difference methods.
No attempt has been made to the incorporation with spectral methods.
448 C. Fu, T. Wang and F. Xu et al. / Journal of the Mechanics and Physics of Solids 124 (2019) 446–470
Fig. 1. Schematic of a stretched rectangular sheet with two clamped-ends. The area enclosed by dotted lines represents the undeformed configuration.
Equations obtained in the quadratic form of the ANM are spatially discretized using a Chebyshev collocation spectral
method (Boyd, 2001; Trefethen, 2000; Weideman and Reddy, 2000). This method is characterized by the fact that the nu-
merical solution is forced to satisfy the governing equations exactly at collocation points. Motivation of using collocation
spectral methods stems from the high precision and ease of programming. Spectral methods, different from finite difference
and finite element methods based on local arguments, are global in nature, which can provide superior accuracy, at the
expense of domain flexibility. Fourier spectral method was previously used to solve nonlinear wrinkling instability problems
(Audoly and Boudaoud, 2008; Fu et al., 2018; Huang et al., 2016; 2005). However, it can only deal with periodic boundary
conditions (BCs), which rather limits its practical application. By contrast, Chebyshev spectral method can cope with general
BCs expediently and we first attempt to apply it for the resolution of nonlinear instability problems with real BCs. Cheby-
shev collocation method is an appropriate choice for problems on a nonperiodic, bounded domain (Trefethen, 20 0 0). Cheby-
shev polynomials are both orthogonal polynomials and trigonometric functions in disguise (Boyd, 2001), which makes them
widely used in numerical computation. Another advantage of using Chebyshev polynomials as a tool for expansion functions
is the good representation of smooth functions by finite Chebyshev expansion provided that the function is infinitely dif-
ferentiable. More importantly, the coupling between the ANM and Chebyshev collocation method is a new procedure, and
gains the best of both worlds by hybridizing spectral and asymptotic numerical methods.
The paper is organized as follows. In Section 2, equilibrium equations and BCs of finite-strain thin sheets are presented
based on a modified Föppl-von Kármán plate theory. Then several representative compressible or incompressible plate mod-
els are derived based on 3D strain energy density functions, e.g., SVK, nHk, MR, Gent model and GG (Gent-Gent) model.
Section 3 gives details on the coupling of ANM and Chebyshev collocation method to the considered buckling problem of
hyperelastic thin sheets. Results and discussions are provided in Section 4, including efficiency and accuracy of the pro-
posed framework, and investigations of Poisson effect and strain-stiffening phenomena on the wrinkling and restabilization
response. Conclusion is given in Section 5. Throughout this paper, without special elucidation, Latin subscript (i, j, . . .) runs
from 1 to 3, while Greek subscript (α , β , . . .) implies summation over 1 and 2. The Einstein summation convention is em-
ployed, where repeated indices are automatically summed over.
2. Finite strain models for hyperelastic thin sheets
In this section, we derive modified Föppl-von Kármán plate models that are extended to general isotropic hyperelastic
materials. The principles are identical to the ones in Healey et al. (2013) and Li and Healey (2016). The bending terms
remain the same as the classical Föppl-von Kármán model and a fully nonlinear membrane model is adopted within the
plane stress framework. We show that the transition “3D → plane stress” can be conducted in a purely analytical way for
several representative hyperelastic constitutive laws (e.g., SVK, nHk, MR, Gent, etc.) so that these extended Föppl-von Kármán
plate models remain rather simple. Discussions on these nonlinear finite strain models will be given for both incompressible
and compressible materials. Of course, a more consistent model could account for the influence of membrane stresses on
bending stiffness, as established in Steigmann (2007).
2.1. Equilibrium equations and boundary conditions
We consider a highly stretched, hyperelastic thin sheet, which would buckle when the tensile strain reaches a certain
condition. Geometry of the plate is shown in Fig. 1, where the left side (x1 = 0) is fixed and the stretch is imposed on the
right side (x1 = L), while the other two edges (x2 = ±W/2) remain free. Let {e1 , e2 , e3 } denotes a fixed orthonormal basis
for the Euclidean vector space E3 . We assume the rectangular film in the reference configuration denoted as , lying in the
plane E2 = span{e1 , e2 }. The width and length of undeformed film are denoted by W and L, respectively. The aspect ratio
of the domain is defined as β = L/W . We describe the deformation of a point x on the midsurface from the undeformed to
the deformed configuration, f : → E3 , via
f(x ) := xα eα + uα (x )eα + w(x )e3 , (1)
in which uα and w denote in-plane displacements and out-of-plane deflection, respectively. Hence, the (3 × 2) planar defor-
mation gradient tensor of a sheet can be written as
∂ uα ∂w
F := ∇s f = eα eα + e eβ + e eα , (2)
∂ xβ α ∂ xα 3
where the operator ∇ s (·) represents the surface gradient, i.e., ∇ s f := ∂ fi /∂ xα ei eα , and the notation “” denotes the tensor
product. This yields the (2 × 2) right Cauchy-Green deformation tensor
Cs := FT · F = Is + ∇s u + ∇s uT + ∇s uT · ∇s u + ∇s w ∇s w, (3)
in which Is := eα eα is the (2 × 2) identity tensor. The Green-Lagrange strain tensor is defined as
1
Es = (Cs − Is ), (4)
2
and the bending strain tensor is linearized as the Föppl-von Kármán model (Healey et al., 2013; Li and Healey, 2016), given
by
Ks ∼
= −∇s ◦ ∇s w, (5)
where ∇ s ◦∇ s (·) stands for the surface second gradient. As the Föppl-von Kármán model, we assume an additive decompo-
sition of the stored energy density, in terms of membrane and bending effects, namely,
Wtot (Cs , Ks ) = Wm (Cs ) + Wb (Ks ), (6)
where the 2D membrane strain energy is given by
Wm (Cs ) = h(Cs ), (7)
in which denotes 3D strain energy density function. The bending energy density Wb (Ks ) can be expressed as (Healey et al.,
2013)
Eh3
Wb (Ks ) = ν tr(Ks )2 + (1 − ν )Ks · Ks , (8)
24 1 − ν 2
where h represents film thickness, E and ν correspond to initial Young’s modulus and Poisson’s ratio in the linear elastic
regime, respectively. The potential energy P of the film reads

P= [Wm (Cs ) + Wb (Ks )]d. (9)

Variation of the energy (9) with respect to the components of displacement field yields the equilibrium (Euler-Lagrange)
equations:
∇s · [ ( Is + ∇s u ) · Ss ] = 0, (10)
s M s + h ∇ s · ( S s · sw ) = 0, (11)
where s (·) is the Laplace operator, and Ss and Ms := ∂ Wb (Ks )/∂ Ks respectively denote the (in-plane) second Piola-Kirchhoff
(P-K) stress tensor and couple-stress tensor. Integration by parts gives free BCs (0 x1 L, x2 = ±W/2) for Eq. (9) so that
the bending moment, membrane stresses and effective transverse shear force respectively equal to zero, which are listed
here for clarity:
⎧ T
⎪e · M · e = 0,
⎨(I2 + ∇s u )2 · S · e = 0,
s s s 2
(12)
⎪ ∂
⎩e2 · hSs · s w + ( M · e ) + ∇s · Ms = 0.
∂ x1 s 1
Table 1
Comparison between ANM-Spectral method and FEM in Abaqus for incompressible nHk model.
ANM-Spectral FEM
No. of Grids 400 1600 3600 6400 416 1600 6477 25908
No. of DOFs 1323 5043 11,163 19,683 2772 10,140 39,936 157,590
No. of Steps – 160 150 140 – 1003 1003 1003
ε cr1 3.37% 3.37% 3.37% 3.37% 4.80% 3.50% 3.33% 3.32%
ε cr2 – 29.22% 29.21% 29.21% – 27.2% 29.14% 29.70%
Peak (w/h) – 3.2851 3.2693 3.2832 – 3.7681 3.4028 3.5442
Fig. 2. Bifurcation diagrams for different incompressible models, i.e., SVK, nHk and MR models. The solid lines represent our numerical results, while the
dots denote finite element results from Li and Healey (2016), in which wmax = max[w(x1 , x2 )]. Geometric parameters: β = 2, W/h = 10 0 0. The material
parameter c2 = μ/22 in Eq. (35) is chosen as Li and Healey (2016).
Fig. 3. Comparison of bifurcation diagrams between our ANM-Spectral method and FEM for different incompressible models: SVK model, nHk model
and MR model. The solid lines represent our numerical results, while the dash lines represent those calculated by FEM. Geometric parameters: β = 2.5,
W/h = 10 0 0. The material parameter c2 = μ/22 in Eq. (35) for MR model.
2.2. Constitutive equations
In general, the mechanical response of hyperelastic materials is derived from a 3D strain energy function . In this
work, we limit this function to isotropic cases. Anisotropic materials such as orthotropic elasticity considered in Sipos and
Fehér (2016) and Zhu et al. (2018) can be straightforward introduced into the following SVK models and thus we omit it
here. Isotropic hyperelastic materials can be written as a function of strain invariants of the right symmetric Cauchy-Green
tensor C. Hence, the strain energy function is formulated as
= [I1 (C ), I2 (C ), I3 (C )], (13)

in which
⎧
⎨I1 = tr(C ),
1
I2 = tr(C )2 − tr C2 , (14)
⎩ 2
I3 = J 2 = det (C ).
We treat the membrane as 3D continuum and thus the extended right Cauchy-Green deformation tensor reads
Cs 0
C= , (15)
0 C33
where Cα 3 = 2Eα 3 = 0 due to the Kirchhoff-Love hypothesis (normal remains normal), and C33 describes the thickness defor-
mation, which can be explicitly computed from the in-plane components Cαβ using the incompressibility condition or plane
stress condition.
2.2.1. Compressible materials

