Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Finite Elements in Analysis and Design

journal homepage: www.elsevier.com/locate/finel

Co-rotational finite element formulation used in the Koiter–Newton

method for nonlinear buckling analyses
Ke Liang a,b,n, Martin Ruess c, Mostafa Abdalla b
a
Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing 100094, China
b
Aerospace Structures and Computational Mechanics, Delft University of Technology, Kluyverweg 1, 2629 HS Delft, The Netherlands
c
School of Engineering, University of Glasgow, Rankine Building, Oakfield Avenue, Glasgow G12 8LT, United Kingdom

art ic l e i nf o a b s t r a c t

Article history: The Koiter–Newton approach is a novel reduced order modeling technique for buckling analysis of
Received 27 July 2015 geometrically nonlinear structures. The load carrying capability of the structure is achieved by tracing
Received in revised form the entire equilibrium path in a stepwise manner. At each step a reduced order model generated from
15 March 2016
Koiter's asymptotic expansion provides a nonlinear prediction for the full model, corrected by a few
Accepted 31 March 2016
Newton steps. The construction of the reduced order model requires derivatives of the strain energy with
respect to the degrees of freedom up to the fourth order, which is two orders more than traditionally
Keywords: needed for a Newton based nonlinear finite element technique. In this paper we adopt the co-rotational
Koiter–Newton formulation to facilitate these complex differentiations. We extend existing co-rotational beam and shell
Geometrically nonlinear structures
element formulations to make them applicable for the high order derivatives of the strain energy. The
Co-rotational formulation
geometrical nonlinearities are taken into account using derivatives of the local co-rotational frame with
High order derivatives
Nonlinear rotation matrix respect to global degrees of freedom. This is done outside the standard element routines and is thus
independent of the element type. We utilize three configurations and the nonlinear rotation matrix to
describe finite rotations of the shell accurately, and profit from the automatic differentiation technique to
optimize the programming of high order derivatives. The performance of the proposed approach using
the co-rotational formulation is demonstrated using benchmark examples of isotropic and laminated
composite structures.
& 2016 Elsevier B.V. All rights reserved.

1. Introduction accurate enough the propensity of a structure to buckle and to

provide the structure's initial postbuckling behavior. In Koiter's
The analysis of geometrically nonlinear responses of structures perturbation technique, his asymptotic expansion is used only
is important for determining their load carrying capability, espe- once at the bifurcation point to construct the reduced order model
cially in the case of buckling where a limit point or a bifurcation (ROM) which can present the initial postbuckling path of the
point exists [1–8]. Utilizing the expanding computational power of structure. Hence, the traditional Koiter reduction method is valid
modern computers nonlinear finite element analysis (FEA) has asymptotically in the neighborhood of the bifurcation point. The
become the standard technique used to obtain the nonlinear majority of research done in this field applies an expansion of the
response of complex structures, however, the repeated analyses displacement field up to the second order which is usually accu-
that are needed for FEA in the design loop are still computationally rate enough to capture the initial postbuckling response of struc-
demanding. Thus, reduced order techniques that can be used to tures [16,21–24]. Damil and Potier-Ferry [25] have adopted higher
reduce a problem's size significantly are attractive. order terms to increase the range of validity of the perturbation
Koiter reduction methods [9–17] use the Koiter's famous per- expansion further. Increasing the number of higher order terms in
turbation technique [18] to reduce the number of degrees of the displacement provides a wider range of validity. Yet, the
freedom in the finite element model. The main advantage of Koi- reduced order model obtained from a single perturbation expan-
ter's theory [2,18–20] is the capability to predict quickly but sion still has a limited range of validity that cannot be determined
a priori. In addition, the prebuckling state is assumed to be linear
n in most of Koiter reduction methods used in [10,15,16,24,26,27],
Corresponding author at: Qian Xuesen Laboratory of Space Technology, China
Academy of Space Technology, Beijing 100094, China. Tel./fax: þ86 01068113099. since these methods are based on an alternative procedure pro-
E-mail addresses: [email protected], [email protected] (K. Liang). posed in Budiansky and Hutchinson [28], in which the assumption

http://dx.doi.org/10.1016/j.finel.2016.03.006
0168-874X/& 2016 Elsevier B.V. All rights reserved.

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
2 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Nomenclature θ angle between nominal configuration and x-axis

θ0 angle between reference configuration and x-axis
List of symbols
Δθ angle between reference and nominal configurations
ra , rb position vectors of nodes a and b in reference
N number of degrees of freedom in the full finite configuration
element model da , db position vectors of nodes a and b in nominal
1þ m number of degrees of freedom in the reduced configuration
order model κ, A, U curvature, area of cross section and strain energy
m number of closely spaced buckling modes f, b
f , KL global and local internal loads, linear stiffness
f ext , f int , f r external load, internal load and residual force    
x0a ; y0a , x0b ; y0b coordinates of nodes a and b in initial
q, q0 displacements in current and nominal configurations configuration
u relative displacement from nominal configuration to
current configuration
Symbols used in co-rotational shell element
λ, λ0 load parameters at current and nominal
configurations
a, a ¼ 1; 2; 3; 4 node number
Δλ increment of load parameter from nominal to current
Tg global coordinate system
configurations
X; Y; Z axes in Tg
L, Q, C linear, quadratic and cubic forms in expansion of
T0 , T co-rotational frames in reference and nominal
equilibrium equations
configurations
L, Q, C 2D, 3D and 4D tensors of L, Q, C 0 0 0
d1 ; d2 ; d3 axes in T0
F, ϕ, f α load matrix, load amplitude and sub-loads
d1 ; d2 ; d3 axes in T
ξ, uα , uαβ , uαβγ generalized displacement and first to third
F, R, U deformation gradient, rotation matrix, tension matrix
order displacement fields
q displacement from reference to current configurations
L, Q, C linear, quadratic and cubic forms in reduced
qn relative displacement from reference to nominal
order model
configurations
L, Q, C 2D, 3D and 4D tensors of L, Q, C
u relative displacement from nominal to current
Kt tangent stiffness matrix
configurations
Eα α-th unit vector
ta translation part of each node in q
L α , Q αβ , C αβγδ column vectors of tensors L, Q, components of
tensor C
θa , θna , ϑa rotation parts of each node in q, qn and u
θ~ , ϑ~
n
q;λ linear displacement of the structure under the exter- a a anti-symmetric matrices formed by three components
in θa and ϑa
n
nal load f ext
b
q local degrees of freedom
Symbols used in co-rotational beam element b b
ta, θ b
translation and rotation parts of each node in q
a
~b
θa anti-symmetric matrix formed by three components
a, b node number b
in θ a
(x,y) global coordinate system
r0a position vector of each node in reference
ðe0x ; e0y Þ co-rotational frame in reference configuration
configuration
ðex ; ey Þ co-rotational frame in nominal configurations
b I 3  3 identity matrix
q, q global and local degrees of freedom
Ra rotation matrix of each node
ua ; va ; θa ; ub ; vb ; θb components in global degrees of freedom q ba
b ;θ b components in local degrees of freedom q R local rotation matrix of each node
bb ; θ
u b
a b
εb, C local strain and material matrix
l, ln initial and current length of beam

is made that the prebuckling is linear. In reality, this linear the numerical effort spent in repeated solution steps are fluid
assumption for the prebuckling state will often overestimate the mechanics [39–42], aeroelasticity [43,44] or optimization [45,46]
buckling load of an important class of engineering problems for to mention a few. An exhaustive overview about reduced order
which the prebuckling is obviously nonlinear. Cohen [26] and Fitch modeling techniques and strategies, about fields of application and
[27], and later Arbocz and Hol [29,30] have derived the mod- the analysis of convergence behavior and error-control is provided
ifications necessary to make Budiansky and Hutchinson's work e.g. in Chinesta et al. [47] or Quaterioni and Rozza [48].
[28] include prebuckling nonlinearity. Recently, Rahman [13,14,17] The aforementioned Koiter reduction methods adopt the
and Zagari [11,12] have made use of Arbocz and Hol's [29,30] solution of the reduced order model as a predictor without using
derivations within a finite element context to consider the non- any correction step based on the full model [40,41,49–52]. Another
linearity of the prebuckling of a structure. family of model reduction techniques combines the prediction
Apart from Koiter reduction methods which are mainly used in stage together with a correction phase [53–55]. These predictor-
the static buckling analysis of structures, the idea of reduced order corrector methods [56–58] are commonly used to trace the non-
models has been studied intensively in the past across various linear equilibrium path of structures. Recently, a novel approach,
disciplines following different approaches. In particular, projection the Koiter–Newton method, has been proposed for the numerical
methods based on Krylov subspace algorithms have a tradition solution of a class of elastic nonlinear structural analysis problems
due to the numerically efficient and relatively stable generation of [59–61]. The range of validity of this approach is not limited to the
orthogonal projection bases [31–33] often with application in small range near the bifurcation point, since Koiter's asymptotic
structural vibration [20,34,35] and dynamics [16,36–38]. Other expansion is applied from the beginning of the equilibrium path
disciplines where reduced order models may significantly alleviate rather than only at the bifurcation point. In a series of expansion

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

steps, a reduced order model is constructed based on Koiter's is adopted and used to achieve equilibrium equations in a third
asymptotic expansion and used to predict the nonlinear response order form. We believe that our approach profits essentially from
of the full model. A Newton method is used to correct the pre- the co-rotational formulation for various reasons:
diction if the accuracy of the reduced order model has been lost. In
the next expansion step, the reduced order model is updated to  the element independent co-rotational frame is computation-
reflect any changes in structural stiffness and load distribution. ally more accurate for the case of large displacements and/or
Structural nonlinearities are considered automatically using this large rotations in a geometrically nonlinear analysis, as the rigid
multi-expansion strategy. The method favorably combines ideas part is purged from the total motion.
from Koiter's initial post-buckling analysis with Newton arc-length  the co-rotational formulation addresses the accuracy and con-
methods to obtain an algorithm that is accurate over the entire vergence problems which often occur in the implementation of
equilibrium path and efficient in the presence of buckling and/or a Koiter perturbation approach in a finite element context
imperfection sensitivity. The Newton-based corrector phase of the [11,14,75].
presented approach uses a few factorizations of the full model.  the proposed Koiter–Newton approach requires derivatives of
Nevertheless, the high stability of the method even for severe test the strain energy with respect to degrees of freedom up to the
cases justifies the increased numerical effort of full model on-line fourth order. This is two orders more than traditionally needed
calculations [47] and still shows a superior numerical efficiency for a Newton's method. Nonlinear elements based on the ele-
compared to standard path-follower methods. In [59] we pre- ment independent co-rotational frame can facilitate the differ-
sented the basic idea of the Koiter–Newton approach and entiations [76,77]. The full nonlinear kinematics are taken into
demonstrated the numerical efficiency in terms of reliability and account when constructing the reduced order model in
accuracy with a number of examples of moderate deformation. A this way.
specialization of the Koiter–Newton method for thin plates and  due to the element independent characteristics of the co-
shells with negligible in-plane rotations was introduced in [60] rotational frame, any type of linear elements in the current
considering von Kármán kinematics which neglect higher order element library can be used for the Koiter–Newton approach,
terms in the strain tensor components thus reducing the compu- and geometric nonlinearities are taken into account by deriva-
tational complexity of the approach. tives of the local co-rotational frame with respect to the
With this contribution we extend the Koiter–Newton method global frame.
to large deformation analysis on the basis of a co-rotational for-
mulation. This allows us to cover a much broader field of appli- Based on these characteristics we present the complete for-
cations including solids and structures with a geometrically highly mulation for a two-dimensional co-rotational beam element and a
nonlinear response [62–64]. Structures usually undergo a large co-rotational shell element for large deformation analysis includ-
motion in the presence of buckling, hence geometrically nonlinear ing the required equilibrium equations of a third order form. The
mechanics is usually employed in a buckling analysis [4,65–67]. formulation considers a large rotation theory to handle the non-
Currently, three Lagrangian kinematic are used for finite element commutativity of the finite rotations involved [78–83]. We
analysis of geometrically nonlinear structures [66,68,63]: (1) a demonstrate the good performance achieved with the developed
total Lagrangian (TL) formulation, (2) an updated Lagrangian (UL) method with several benchmark tests. We show that our approach
formulation, and (3) a co-rotational formulation. The latter was the can exactly re-produce large deformation and buckling analysis
last to be developed [69,70], and the co-rotational formulation results of standard path-follower or Newton-type methods with
which is adopted in this paper has recently received considerable far less effort and with equivalent reliability and accuracy. With a
attention. In the beginning, the co-rotational formulation was not comparison between sophisticated solution schemes from litera-
designed to be implemented in a general finite element for the ture and our method we illustrate the competitiveness of our
nonlinear analysis of structures, since the application of this for- approach.
mulation is limited to an a priori kinematics assumption [66,71] The paper is organized as follows: a brief review of the Koiter–
which allows arbitrary large displacements and rotations but Newton approach is given in Section 2. The finite element for-
which is limited to small deformations [66]. To make the co- mulation of co-rotational beam and shell elements are presented
rotational formulation more widely applicable, the idea of multiple in Sections 3 and 4, respectively. Numerical examples used to
frames [72] has been proposed for the CR formulation. Instead of demonstrate the success of the method are provided in Section 5.
one co-rotational frame for a whole structure, each element of a We summarize the paper and draw conclusions in Section 6.
structure is given on independent frame called a co-rotational
element frame. This modification helps to ensure the above
assumption, achieved by making the element deformational dis- 2. Review of the Koiter–Newton approach
placements and rotations small with respect to the co-rotational
element frame. The application of the co-rotation formulation in Let us start with a brief review of the Koiter–Newton approach
Koiter's analysis has been studied before by some researchers. outlining the essential strategy behind this novel path-tracing
Pacoste and Eriksson [73] have investigated the possibility of using technique.
co-rotational formulation. Garcea [11] has studied implementing
Koiter's asymptotic approach in nonlinear structural finite element 2.1. Basic algorithmic design
models which are based on a co-rotational frame. Recently, Zagari
[12,74] has presented a co-rotational formulation, suitable for a Similar to classical path-following techniques the Koiter–
nonlinear, fourth order accurate asymptotic postbuckling analysis Newton method is based on a step by step procedure to trace the
of shell structures, which exploits three dimensional finite rota- complete equilibrium path of the deforming structure. The unique
tions. However, the above applications are limited to some tradi- properties and algorithmic differences compared to standard
tional Koiter reduction methods, which can only exploit the initial methods are listed in the following with the help of the flow chart
post-buckling response of structures. In the Koiter–Newton depicted in Figs. 1 and 2.
approach addressed in this contribution, Koiter's asymptotic As shown in Fig. 2, starting from a known nominal equilibrium
expansion is applied at any known equilibrium state to construct a state, usually the undeformed reference configuration of a struc-
reduced order model. In this paper, the co-rotational formulation ture, a reduced order model is constructed based on Koiter's