Let us begin by considering a simple case, i.e., SVK model where only geometric nonlinearity is taken into account:
λ
SVK = tr(E )2 + μtr E2 , (16)
2
where λ = E ν /[(1 + ν )(1 − 2ν )] and μ = E/[2(1 + ν )] denote Lamé’s material constants, and the 3D Green-Lagrange strain
tensor E is defined as
1 E 0
E= (C − I ) = 0s E33
, (17)
2
in which E33 = (C33 − 1 )/2 and I is the (3 × 3) identity tensor. Then the second Piola-Kirchhoff stress tensor S can be ex-
pressed as
∂ ∂
S=2 = = λtr(E )I + 2μE. (18)
∂C ∂E
To satisfy the plane stress condition (S33 = 0), E33 can be calculated by
λ
E33 = − tr(Es ). (19)
λ + 2μ
Subsequently, the in-plane second P-K stress tensor is given as
E
Ss = [ν tr(Es )Is + (1 − ν )Es ]. (20)
1 − ν2
When ν = 0.5, Eq. (20) tends to be an incompressible SVK stress tensor.
For the compressible nHk model (Ogden, 1985; Simo and Hughes, 1998) which is an extension of linear isotropic consti-
tutive law to large deformations, the corresponding strain energy density function can be written as
λ 1 2 μ
nHk (I1 , I3 ) = J − 1 − ln J + (I1 − 3 − 2 ln J ). (21)
2 2 2
Then the second Piola-Kirchhoff stress tensor S can be derived as

λ λ μ
S= μI + − − I3 C−1
2 2I3 I3
λ
= μ I − C−1 + J2 − 1 C−1 . (22)
2
Under plane stress assumption (S33 = 0), C33 can be calculated by
λ + 2μ
C33 = , (23)
λA + 2μ
where A = det(Cs ). Hence, the in-plane second P-K stress tensor is given as
λ 2
Ss = μ Is − C−1
s + J − 1 C−1
s , (24)
2
in which J 2 = AC33 = A(λ + 2μ )/(λA + 2μ ).
Fig. 4. Evolution of wrinkles for incompressible nHk model. (a) Load-displacement bifurcation diagram. Each blue point corresponds to one ANM step. The
green square represents the analytic critical strain given by Puntel et al. (2011). (b) Wrinkled configurations with increasing stretching: ε = 0.034, 0.06,
0.10, 0.20 and 0.29, corresponding to the red circles marked on the bifurcation diagram (a). (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
Next we consider a more complex model, i.e., MR model, in which the strain energy depends on three invariants I1 , I2
and I3 . One version of MR model can be written as

MR (I1 , I2 , I3 ) = c1 (I1 − 3 ) + c2 (I2 − 3 ) + c J2 − 1 − d ln J, (25)
in which c and d are temperature-dependent material constants, and μ = 2(c1 + c2 ) is the shear modulus. By prescribing
that the reference configuration is stress-free, one obtains d = 2c + 2c1 + 4c2 . The nHk model (21) can be deduced from the
Fig. 5. Bifurcation curves for incompressible Gent model with a variety of Jm . The limiting strain is denoted as ε m . In the limit case of Jm → ∞, the Gent
model recovers nHk model.
strain energy (25) by taking c2 = 0, which yields d = λ/2 + μ. The last two terms in (25) were proposed by Ciarlet (1988),
slightly different from Ciarlet and Geymonat (1982). The second Piola-Kirchhoff stress tensor S is then given by

S = 2(c1 + c2 I1 )I − 2c2 C + 2cJ 2 − d C−1 . (26)
According to S33 = 0, C33 can be calculated by
d
C33 = , (27)
2(c1 + c2 B + cA )
where B = tr(Cs ). Then the in-plane second P-K stress tensor is expressed as

Ss = 2(c1 + c2 B + c2C33 )Is − 2c2 Cs + 2cJ 2 − d C−1
s , (28)
where = AC33 = dA/[2(c1 + c2 B + cA )].
J2
Lastly, we present the compressible version of Gent model, which is particularly suitable for modeling severe strain-
stiffening phenomena. The strain-energy function is given as
λ 1 2 μ I1 − 3

G (I1 , I3 , Jm ) = J − 1 − ln J − Jm ln 1 − + 2 ln J , (29)
2 2 2 Jm
in which I1 < Jm + 3. In the limit case of Jm → ∞, Eq. (29) recovers the (parent) nHk material in Eq. (21). The second Piola-
Kirchhoff stress tensor S can be derived as

μJm λ λ μ
S= I+ − − I3 C−1
Jm − I1 + 3 2 2I3 I3
Jm λ 2
= μ I − C−1 + J − 1 C−1 . (30)
Jm − I1 + 3 2
Following the same way, the plane stress condition (S33 = 0) can be used to eliminate the thickness deformation C33 . Then
through substituting C33 into Eq. (30), one can obtain the in-plane second Piola-Kirchhoff stress tensor Ss as
Jm λ
Ss = μ Is − C−1
s + (AC33 − 1 )C−1
s . (31)
Jm − B − C33 + 3 2
2.2.2. Incompressible materials

Some hyperelastic materials can support large strains with a negligible volume change and satisfy the incompressibility
condition:
J = 1. (32)
Note that the compressibility is represented by adding the strain energy vol which describes the so-called volumetric (or
dilational) elastic response, while it is null for the incompressible materials. Hence, in order to deduce the constitutive
equations for the incompressible hyperelastic plates (I3 = 1), one can consider the strain energy function in the following
form:
= [I1 (C ), I2 (C )] − p(J − 1 ), (33)
Fig. 6. Wrinkling evolution of incompressible Gent model with Jm = 1. (a) Bifurcation diagram. The limiting strain is indicated as ε m . Each blue point
corresponds to one ANM step. (b) Wrinkled configurations with increasing stretching: ε = 0.034, 0.05, 0.10, 0.16 and 0.20, corresponding to the red circles
marked on the bifurcation diagram (a). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this
article.)
in which p denotes a Lagrange multiplier, which can be identified as a hydrostatic pressure. Generally, for a 3D constitutive
model, the value of p needs to be computed at each iterative step. However, for the proposed thin plate model, analytical
formulations for the Lagrange multiplier p can be solved under plane stress condition. The second Piola-Kirchhoff stress
tensor is given by
∂ ∂
S=2 =2 (I , I ) − pC−1 . (34)
∂C ∂C 1 2
Fig. 7. Comparison of bifurcation diagrams between compressible nHk and MR models with varying Poisson’s ratios. The solid lines represent nHk model,
while the dash lines represent MR model. Material parameters for MR model: d = λ/2 + μ, c1 = 5μ/11, c2 = μ/22 and c = d/2 − c1 − 2c2 .
Fig. 8. Load-displacement bifurcation diagram for compressible Gent model with varying Jm but the same Poisson’s ratio ν = 0.4. The limiting strain is
marked as ε m . When Jm → ∞, the Gent model recovers nHk model.
For incompressible MR material, the corresponding strain energy density function reads
MR (I1 , I2 , I3 ) = c1 (I1 − 3 ) + c2 (I2 − 3 ) − p(J − 1 ), (35)
where c1 and c2 are material constants with c1 + c2 = μ/2. Note that the incompressible MR strain energy in Li and
Healey (2016) (see Eq. (10)) is identical to h MR as c2 = μ/22 and ν = 0.5. When c2 = 0, it is degenerated into the in-
compressible nHk model with one single material constant. The second P-K stress tensor is expressed as
S = 2(c1 + c2 I1 )I − 2c2 C − pC−1 . (36)
The incompressibility condition J = 1 yields C33 = 1/A. By enforcing the plane stress condition (S33 = 0), one obtains
p = 2(c1 + c2 B )C33 . (37)
Substituting Eq. (37) into Eq. (36), the in-plane second P-K stress tensor is then expressed as
Ss = 2[c1 + c2 (1/A + B )]Is − 2c2 Cs − pC−1
s . (38)
Next we present the incompressible version of the Gent strain-energy function (Gent, 1996):
μ I −3

G (I1 , I3 , Jm ) = − Jm ln 1 − 1 − p(J − 1 ), (39)
2 Jm
Fig. 9. Stretching-induced transverse compressive stress for compressible MR model. (a) Contour of dimensionless transverse stress S22 /E at ε = 0.15, with
only the compressive stress region in color. Poisson’s ratio is taken as ν = 0.45. (b) Evolution of transverse compressive stress with various Poisson’s ratios.
The maximum dimensionless compressive stress is defined as S̄22 = min(S22 (x1 , 0 ))/E.
where I1 < Jm + 3. In the limit case of Jm → ∞, Eq. (39) recovers the parent material (nHk model). The second P-K stress
associated with Eq. (39) is given by
μJm
S= I − pC−1 . (40)
Jm − I1 + 3
The Gent model which involves only two material parameters has the advantage of mathematical simplicity. Nevertheless,
Eq. (40) might not give an accurate prediction of mechanical properties of rubber-like materials for the full range of finite
deformations. Modifications of the Gent mode involving three constitutive parameters have been proposed by Yeoh (1997),
Pucci and Saccomandi (2002), and Ogden et al. (2004). Such a model named Gent-Gent (GG) model (Ogden et al., 2004)
reads
I1 − 3
I2

GG (I1 , I2 , I3 , Jm ) = −c1 Jm ln 1 − + c2 ln − p(J − 1 ), (41)
Jm 3
with 2c1 + 2c2 /3 = μ. When c2 = 0, it is degenerated into the incompressible Gent model (39). In the limit case of Jm → ∞,
it recovers the Gent-Thomas model (Gent and Thomas, 1958). The second P-K stress is then given by
2c1 Jm 2c2 I1

2 c2
S= + I− C − pC−1 . (42)
Jm − I1 + 3 I2 I2
Fig. 10. Evolutions of longitudinal and transverse stresses for incompressible Gent model with different material parameters Jm . (a) Variation of the maxi-
mum tensile stresses with increasing stretching, defined as S̄11 = max(S11 (x1 , x2 ))/E. (b) Variation of transverse compressive stresses. The limiting strain is
denoted as ε m .
Similarly, by enforcing the plane stress condition (S33 = 0), one obtains
1 2c1 Jm 2c2 I1 2 c2

p= + − . (43)
A Jm − I1 + 3 I2 AI2
Following the same way, the in-plane second P-K stress tensor Ss can also be obtained by substituting p into Eq. (42):
2c1 Jm 2c2 I1
2 c2
Ss = + Is − Cs − pC−1
s , (44)
Jm − I1 + 3 I2 I2
where I1 = B + 1/A and I2 = B/A + A.
3. Numerical implementation
New hyperelastic plate models have been presented above, including the approximation (8) for bending stiffness and
exact hyperelastic models for membrane stiffness. We have established that the corresponding plane stress constitutive law
can be written explicitly in many cases, including compressible and incompressible MR models as well as the incompressible
GG model. In this section, we develop a novel numerical algorithm to solve the highly nonlinear problems involving sec-
ondary bifurcations with the hyperelastic models described in Section 2, and show how to apply the resolution framework
to various hyperelastic constitutive laws involving both geometric and material nonlinearities. The ANM that allows trans-
forming the nonlinear problems into a sequence of linear ones is applied. As compared to the classical predictor-corrector
algorithms such as Newton-Raphson method, the number of tangent matrix inversion is largely reduced due to high-order
arc-length prediction. The main interests of the ANM lie in step length adaptation with fewer incremental steps and its
ability to trace the post-buckling evolution on the equilibrium path. The resulting continuous problems are discretized by a
spectral method, which generates algebraic equations with full matrices, but in compensation, the high order of the basis
functions can provide high accuracy. Especially when the geometry of the problem is fairly smooth and regular, spectral
methods appear to be much more efficient than finite element method and finite difference method. Moreover, for bounded
non-periodic domains, Chebyshev polynomials on irregular grids are a proper choice (Trefethen, 20 0 0). Chebyshev polyno-
mials are well-known family of orthogonal polynomials on the interval [−1, 1]. They are independent of BCs, yet it is easy
to enforce explicit constraints such as Dirichlet, Neumann and Robin BCs.
3.1. Asymptotic formulations for hyperelasticity
ANM (Cochelin et al., 1994; 2007; Damil and Potier-Ferry, 1990) is applied to solve the resulting nonlinear equations.
It consists in expanding the main variables of the problem to high-order truncated power series (perturbation technique)
with respect to a well chosen path parameter. In this way, the perturbation procedure allows one to transform the nonlinear
problem into a sequence of linear ones having the same tangent operator. Compared with the classical Newton-Raphson al-
gorithms, ANM is considered as a high-order predictor which generally does not need any correction procedure. In addition,
an adaptive step length is determined a posteriori in a simple and optimal way, which is a key point as for the efficiency
and robustness of ANM.
Here we extend the ANM algorithm to large deformation of hyperelastic plates. For the static or quasi-static analysis of
plate structures, Eqs. (10) and (11) can be expressed as
R ( U, λ ) = L ( U ) + Q ( U, U ) − λ
˜ Pext = 0, (45)
in which R is the residual vector, U = {w, u, F, Ss , Cs , A, · · · } denotes a mixed vector of unknown variables, L is a linear
operator, Q is a quadratic one and Pext represents a force vector. The external load parameter is denoted as a scalar λ ˜ . The
principle of the ANM continuation method consists in describing the solution path by computing a succession of truncated
power series expansions. From a known solution point (U0 , λ ˜ 0 ), the solution (U, λ
˜ ) is expanded into high-order truncated
power series of a perturbation parameter a:

U (a ) U0
N
Ui
i
= + a , (46)
λ˜ (a ) λ˜ 0 i=1
λ˜ i
a = u − u0 , u1 + w − w0 , w1 + ( λ
˜ −λ
˜ 0 )λ
˜ 1, (47)
where N is the truncation order of the series which is generally chosen in the range 10 ∼ 20. According to Eqs. (10) and (11),
one can easily derive quadratic component equations, and thus one just needs to express the second P-K stress in quadratic
form. Considering the simplest case of SVK model, the constitutive relation (20) can be written as
⎧
⎪F = Is + ∇s u + e3 ∇s w,
⎪
⎨ 1 T
s E = F · F − Is , (48)
2
⎪
⎪
⎩Ss = E [ν tr(Es )Is + (1 − ν )Es ].
1 − ν2
Introducing the Eq. (46) in the problem (48) gives
⎧
⎪
⎪ N N N
⎪ a Fi = Is + ∇s
⎪ i
a ui + e3 ∇s
i i
a wi ,
⎪
⎪
⎪ i=0
⎪ i =0 i =0
⎪
⎨ N
1 N N
a (Es )i =
i i T
a Fi · i
a Fi − Is , (49)
⎪
⎪ 2
⎪
⎪
i=0 i=0 i=0
⎪
⎪ N N N
⎪
⎪ E
⎪ a (Ss )i =
⎩
i
ν tr a (Es )i Is + (1 − ν )
i
a (Es )i .
i
1 − ν2
i=0 i=0 i=0
By identifying the terms with the same power of “a”, we obtain a recursive sequence of linear problems. The terms at order
0 in Eq. (49) correspond to the initial known solution for which the residual of each equation is assumed to be infinitesimal.
The linear problems for each order, therefore, can be written as follows:
At Order 1:
⎧
⎪F1 = ∇s u1 + e3 ∇s w1 ,
⎪
⎨ 1
(Es )1 = FT0 · F1 + FT1 · F0 , (50)
2
⎪
⎪
⎩(Ss )1 = E {ν tr[(Es )1 ]Is + (1 − ν )(Es )1 }.
1 − ν2
At Order n (2 ≤ n ≤ N):
⎧
⎪Fn = ∇s un+ e3 ∇s wn ,
⎪
⎪
⎪
⎨ 1
n−1

(Es )n = FT0 · Fn + FTn · F0 + FTr · Fn−r ,
(51)
2
⎪
⎪ r=1
⎪
⎪
⎩(Ss )n = E {ν tr[(Es )n ]Is + (1 − ν )(Es )n }.
1 − ν2
Since hyperelastic models usually involve strong nonlinear relations, application of ANM algorithm to these models is
not straightforward. To illustrate the application of ANM to hyperelastic plates, we take two representative MR models as
examples. Details of this implementation on incompressible Gent models can be found in Appendix A.
Compressible MR model
Using the approach presented in Section 2.2.1, the constitutive relation (28) of the compressible MR model can be ex-
pressed in a quadratic form:
⎧
⎪Cs = FT · F,
⎪
⎪
⎪C−1
⎪ s · Cs = Is ,
⎪
⎨A = C11 · C22 − C12
2
,
B = C11 + C22 , (52)
⎪
⎪C33 = d/[2(c1 + c2 B + cA )],
⎪
⎪
⎪
⎪ 2
⎩J = AC33 ,
Ss = 2(c1 + c2 B + c2C33 )Is − 2c2 Cs + 2cJ 2 − d C−1
s .
Application of the perturbation technique to Eq. (52) allows deducing a series of linear problems for each order n:
⎧
⎪
⎪
n−1

⎪
⎪ ( C ) = F T
· F + F T
· F + FTr · Fn−r ,
⎪
⎪
s n 0 n n 0
⎪
⎪ r=1
⎪
⎪ −1 −1 −1 n−1
⎪
⎪ Cs = − Cs · (Cs )n · C−1 − Cs · (Cs )r · C−1 ,
⎪
⎪ s s
⎪
⎪
n 0 0 0 n−r
⎪
⎪
r=1
⎪An = (Cn11
⎪
⎪
)0 · (C22 )n + (C11 )n · (C22 )0 − 2(C12 )0 (C12 )n
⎪
⎪ −1
⎪
⎪ + (C11 )r · (C22 )n−r − (C12 )r · (C12 )n−r ,
⎪
⎪
⎪
⎪ r=1
⎪
⎨Bn = (C11 )n + (C22 )n ,
n−1
d (C33 )r · (c2 Bn−r + cAn−r ) + (C33 )n−r · (c2 Br + cAr ) (53)
⎪
⎪ (C33 )n = − (n − r )
⎪
⎪ 2 n c1 + c2 B0 + cA0
⎪
⎪ r=1
⎪
⎪ d ( C ) · ( c B + cA )
⎪
⎪ −
33 0 2 n n
,
⎪
⎪ 2(c1 + c2 B0 + cA0 )
⎪
⎪ −1
⎪
⎪
n
⎪
⎪ n J 2
= A (C ) + A (C ) + Ar (C33 )n−r ,
⎪
⎪
0 33 n n 33 0
⎪(S ) = 2[c B + c (C ) ]I − 2c (C ) + 2cJ2 − d C−1