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
4 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎

reduced order model is stopped for f r being above a given

threshold. In a subsequent corrector phase, the residual is driven
toward zero in a manner similar to that used in traditional Newton
arc-length methods, as indicated in Fig. 1 using a small red arrow.
Based on the new found equilibrium state of the structure an
updated reduced order model is constructed which is used for
further load steps.
There are three basic parts in each expansion step of the Koi-
ter–Newton approach: (1) construction of the reduced order
model, (2) iterative solution of the reduced order model, (3) cor-
rection of the predicted load-displacement step produced by the
reduced order model. During the construction of the reduced
order model, a set of augmented linear FEM equations, which will
be addressed as (10) and (11) in Section 2.2, needs to be solved.
Since all these systems share the same coefficient matrix, only one
stiffness matrix factorization is needed. Hence, the computational
cost for constructing the reduced order model is almost equivalent
Fig. 1. The path-following strategy of the Koiter–Newton approach. to the cost of solving one linear FEM system, assuming that the
computational cost is dominated by a factorization and not by a
forward and back substitution. Since the scale of the reduced order
model is very small, i.e. usually less than ten, the computational
effort for the iterative solution using the classical Newton path-
following method can be neglected. During the correction of the
predicted load-displacement step produced by the reduced order
model, some general Newton iterations are used to increase the
numerical stability of the procedure. The computational cost for
each iteration is equal to the cost of solving a linear FEM system.
The total number N of the linear FEM systems which need to be
solved in the Koiter–Newton approach is assessed with:
X
n
p
N¼ ð1 þ N c Þ ð2Þ
p¼1

where n is the total number of steps taken by the method, and N pc

is the number of linear FEM systems for correction in step p. The
proposed Koiter–Newton approach will significantly improve the
efficiency of nonlinear static finite element analysis, especially for
Fig. 2. Algorithmic concept of the reduced order model solution procedure. buckling sensitive structures as demonstrated in Section 5.

theory which is used to make an initial nonlinear prediction of the

2.2. Construction of the reduced order model
response of the structure being analyzed. Compared to traditional
Newton methods which use a linear prediction, a significantly
In this section, we briefly illustrate how the reduced order
larger step size can be achieved using the Koiter–Newton
model is constructed based on Koiter's theory. Nonlinear equations
approach due to a better nonlinear prediction provided by the
of equilibrium are used to describe the nonlinear response of a
reduced order model, see the red, dashed curve plotted in Fig. 1.
structure. The discretized equilibrium equations are reduced to a
Furthermore, the proposed non-linear prediction detects and tra-
set of nonlinear algebraic equations of the form f int ðqÞ ¼ λf ext ,
ces accurately the branching into secondary stability paths at
where f int and f ext are the internal force vector and external force
bifurcation points, cf. Fig. 1. This efficient and reliable branch
vector, respectively, λ is the load parameter of the analysis, and q is
detection is attributed to the implementation of the perturbation
the vector of primal values, the unknown displacements in
loads during the construction of the reduced order model. The
our model.
perturbation loads related to the buckling modes of the structure
In the proposed Koiter–Newton approach, a reduced order
trigger a buckling mode which is used to perturb numerically the
model is established to approximate equilibrium equations in the
path-following algorithm and to ensure transition at the bifurca-
neighborhood of a known equilibrium state ðq0 ; λ0 Þ. The vector q0
tion point to the corresponding secondary equilibrium path.
During the simulation of the reduced order model, the unba- describes the configuration at this equilibrium state, which we will
lanced force residual at the new predicted location on the equili- refer to as the known nominal configuration. Let q be the
brium path is calculated using (1), based on the full finite element unknown displacement vector near this nominal state, and u
model. describes the current configuration with respect to the nominal
configuration:
f r ¼ λf ext  f int ð1Þ
q ¼ u○q0 ð3Þ
where f int is the internal force of the current iteration step, f ext is
the total external force and λ is the current load parameter. The where the composition of q0 and u is not always a simple addition,
internal force calculation is very accurate as it is calculated using i.e. in the case of 3D large rotations for shells, the composition of
the full finite element model. rotations is not found by simple addition, see Section 4.
This residual force f r is used to judge whether the applied The nonlinear relations are approximated by a Taylor series
reduced order model is sufficiently accurate. The simulation of the expansion to the third order with respect to u in the nominal

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5

configuration: the various powers of ξ with zero, and taking orthogonality con-
ditions (8) into account results in the following two sets of gov-
LðuÞ þQðu; uÞ þ Cðu; u; uÞ þ Oðkuk Þ ¼ Δλf ext
4
ð4Þ
erning system of equations:
where L, Q and C denote the linear, quadratic and cubic forms of  " # " #
Kt F uα 0
the approximated elastic forces. These forms can also be con- ¼ ð10Þ
 Ft 0 Lα  Eα
sidered as a two-dimensional tensor L, a three-dimensional tensor
Q and a four-dimensional tensor C of order N, where N is the total  " u #  
number of degrees of freedom in the finite element model. For Kt F αβ Qðuα ; uβ Þ
¼ ð11Þ
example, Qðu; vÞ ¼ Q pij ui vj , where the Einstein summation con-  Ft 0 Q αβ 0
vention is applied. We further use Δλ ¼ λ  λ0 and the fact that
f int ðq0 Þ ¼ λ0 f ext . where Kt ¼ L is the tangent stiffness matrix, where Eα is the α-th
The proposed Koiter–Newton technique is aimed at being unit vector with coefficient eα ¼ 1 and all other coefficients being
applicable for buckling sensitive structures. In the presence of zero, and where L α and Q αβ represent column vectors of order
buckling, multiple secondary equilibrium branches that intersect ð1 þ mÞ of the multi dimensional tensors L and Q, respectively.
with the primary path at bifurcation points exist. For the method The two sets of linear systems of equations, (10) and (11), have
to work in such cases, the possibility of some perturbation loads identical coefficient matrices, hence factorization is needed only
that may excite the secondary branches needs to be taken into once to construct the reduced order model. The dimension of the
account. Thus, the equations of equilibrium (4) may be extended coefficient matrix is ðN þ1 þ mÞ, where N is the number of degrees
to consider multiple loading of the form: of freedom in the full finite element model and ð1 þ mÞ is the
number of degrees of freedom in the reduced order model. Since
LðuÞ þQðu; uÞ þ Cðu; u; uÞ þ Oðkuk4 Þ ¼ Fϕ ð5Þ 1 þm is generally very small compared to N, these two sets of
where F is a load matrix column-wise formed by m þ1 load vec- systems can be regarded as augmented linear FEM equations [84].
tors f α , denoted as sub-loads. The first sub-load vector f 1 ¼ f ext In addition to (10) and (11) the load and displacement expan-
corresponds the external load. The remaining m load vectors are sion results in the relation:
perturbation loads derived from m closely spaced linear buckling 2h i
C αβγδ ¼ Cðuα ; uβ ; uγ ; uδ Þ  utαβ Lðuδγ Þ þ utβγ Lðuδα Þ þ utγα Lðuδβ Þ
modes of the structure. The selection of perturbation loads will be 3
addressed later. The vector ϕ represents load amplitudes, the form ð12Þ
of which will be presented in (14).
where C αβγδ denote the components of C.
The solution u of (5) lies, in general, on an m þ1 dimensional The relation (12) does not need a factorization. The solution of
surface. The numerical construction of such a surface would be (10)–(12) obtains the unknown displacements uα , uαβ and the
computationally prohibitive. To circumvent this, an approximate
unknown components of L, Q, C. At this point, the expansion (9)
solution is obtained using a series expansion. The equilibrium
of load amplitudes is achieved. We rewrite it in the following form,
surface is parametrized in terms of generalized displacements ξ,
to get the reduced order model of order 1þ m:
and the equilibrium displacement is expanded to the third order
with respect to ξ as follows: LðξÞ þ Qðξ; ξÞ þCðξ; ξ; ξÞ ¼ ϕ ð13Þ

u ¼ uα ξα þ uαβ ξα ξβ þ uαβγ ξα ξβ ξγ ð6Þ where the left hand side of (13) has already been obtained after
solving (10)–(12). The right hand side, which is the load ampli-
where the subscripts α; β; γ ¼ 1; 2; …; m þ1, and the summation tudes ϕ, can be expressed by the load parameter Δλ in the equi-
convention is applied. The first order displacements uα define the librium Eq. (4) as:
tangent plane to the equilibrium surface at the approximation
point. The second order displacements uαβ and third order dis- ϕ ¼ ΔλE1 ð14Þ
placements uαβγ describe the interactions among first and second where E1 is the 1st unit vector. The dimension of the load
order displacement fields, respectively. amplitudes ϕ is 1 þ m. The first component Δλ is the increment of
The equilibrium surface may be parametrized with an infinite the parameter for external load f 1 . The other components are
number of choices for ξ. To fix the parameterization, we choose parameters of the perturbation loads f α ; α ¼ 2; …; m þ 1 and all are
the vector ξ such that it is work conjugate to the load amplitudes set equal to zero to simulate the response of the structure to
ϕ, as given by: actual loads.
As shown in Fig. 2, if the buckling does not happen in the
ðFϕÞt δu  ϕ δξ:
t
ð7Þ
analysis, only the external load and Δλ need to be considered and
This leads to the following constraint equations: there will be only one degree of freedom in the reduced order
8 model. Otherwise, the perturbation loads f α ; α ¼ 2; …; m þ 1 are
> f u ¼ δαβ
t
>
< α β
> used to excite the buckling branches, where m is the number of
t
f α uβγ ¼ 0 ð8Þ closely spaced buckling modes of the analyzed structure.
>
>
>
: ft u ¼0 The construction of the perturbation loads is related to the
α βγδ
buckling modes of the structure. The buckling modes are found
where δαβ is the Kronecker delta. from a linear buckling analysis which requires the solution of the
Consistent with the displacement expansion, we assume the generalized eigenvalue problem ðKt þ μα Kg Þx ¼ 0 at the approx-
expansion for the load amplitudes ϕ to be: imation point, in which Kt is the tangent stiffness matrix, Kg is the
geometric stiffness matrix, in which μα is the coefficient of the
ϕ ¼ LðξÞ þ Qðξ; ξÞ þ Cðξ; ξ; ξÞ ð9Þ
buckling load ðμα f ext Þ and x is the corresponding buckling mode.
where L, Q and C are still to be determined linear, quadratic and The geometric matrix Kg is calculated using the form 2Qðq;λ Þ,
cubic forms of the approximated load amplitudes and represented where q;λ is the linear displacement of the structure under the
by a two-dimensional tensor L, a three-dimensional tensor Q and external load f ext and Q is the quadratic form of the approximated
a four-dimensional tensor C of order ð1 þ mÞ, respectively. elastic forces in (4). The perturbation load is calculated by the
Using the displacement expansion (6) and the load expansion multiplication of the geometric stiffness and the buckling mode. In
(9) into equilibrium equations (5) and equating the coefficients of this way, the buckling shapes of the structure are introduced into