⎪
⎪
r=1
⎪
⎪ s n 2 n 2 33 n s 2 s n s
⎪
⎪ n−1
0 n
⎪
⎪
⎪
⎩ +2c J 2
n
Cs−1
0
+ 2c J 2
r
Cs−1
n−r
.
r=1
Incompressible MR model
Following the same way, the constitutive law (38) of the incompressible MR model can be written in a quadratic form
as follows:
⎧
⎪Cs = FT · F,
⎪
⎪
⎪
⎪C−1
s · Cs = Is ,
⎪
⎨A = C11 · C22 − C12
2
,
B = C +C ,
11 22 (54)
⎪
⎪C33 = 1/A,
⎪
⎪
⎪ p = 2(c1 + c2 B )C33 ,
⎪
⎩
Ss = 2[c1 + c2 (C33 + B )]Is − 2c2 Cs − pC−1
s .
The order n of this system (54) reads

⎧ n−1
⎪
⎪
⎪ ( C ) = F T
· F + F T
· F + FTr · Fn−r ,
⎪
⎪
s n 0 n n 0
⎪
⎪ r=1
⎪
⎪ −1 −1 n−1
⎪
⎪
⎪
⎪ Cs = − C−1 · (Cs )n · C−1 − Cs · (Cs )r · C−1 ,
⎪
⎪
n s 0 s 0 0 s n−r
⎪
⎪
r=1
⎪
⎪An = (C11 )0 · (C22 )n + (C11 )n · (C22 )0 − 2(C12 )0 (C12 )n
⎪
⎪ n−1
⎪
⎪
⎪
⎪ + (C11 )r · (C22 )n−r − (C12 )r · (C12 )n−r ,
⎪
⎨ r=1
Bn = (C11 )n + (C22 )n , (55)
⎪
⎪ n−1
⎪
⎪ (C ) · An 1
⎪
⎪ (C33 )n = − 33 0 − (n − r ) (C33 )r · An−r + (C33 )n−r · Ar ,
⎪
⎪ A0 nA0
⎪
⎪
r=1
⎪
⎪
n−1

⎪
⎪ pn = 2(c1 + c2 B0 )(C33 )n + 2c2 Bn (C33 )0 + 2c2 Br (C33 )n−r ,
⎪
⎪
⎪
⎪ r=1
⎪
⎪ (Ss )n = 2[c2 Bn + c2 (C33 )n ]Is − 2c2 (Cs )n − p0 C−1 − pn C−1
⎪
⎪ s s
⎪
⎪ n−1 −1
n 0
⎪
⎪
⎩ − pr Cs
n−r
.
r=1
Detailed algorithm to compute the series can be found in Cochelin et al. (2007) and thus we limit ourselves to a brief
description in what follows. Introducing Eq. (55) (or Eq. (53) for compressible MR model) into Eqs. (10), (11) and (47), one
obtains a recurrent sequence of linear problems with respect to the perturbation parameter a:
At Order 1:
Lt (U1 ) = λ
˜ 1 Pext ,
u1 , u1 + w1 , w1 + λ
˜ 2 = 1,
1 (56)
At Order n (2 ≤ n ≤ N):
n−1

Lt (Un ) = λ
˜ n Pext − Q(Ur , Un−r ) = λ
˜ n Pext + Pnl ,
n
(57)
r=1
u1 , un + w1 , wn + λ
˜ 1λ
˜ n = 0,
in which Lt (· ) = L(· ) + Q(U0 , · ) is the tangent operator defined at the starting point, and Pnl
n represents the residual vector.
Note that the ANM formulation at order 1 is identical to the classical numerical methods based on linearization of the
nonlinear problem. As a result, the tangent operator is exactly the same as obtained in the prediction stage of classical
incremental-iterative algorithm. To check whether the global tangent stiffness matrix Lt is positive definite or not, Crout
decomposition is applied at each step during the nonlinear resolution to evaluate the stability of solutions (Xu et al., 2015b).
The cost of this stability test is negligible since Crout decomposition has also been done to solve the problems (56) and (57).
The computation can be managed by prescribing a displacement, a force or a combination of both. In all computations in
this paper, we adopted displacement loading.
The maximum value of the path parameter amax should be automatically defined by analyzing the convergence of the
power series at each step. This amax can be based on the difference of displacements at two successive orders that must be
smaller than a given precision parameter δ . The validity range is given as
1/(N−1)
u1 + w1 e3
amax = δ , (58)
uN + wN e3
where · denotes the Euclidean norm. Unlike classical incremental-iterative methods, the arc-length step size amax is
adaptive since it is determined a posteriori by the algorithm. When there is a bifurcation point on the solution path, the
radius of convergence is defined by the distance to the bifurcation. The step length defined in Eq. (58), therefore, becomes
smaller and smaller, which looks as if the continuation process “knocks” against the bifurcation. This accumulation of small
steps is an effective indicator of the presence of a singularity on the path. All the bifurcations can be easily identified in this
way by the user without any special tool (Xu et al., 2015a; Xu and Potier-Ferry, 2016).
3.2. Spectral collocation method
Differentiation matrices are derived from the spectral collocation (also known as pseudospectral) method for solving
differential equations of boundary value type. This method is briefly discussed below and for more complete descriptions,
one refers to Canuto et al. (1988), Funaro (1992), and Fornberg (1996). In the spectral collocation method, the unknown
solution to the differential equation is expanded as a global interpolant, e.g., trigonometric or polynomial interpolant. In
other methods such as finite elements or finite differences, the underlying expansion involves local interpolants such as
piecewise polynomials. In practice this means the √accuracy of the spectral method is superior: for the problems with smooth
solutions, convergence rates of O (e−cM ) or O (e−c M ) are normally achieved, where M is the number of degrees of freedom
(DOFs) in the expansion. In contrast, finite elements or finite differences yield convergence rates that are only algebraic in
M, typically O (M−2 ) or O (M−4 ).
To introduce the idea of differentiation matrix, we recall that the spectral collocation method for solving differential
equations is based on weighted interpolants of the following form (Canuto et al., 1988; Fornberg, 1996):

M
α (x )
f (x ) ≈ pM−1 (x ) = φ j (x ) f j , (59)
j=1
α xj

where {x j }M
j=1
is a set of distinct interpolation nodes, α (x) denotes a weight function and f j = f x j . The set of interpolating
functions {φ j (x )}M
j=1
satisfies φ j (xk ) = δ jk (the Kronecker delta), and pM−1 (x ) is defined as an interpolant of the function
f(x), namely
pM−1 (xk ) = f (xk ), k = 1, · · · , M. (60)
Derivative operator of the interpolant is used to approximate that of f(x). This operator is generated by taking lth derivative
of Eq. (59) and evaluating the result at the nodes xk :

M
dl α (x )
f ( l ) ( xk ) ≈ φ j (x ) f j, k = 1, · · · , M. (61)
dxl
j=1
α xj
x=xk
The derivative operator can be represented by a differentiation matrix, namely

(l ) dl α (x )
Dk, j = φ j (x ) . (62)
dxl α xj
x=xk
The numerical differentiation process can therefore be expressed as the matrix-vector product:
f ( l ) = D ( l ) f, (63)
where f(l)
denotes the vector of approximate derivative values at the nodes {x j }Mj=1 .
Since the Chebyshev interpolating function is elegant, useful, and more importantly it is capable of dealing with arbitrary
BCs, we employ it here. This function is constrained to Chebyshev nodes and constant weights. Considering a non-uniform
grid {xk } by

k−1
xk = cos π , k = 1, · · · , M, xk ∈ [−1, 1]. (64)
M−1
The weight function is defined as
α ( x ) = 1. (65)
Thus, according to Eq. (59) the interpolant can be written as

M
pM−1 (x ) = φ j (x ) f j , (66)
j=1
where
(x )
(−1 ) j 1 − x2 TM−1
φ j (x ) = , (67)
c j (M − 1 )2 x − x j
where TM−1 (x ) is the Chebyshev polynomial of degree M − 1, c1 = cM = 2 and c2 = · · · = cM−1 = 1. The interpolant can be
rewritten in a barycentric form:
M (−1 ) j f j
(
j=1 c j x−x j )
pM−1 (x ) = . (68)
M (−1 ) j
(
j=1 c j x−x j )
Hence, the Chebyshev differentiation matrix is given as (Canuto et al., 1988)
⎧
⎪ k( ) ,
⎪ c −1 j+k
⎪ j = k,
⎪
⎪
⎪ c j xxk − x j
⎪
⎪
⎨ k , j = k = 1, M,
(1 ) 2 xk − 1
2
Dk, = (69)
j
⎪
⎪ 2(M − 1 )2 + 1
⎪
⎪ , j = k = 1,
⎪
⎪ 6
⎪
⎪
⎩− 2 ( M − 1 )2
+ 1
, j = k = M,
6
l
D (l ) = D (1 ) . (70)
For 2D grids there are various options for the storage layout of the data and for the computation of partial derivatives in
each of the two directions. We store the field values on the grid as a 1D array of size N2D = n1 n2 , where n1 and n2 represent
the numbers of nodes in the x1 and x2 directions, respectively. We denote the differentiation matrices in the two spatial
directions by D1(l ) = Dn(l1)×n1 and D2(l ) = Dn(l2)×n2 . One elegant way (Trefethen, 20 0 0) is to construct 2D differentiation matrices
acting on 1D arrays of size N2D using the Kronecker product:
D˜ 1(l ) = D1(l ) I2 , D˜ 2(l ) = I1 D2(l ) , (71)
where Iα is the identity matrix of size nα × nα , and D˜ (l ) denotes the 2D differentiation matrix of size N2D × N2D . The compu-
tation of the spectral derivative of a 2D field given as a 1D vector v using (71) is given by the matrix multiplication, namely
∂α(l ) v = D˜ α(l ) · v. (72)
With the assistance of 2D differentiation matrices, we can easily deduce the tangent operator in Eq. (57).
We now focus on BCs. Numerical BCs can be explicitly imposed through two strategies: (i) reducing the number of
collocation conditions on the residual of the differential equation, and using rows of the pseudospectral matrix to explicitly
enforce the constraint or (ii) modifying the problem so that the BCs of the modified problem are homogeneous and then
altering the basis set so that the basis functions individually meet these conditions. These two strategies are respectively
referred to “boundary-bordering” and “basis recombination” (Boyd, 2001). The reason for the name “boundary-bordering”
is that the “border” of the pseudospectral matrix explicitly imposes the BCs, and the residual of the differential equation
should vanish at interpolation points on the interior of the interval. Compared with “basis recombination”, this method
is more flexible and is often better for more complicated problems. For “basis recombination”, it is conceptual simplicity
and almost always the method of choice for analytical, paper-and-pencil or algebraic manipulation language calculations.
After having shifted to the modified basis set, we can thenceforth ignore the BCs and concentrate solely on the differential
equation. Details on the implementation of the two strategies are given in the literature (Boyd, 2001; Trefethen, 20 0 0;
Weideman and Reddy, 20 0 0).
The “basis recombination” method is superior to “boundary-bordering” for eigenvalue problems whose BCs do not involve
the eigenvalue, while “basis recombination” gives a spectral matrix which contains the eigenvalue in every row. But in
practice, the matrix elements for the modified basis functions are more complicated to compute than those involving the
unmodified basis functions of the “boundary-bordering” method. Therefore, the “basis recombination” method is always
used to handle simple and linear BCs. In many situations, the differences between “basis recombination” and “boundary-
bordering” have little practical significance. Karageorghis (1993) discussed the relationship between “basis recombination”
and “boundary-bordering” formulations for the spectral collocation method in rectangles.
In this work, we propose an effective strategy to deal with the BCs of the clamped-clamped rectangular sheet through
a combination of the two approaches mentioned above. For the clamped BCs (w = ∂ w/∂ x1 = 0 at x1 = 0, L), the “basis re-
combination” method is adopted due to its good convergence and easy implementation. Details on this method applied to a
biharmonic problem with clamped BCs are given in Appendix B. For the free BCs (x2 = ±W/2, see Eq. (12)), we find that it
is difficult to construct a shift function that satisfies these BCs. Therefore, the “boundary-bordering” method seems a better
choice for free BCs.
4. Results and discussion
We perform numerical computations to explore the post-buckling response of a highly stretched hyperelastic rectangular
sheet, and demonstrate the effectiveness of the proposed modeling and resolution framework. Different hyperelastic con-
stitutive laws are considered for both compressible and incompressible materials. Results of the algorithm associating ANM
and spectral collocation method are compared with Li and Healey (2016) and those obtained by commercial finite element
software Abaqus (ABAQUS, 2013), where a four-node quadrilateral shell element with reduced integration (S4R) is adopted.
In computations, we consider one half of the membrane ((0, L) × (0, W/2)) enforcing anti-symmetric conditions along
x2 = 0. As noted by Healey et al. (2013) and Li and Healey (2016), the reflection-symmetric conditions along x2 = 0 are
also reasonable on account of apparent “asymptotic” symmetry in the presence of high stretching and fine thicknesses.
The fact that both arise at the same eigenvalue (Zheng, 2009) shows that either mode is equally likely to appear once the
critical wrinkling threshold is reached. Quantitative comparisons of bifurcation diagram between anti-symmetric solutions
and reflection symmetric ones for different incompressible constitutive models are provided in Supplementary Material.
Distinct sets of both BCs are listed below.
Reflection symmetric BCs:
u2 = w,2 = eT1 · (Is + ∇s u ) · Ss · e2
∂
= e2 · hS s · sw + ( M · e ) + ∇s · Ms = 0, x2 = 0. (73)
∂ x1 s 1
Anti-symmetric BCs:
u2 = w = eT1 · (Is + ∇s u ) · Ss · e2 = eT2 · Ms · e2 = 0, x2 = 0. (74)
Owing to high precision of the spectral method, we use a 60 (length) × 60 (width) nonuniform grid on the half domain,
which refers to Chebyshev nodes. The number of grids is much less than that in the finite element method and finite
difference method, e.g., 70 (length) × 170 (width) in Li and Healey (2016), and 60,0 0 0 in Taylor et al. (2014), respectively.
Note that the length of the grid in the x1 direction is much larger than that in the x2 direction, since only half a wave
appears in the x1 direction in the post-wrinkling stage. For all the calculations, the ANM (perturbation technique) is used
with a truncation order N = 15 of the series and with the precision parameter δ = 10−8 , and the width-to-thickness ratio of
the film is fixed as W/h = 10 0 0. In order to trigger a transition from the fundamental branch to the bifurcated one, a small
perturbation displacement, w p = h/20 0 0, is imposed on the membrane.
To assess the efficiency of ANM-Spectral method in comparison with finite element method (FEM) in Abaqus, we perform
nonlinear post-buckling calculations using different meshes densities for incompressible nHk model. In Table 1, comparisons
among grid number, DOF number, step number, critical buckling strain, restabilization point (the second bifurcation) and
maximum amplitude are listed with different meshes. One can observe that spectral spatial discretization offers a higher
precision with smaller number of grids as well as DOFs, and the ANM needs fewer predictor steps due to its step length
adaptation and high order series approximation. This results in a significant reduction of computational cost by using ANM-
Spectral algorithm. It is worth mentioning that neither method can obtain a complete isola-center bifurcation curve when
the grids become very sparse, e.g., 400 spectral grids. The first critical buckling strain computed by ANM-Spectral method,
however, is still quite accurate.
4.1. Incompressible materials
Let us begin by considering a highly stretched incompressible rectangular sheet (Li and Healey, 2016), with the aspect
ratio β = 2. Fig. 2 shows a load-displacement bifurcation diagram with three hyperelastic models: SVK, nHk and MR models.
The solid lines stand for our numerical results, while the dots represent those in Li and Healey (2016). We find a good
quantitative agreement between the two. Our models are capable of predicting the “birth and death” of the wrinkles, even
though the SVK model does not agree well with the predictions from the other two more accurate finite strain models (nHk
and MR) whose results match well with each other.
Inspired by theoretical investigations by Cerda and Mahadevan (2003) and Puntel et al. (2011) and numerical studies
by Nayyar et al. (2011), the aspect ratio of the sheet in the following examples is fixed as β = 2.5. Bifurcation diagram
is plotted in Fig. 3, where the solid lines represent our numerical results, while the dash lines stand for those calculated
based on finite element method (shell element S4R) in Abaqus (ABAQUS, 2013). One can observe that the bifurcation points
obtained by the two approaches are highly consistent, even though there exists a slight deviation in magnitude of wrinkling
amplitudes. The results of Fig. 3 can be viewed as a validation of the assumption (8) on the bending stiffness, since the
finite element code does not rely on this assumption. Discussions on material parameters c1 and c2 (MR model) is further
provided in Appendix C. Fig. 4 demonstrates a bifurcation diagram and instability mode evolution for incompressible nHk
model. The ANM takes relatively big step lengths in the pre-buckling stage, and then the step automatically shortens at
the bifurcation (see Fig. 4a). Two bifurcations have been captured. In Fig. 4b, we show the equilibrium configurations and
wrinkled cross-sections of the sheet at overall strains of 0.034, 0.06, 0.1, 0.2 and 0.29, as marked on the bifurcation diagram
4a. Wrinkles first occur at ε ∼ 0.034 and grow as the overall strain increases, peaking at ε ∼ 0.1. Upon further stretching, the
amplitude of wrinkles decreases, and finally vanishes at ε ∼ 0.29. Our computations agree well with Nayyar et al. (2011) and
analytical predictions in Cerda and Mahadevan (2003) and Puntel et al. (2011).
Fig. 11. Phase diagram for incompressible Gent model. The limiting strain, critical buckling strain and restabilization strain are denoted as ε m , ε cr1 and ε cr2 ,
respectively. Three colored regions represent flat (blue), wrinkle (red) and isola-center bifurcation (green), respectively. (For interpretation of the references
to colour in this figure legend, the reader is referred to the web version of this article.)
Soft biological tissue is usually a composite of a compliant matrix and stiff fibers. When the tissue is subjected to small
tension, the fibers are not tight, and the soft matrix carries much of the load so that the tissue remains soft. As the load
increases, the fibers gradually tighten and rotate to the loading direction and thus the tissue stiffens greatly. Another repre-
sentative “limited elastic” material recalls elastomer which is a 3D network of flexible polymer chains. When the elastomer
is under moderate stretches, these chains uncoil and finally become almost straight so that they no longer obey the Gaussian
statistics, which makes the stress-strain curve rise steeply and deviate significantly from the nHk model. Some phenomeno-
logical models such as Gent model and GG model can account for such severe strain-stiffening phenomena.
Bifurcation curves for incompressible Gent model with various Jm are plotted in Fig. 5. One can see that the material
parameter Jm has no apparent influence on the first bifurcation and rather limited effect on the wrinkling amplitude. When
0.1 ≤ Jm ≤ 1.5, the displacement “locks up” at a limiting finite value of stretching strain (less than 0.3) denoted as ε m before
the disappearance of wrinkles. When Jm → ∞, the material recovers the nHk material (see Fig. 4a). Fig. 6 presents bifurcation
diagram and wrinkling pattern evolution for the incompressible Gent model with Jm = 1. The established incompressible
Gent model based on the ANM offers a very fast computing speed to reach the critical point with only ten steps (see Fig. 6a),
with an adaptive reduction of step length at the critical bifurcation. In particular, when the overall tensile strain exceeds
0.2, the stretch reaches the material limit so that the step of ANM accumulates. The bifurcation diagram for incompressible
GG model is given in Supplementary Material. The overall trend remains the same as Gent model, although the second
bifurcation and the wrinkle amplitude in the GG model are slightly different from those in Gent model.
4.2. Compressible materials
Poisson effect that induces transverse compression during the longitudinal stretching is the origin of wrinkling instabil-
ity. When the transverse compression exceeds the critical value, membrane wrinkles. This naturally inspires us to explore
the effect of Poisson’s ratio on the wrinkling and restabilization behavior of uniaxially stretched compressible hyperelastic
sheets. The existence of elastic materials with small Poisson’s ratio ν ≤ 0.45 can be found in porous membranes (Zhu et al.,
2018) and metamaterials (Chen et al., 2017).
Quantitative comparisons with the finite element results calculated by Abaqus (ABAQUS, 2013) for the compressible nHk
model, and the wrinkling evolution with ν = 0.4 are given in Supplementary Material. Our numerical results agree well
with those obtained by FEM, even though a slight difference exists in the magnitude of wrinkling amplitudes. The effects
of Poisson’s ratios and different compressible constitutive models on the post-buckling response are demonstrated in Fig. 7.
Poisson’s ratio has a significant effect on the wrinkling and restabilization. More specifically, smaller Poisson’s ratio makes
smaller instability cycle, i.e., later appearance of wrinkles, smaller amplitude and earlier wrinkling elimination. There exists
a threshold that no wrinkle forms when ν < 0.35. Moreover, constitutive models may lead to quantitative differences in the
nonlinear evolution. The smaller the Poisson’s ratio is, the more significant the differences exist, especially for the second bi-
furcation. This indicates that the material parameter c2 in Eq. (25) has a considerable impact on the post-wrinkling behavior
of compressible materials.
Fig. 12. Distribution and evolution of transverse compressive stresses for compressible Gent model with various Jm but the same Poisson’s ratio ν = 0.4. In
the limit case of Jm → ∞, the Gent model recovers nHk model. (a) Distributions of S22 along the horizontal center line of the sheet at ε = 0.1, showing only
the compressive stress region. (b) Variation of the maximum compressive stresses with increasing stretching. Each point corresponds to one ANM step.
Next we look into Gent model that exhibits a compressible strain-stiffening response, as shown in Fig. 8 where the
maximum wrinkled amplitude vs. tensile strain ε is depicted. Poisson’s ratio is set to be 0.4. It is found that bifurcation
thresholds and maximum deflections vary with different Jm . Larger Jm makes later onset of wrinkles and lower peak, which
differs from the incompressible cases (compare Figs. 5 and 8). When Jm → ∞, the bifurcation diagram recovers compressible
nHk model. By contrast, when Jm ≤ 1.5, deflections “lock up” at a limiting finite value of uniaxial stretch so that the wrinkles
do not disappear eventually.
4.3. Stress analysis
Distribution of transverse compressive stresses is a crucial factor to initiate instability and restabilize wrinkles, which was
early pointed out by Friedl et al. (20 0 0). Here we investigate on effects of Poisson’s ratio and strain stiffening parameter Jm
on the transverse compressive stresses.
The contour of dimensionless transverse stress S22 /E with Poisson’s ratio ν = 0.45 and overall stretching strain ε = 0.15
is presented in Fig. 9a. The colorful part represents the compression region, while the grey one denotes the tensile domain.
Stress localization is observed in the center and particularly along the centerline. To examine Poisson effect, we plot in
Fig. 9b the maximum dimensionless compressive stress (S̄22 = min(S22 (x1 , 0 ))/E) in function of overall stretching strain with
a variety of Poisson’s ratios for MR model. When Poisson’s ratio declines from 0.5 to 0.35, the transverse stress decreases
accordingly, which explains why the wrinkles occur later and more difficult, while the restabilization tends to be earlier and
easier (see Fig. 7).
Strain stiffening effect is investigated for both incompressible and compressible materials. Evolutions of the longitudinal
and transverse stresses for incompressible Gent model are shown in Fig. 10. For a given value of Jm (except Jm = ∞), the
S̄11 − ε curves tend to be vertical at a limiting strain which increases with Jm (see Fig. 10a). However, transverse stresses
remain almost the same with varying Jm below the limiting strain (see Fig. 10b). Therefore, the material parameter Jm has
limited effect on the critical thresholds and wrinkling amplitude for the incompressible case. To quantitatively reveal the ef-
fect of Jm on wrinkling and restabilization behavior, a phase diagram is drawn in Fig. 11. The limiting strain ε m varies almost
linearly with Jm . When Jm < 0.07, the corresponding ε m is below the first critical threshold so that the strain-stiffening mem-
brane remains flat. When 0.07 ≤ Jm < 1.89, the corresponding ε m is smaller than the second bifurcation threshold but larger
than the first one, and thus the membrane wrinkles. While Jm > 1.89, a complete wrinkling and restabilization response
with an isola-center bifurcation is observed. Unlike the incompressible materials, the parameter Jm plays an important role
in the critical thresholds and wrinkling amplitude in the compressible case (see Fig. 8). As shown in Figs. 12a and 12b,
the distribution and evolution of transverse stresses are calculated using Gent model with ν = 0.4. Larger Jm makes smaller
transverse compressive stress, leading to later onset of wrinkles and lower peak of deflection (see Fig. 8). In addition, the
second bifurcation points are clearly observed when Jm ≥ 2 (Fig. 12b).
5. Conclusion
Stretch-induced wrinkling and restabilization of a hyperelastic membrane is an interesting phenomenon increasingly at-
tracting attention lately. Here we have developed a modeling and resolution framework for this strongly nonlinear phe-
nomenon in both compressible and incompressible hyperelastic sheets upon highly stretching. This hyperelastic thin plate
formulations are completely described by the midsurface metric, while the thickness contraction is statically condensed us-
ing the plane stress condition or incompressibility constraint. We have systematically derived 2D hyperelastic constitutive
models for both compressible and incompressible materials from general 3D strain energy density functions, e.g., SVK, nHk,
MR, Gent and GG models. Moreover, a novel, efficient and robust numerical algorithm through coupling of ANM and a spec-
tral method is proposed to solve large deformations of hyperelastic sheets, the ANM being a path-following continuation
technique that is efficient to solve bifurcation problems and the spectral collocation method permitting exponential con-
vergence to greatly reduce the number of unknowns. We have also demonstrated the detailed derivation of the proposed
asymptotic formulations incorporating with spectral spatial discretization approach for nonlinear problem resolution. Com-
putational cost can be greatly reduced through a combination of the two methods (ANM & Spectral) as compared with
finite element method. The accuracy and efficiency of the proposed framework were verified by comparing with several
benchmarks.
For incompressible materials, different constitutive laws mainly affect restabilization points and wrinkling amplitudes
quantitatively. For compressible materials, Poisson’s ratio plays a significant role in the wrinkling and restabilization re-
sponse. Smaller the Poisson’s ratio is, the less likely the film would wrinkle, which can be explained by the decreasing
transverse compressive stresses with respect to the reducing Poisson’s ratio. Besides, other material parameters, e.g., c2 in
Eq. (25) and Jm in Eq. (29), can affect the two bifurcation thresholds and wrinkling amplitudes. Moreover, Gent and GG
material laws were taken into account to model severe strain-stiffening phenomena. For example, when a soft biological
material is subjected to an increasing loading, its fibers gradually tighten and rotate to the loading direction, which makes
the material stiffen significantly. Another representative strain-stiffening material is elastomer that is a 3D network of poly-
mer chains. These chains uncoil and finally become nearly straight when the elastomer is under moderate stretches, which
makes the stress-strain curve rise steeply. We find that wrinkles of the strain-stiffening membrane would not eventually
disappear when the limiting extensibility parameter Jm is relatively small.
Acknowledgements
Supports from the National Natural Science Foundation of China (Grants No. 11461161008, No. 11602058, No. 11772094
and No. 11872150), Shanghai Education Development Foundation and Shanghai Municipal Education Commission (Shanghai
Chenguang Program, Grant No. 16CG01), State Key Laboratory for Strength and Vibration of Mechanical Structures (Grant
No. SV2018-KF-17), start-up fund from Fudan University, and National Key Research and Development Program of China
(2016YFB0700103) are acknowledged. FX and MPF thank for the financial support from the French National Research Agency
ANR (LabEx DAMAS, Grant No. ANR-11-LABX-0 0 08-01). MPF is grateful for support from the Fudan Fellowship.
Appendix A. Asymptotic formulations for incompressible Gent model
The same approach as in Section 3.1 is adopted to rewrite the incompressible Gent model in a quadratic form, which
allows obtaining the following system:
⎧
⎪Cs = FT · F,
⎪
⎪
⎪C−1
⎪ s · Cs = Is ,
⎪
⎪
⎪ 2
⎨ = C11 · C22 − C12 ,
A
B = C11 + C22 ,
(75)
⎪
⎪C33 = 1/A,
⎪
⎪T = 1/(Jm − C33 − B + 3 ),
⎪
⎪
⎪ p = μJmC33 T ,
⎪
⎩
Ss = μJm T Is − pC−1
s .
The asymptotic expansion at order n of Eq. (75) gives the following linear problems:
⎧ n−1
⎪
⎪
⎪ (Cs )n = FT0 · Fn + FTn · F0 + FTr · Fn−r ,
⎪
⎪
⎪
⎪ r=1
⎪
⎪ −1 −1 −1 −1 n−1
⎪
⎪
⎪
⎪ C = − C · ( C s ) · C − C · (Cs )r · C−1 ,
⎪
⎪
s n s 0 n s 0 s 0 s n−r
⎪
⎪ r=1
⎪
⎪An = (C11 )0 · (C22 )n + (C11 )n · (C22 )0 − 2(C12 )0 (C12 )n
⎪
⎪
⎪
⎪
n−1