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
6 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎

the algorithm as a kind of numerical perturbation, due to the use the nominal configuration and the global x-axis, θ0 is the angle
of the perturbation loads. For the interested reader this part is between the reference configuration and the global x-axis, and Δθ
discussed in detail in [59]. is the angle between the initial and current configurations.
If the structure has many closely spaced buckling modes, In the nominal configuration the beam element is described in
buckling modes associated with simultaneous or nearly simulta- an axis aligned local coordinates system with the unit vectors ex ,
neous buckling loads, these closely spaced modes will interact and ey as basis vectors. The basis vector ex is aligned with the coor-
influence the buckling characteristics of the structure. So the dinate direction from node a to node b. The local coordinate sys-
selection of perturbation loads should also consider this case. As a tem co-rotates with the element movement in the 2D plane.
simple rule, buckling modes with buckling loads within 120% of The global degrees of freedom of the beam element in the
the first buckling load are considered to be closely spaced buckling global coordinate system (x,y) are given as:
modes and used to construct the perturbation loads.  t
Using the solution of the reduced order model the relation q ¼ ua va θ a ub vb θ b : ð15Þ
between the load parameter Δλ and the displacement parameter ξ
In the case of large displacements and/or large rotations, the
can be found from (13). In order to get the nonlinear response
global degrees of freedom of the structure contain the rigid motion
about λ and q, the definition λ ¼ λ0 þ Δλ and Eq. (3) are used and a
and the deformational part. Since the strain energy is only related
displacement expansion (6) only up to the second order is
to the deformational part, we need to pick up the local degrees of
invoked.
freedom which are related to the deformation, using the co-
From the discussion above, it can be seen that the difficulty for
rotational frame, which results in:
a finite element implementation of the Koiter–Newton approach is
to obtain the expression of equilibrium equations in a third order h i
b¼ 0 0 θ b u b t
q a bb 0 θ b ð16Þ
form (5) in the known nominal configuration, which needs the
derivatives of the strain energy up to the fourth order, since in the ba , v
ba of node a and the local
where, the two local translations u
general nonlinear finite element technique we only use the deri- bb of node b are always equal to zero during
vertical translation v
vatives up to the second order. In this paper, we focus on dis-
the deformation of the element, due to the definition of the local
cussing how these high-order derivatives are achieved using the b can
b and θ
co-rotational element frame. In (16), local rotations θ a b
co-rotational formulation in the nominal configuration, in the
also be called local co-rotational slopes [68].
following Sections 3 and 4.
The local degrees of freedom q b of the beam element can be
reduced to:
h i
3. Co-rotational beam model b¼ u
q b θ
bb θ b t: ð17Þ
a b

A classical linear Euler beam element together with an It can be seen from Fig. 3 that global and local degrees of
element-independent co-rotational frame is used to demonstrate
freedom should satisfy the following geometrical relations, given
the finite element implementation of the Koiter–Newton approach
by:
in this section. 8
>
> ub ¼ ln l
< b
3.1. Governing equations of the 2D beam model θb a ¼ θa  Δθ ¼ θa ðθ  θ0 Þ ¼ θa  θ þ θ0 ð18Þ
>
>
:θ b ¼ θ  Δθ ¼ θ  ðθ  θ Þ ¼ θ  θ þ θ
b b b 0 b 0
We introduce the descriptions of the beam element in a co-
rotational frame. Based on this frame we provide local and global where the l and ln are the initial and current length of the beam
degrees of freedom for the beam element which allows us to element, respectively, θ is the angle between the current config-
extract the deformational part of the total motion by purging the uration and the x-axis of the global coordinate system, θ0 is the
rigid part before the element computations. angle between the initial configuration and the x-axis, and Δθ is
A complete description of a co-rotational beam element using
the angle between the initial and current configurations, as shown
Kirchhoff theory is presented in Crisfield's book [68]. The defor-
in Fig. 3.
mation of a 2-D beam element from the initial undeformed con-
We denote the position vectors of nodes a and b in the refer-
figuration to the nominal deformed configuration is shown in
ence configuration with ra and rb , respectively, and the position
Fig. 3. The two nodes of the beam element are denoted a and b.
vectors in the nominal configuration with da and db , respectively,
Three degrees of freedom are specified at each node which are two
see Fig. 4. Then, the current length ln of the beam element can be
translations u, v and one rotation θ. The beam configurations are
referred to the global Cartesian coordinates system (x,y) with basis
vectors x, y as depicted in Fig. 3. In Fig. 3, θ is the angle between

Fig. 3. Sketch of the beam element in a co-rotational frame. Fig. 4. Position vectors of the two nodes in a beam element.

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

expressed as: ∂3 U b n ∂q
∂2 q bm b ∂2 q
∂q bm b ∂2 q
∂q bm
2Q pij ¼ ¼ KL þ KL n þ KL n
2 2 ∂qp ∂qi ∂qj ∂qi ∂qj ∂qp ∂qi ∂qp ∂qj ∂qj ∂qp ∂qi
ln ¼ l  ðrb  ra Þt ðrb  ra Þ þ ðdb  da Þt ðdb  da Þ: ð19Þ
bm
∂3 q
bm
þ KL q ð27Þ
The strain of the beam element is expressed as: ∂qp ∂qi ∂qj
b b ln l l2n  l2
u  ðrb  ra Þt ðrb  ra Þ þ ðdb  da Þt ðdb da Þ
ε¼ ¼ ¼ ¼ : ∂4 U b n ∂q
∂3 q bm b n ∂2 q
∂2 q bm
l l lðln þ lÞ lðln þ lÞ 6C pijk ¼ ¼ KL þKL
∂qp ∂qi ∂qj ∂qk ∂qi ∂qj ∂qk ∂qp ∂qi ∂qj ∂qp ∂qk
ð20Þ
b n ∂2 q
∂2 q bm b
∂q bm
∂3 q bn ∂2 q
∂2 q bm
The curvature of the beam element can be given as [85]: þ KL þ KL n þ KL
∂qi ∂qk ∂qp ∂qj ∂qi ∂qp ∂qj ∂qk ∂qj ∂qk ∂qp ∂qi
 
4 x 2 x b n ∂3 qbm b bm bm
κ ¼  þ6 2 θb a þ  þ 6 2 θb b ð21Þ þ KL
∂q ∂q
þ KL n
∂3 q
bm
þ KL q
∂4 q
ð28Þ
l l l l ∂qj ∂qp ∂qi ∂qk ∂qk ∂qp ∂qi ∂qj ∂qp ∂qi ∂qj ∂qk

where the subscripts p, i, j, k, m and n run from 1 to 6, with 6 being

3.2. Equilibrium equations of a third order form – beam model the number of total degrees of freedom in a beam element. Note
the summation convention is only applied on m and n.
The strain energy of the beam element is first given based on Eqs. (25)–(28) demonstrate that any linear beam element with
the descriptions of the beam element in Section 3.1, then L, Q and a given KL can be implemented into the Koiter–Newton approach,
C of equilibrium equations of a third order form (5) are obtained by and the geometric nonlinearity of the element is taken into
differentiating the strain energy with respect to the global degrees account by derivatives of the local co-rotational frame q b with
of freedom. respect to global degrees of freedom q, which is independent of
Strain energy – beam model: The strain energy of the beam the element type.
element can be obtained by introducing the strain (20) and the After obtaining expressions of L, Q and C from (25)–(28) and
curvature (21) into the expression of the strain energy and inte- substituting them into (5), equilibrium equations in a third order
grating along the length of the beam, which results in: form for the beam element are achieved, however, the derivatives
Z l of local degrees of freedom q b with respect to global degrees of
b ;θ b ¼1  t  1
U ¼U u bb ; θ a b κ EIκ þ εt EAε dx ¼ EAlε2 freedom q in (25)–(28) are still unknown, these are discussed in
2 0 2
  the following subsection.
2
2EI b 2 2 1 u b 2EI b 2 2 Derivatives of local degrees of freedom: The first order derivative
þ θ a þ θb a θb b þ θb b ¼ EA b þ θ a þ θb a θb b þ θb b ð22Þ
l 2 l l of local degrees of freedom q b (17) with respect to global degrees of
freedom q is a ð3  6Þ matrix:
where E is the Young's modulus, and A is the area of the cross 2 3
section. Since the strain energy (22) is a quadratic function of local ∂ubb ∂ubb ∂ubb
6 ∂q ⋯ ⋯
degrees of freedom, the derivatives of the strain energy are only 6 1 ∂qi ∂q6 77
6 7
up to the second order. The first order derivative of the strain b
∂q 6 a6 ∂ b
θ ∂ b
θ ∂ b
θ 7
a a 7
energy with respect to local degrees of freedom is the local ¼6 ⋯ ⋯ 7 i ¼ 1; 2; …; 6: ð29Þ
∂q 6 ∂q1 ∂qi ∂q6 7
internal load: 6 7
6 b b b 7
4 ∂θ b ∂θ b ∂θ b 5
 t ⋯ ⋯
b ∂U b 2EI
u b þθb ; 2EI θ b þ2θ b ∂q1 ∂qi ∂q6
f¼ ¼ EA b ; 2θ a b a b : ð23Þ
b
∂q l l l
In the same way we can now derive the second to fourth order
The second order derivative of the strain energy provides the derivatives of the local degrees of freedom with respect to the
linear stiffness matrix of the beam element: global degrees of freedom which will result in a three-dimensional
2 3 matrix of dimension ð3  6  6Þ, a four-dimensional matrix ð3 
EA
0 0 6  6  6Þ and a five-dimensional matrix of ð3  6  6  6  6Þ.
6 l 7
6 7 These matrices are formed by derivatives of the three components
∂2 U 6 4EI 2EI 7
KL ¼ 2 ¼ 6 6 0 7:
7 ð24Þ b , which are here listed only to the second order forms:
in q
∂qb 6 l l 7
4 2EI 4EI 5
0 b b ∂ln
∂u bb
∂2 u ∂ 2 ln
l l ¼ ; ¼ ;… ð30Þ
∂qi ∂qi ∂qi ∂qj ∂qi ∂qj
After obtaining the expression of the strain energy U , the 8
internal load vector f, the tangent stiffness matrix L, the three- > 1; if i ¼ 3 8
> if i; j ¼ 3; 6
b
∂θ
>
< 0; if i ¼ 6 b
∂2 θ < 0;
>
dimensional matrix Q and the four-dimensional matrix C can be a a
∂2 θ
¼ ; ¼ ;…
achieved using the first to fourth order derivatives of the strain ∂qi > > ∂θ ∂qi ∂qj > : ∂q ∂q ; if i; j ¼ others
>
: ∂q ; if i ¼ others i j
energy with respect to global degrees of freedom q. Since the i
b,
strain energy is an explicit function of local degrees of freedom q ð31Þ
we should differentiate the strain energy with respect to local
8
degrees of freedom first and then multiply the derivatives of the > 1; if i ¼ 3 8
> if i; j ¼ 3; 6
local degrees of freedom with respect to the global degrees of b
∂θ
>
< 0; if i ¼ 6 b
∂2 θ < 0;
>
freedom. Using (23) and (24), components of f, L, Q and C can be
b
¼ ; b
¼ ∂2 θ ;…
∂qi > > ∂θ ∂qi ∂qj > : ∂q ∂q ; if i; j ¼ others
expressed as follows: >
: ∂q ; if i ¼ others i j
i