⎪
⎪ + (C11 )r · (C22 )n−r − (C12 )r · (C12 )n−r ,
⎪
⎪
⎪
⎪ r=1
⎪
⎪Bn = (C11 )n + (C22 )n ,
⎪
⎪
⎨ (C ) · An (n − r ) (C33 )r · An−r + (C33 )n−r · Ar
n−1
(C33 )n = − 33 0 − , (76)
⎪ A0 nA0
⎪
⎪ r=1
⎪
⎪
n−1
(n − r ) Tr · Bn−r + (C33 )n−r + Tn−r · [Br + (C33 )r ]
⎪
⎪T =
⎪
⎪
n
n[Jm − (C33 )0 − B0 + 3]
⎪
⎪ r=1
⎪
⎪ T [ B + ( C )
33 n ]
⎪
⎪ +
0 n
,
⎪
⎪ J −
m ( C ) − B0 + 3
⎪
⎪
33 0
⎪
⎪ n−1
⎪
⎪ pn = μJm T0 (C33 )n + Tn (C33 )0 + Tr (C33 )n−r ,
⎪
⎪
⎪
⎪
⎪
⎪
r=1
⎪
⎪ −1 −1 n−1 −1
⎪
⎪
⎩(Ss )n = μJm Tn Is − p0 Cs n + pn Cs 0 + pr Cs
n−r
.
r=1
Appendix B. “Basis recombination” for a biharmonic problem
We consider an example of a 1D biharmonic problem:
u,xxxx = f (x ), −1 < x < 1, u(±1 ) = u,x (±1 ) = 0. (77)

Let p(x) be an interpolant of the function u(x), namely,

p xj = u xj , p(±1 ) = p,x (±1 ) = 0, (78)
in which {x j }M
j=1
is a set of Chebyshev nodes (see Eq. (64)). In order to simplify the complex BCs, we set

p(x ) = 1 − x2 q(x ), −1 < x < 1. (79)
Clearly, the polynomial q(x) with q(±1 ) = 0 corresponds to the interpolant p(x) with p(±1 ) = p,x (±1 ) = 0. Thus, one can
obtain the fourth derivative of p(x) in the following form:

p,xxxx = 1 − x2 q,xxxx − 8xq,xxx − 12q,xx . (80)
Then the spectral biharmonic operator is

L4 = diag 1 − x2j Dx(4) − 8diag x j Dx(3) − 12Dx(2) × diag 1/ 1 − x2j , (81)
where Dx(l ) is the lth-order Chebyshev differentiation matrix and diag(x) returns a square diagonal matrix with the elements
of vector x on the main diagonal.
Appendix C. Effect of material parameters for MR model
Effect of material parameter c2 in MR model (see Eq. (35)) on post-wrinkling behavior of a stretched membrane are
explored in Fig. C1a. It shows that larger parameter c2 makes later disappearance of wrinkles and higher deflections, while
this parameter has no apparent influence on the initial wrinkling stage. The same conclusion is made for compressible MR
model as well (see Fig. C1b).
Fig. C1. Bifurcation diagram of MR model with varying material parameters. (a) Incompressible MR model. (b) Compressible MR model (Poisson’s ratio
ν = 0.4).
Supplementary material
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.jmps.2018.11.005.
References
ABAQUS, 2013. ABAQUS Analysis User’s Manual, version 6.13.