∂U b
∂q ð32Þ
fp ¼ bm m
¼ KL q ð25Þ
∂qp ∂qp
where the subscripts i and j run from 1 to 6, and where we con-
sider the fact that the geometrical relations (18) are used.
∂2 U b ∂q
∂q b bm
∂2 q
Lpi ¼ bm
¼ KL n m þ KL q ð26Þ From Eqs. (30)–(32) it can be seen that the unknown terms are
∂qp ∂qi ∂qi ∂qp ∂qp ∂qi converted into the derivatives of the current length ln and the rigid

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
8 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎

rotation θ of the beam element with respect to global degrees of

freedom. A detailed description of how to calculate these two
unknown groups of derivatives is given below.
Referring to Fig. 3, the following geometrical relation can be
easily found in a triangle formed using the current configuration of
the beam and the x-axis of the coordinate system, which results
in:
 
ln cos θ ln sin θ ¼ db  da ¼ ΔrðqÞ ð33Þ
where the vector ΔrðqÞ is related to global degrees of freedom
ua ; va ; ub ; vb , and can be expressed by:
 
ΔrðqÞ ¼ ðx0b  x0a Þ þ ðub ua Þ ðy0b  y0a Þ þðvb  va Þ ð34Þ
   
where x0a ; y0a and x0b ; y0b are the coordinates of nodes a and b,
Fig. 5. Configurations of the shell element in a co-rotational frame.
respectively, in the reference configuration.
Differentiating (33) with respect to global degrees of freedom,
the following equation is obtained: freedom of the shell into a translational part and a rotational part
to handle this issue and to obtain a proper presentation of the
∂ln     ∂θ ΔrðqÞ
cos θ sin θ þ ln  sin θ cos θ ¼ ð35Þ derivatives needed. We use a classical linear shell element
∂qi ∂qi ∂qi according to the theory of Kirchhoff with an element-independent
where the right hand side is a constant vector obtained by dif- co-rotational frame.
ferentiating (34), which leads to:
8 4.1. Governing equations
> ½ 1 0 if i ¼ 1
>
>
>
>
> 0 1 if i ¼ 2
½
ΔrðqÞ < ½1 0 if i ¼ 4 :
Deformation gradient: A quadrilateral four-node flat shell ele-
¼ ð36Þ ment with six degrees of freedom per node is shown in Fig. 5. Two
∂qi >
>
>
> ½ 0 1  if i¼5
>
> known configurations of this shell element during the deforma-
: ½ 0 0 if i ¼ 3; 6 tion are defined, where the reference configuration is the initial
undeformed configuration and the nominal configuration is the
Then, if both sides of (35) are multiplied by a vector
 t latest known configuration. In the initial step, the nominal con-
cos θ sin θ , the first order derivative of the current length ln of
figuration is just the reference configuration.
the beam element with respect to global degrees of freedom is
There are several choices to define the co-rotational element
obtained as:
frame [68,69]. Here we use the straight lines which are defined by
∂ln ΔrðqÞ t the mid-points of two opposite sides as the coordinate axes 1 and
¼ cos θ sin θ : ð37Þ
∂qi ∂qi 2, the third axis is naturally orthogonal to the plane formed by the
Next, if both sides of (35) are multiplied by a vector axes 1 and 2. As shown in Fig. 5, we assume matrix T0 ¼
 t h i  
 sin θ cos θ , the first order derivative of the rigid rotation θ 0 0 0
d1 d2 d3 and matrix T ¼ d1 d2 d3 to be the co-rotational
with respect to global degrees of freedom can be achieved:
element frames in the reference and nominal configurations,
∂θ ΔrðqÞ t respectively, which satisfies the following relationships:
ln ¼  sin θ cos θ : ð38Þ
∂qi ∂qi ( 0 0 0
d3 ¼ d1  d2
In the same way, the other derivatives of the current length ln ð39Þ
d3 ¼ d1  d2
and the rigid rotation θ with respect to global degrees of freedom
can be achieved. We provide the higher order derivations in where d1 , d2 , d3 denote the basis vectors in the nominal config-
0 0 0
Appendix A. uration and where d1 , d2 , d3 denote the basis vectors in the
At this point all of the essential derivatives have been obtained, reference configuration.
and introducing them into Eqs. (25)–(28) will give the equilibrium The deformation gradient F represents the gradient of the
equations in a third order form for the co-rotational beam ele- mapping function related to both, the reference and nominal
ment. Following the procedures outlined in Section 2.2, the configurations. In order to find the deformation gradient, we
reduced order model using the co-rotational beam element can assume that two arbitrary vectors, a0 and a, are located in the
now be constructed using the L, Q and C in the equilibrium reference and nominal configurations, respectively, and that they
equations (5). are in the same position relative to their own co-rotational frames,
given as:
( 0 0 0
4. Co-rotational shell model a0 ¼ a1 d1 þa2 d2 þ a3 d3
ð40Þ
a ¼ a1 d1 þ a2 d2 þ a3 d3
As pointed out in Section 1, 3D large rotations may appear for
where a1 ; a2 ; a3 are the vector coordinates.
shells that are non-commutative and thus introduce an additional
Using (40), vector a can be expressed by a0 , as follows:
complexity in the formulation. Argyris discussed this challenge
stating that “commutativity, which is one of the three essential  h 0 0 0 i  1 0
a ¼ d1 d2 d3 d1 d2 d3 a : ð41Þ
properties of a vector, is not satisfied for two or more finite rota-
tions about arbitrary axes in space since the sequential order of In (41), the mapping function related to these two vectors is the
their imposition determines in each case a different result and this deformation gradient, as given by:
non-commutativity of finite rotations is, however, not only char-
 h 0 0 0 i  1
acteristic of truly large rotations but it holds even when second F ¼ d1 d2 d3 d1 d2 d3 : ð42Þ
order effects have to be considered” [78]. We split the degrees of

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 9

The deformation gradient can be decomposed, using the polar

decomposition theorem, into a product of two second-order ten-
sors: an orthogonal rotation tensor and a positive definite sym-
metric deformation tensor:
F ¼ RU ð43Þ
where the orthogonal matrix R represents the pure rotational part
of the total motion, and where the symmetric tension matrix U
indicates the pure deformation part of the total motion.
Matrix U is obtained by multiplying Ft on both sides of (43):
Ft F ¼ Ft RU ¼ URt RU ¼ UU ð44Þ
t
where the orthonormal relation R R ¼ I is used. Fig. 6. Transformations between the configurations of the co-rotational shell
Using (43) the rotation matrix R is: element.

R ¼ FU  1 : ð45Þ
displacement is not a simple addition in the case of finite rota-
Local translations: Global and local degrees of freedom of the tions, and will depend on the parametrization of the rotations.
shell element are expressed by two vectors of dimension ð24  1Þ, In the Cartesian coordinate system, an arbitrary rotation ϑ ¼
q and qb , respectively. Global degrees of freedom are measured in  
φ χ ψ can be represented by a corresponding rotation matrix R.
the global coordinate system Tg ¼ ½X Y Z in Fig. 5, and local Then, after rotating a vector p by ϑ , a new vector p0 can be
degrees of freedom are measured in the co-rotational shell frame obtained [78]:
by purging the rigid motion.
For a convenient description, degrees of freedom are divided p0 ¼ Rp ð50Þ
into two parts, a translational and a rotational part. For each node
where the rotation matrix R is an orthogonal matrix, and this
a ða ¼ 1; 2; 3; 4Þ, the (3  1) vectors ta and θa are used to denote the
nonlinear rotation matrix is a nonlinear function of the rotation ϑ ,
translation and rotation parts of the global degrees of freedom,
b represent the translation and rotation as given by:
respectively, and b t a and θ a
parts of the local degrees of freedom, respectively, both discussed 1 2 1 3
R ¼ exp ϑ~ ¼ Iþ ϑ~ þ ϑ~ þ ϑ~ þ⋯ ð51Þ
in the following. 2! 3!
We assume a vector r0a in the global frame Tg to be the position
where it can degenerate to the familiar infinitesimal rotation
vector of node a in the reference configuration, see Fig. 5, and ta is
matrix, given by:
the global translation of this vector from the reference config-
uration to the nominal configuration, hence the position of the
R ¼ exp ϑ~  Iþ ϑ~ ð52Þ
new vector in the nominal configuration is r0a þ ta . Then, Tt ðr0a þ ta Þ
is the local position of this new vector under the nominal co-
where matrix I is a (3  3) identity matrix, the operator expð…Þ
rotational frame T, which is equal to:
means applying an exponent operation on the matrix in the
Tt ðr0a þ ta Þ ¼ Tt0 r0a þ b
ta ð46Þ bracket, and matrix ϑ~ is an antisymmetric matrix formed by the
three components of the rotation vector ϑ, which results in:
where Tt0 r0a
is the local position of the initial vector r0a in the
reference configuration, and b 2 3
t a is the local translation. The co- 0 ψ χ
rotational frame T0 in the reference configuration and the co-
ϑ~ ¼ 6
4 ψ 0 φ 7 5: ð53Þ
rotational frame T in the nominal configuration satisfy the fol-
χ φ 0
lowing relationship:
T ¼ RT0 ð47Þ If a sequence of n successive rotations ϑ1 ; ϑ 2 ; …; ϑn is applied
on the vector p, and the corresponding rotation matrices of them
where R is the average rotation matrix of the shell element.
are fR1 ; R2 ; …; Rn g, the final total rotation matrix R of this
Substituting (47) into (46), the expression of the local transla-
sequence of rotations can be written as:
tion b
t a is obtained:
b R ¼ Rn Rn  1 ⋯R1 : ð54Þ
t a ¼ Tt0 Rt ta þ Tt0 ðRt IÞT0 Tt0 r0a ð48Þ
where r0a ; a ¼ 1; 2; 3; 4 is the position vector of each node in the Eqs. (51)–(54) show some basic properties of the rotation
reference configuration. matrix. In this study, R is the average global rotation matrix for the
Local rotations: As shown in Fig. 6, an unknown current con- shell element, and in addition four nodes also possess their own
figuration which is close to the nominal configuration is intro- global rotation matrices Ra , a ¼ 1; 2; 3; 4. According to (51), the
duced, excepting the reference and nominal configurations, to local rotation matrix Rb related to the local rotation θ b of node a
a a
make a better description of the 3D finite rotation case. Based on can be expressed as:
these three configurations, the global degrees of freedom q that 
b a ¼ exp θ
R b~ t t
present the total displacement from the reference configuration to a ¼ T 0 R Ra T 0 ð55Þ
the current configuration can be divided into two parts, which are
the relative displacement qn from the reference configuration to b~ is formed by three compo-
where the anti-symmetric matrix θ a
the nominal configuration and the relative displacement u from
b , using (53).
nents in the local rotation vector θ
the nominal configuration to the current configuration, as given a

by: The global rotation θa from the reference to the current con-
figurations may be a finite rotation, while the corresponding local
q ¼ u○qn ð49Þ b after removing the rigid motion can seem to be infi-
rotation θ a
where, as mentioned in (3), the composition ○ for rotations in the b a (55) is rewritten
nitesimal small. Thus, the local rotation matrix R