Abdelkhalek, S., Zahrouni, H., Legrand, N., Potier-Ferry, M., 2015. Post-buckling modeling for strips under tension and residual stresses using asymptotic
numerical method. Int. J. Mech. Sci. 104, 126–137.
Alioli, M., Masarati, P., Morandini, M., Albertani, R., Carpenter, T., 2017. Modeling effects of membrane tension on dynamic stall for thin membrane wings.
Aerosp. Sci. Technol. 69, 419–431.
Attipou, K., Hu, H., Mohri, F., Potier-Ferry, M., Belouettar, S., 2015. Thermal wrinkling of thin membranes using a Fourier-related double scale approach. Thin
Wall. Struct. 94, 532–544.
Audoly, B., Boudaoud, A., 2008. Buckling of a stiff film bound to a compliant substrate–Part I: Formulation, linear stability of cylindrical patterns, secondary
bifurcations. J. Mech. Phys. Solids 56, 2401–2421.
Boyd, J.P., 2001. Chebyshev and Fourier Spectral Methods. Dover Publications, New York.
Canuto, C., Hussaini, M.Y., Quarteroni, A., Zang, T.A., 1988. Spectral Methods in Fluid Dynamics. Springer-Verlag, Berlin.
Cerda, E., Mahadevan, L., 2003. Geometry and physics of wrinkling. Phys. Rev. Lett. 90, 074302–1–074302–4.
Cerda, E., Ravi-Chandar, K., Mahadevan, L., 2002. Wrinkling of an elastic sheet under tension. Nature 419, 579–580.
Chen, Y., Li, T., Scarpa, F., Wang, L., 2017. Lattice metamaterials with mechanically tunable Poisson’s ratio for vibration control. Phys. Rev. Lett. 7,
024012–1–024012–11.
Ciarlet, P.O., 1988. Mathematical Elasticity, Vol. I.. Three-Dimensional Elasticity. North-Holland, Amsterdam.
Ciarlet, P.O., Geymonat, G., 1982. Sur les lois de comportement en élasticité non linéaire compressible. C. R. Acad. Sc. 295, 423–426.
Cochelin, B., Damil, N., Potier-Ferry, M., 1994. Asymptotic-numerical methods and Padé approximants for non-linear elastic structures. Int. J. Numer. Methods
Eng. 37, 1187–1213.
Cochelin, B., Damil, N., Potier-Ferry, M., 2007. Méthode Asymptotique Numérique. Hermès Science Publications, Paris.
Damil, N., Potier-Ferry, M., 1990. A new method to compute perturbed bifurcation: Application to the buckling of imperfect elastic structures. Int. J. Eng.
Sci. 26, 943–957.
Dervaux, J., Ben Amar, M., 2008. Morphogenesis of growing soft tissues. Phys. Rev. Lett. 101, 068101–068104.
Destrade, M., Fu, Y., Nobili, A., 2016. Edge wrinkling in elastically supported pre-stressed incompressible isotropic plates. Proc. R. Soc. A 472,
20160410–1–20160410–17.
Elsabbagh, A., 2015. Nonlinear finite element model for the analysis of axisymmetric inflatable beams. Thin Wall. Struct. 96, 307–313.
Fischer, F.D., Rammerstorfer, F.G., Friedl, N., Wieser, W., 20 0 0. Buckling phenomena related to rolling and levelling of sheet metal. Int. J. Mech. Sci. 42,
1887–1910.
Fornberg, B., 1996. A Practical Guide to Pseudospectral Methods. Cambridge University Press, New York.
Friedl, N., Rammerstorfer, F.G., Fischer, F.D., 20 0 0. Buckling of stretched strips. Comput. Struct. 78, 185–190.
Fu, B., Sperber, E., Eke, F., 2016. Solar sail technology–A state of the art review. Prog. Aerosp. Sci. 86, 1–19.
Fu, C., Xu, F., Huo, Y., 2018. Photo-controlled patterned wrinkling of liquid crystalline polymer films on compliant substrates. Int. J. Solids Struct. 132–133,
264–277.
Funaro, D., 1992. Polynomial Approximation of Differential Equations. Springer Verlag, New York.
Gent, A.N., 1996. A new constitutive relation for rubber. Rubber Chem. Technol. 69, 59–61.
Gent, A.N., Thomas, A.G., 1958. Forms for the stored (strain) energy function for vulcanized rubber. J. Polym. Sci. 28, 625–628.
Healey, T.J., Li, Q., Cheng, R.B., 2013. Wrinkling behavior of highly stretched rectangular elastic films via parametric global bifurcation. J. Nonlinear Sci. 23,
777–805.
van der Heijden, A.M.A., 2009. W.T. Koiter’s Elastic Stability of Solids and Structures. Cambridge University Press, Cambridge.
Holzapfel, G.A., 2001. Biomechanics of soft tissue. In: Lemaitre, J. (Ed.), Handbook of Materials Behavior Models, Volume III, Multiphysics Behaviors. Aca-
demic Press, New York.
Huang, X., Li, B., Hong, W., Cao, Y., Feng, X.Q., 2016. Effects of tension-compression asymmetry on the surface wrinkling of film-substrate systems. J. Mech.
Phys. Solids 94, 88–104.
Huang, Z.Y., Hong, W., Suo, Z., 2005. Nonlinear analyses of wrinkles in a film bonded to a compliant substrate. J. Mech. Phys. Solids 53, 2101–2118.
Jacques, N., Potier-Ferry, M., 2005. On mode localisation in tensile plate buckling. C.R. Mec. 333, 804–809.
Karageorghis, A., 1993. Conforming Chebyshev spectral methods for Poisson problems in rectangular domains. J. Sci. Comput. 8, 123–133.
Kim, T.Y., Puntel, E., Fried, E., 2012. Numerical study of the wrinkling of a stretched thin sheet. Int. J. Solids Struct. 49, 771–782.
Krivoshapko, S.N., 2017. Thin sheet metal suspended roof structures. Thin Wall. Struct. 119, 629–634.
Lecieux, Y., Bouzidi, R., 2010. Experimental analysis on membrane wrinkling under biaxial load – Comparison with bifurcation analysis. Int. J. Solids Struct.
47, 2459–2475.
Lecieux, Y., Bouzidi, R., 2012. Numerical wrinkling prediction of thin hyperelastic structures by direct energy minimization. Adv. Eng. Softw. 50, 57–68.
Li, Q., Healey, T.J., 2016. Stability boundaries for wrinkling in highly stretched elastic sheets. J. Mech. Phys. Solids 97, 260–274.
Li, Y., 2018. Roll up your sleeves. Nat. Phys. 14, 534.
Luo, Y., Xing, J., Niu, Y., Li, M., Kang, Z., 2017. Wrinkle-free design of thin membrane structures using stress-based topology optimization. J. Mech. Phys.
Solids 102, 277–293.
Nayyar, V., Ravi-Chandar, K., Huang, R., 2011. Stretch-induced stress patterns and wrinkles in hyperelastic thin sheets. Int. J. Solids Struct. 48, 3471–3483.
Nayyar, V., Ravi-Chandar, K., Huang, R., 2014. Stretch-induced wrinkling of polyethylene thin sheets: Experiments and modeling. Int. J. Solids Struct. 51,
1847–1858.
Nezamabadi, S., Zahrouni, H., Yvonnet, J., 2011. Solving hyperelastic material problems by asymptotic numerical method. Comput. Mech. 47, 77–92.
Nguyen, Q.T., Thomas, J.C., Le Van, A., 2015. Inflation and bending of an orthotropic inflatable beam. Thin Wall. Struct. 88, 129–144.
Ogden, R.W., 1985. Non-linear elastic deformations. Z. Angew. Math. Mech. 65, 404.
Ogden, R.W., Saccomandi, G., Sgura, I., 2004. Fitting hyperelastic models to experimental data. Comput. Mech. 34, 484–502.
Plucinsky, P., Bhattacharya, K., 2017. Microstructure-enabled control of wrinkling in nematic elastomer sheets. J. Mech. Phys. Solids 102, 125–150.
Pucci, E., Saccomandi, G., 2002. A note on the Gent model for rubber-like materials. Rubber Chem. Technol. 75, 839–851.
Puntel, E., Deseri, L., Fried, E., 2011. Wrinkling of a stretched thin sheet. J. Elast. 105, 137–170.
Rammerstorfer, F.G., 2018. Buckling of elastic structures under tensile loads. Acta Mech. 229, 881–900.
Sharon, E., Marder, M., Swinney, H.L., 2004. Leaves, flowers and garbage bags: Making waves. Am. Sci. 92, 254–261.
Silvestre, N., 2016. Wrinkling of stretched thin sheets: Is restrained Poisson’s effect the sole cause? Eng. Struct. 106, 195–208.
Simo, J.C., Hughes, T.J.R., 1998. Computational Inelasticity. Springer, New York.
Sipos, A.A., Fehér, E., 2016. Disappearance of stretch-induced wrinkles of thin sheets: A study of orthotropic film. Int. J. Solids Struct. 97–98, 275–283.
Steigmann, D.J., 2007. Thin-plate theory for large elastic deformations. Int. J. Nonlinear Mech. 42, 233–240.
Taylor, M., Bertoldi, K., Steigmann, D.J., 2014. Spatial resolution of wrinkle patterns in thin elastic sheets at finite strain. J. Mech. Phys. Solids 62, 163–180.
Taylor, M., Davidovitch, B., Qiu, Z., Bertoldi, K., 2015. A comparative analysis of numerical approaches to the mechanics of elastic sheets. J. Mech. Phys.
Solids 79, 92–107.
Trefethen, L.N., 20 0 0. Spectral Methods in Matlab. SIAM, Philadelphia.
Vandeparre, H., Pineirua, M., Brau, F., Roman, B., Bico, J., Gay, C., Bao, W.Z., Lau, C.N., Reis, P.M., Damman, P., 2011. Wrinkling hierarchy in constrained thin
sheets from suspended graphene to curtains. Phys. Rev. Lett. 106, 224301–1–224301–4.
Weideman, J.A.C., Reddy, S.C., 20 0 0. A MATLAB differentiation matrix suite. ACM Trans. Math. Softw. 26, 465–519.
Wong, Y.W., Pellegrino, S., 2006a. Wrinkled membranes. Part I: Experiments. J. Mech. Mater. Struct. 1, 3–25.
Wong, Y.W., Pellegrino, S., 2006b. Wrinkled membranes. Part II: Analytical models. J. Mech. Mater. Struct. 1, 27–61.
Wong, Y.W., Pellegrino, S., 2006c. Wrinkled membranes. Part III: Numerical simulations. J. Mech. Mater. Struct. 1, 63–95.
Xu, F., Koutsawa, Y., Potier-Ferry, M., Belouettar, S., 2015a. Instabilities in thin films on hyperelastic substrates by 3D finite elements. Int. J. Solids Struct.
69–70, 71–85.
Xu, F., Potier-Ferry, M., 2016. A multi-scale modeling framework for instabilities of film/substrate systems. J. Mech. Phys. Solids 86, 150–172.
Xu, F., Potier-Ferry, M., Belouettar, S., Hu, H., 2015b. Multiple bifurcations in wrinkling analysis of thin films on compliant substrates. Int. J. Nonlinear Mech.
76, 203–222.
Yan, D., Zhang, K., Peng, F., Hu, G., 2014. Tailoring the wrinkle pattern of a microstructured membrane. Appl. Phys. Lett. 105, 071905–1–071905–4.
Yang, Y., Dai, H.H., Xu, F., Potier-Ferry, M., 2018. Pattern transitions in a soft cylindrical shell. Phys. Rev. Lett. 120, 215503–1–215503–5.
Yeoh, O.H., 1997. Hyperelastic material models for finite element analysis of rubber. J. Nat. Rubber Res. 12, 142–153.
Yoon, H., Ghosh, A., Han, J.Y., Sung, S.H., Lee, W.B., Char, K., 2012. Nanowalls: Lateral buckling of high aspect ratio janus nanowalls. Adv. Funct. Mater. 22,
3530.
Zhao, Y., Shao, Z.C., Li, G.Y., Zheng, Y., Zhang, W.Y., Li, B., Cao, Y., Feng, X.Q., 2017. Edge wrinkling of a soft ridge with gradient thickness. Appl. Phys. Lett.
110, 231604–1–231604–5.
Zheng, L., 2009. Wrinkling of Dielectric Elastomer Membranes. California Institute of Technology, Pasadena, USA. Ph.D. thesis.
Zhu, J., Zhang, X., Wierzbicki, T., 2018. Stretch-induced wrinkling of highly orthotropic thin film. Int. J. Solids Struct. 139–140, 238–249.

A modeling and resolution framework for wrinkling in hyperelastic sheets at finite membrane strain

Uploaded by

Copyright:

Available Formats

A modeling and resolution framework for wrinkling in hyperelastic sheets at finite membrane strain

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A modeling and resolution framework for wrinkling in hyperelastic sheets at finite membrane strain

Uploaded by

Copyright:

Available Formats

Journal of the Mechanics and Physics of Solids 124 (2019) 446–470

Contents lists available at ScienceDirect

Journal of the Mechanics and Physics of Solids

A modeling and resolution framework for wrinkling in

2. Finite strain models for hyperelastic thin sheets

2.1. Equilibrium equations and boundary conditions

f(x ) := xα eα + uα (x )eα + w(x )e3 , (1)

Wtot (Cs , Ks ) = Wm (Cs ) + Wb (Ks ), (6)

where the 2D membrane strain energy is given by

Wm (Cs ) = h(Cs ), (7)

2.2. Constitutive equations

tensor C. Hence, the strain energy function is formulated as

 =  [I1 (C ), I2 (C ), I3 (C )], (13)

2.2.1. Compressible materials

2.2.2. Incompressible materials

3.1. Asymptotic formulations for hyperelasticity

⎪(S ) = 2[c B + c (C ) ]I − 2c (C ) + 2cJ2 − d C−1

The order n of this system (54) reads

3.2. Spectral collocation method

The derivative operator can be represented by a differentiation matrix, namely

D˜ 1(l ) = D1(l ) I2 , D˜ 2(l ) = I1 D2(l ) , (71)

∂α(l ) v = D˜ α(l ) · v. (72)

4. Results and discussion

4.1. Incompressible materials

4.2. Compressible materials

4.3. Stress analysis

Appendix A. Asymptotic formulations for incompressible Gent model

Appendix B. “Basis recombination” for a biharmonic problem

We consider an example of a 1D biharmonic problem:

u,xxxx = f (x ), −1 < x < 1, u(±1 ) = u,x (±1 ) = 0. (77)

Appendix C. Effect of material parameters for MR model

ABAQUS, 2013. ABAQUS Analysis User’s Manual, version 6.13.

You might also like

Wm (Cs ) = h(Cs ), (7)

= [I1 (C ), I2 (C ), I3 (C )], (13)