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
10 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎

in the following form according to (52): convention is applied on m and n. Similar to the beam element,
 derivatives of local degrees of freedom q b with respect to global
b a ¼ exp θ b~ b~
a ¼ T0 R Ra T0  Iþ θ a :
t t
R ð56Þ degrees of freedom q are needed. Since local degrees of freedom
have been split into local translations and local rotations, their
Since any matrix can be represented as the summation of a derivatives are presented in following two parts.
symmetric matrix and an anti-symmetric matrix, the anti- Derivatives of local translations: The first order derivatives of
b a should be equal
symmetric part of the local rotation matrix R local translations with respect to global degrees of freedom are
~
b
to the anti-symmetric matrix θ in (56), which leads to:
a
introduced below, the second to fourth order derivatives can be
 A easily achieved following the same method.
~
θb a ¼ Tt0 Rt Ra T0 ð57Þ According to the expression of local translations (48), its first
order derivatives result in:
where the operator ½…A means picking up the anti-symmetric
part of the matrix in the bracket. ∂b
ta ∂Rt ∂ta ∂Rt
¼ Tt0 ta þ Tt0 Rt þ Tt0 T0 Tt0 r0a j ¼ 1; …; 24 ð64Þ
The global rotation matrix Ra in (57) is related to the global ∂qj ∂qj ∂qj ∂qj
rotation θa of each node, and is composed of two successive ∂ta ∂Rt
rotations which are the rotation θa between the reference and
n with the unknown quantities, ∂qj and ∂qj , on the right hand side.
∂ta
nominal configurations and the rotation ϑ a between the nominal Term ∂qj is obtained as:
and current configurations. Since these rotations may be finite 8
>
> ½1 0 0t ; if qj ¼ t a ð1Þ
rotations, the composition here is not a simple addition and it >
>
>
< ½0 1 0t ; if qj ¼ t a ð2Þ
results in: ∂ta
¼ ð65Þ
∂qj >
> ½0 0 1t ; if qj ¼ t a ð3Þ
Ra ¼ exp ϑ~ a exp θ~ a
n
ð58Þ >
>
>
: ½0 0 0t ; if qj ¼ others
where Eqs. (51) and (54) are used.
b~ where t a ðiÞ; i ¼ 1; 2; 3 are the three components of global transla-
Introducing (58) into (57), the anti-symmetric matrix θ t

b of
a
tion ta . The calculation of term ∂R ∂qj needs some simple forward and
formed by the three components of the local rotation vector θ a
back substitutions, which we provide in Appendix B.1.
node a can be rewritten as:
Derivatives of local rotations: Based on (60), the first order
~ h iA
θb a ¼ Tt0 Rt exp ϑ~ a exp θ~ a T0 :
n derivatives of the local rotations with respect to global degrees of
ð59Þ
freedom can be achieved as follows:
b can be obtained by
Then using (53), the local rotation vector θ 2 0 1 3A  V
∂ exp ϑ~ a
a
~b b
∂θ ∂Rt
exp ϑ~ a þ Rt Aexp θ~ T0 5
n
its corresponding anti-symmetric matrix θ a , given by: ¼ 4Tt0 @
a
a j ¼ 1; …; 24:
∂qj ∂qj ∂qj
h iA  V
θb a ¼ Tt0 Rt exp ϑ~ a exp θ~ a T0
n
ð60Þ ð66Þ

where the operator ½ A  V means picking up the anti-symmetric Further high order derivatives are obtained analogously. We
part of the matrix in the bracket and then generating a vector provide the derivatives of the rotation R in Appendix B.1.
using the three independent components of this anti-symmetric The derivatives of exp ϑ~ a are obtained by differentiating (51):
matrix.
∂ exp ϑ~ a ∂ϑ~ a ∂ϑ~ a 1 ~ 2 ∂ϑ~ a ∂ϑ~ a
¼ þ ϑ~ a þ ϑa þ⋯  ð67Þ
4.2. Equilibrium equations of a third order form – shell model ∂qj ∂qj ∂qj 2 ∂qj ∂qj

where the rotation ϑa between the nominal and current config-

Strain energy – shell model: In a co-rotational frame, the ele-
urations approaches zero, since these derivatives are calculated in
ment strain energy only depends on local degrees of freedom q b
the nominal configuration. The following Eqs. (68) and (69) are
which are related to the deformational part of the total motion,
used to build the derivatives:
given by: 2 3
0 ψa χa
1 t 1 t   6
U¼ εb Cε
b¼ qb KL q
b ð61Þ ϑa ¼ φa χ a ψ a ) ϑ~ a ¼ 4 ψ a 0  φa 75; φ a ; χ a ; ψ a A q
2 2
 χa φa 0
where the vector ε b is the local strain of the element, C is the
material matrix, and KL is the linear stiffness matrix of a quad- ð68Þ
rilateral element. From (61), it can be seen that the element strain 2 3 2 3 2 3
b, 0 0 0 0 0 1 0 1 0
energy U is a quadratic function of local degrees of freedom q ∂ϑ~ a 6 7 ∂ϑ~ a 6 7 ∂ϑ~ a 6 7
hence we can differentiate U up to a second order form, as follows: ¼40 0 1 5; ¼4 0 0 0 5; ¼41 0 0 5:
∂ φa ∂χ a ∂ψ a
0 1 0 1 0 0 0 0 0
b ∂U
f¼ b
¼ KL q ð62Þ ð69Þ
b
∂q
At this point, the derivatives of local degrees of freedom with
∂2 U respect to global degrees of freedom have been obtained by
KL ¼ 2
ð63Þ
b
∂q combining both, the translation part of the shell element, Section
4.2, and the rotation part of the shell element, Section 4.2. Intro-
where the first and second order derivatives of the strain energy
ducing these derivatives into (25)–(28), the equilibrium equations
are the local internal load b f and linear stiffness matrix KL , (5) of the shell element in a co-rotational frame are achieved. Now,
respectively. Substituting (62) and (63) into Eqs. (25)–(28), we the reduced order model of the CR shell element can be con-
obtain the components of the internal load vector f, tangent
structed using the L, Q and C in equilibrium equations (5).
stiffness matrix L, three-dimensional matrix Q and four-
dimensional matrix C of the equilibrium equations (5) for the Remark 1. The expressions of high order derivatives of local
shell element, where p, i, j, k, m, n ¼1,…, 24, and the summation degrees of freedom are complicated, as discussed in Section 4.2.

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 11

Thus, to facilitate the programming and to optimize the code, an compared with those obtained in former works using different
automatic differentiation technology [86] is adopted during the algorithms [53,56,88–90]
implementation of the co-rotational shell element. In Tsukanov's The selected examples are classical benchmark studies of
paper [86] the data structure and algorithms are developed for the buckling analysis with large deformations, and some of the
direct application of generalized Leibniz rules to the numerical structures are made of laminated composites. They represent
computation of partial derivatives in forward mode. The proposed severe test cases due to the highly nonlinear equilibrium path and/
data structure provides constant time access to the partial deri- or a number of closely spaced buckling modes which are present
vatives, which accelerates the automatic differentiation computa- in the structure.
tions. In our work, this automatic differentiation technique, based
on generalized Leibniz rules, is used to differentiate a composite 5.1. Buckling of shallow and high curvature arches
function in a forward mode, as given in Appendix B.2.
Two arch structures with a different curvature are fetched from
a whole circle, as shown in Fig. 7. These structures are character-
ized by highly geometric nonlinearities in the prebuckling state
5. Numerical examples [91,92]. In this paper, we discuss the performance of the devel-
oped co-rotational beam element using the Koiter–Newton
In the following section we provide several examples that
approach. We modeled the shallow arch with 80 beam elements
demonstrate the potential of the Koiter–Newton approach using which results in 234 degrees of freedom. The deep arch was
the co-rotational formulation for buckling analysis of geome- modeled with 100 element resulting in 303 degrees of freedom.
trically nonlinear structures. We compare our results to classical For both arch structures, the used reduced order model has two
structural nonlinear analysis results obtained from ABAQUS [87] degrees of freedom, where one degree of freedom represents the
computations using the full number of equations of the finite primary path, and another one indicates the first buckling mode.
element model. The discrete models in our computations corre- We show the vertical displacement of the loading point versus the
spond the models used for the ABAQUS reference solution, with applied load in Fig. 8 for the shallow arch. The solid line represents
the same number of nodes and elements. The step size of the an ABAQUS reference solution, the triangle markers indicate the
proposed approach is indicated in the load-displacement diagrams solution steps of the proposed Koiter–Newton approach and cor-
of the reference solution with markers. It should be pointed out respondingly represent the accepted equilibrium state. Each of the
that the path-following method implemented in ABAQUS is a marked solution steps follows from a nonlinear prediction pro-
variant of the modified Riks method, also known as Riks/Wempner vided by the reduced order model which is indicated by the red,
method, and this method may be not optimal in terms of number green and blue curve, respectively. It can be seen that the range of
of linear systems that need to be solved. Hence, the results are also validity of the reduced order model is limited, the prediction is

Fig. 7. Two arch structures.

Fig. 8. Response curves of the two arch models. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this paper.)

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
12 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 9. Deformations near the limit point, deep arch model.

Fig. 10. 7 member frame.

Fig. 11. Cylindrical roof model.

considered accurate up to the triangle marker on the equilibrium

Table 1
path. A few consecutive corrector steps are used to recover full Material specification of the cylindrical roof models.
equilibrium and to fully coincide with the reference solution. It is
one of the essential properties of the Koiter–Newton approach Case Material properties Lamination

that the predictor step closely follows the equilibrium path thus Isotropic E ¼3102.75 MPa, μ¼ 0.3 None
allowing for a significantly larger step size compared to classical Laminate 1 E1 ¼ 3300 MPa; E2 ¼ 1100 Mpa; G12 ¼ 660 MPa, ½0=90=0
methods. The range of validity is essentially determined by the μ¼ 0.25
Laminate 2 E1 ¼ 3300 MPa; E2 ¼ 1100 Mpa; G12 ¼ 660 MPa, ½90=0=90
projected buckling information into the reduced order model μ¼ 0.25
which is adapted according to the complexity of the considered
structure and expected solution. As an indicator for a reliable sufficiently enriched reduced order model which represents the
choice of the size of the reduced order model the distribution of desired buckling response.
the buckling load spectrum is considered. All modes correspond- The practical relevance of the large step-size used in the
ing to closely spaced buckling modes are used to provide a method becomes obvious in all stability-related problems from

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 13

Fig. 12. C-shaped beam.

Table 2 one indicates the first buckling mode. From the response curve
Material specification of the C-shape beam models. about the lateral displacement of the loading point vs. the applied
load in Fig. 10(b), it can be observed that the present analysis is in a
Case Material properties Lamination
good agreement with the result from ABAQUS, however, only four
Isotropic E ¼ 1e7 MPa, μ¼0.333 None steps are needed in the Koiter–Newton analysis to achieve the
Laminate E1 ¼ 1:5e7 MPa; E2 ¼ 1e7 Mpa; G12 ¼ 5:555e7 MPa, ½0=90=0 response curve, compared to 74 steps needed by ABAQUS.
μ ¼0.25
5.3. Buckling of a cylindrical roof – isotropic and laminated com-
posite models
engineering, in particular, where a systematic analysis or a prob-
abilistic analysis requires the consideration of a larger number of
A hinged semi-cylindrical roof subjected to a central pinching
models for a reliable prediction of the structural response [93,94].
force, see Fig. 11(a), is a popular benchmark problem for geometric
But also standard stability analyses profit significantly from the
nonlinear analysis of shells due to the snap-through behavior of
reduced numerical effort of the approach which still constitutes a
the structural response, and this benchmark has received con-
limiting factor for many standard methods where the complete
siderable attention particular with respect to the Newton arc-
numerical model is used and often a large number of load steps is
length methods [96,97]. An isotropic material and two different
needed to detect reliably the singular points of a complex
laminates are considered in this example, with the properties
equilibrium path.
listed in Table 1, 100 co-rotational shell elements were used to
In the remaining example of this paper we will use only markers
mesh the structure. The first buckling load is F pcr ¼ 4:56 KN for the
on the reference solution to represent the step size of each Koiter–
isotropic case, and F pcr ¼ 3:58 KN and F pcr ¼ 2:22 KN for [0/90/0]
Newton step. The nonlinear response curve of the shallow arch
and [90/0/90] laminations, respectively. Since the first buckling
model was obtained within 3 load steps which required 9 solutions
loads are well separated from the others, reduced order models
of the global governing linear system of equations. In comparison,
the ABAQUS solution adopts 41 load steps which needed 84 solu- with 2 degrees of freedom are constructed in the Koiter–Newton
tion steps of the governing linear system of equations, cf. Fig. 8(a). approach. The nonlinear response curves, about the lateral dis-
The deep arch needs 5 load steps to solve 15 full FEM systems due placement of the loading point versus the applied load, are
to the more severe geometrical nonlinearities, compared to 89 load reported in Fig. 11(b) for the isotropic material model and the two
steps with 182 full FEM systems needed for ABAQUS, as shown in laminated composite structures, and compared with correspond-
Fig. 8(b). The large deformation of the deep arch near the limit point ing ABAQUS analysis results. Three steps with the solutions of
is shown in Fig. 9 and compared with the result from ABAQUS. 9 full FEM systems are used for all three material cases, compared
The deep arch example has been tested in many studies using to 44 steps solving 89 linear FEM systems needed for ABAQUS.
different algorithms [88–90,53]. Simo [90] needed 155 steps using This example is a classical benchmark and has been tested in
the Newton–Raphson method, and Kouhia [88] used different many researches [56,53,88]. The Newton Raphson method or
strategies with at least 34 steps and 220 iterations. Lahmam [53] variants of this method need to solve 40–70 FEM systems whereas
proposed a high-order predictor-corrector algorithm based on the other methods such as asymptotic numerical methods [98] only
asymptotic numerical method which needs 8 load steps. need to solve less than 10 full systems of equations [53].

5.2. Buckling of a plane frame model 5.4. Buckling of a C-shaped beam – isotropic and laminated com-
posite models
In the following example we apply our method to a frame which
is composed of 7 beams, see Fig. 10(a). A similar classical bench- A C-shaped beam is a commonly used structure in aerospace and
mark, a 3D 12 member frame, has been presented and discussed in marine engineering. As discussed in [12], a cantilever C-shaped beam
[92,95] using analytical methods or the Newton–Raphson method provides a severe buckling test due to its folding plate, interaction
and its variants. Six elements per beam member are used in the modes and serious nonlinear prebuckling behavior [99,100]. The
current Koiter–Newton analysis which results in 249 degrees of parameters of the C-shaped beam are shown in Fig. 12(a). We con-
freedom. A reduced order model with 2 degrees of freedom is sider both, isotropic material properties and a laminate composite
constructed, where one represents the primary path, and another structure, with the properties given in Table 2. The beam is meshed

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
14 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 13. The first four buckling modes of the C-shaped beam.

theory and a Newton arc-length method for the buckling analysis

of geometrically nonlinear structures. The proposed approach can
be used to trace the entire nonlinear equilibrium path of a struc-
ture in a stepwise manner and demonstrates high effectiveness
and stability in the presence of buckling.
In this paper, we lay stress on the finite element formulation of
beam and shell elements and implementation aspects of the Koi-
ter–Newton approach using a co-rotational formulation. We
extended the element formulations by a co-rotational formulation
which facilitates the computation of the up to the fourth order
derivatives of the strain energy with respect to the degrees of
freedom which are needed to construct the reduced order model.
Since in this formulation the geometrical nonlinearities are taken
into account using derivatives of the local co-rotational frame with
respect to the global degrees of freedom which are independent of
the element type, we are able to apply our formulation reliably to
any type of linear finite elements from the current element library.
Besides a 2D beam element we provided a 3D shell element for-
mulation which is the more challenging case due to large rotations
in 3D. We first introduced three configurations to describe the
deformation of the shell element, which serves the description of
the large rotations. We then adopted the nonlinear rotation matrix
to describe accurately the finite rotations of the shell. Finally, we
used an automated differentiation technique which significantly
simplifies the programming of the needed high order derivatives.
Using several benchmark problems for beam and shell models
we assessed the accuracy of our method and demonstrated high
Fig. 14. The first order displacement fields of the C-shaped beam.
computational efficiency compared to established state-of-the-art
finite element software packages. The high level of accuracy of the
with 1440 co-rotational shell elements. The reduced order model is
Koiter–Newton method using a co-rotational formulation is also
constructed with 5 degrees of freedom, five of which are related to
ensured for geometrically highly nonlinear cases accompanied by
the first four closely spaced buckling modes, shown in Fig. 13. The
buckling phenomena as demonstrated for isotropic and laminate
first order displacement fields, u1 and u2 , are shown in Fig. 14. The
composite shell models. Finally we showed the high efficiency of
nonlinear response curves, about the lateral displacement of loading
point vs. load, for isotropic material and laminated composite are the method with regard to computational cost for tracing the
plotted in Fig. 12(b) and compared with the ABAQUS analysis. Five complete nonlinear equilibrium response path of a structure.
and six steps are used, respectively, for the two material cases, Compared to the computational cost of ABAQUS analysis of full
compared to the 105 steps needed for the ABAQUS analysis. dimension the proposed method reduces the effort by more than a
factor 15 in almost all cases, independent of the geometric
dimension. We believe that the proposed Koiter–Newton method
6. Conclusions is a successful representative of the class of reduced order models
with a distinct potential to soften computational limitations dur-
The Koiter–Newton approach is a hybrid reduced order mod- ing the design process of structures with a geometrically highly
eling technique inspired by Koiter's classical initial post-buckling nonlinear response.

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 15

Acknowledgment ∂2 ln ∂2 θ ∂ln ∂3 θ ∂ln ∂3 θ

  
∂qj ∂ql ∂qi ∂qk ∂qk ∂qi ∂qj ∂ql ∂ql ∂qi ∂qj ∂qk
The research leading to these results received funding from the ∂ln ∂3 θ ∂ln ∂3 θ ∂ln ∂θ ∂θ ∂θ
European Community's Seventh Framework Programme ([FP7/   þ
∂qi ∂qj ∂qk ∂ql ∂qj ∂qi ∂qk ∂ql ∂ql ∂qi ∂qj ∂qk
2007-2013]) under grant agreement no. 282522. This research
∂ln ∂θ ∂θ ∂θ ∂ln ∂θ ∂θ ∂θ ∂ln ∂θ ∂θ ∂θ
work was also sponsored by Laboratory Independent Innovation þ þ þ
∂qk ∂qi ∂qj ∂ql ∂qi ∂qj ∂qk ∂ql ∂qj ∂qi ∂qk ∂ql
project of Qian Xuesen Laboratory of Space Technology. Thank you
Miranda Aldham-Breary for reviewing the English text carefully. ∂2 θ ∂θ ∂θ ∂2 θ ∂θ ∂θ ∂2 θ ∂θ ∂ θ
þ ln þ ln þ ln
∂qi ∂ql ∂qj ∂qk ∂qk ∂ql ∂qi ∂qj ∂qj ∂ql ∂qi ∂qk
∂2 θ ∂θ ∂θ ∂2 θ ∂θ ∂θ ∂ 2 θ ∂θ ∂θ
þ ln þ ln þ ln :
Appendix A ∂qi ∂qk ∂qj ∂ql ∂qj ∂qk ∂qi ∂ql ∂qi ∂qj ∂qk ∂pl
ðA:4Þ
Second order derivatives of the current length ln of the beam
element and the rigid rotation θ with respect to the global degrees
of freedom:
8 Appendix B
>
> ∂ 2 ln ∂θ ∂θ
>
> ¼ ln
>
< ∂qi ∂qj ∂qi ∂qj B.1
!: ðA:1Þ
>
> ∂2 θ ∂ln ∂θ ∂ln ∂θ
> ln
> ¼  þ The component of the first derivative of the rotation matrix R
>
: ∂qi ∂qj ∂qi ∂qj ∂qj ∂qi t
with respect to the global degrees of freedom q, that is the ∂R∂qj , is
Third order derivatives: calculated as follows.
8 First, we differentiate both sides of (43) with respect to global
>
> ∂3 ln ∂2 θ ∂θ ∂2 θ ∂θ ∂2 θ ∂θ ∂ln ∂θ ∂θ degrees of freedom, and then multiply the transpose of the rota-
>
> ¼ ln þ ln þ ln þ
>
> ∂q ∂q ∂q ∂q ∂q ∂q ∂q ∂q ∂q ∂q k ∂qi ∂qj ∂qi ∂qj ∂qk tion matrix Rt on both sides, which results in:
>
> i j k i j k j k i
>
>
>
> ∂F ∂R ∂U
>
> ∂ln ∂θ ∂θ ∂ln ∂θ ∂θ
>
> þ þ Rt ¼ Rt Uþ ðB:1Þ
>
< ∂qj ∂qk ∂qi ∂qk ∂qi ∂qj ∂qj ∂qj ∂qi
: ðA:2Þ
>
> ∂3 θ ∂2 ln ∂θ ∂2 ln ∂θ ∂2 ln ∂θ ∂ln ∂2 θ where the fact of Rt R ¼ I is used, since the rotation matrix R is an
>
> ln ¼   
>
> ∂pi ∂pj ∂pk ∂qi ∂qj ∂qk ∂qj ∂qk ∂qi ∂qk ∂qi ∂qj ∂qi ∂qj ∂qk orthogonal matrix.
>
>
>
> Any matrix can be represented as a summation of a symmetric
>
>
>
> ∂ln ∂2 θ ∂ln ∂2 θ ∂θ ∂θ ∂ θ matrix and an anti-symmetric matrix. The derivatives of the
>
>   þ ln
>
: ∂qj ∂qk ∂qi ∂qk ∂qi ∂qj ∂qi ∂qj ∂qk symmetric matrix U is still a symmetric matrix, which indicates
that the anti-symmetric part of (B.1) is given by:
Fourth order derivatives: " #A " #A
∂R ∂F
∂ 4 ln ∂ 2 ln ∂ θ ∂ θ ∂2 ln ∂θ ∂θ ∂2 ln ∂θ ∂θ R t U ¼ Rt ðB:2Þ
¼ þ þ ∂qj ∂qj
∂qi ∂qj ∂qk ∂ql ∂qi ∂qj ∂qk ∂ql ∂qj ∂qk ∂ql ∂qi ∂qk ∂ql ∂qi ∂qj
∂2 ln ∂θ ∂θ ∂2 ln ∂θ ∂θ ∂2 ln ∂θ ∂θ where the operator ½ A is used to pick up the anti-symmetric part
þ þ þ of the matrix in the bracket.
∂ql ∂qi ∂qj ∂qk ∂qi ∂qk ∂qj ∂ql ∂qj ∂ql ∂qi ∂qk
∂F
we use P to indicate matrix Rt ∂q in the right hand side of (B.2),
∂ln ∂2 θ ∂θ ∂ln ∂2 θ ∂θ ∂ln ∂2 θ ∂θ j

þ þ þ that is:
∂qi ∂qj ∂qk ∂ql ∂qi ∂qj ∂ql ∂qk ∂qi ∂qk ∂ql ∂qj
∂F
∂ln ∂2 θ ∂θ ∂ln ∂2 θ ∂θ ∂ln ∂2 θ ∂θ P ¼ Rt ðB:3Þ
þ þ þ ∂qj
∂qj ∂qi ∂ql ∂qk ∂qj ∂qk ∂ql ∂qi ∂qj ∂qi ∂qk ∂ql
∂F
where term ∂q is achieved by differentiating (42), as given by:
∂ln ∂2 θ ∂θ ∂ln ∂2 θ ∂θ ∂ln ∂2 θ ∂θ "
j
#
þ þ þ ∂d1 ∂d2 ∂d3 h 0 0 0 i  1
∂qk ∂qi ∂ql ∂qj ∂qk ∂qj ∂ql ∂qi ∂qk ∂qi ∂qj ∂ql ∂F
¼ ∂q ∂qj ∂qj d1 d2 d3 ðB:4Þ
∂ln ∂2 θ ∂θ ∂ln ∂2 θ ∂θ ∂ln ∂2 θ ∂θ ∂qj j
þ þ þ h i
∂ql ∂qi ∂qk ∂qj ∂ql ∂qj ∂qk ∂qi ∂ql ∂qi ∂qj ∂qk
where term ∂d 1 ∂d2 ∂d3
∂qj ∂qj ∂qj can be easily obtained by geometry and
∂3 θ ∂θ ∂3 θ ∂θ ∂3 θ ∂θ deformation relations.
þ ln þ ln þ ln ∂R
∂qi ∂qj ∂qk ∂ql ∂qi ∂qj ∂ql ∂qk ∂qj ∂qk ∂ql ∂qi The matrix Rt ∂q in the left hand side of (B.2) is an anti-
j

∂3 θ ∂θ ∂2 θ ∂2 θ ∂2 θ ∂2 θ symmetric matrix, and this conclusion is obtained by differ-

þ ln þ ln þ ln entiating the equation Rt R ¼ I on both sides, which is given as:
∂qi ∂qk ∂ql ∂qj ∂qi ∂qj ∂qk ∂ql ∂qi ∂ql ∂qj ∂qk
!t
∂2 θ ∂2 θ ∂θ ∂θ ∂ θ ∂θ ∂Rt t ∂R t ∂R t ∂R
þ ln  ln ðA:3Þ RþR ¼0 ) R ¼ R : ðB:5Þ
∂qi ∂qk ∂qj ∂ql ∂qi ∂qj ∂qk ∂ql ∂qj ∂qj ∂qj ∂qj
∂R
Hence, The matrix Rt ∂q can be represented by a anti-symmetric
j

∂ θ
4
∂ ln ∂ θ
3
∂ ln ∂ θ
3
∂ ln ∂θ 3 matrix Sj , as:
ln ¼   2 3
∂qi ∂qj ∂qk ∂ql ∂qi ∂qj ∂qk ∂ql ∂qi ∂qj ∂ql ∂qk ∂qi ∂qk ∂ql ∂qj 0 s1j s2j
∂R 6  s s3j 7
∂ 3 ln ∂ θ ∂ 2 ln ∂ 2 θ ∂2 ln ∂2 θ Sj ¼ Rt ¼4 1j 0 5 ðB:6Þ
   ∂qj
∂qj ∂qk ∂ql ∂qi ∂qi ∂qj ∂qk ∂ql ∂qk ∂ql ∂qi ∂qj  s2j  s3j 0
∂ 2 ln ∂ 2 θ ∂2 ln ∂2 θ ∂ 2 ln ∂ 2 θ where s1j, s2j and s3j are the three independent components in this
  
∂qi ∂qk ∂qj ∂ql ∂qi ∂ql ∂qj ∂qk ∂qj ∂qk ∂qi ∂ql anti-symmetric matrix Sj .

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
16 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Any anti-symmetric matrix can be expressed by the three [7] H. Kalathur, T.M. Hoang, R.S. Lakes, W.J. Drugan, Buckling mode jump at very
independent components, hence the independent components of close load values in unattached flat-end columns: theory and experiment, J.
Appl. Mech. 81 (4) (2013) 041010–1–041010–8.
the two anti-symmetric matrices in (B.2) should be equal, which [8] Y. Guo, M. Ruess, Z. Gürdal, A contact extended isogeometric layerwise
leads to a linear system of equations: approach for the buckling analysis of delaminated composites, Compos.
2 32 3 2 3 Struct. 116 (2014) 55–66.
U 11 þU 22 U 23  U 13 s1j P 12  P 21 [9] A.D. Lanzo, G. Garcea, Koiter's analysis of thin-walled structures by a finite
6 U 11 þ U 33 7 6 7 6 7
4 U 23 U 12 54 s2j 5 ¼ 4 P 13  P 31 5 ðB:7Þ element approach, Int. J. Numer. Methods Eng. 39 (1996) 3007–3031.
[10] E.G. Carnoy, Asymptotic study of the elastic postbuckling behavior of struc-
 U 13 U 12 U 22 þ U 33 s3j P 23  P 32 tures by the finite element method, Comput. Methods Appl. Mech. Eng. 29
(1981) 147–173.
where U11, U22, U33, U12, U13 and U23 are components of symmetric [11] G. Garcea, A. Madeo, G. Zagari, R. Casciaro, Asymptotic postbuckling fem
matrix U, and P12, P21, P13, P31, P23 and P32 are components of analysis using corotational formulation, Int. J. Solids Struct. 46 (2009)
377–397.
matrix P defined in (B.3). [12] G. Zagari, Koiter's asymptotic numerical methods for shell structures using a
After solving this linear system of equations (B.7) and intro- corotational formulation (Ph.D. thesis), Università Della Calabria, 2009.
 
ducing the solution vector s1j s2j s3j into (B.6), we can obtain the [13] T. Rahman, E.L. Jansen, Finite element based multi-mode initial post-
buckling analysis of composite cylindrical shells, in: 50th AIAA/ASME/
anti-symmetric matrix Sj . Then, according to (B.6), the component ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference,
of the first derivative of the rotation matrix with respect to global 2009.
t
degrees of freedom, ∂R∂q , is achieved by:
[14] T. Rahman, A perturbation approach for geometrically nonlinear structural
j
analysis using a general purpose finite element code (Ph.D. thesis), Delft
∂Rt University of Technology, 2009.
¼ RSj : ðB:8Þ [15] R.T. Haftka, R.H. Mallett, W. Nachbar, Adaption of Koiter's method to finite
∂qj
element analysis of snap-through buckling behavior, Int. J. Solids Struct. 7
(10) (1971) 1427–1445.
[16] P. Tiso, Finite element based reduction methods for static and dynamic
B.2 analysis of thin-walled structures (Ph.D. thesis), Delft University of Tech-
nology, 2006.
[17] T. Rahman, E.L. Jansen, Finite element based initial post-buckling analysis of
In our work, local translations (48) and local rotations (60) both
shells of revolution, in: 49th AIAA/ASME/ASCE/AHS/ASC Structures, Struc-
are products of two variable functions, which can be simply tural Dynamics and Materials Conference, Schaumburg, Illinois, 2008.
represented by yðxÞ ¼ pðxÞqðxÞ. In Tsukanov's work, a partial deri- [18] W.T. Koiter, On the stability of the elastic equilibrium (Ph.D. thesis), Delft
vative of a product of two functions is a linear combination of University of Technology, 1945.
[19] E. Schmidt, Zur theorie linearen und nichtlinearen Integralgleichungen. Theil
products of partial derivatives of p and q with binomial coeffi- 3: Über die Auflösuug der nichtlinearen Integralgleichungen und Verzwei-
cients, which gives [86]: gung ihrer Losungen, Math. Ann. 65 (1908) 370–399.
! ! ! [20] S.W. Shaw, C. Pierre, Normal modes for non-linear vibratory systems, J.
δ1 δ2 δn δ1 δ2 δn
∂jδj y X X X Sound Vib. 164 (1) (1993) 85–124.
¼ ⋯ ⋯
∂xδ11 ∂xδ22 ⋯∂xδnn α1 ¼ 0 α2 ¼ 0 αn ¼ 0
α1 α2 αn [21] G. Salerno, R. Casciaro, Mode jumping and attractive paths in multimode
elastic buckling, Int. J. Numer. Methods Eng. 40 (5) (1997) 833–861.
jα j
∂ p ∂ jδ  αj
q [22] G. Garcea, R. Casciaro, G. Attanasio, F. Giordano, Perturbation approach to
n α1 α 2 n ðB:9Þ elastic post-buckling analysis, Comput. Struct. 66 (5) (1998) 585–595.
∂x1 ∂x2 ⋯∂xαn n ∂xδ1  α1 ∂xδ2  α2 ⋯∂xδnn  αn [23] R. Kouhia, M. Mikkola, Tracing the equilibrium path beyond compound cri-
1 2
tical points, Int. J. Numer. Methods Eng. 46 (7) (1999) 1049–1074.
where [24] N. Rizzi, A. Tatone, Symbolic manipulation in buckling and postbuckling
analysis, Comput. Struct. 21 (1985) 691–700.
  X n   [25] N. Damil, M. Potier-Ferry, A new method to compute perturbed bifurcations:
δ ¼ δi ; 0 o δ rm ðB:10Þ application to the buckling of imperfect elastic structures, Int. J. Eng. Sci. 9
i¼1 (1990) 943–957.
[26] G. Cohen, Effect of a nonlinear prebuckling state on the postbuckling beha-
X
n vior and imperfection sensitivity of elastic structures, AIAA J. 6 (1968)
jαj ¼ αi ðB:11Þ 1616–1619.
i¼1 [27] J. Fitch, The buckling and postbukling behavior of spherical caps under
concentrated loads, Int. J. Solids Struct. 4 (1968) 421–446.
where m is the order of the highest derivative, α is a multi-index [28] B. Budiansky, J. Hutchinson, Dynamic buckling of imperfection sensitive
the elements of which serve as counters of summation loops, αδii structures, in: Proceedings of the 11th IUTAM Congress, Springer-Verlag,
Berlin/Gottingen/Heidelberg/Newyork, 1964, pp. 636–651.
; i ¼ 1; …; n are binomial coefficients. In this way, higher order [29] J. Arbocz, J. Hol, Koiter's stability theory in a computer-aided engineering
derivatives can be achieved using the known lower order deriva- (cae) environment, Int. J. Solids Struct. 26 (1990) 945–975.
[30] J. Arbocz, J. Hol, ANILISA – Computational Module for Koiter's Imperfection
tives efficiently in a forward mode.
Sensitivity Theory, Technical Report LR-582, Delft University of Technology,
1989.
[31] P. Heres, P. Deschrijver, W. Schilders, T. Dhaene, Combining Krylov subspace
References methods and identification-based methods for model order reduction, Int. J.
Numer. Model.: Electron. Netw. Devices Fields 20 (6) (2007) 271–282.
[32] Z. Bai, B. Li, Y. Su, A unified Krylov projection framework for structure-
[1] J. Arbocz, M. Potier-Ferry, J. Singer, V. Tvergaard, Post-buckling behaviour of preserving model reduction, in: W.H.A. Schilders, H.A. van der Vorst,
structures numerical techniques for more complicated structures, Buckling J. Rommes (Eds.), Model Order Reduction: Theory, Research Aspects and
Post-Buckling: Lect. Notes Phys. 288 (1987) 83–142. Applications. Mathematics in Industry, vol. 13, Springer-Verlag, Berlin, Hei-
[2] R. Peek, M. Kheyrkhahan, Postbuckling behavior and imperfection sensitivity delberg, 2008, pp. 75–93.
of elastic structures by the Lyapunov–Schmidt–Koiter approach, Comput. [33] B. Haasdonk, M. Ohlberger, Efficient reduced models and a posteriori error
Methods Appl. Mech. Eng. 108 (3–4) (1993) 261–279. estimation for parametrized dynamical systems by offline/online decom-
[3] R. Degenhardt, A. Kling, A. Bethge, J. Orf, L. Kärger, R. Zimmermann, position, Math. Comput. Model. Dyn. Syst. 17 (2) (2011) 145–161.
K. Rohwer, A. Calvi, Investigations on imperfection sensitivity and deduction [34] A. Lazarus, O. Thomas, J.-F. Deü, Finite element reduced order models for
of improved knock-down factors for unstiffened CFRP cylindrical shells, nonlinear vibrations of piezoelectric layered beams with applications to
Compos. Struct. 92 (8) (2010) 1939–1946. NEMS, Finite Elem. Anal. Des. 49 (1) (2012) 35–51.
[4] C. Schenk, G. Schuëller, Buckling analysis of cylindrical shells with cutouts [35] E. Jansen, R. Rolfes, Non-linear free vibration analysis of laminated cylind-
including random boundary and geometric imperfections, Comput. Methods rical shells under static axial loading including accurate satisfaction of
Appl. Mech. Eng. 196 (2007) 3424–3434. boundary conditions, Int. J. Non-Linear Mech. 66 (2014) 66–74.
[5] C. Dym, N. Hoff, Perturbation solutions for the buckling problems of axially [36] T. Rahman, E.L. Jansen, Z. Gürdal, Dynamic buckling analysis of composite
compressed thin cylindrical shells of infinite or finite length, J. Appl. Mech. cylindrical shells using a finite element based perturbation method, Non-
35 (4) (1968) 754–762. linear Dyn. 66 (3) (2011) 389–401.
[6] N.P. Semenyuk, A.I. Morenko, M.J.A. Smith, On the stability and postbuckling [37] S.E. Azam, S. Mariani, Investigation of computational and accuracy issues in
behavior of shells with corrugated cross sections under external pressure, J. POD-based reduced order modeling of dynamic structural systems, Eng.
Appl. Mech. 81 (1) (2013) 011002–1–011002–8. Struct. 54 (2013) 150–167.

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i
 K. Liang et al. / Finite Elements in Analysis and Design ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 17

[38] D. Amsallem, K. Carlberg, J. Cortial, C. Farhat, A method for interpolating on [69] K. Bathe, Finite Element Procedures, Prentice Hall, Englewood Cliffs, New
manifolds structural dynamics reduced-order models, Int. J. Numer. Methods Jersey, 1996.
Eng. 80 (2009) 1241–1258. [70] N. Cortivoa, C. Felippa, H. Bavestrelloc, W. Silva, Plastic buckling and collapse
[39] N. Akkari, A. Hamdouni, E. Liberge, M. Jazar, A mathematical and numerical of thin shell structures, using layered plastic modeling and co-rotational
study of the sensitivity of a reduced order model by POD (ROMCPOD), for a ANDES finite elements, Comput. Methods Appl. Mech. Eng. 198 (5–8) (2009)
2D incompressible fluid flow, J. Comput. Appl. Math. 270 (2014) 522–530. 785–798.
[40] A. Ammar, D. Ryckelynck, F. Chinesta, R. Keunings, On the reduction of [71] P. Norachan, S. Suthasupradit, K.-D. Kim, A co-rotational 8-node degenerated
kinetic theory models related to finitely extensible dumbbells, J. Non- thin-walled element with assumed natural strain and enhanced assumed
Newton. Fluid Mech. 134 (2006) 136–147. strain, Finite Elem. Anal. Des. 50 (2012) 70–85.
[41] A. Ammar, E. Pruliere, F. Chinesta, M. Laso, Reduced numerical modeling of [72] B.F. de Veubeke, The dynamics of flexible bodies, Int. J. Eng. Sci. 14 (10)
flows involving liquid-crystalline polymers, J. Non-Newton. Fluid Mech. 160 (1976) 895–913.
(2009) 140–156. [73] C. Pacoste, A. Eriksson, Beam elements in instability problems, Comput.
[42] C. Rowley, Model reduction for fluids, using balanced proper orthogonal Methods Appl. Mech. Eng. 144 (1997) 163–197.
decomposition, Int. J. Bifurc. Chaos 15 (3) (2005) 997–1013. [74] G. Zagari, A. Madeoa, R. Casciaro, S. de Mirandab, F. Ubertini, Koiter analysis
[43] B.A. Winther, P.J. Goggin, J.R. Dykman, Reduced-order dynamic aeroelastic of folded structures using a corotational approach, Int. J. Solids Struct. 50 (5)
model development and integration with nonlinear simulation, J. Aircr. 37 (2013) 755–765.
(5) (2000) 833–839. [75] R. Gallagher, Perturbation procedures in nonlinear finite element structural
[44] W. Zhang, B. Wang, Z. Ye, High efficient numerical method for limit cycle analysis, in: Computational Mechanics, Lecture Notes in Mathematics, vol.
flutter analysis based on nonlinear aerodynamic reduced order, in: 51st 461, Springer Verlag, Berlin, 1975.
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials [76] C. Rankin, B. Omid, The use of projectors to improve finite element perfor-
Conference, 12–15 April, Orlando, Florida, 2010. mance, Comput. Struct. 30 (1–2) (1988) 257–267.
[45] Y. Yue, The use of model order reduction in design optimization algorithms [77] C. Rankin, F. Brogan, E. Riks, Some computational tools for the analysis of
(Ph.D. thesis), KU Leuven, 2012. through cracks in stiffened fuselage shells, Comput. Mech. 13 (3) (1993)
[46] T. Bui-Thanh, K. Willcox, O. Ghattas, B. van Bloemen Waanders, Goal-orien- 143–156.
ted, model-constrained optimization for reduction of large-scale systems, J. [78] A. Cardona, M. Geradin, An excursion into large rotations, Comput. Methods
Comput. Phys. 224 (2) (2007) 880–896. Appl. Mech. Eng. 32 (1982) 85–155.
[47] F. Chinesta, A. Leygue, F. Bordeu, J. Aguado, E. Cueto, D. Gonzalez, I. Alfaro, [79] A. Ibrahimbegović, F. Frey, I. Kožar, Computational aspects of vector-like
A. Ammar, A. Huerta, PGD-based computational vademecum for efficient parametrization of three-dimensional finite rotations, Int. J. Numer. Methods
design, Optim. Control Arch. Comput. Methods Eng. 20 (2013) 31–59. Eng. 38 (21) (1995) 3653–3673.
[48] A. Quarterioni, G. Rozza (Eds.), Reduced Order Methods for Modeling and [80] A. Ibrahimbegović, On the choice of finite rotation parameters, Comput.
Computational Reduction. MS&A – Modeling, Simulation & Applications, Methods Appl. Mech. Eng. 149 (1–4) (1997) 49–71.
Springer International Publishing, Switzerland, 2004. [81] M. Correa, D. Camotim, On the differentiation of the Rodrigues formula and
[49] M. Gunzburger, J. Peterson, J. Shadid, Reduced-order modeling of time- its significance for the vector-like parameterization of Reissner–Simo beam
dependent PDEs with multiple parameters in the boundary data, Comput. theory, Int. J. Numer. Methods Eng. 55 (9) (2002) 1005–1032.
Methods Appl. Mech. Eng. 196 (2007) 1030–1047. [82] M. Correa, D. Camotim, Work-conjugacy between rotation-dependent
[50] Y. Maday, E. Ronquist, The reduced basis element method: application to a moments and finite rotations, Int. J. Solids Struct. 40 (11) (2003) 2851–2873.
thermal fin problem, SIAM J. Sci. Comput. 26 (1) (2004) 240–258. [83] A. Cardona, M. Geradin, Beam finite element non-linear theory with finite
[51] S. Niroomandi, I. Alfaro, E. Cueto, F. Chinesta, Model order reduction for rotations, Int. J. Numer. Methods Eng. 26 (11) (1988) 2403–2438.
hyperelastic materials, Int. J. Numer. Methods Eng. 81 (9) (2010) 1180–1206. [84] R. Seydel, Numerical computation of branch points in nonlinear equations,
[52] S. Niroomandi, I. Alfaro, E. Cueto, F. Chinesta, Real time simulation of surgery Numer. Math. 33 (3) (1979) 339–352.
by reduced order modelling and X-FEM techniques, Int. J. Numer. Methods [85] J.-M. Battini, Co-rotational beam elements in instability problems (Ph.D.
Eng. 28 (5) (2012) 574–588. thesis), Royal Institute of Technology, 2002.
[53] H. Lahmam, J. Cadou, H. Zahrouni, N. Damil, M. Potier-Ferry, High-order [86] I. Tsukanov, M. Hall, Data structure and algorithms for fast automatic dif-
predictor-corrector algorithms, Int. J. Numer. Methods Eng. 55 (6) (2002) ferentiation, Int. J. Numer. Methods Eng. 56 (2003) 1792–1949.
685–704. [87] Abaqus 6.11, ABAQUS Theory Manual, Section 2.3.2, 2011.
[54] E.H. Boutyour, H. Zahrouni, M. Potier-Ferry, M. Boudi, Bifurcation points and [88] R. Kouhia, M. Mikkola, Some aspects of efficient path-following, Comput.
bifurcated branches by an asymptotic numerical method and Padé approx- Struct. 72 (1999) 509–524.
imants, Int. J. Numer. Methods Eng. 60 (12) (2004) 1987–2012. [89] A. Noor, J. Peters, Tracing post-limit-point paths with reduced basis techni-
[55] H. Mottaqui, B. Braikat, N. Damil, Discussion about parameterization in the que, Comput. Methods Appl. Mech. Eng. 28 (1981) 217–240.
asymptotic numerical method: application to nonlinear elastic shells, Com- [90] J. Simo, L. Vu-Quoc, A three-dimensional finite-strain rod model. Part II:
put. Methods Appl. Mech. Eng. 199 (2010) 1701–1709. computational aspects, Comput. Methods Appl. Mech. Eng. 58 (1986)
[56] A. Eriksson, Derivatives of tangential stiffness matrices for equilibrium path 79–116.
descriptions, Int. J. Numer. Methods Eng. 32 (5) (1991) 1093–1113. [91] S. Chang, J. Chen, Effectiveness of linear bifurcation analysis for predicting
[57] W. Wagner, A path-following algorithm with quadratic predictor, Comput. the nonlinear stability limits of structures, Int. J. Numer. Methods Eng. 23 (5)
Struct. 39 (3–4) (1991) 339–348. (1986) 831–846.
[58] W. Aggoune, H. Zahrouni, M. Potier-Ferry, High-order prediction-correction [92] J. Meek, H. Tan, Geometrically nonlinear analysis of space frames by an
algorithms for unilateral contact problems, J. Comput. Appl. Math. 168 (1–2) incremental iterative technique, Comput. Methods Appl. Mech. Eng. 47
(2004) 1–9. (1984) 261–282.
[59] K. Liang, M. Abdalla, Z. Gürdal, A Koiter–Newton approach for nonlinear [93] J. Kepple, M.T. Herath, G. Pearce, B.G. Prusty, R. Thomson, R. Degenhardt,
structural analysis, Int. J. Numer. Methods Eng. 96 (12) (2013) 763–786. Stochastic analysis of imperfection sensitive unstiffened composite cylinders
[60] K. Liang, M. Ruess, M. Abdalla, The Koiter–Newton approach using von using realistic imperfection models, Compos. Struct. 126 (2015) 159–173.
Kármán kinematics for buckling analyses of imperfection sensitive struc- [94] M. Alfano, C. Bisagni, Probabilistic buckling analysis of composite and
tures, Comput. Methods Appl. Mech. Eng. 279 (2014) 440–468. sandwich cylindrical shells, in: 55st AIAA/ASME/ASCE/AHS/ASC Structures,
[61] K. Liang, A Koiter–Newton arc-length method for buckling sensitive struc- Structural Dynamics, and Materials Conference, 13–17 January, National
tures (Ph.D. thesis), Delft University of Technology, 2009. Harbor, Maryland, 2014.
[62] C. Felippa, B. Haugen, A unified formulation of small-strain corotational finite [95] M. Papadrakakis, Post-buckling analysis of spatial structures by vector
elements: I. Theory, Comput. Methods Appl. Mech. Eng. 194 (21–24) (2005) iteration methods, Comput. Struct. 14 (5–6) (1981) 393–402.
2285–2335. [96] K. Sze, X. Liu, S. Lo, Popular benchmark problems for geometric nonlinear
[63] A. Saada, Elasticity: Theory and Applications, Pergamon Press, Inc, New York, analysis of shells, Finite Elem. Anal. Des. 40 (11) (2004) 1551–1569.
1974. [97] L. Andrade, A. Awruch, I. Morsch, Geometrically nonlinear analysis of lami-
[64] Z. Xia, F. Ellyin, A finite elastoplastic constitutive formulation with new co- nate composite plates and shells using the eight-node hexahedral element
rotational stress-rate and strain-hardening rule, J. Appl. Mech. 62 (3) (1995) with one-point integration, Compos. Struct. 79 (2007) 571–580.
733–739. [98] A. Najah, B. Cochelin, N. Damil, M. Potier-Ferry, A critical review of asymp-
[65] H.H. Chen, W.Y. Lin, K.M. Hsiao, Co-rotational finite element formulation for totic numerical methods, Arch. Comput. Methods Eng. 5 (1) (1998) 31–50.
thin-walled beams with generic open section, Comput. Methods Appl. Mech. [99] A. Ibrahimbegović, Stress resultant geometrically nonlinear shell theory with
Eng. 195 (19–22) (2006) 2334–2370. drilling rotations: Part I. A consistent formulation, Comput. Methods Appl.
[66] C. Felippa, Nonlinear Finite Element Methods, Course Materials for the Mech. Eng. 118 (3–4) (1994) 265–284.
Department of Aerospace Engineering Sciences University of Colorado at [100] J. Chroscielewski, J. Makowski, H. Stumpf, Genuinely resultant shell finite
Boulder, 1999. elements accounting for geometric and material non-linearity, Int. J. Numer.
[67] M.W. Hilburger, J.H. Starnes Jr., Buckling behavior of compression-loaded Methods Eng. 35 (1) (1992) 63–94.
composite cylindrical shells with reinforced cutouts, Int. J. Non-Linear Mech.
40 (7) (2005) 1005–1021.
[68] M. Crisfield, Nonlinear Finite Element Analysis of Solids and Structures, vol. 1,
Wiley, Chichester, 1997.

Please cite this article as: K. Liang, et al., Co-rotational finite element formulation used in the Koiter–Newton method for nonlinear
buckling analyses, Finite Elem. Anal. Des. (2016), http://dx.doi.org/10.1016/j.finel.2016.03.006i