the mitc4+ shell element in geometric nonlinear analysiscmss.kaist.ac.kr/cmss/papers/2017 the mitc4+...
TRANSCRIPT
Computers and Structures 185 (2017) 1–14
Contents lists available at ScienceDirect
Computers and Structures
journal homepage: www.elsevier .com/locate /compstruc
The MITC4+ shell element in geometric nonlinear analysis
http://dx.doi.org/10.1016/j.compstruc.2017.01.0150045-7949/� 2017 Elsevier Ltd. All rights reserved.
⇑ Corresponding author.E-mail address: [email protected] (P.-S. Lee).
Yeongbin Ko a, Phill-Seung Lee a,⇑, Klaus-Jürgen Bathe b
aDepartment of Mechanical Engineering, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu, Daejeon 34141, Republic of KoreabDepartment of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
a r t i c l e i n f o
Article history:Received 27 December 2016Accepted 27 January 2017
Keywords:Shell structures4-node shell finite elementsMITC methodLarge displacements and rotationsShear and membrane lockingReliability in geometric nonlinear analysis
a b s t r a c t
We present the large displacement and rotation formulation of the new MITC4+ shell finite elementrecently proposed by Ko, Lee and Bathe for linear analysis (Ko et al., 2017) and demonstrate the perfor-mance in geometric nonlinear analysis. The element shows in linear analysis an almost ideal convergencebehavior since shear and membrane locking is alleviated using the MITC approach. We show now thatusing the total Lagrangian formulation for large displacements and large rotations, the element is alsorobust and efficient in nonlinear analysis. We demonstrate the element performance through the solu-tions of various benchmark problems and reach the important conclusion that the MITC4+ shell elementperforms reliably and well even when the mesh undergoes large displacements and significant distor-tions during the response.
� 2017 Elsevier Ltd. All rights reserved.
1. Introduction
For the analysis of shell structures, developing ‘‘ideal” shellfinite elements that satisfy the ellipticity, consistency and inf-supconditions has been of great interest [1–10]. Such shell elementsshould pass the basic tests (the isotropy, zero energy mode andpatch tests), show uniformly optimal convergence behavior inany shell problem irrespective of the shell geometry, loading andboundary conditions, and do so when regular and even distortedmeshes are used [7–10]. Also, the shell elements need to performequally well in geometric nonlinear analysis where an effectivebehavior in the nonlinear response predictions is important [4].
In geometric nonlinear analysis of shell structures, significantmesh distortions can occur as the geometry of the elementschanges during the response [11–15]. These element geometricchanges can lead to locking in bending-dominated shell problems[4,8], that is, an overly stiff behavior of the shell discretization isseen, which can be particularly severe when 4-node shell elementsare used to model thin shell structures [1,2,4,8]. Hence, in geomet-ric nonlinear analysis, locking due to the discretization undergoinglarge displacements can lead to erroneous predictions of load-displacement trajectories and critical loads.
The MITC (Mixed Interpolation of Tensorial Components)method [1–4,16–25] has been used effectively to remedy shearand membrane locking. The classical 4-node MITC shell element(labeled as MITC4 element) has been widely used in practice for
both linear and nonlinear analyses. However, in the original formu-lation of the MITC4 shell element, membrane locking was not trea-ted, and thus solution accuracy can deteriorate when curvedgeometries are modeled with distorted meshes [1,2]. Followingvarious attempts to alleviate membrane locking of 4-node shellelements [26–28], we recently presented the new MITC4+ shellelement for general linear analysis [1]. This element satisfies allthe basic element tests, contains no parameter to adjust, andshows an almost optimal convergence behavior in the solutionsof a ‘behavior-encompassing’ set of benchmark problems usingregular and distorted meshes. The fact that the element perfor-mance is also very good in highly distorted meshes is particularlynoteworthy and makes this element an excellent candidate forgeneral large displacement and rotation nonlinear analysis.
This expectation is reasonable because our experience is that ifa well-formulated MITC element has been established for linearanalysis, then this element formulation can directly be extendedto geometric nonlinear analysis without introducing instabilitiesin the element, like seen in formulations based on incompatiblemodes and enhanced assumed strains [4,29,30]. In addition, theincompatible modes and enhanced assumed strain elements arecomputationally more expensive.
In this paper, we present the geometric nonlinear formulationof the newMITC4+ shell. We develop the assumed shear and mem-brane fields of the element for the total Lagrangian formulationusing the Green-Lagrange strains and incremental Green-Lagrange strains to obtain the tangent stiffness matrix and internalforce vector [4]. We demonstrate the performance of the elementin geometric nonlinear solutions by solving various shell problems
2 Y. Ko et al. / Computers and Structures 185 (2017) 1–14
with uniform and distorted meshes. To assess the accuracy of thesolutions, we compare the predicted response with analytical dataand finite element solutions obtained using the MITC4 and MITC9shell elements. An important conclusion is that the new MITC4+shell element performs well even when a mesh undergoes signifi-cant displacements that could induce some locking, like might beseen in solutions using the classical MITC4 shell element.
2. Geometric nonlinear formulation
In this section, we present the geometric nonlinear formulationof the MITC4+ shell element. In the total Lagrangian formulation,the left superscript t denotes ‘‘time” for a general analysis (in staticsolutions ‘‘time” simply denotes the load step and configuration)and the left subscript 0 is used to denote the initial (reference) con-figuration [4].
2.1. Geometry and displacement interpolations
The geometry of the MITC4+ shell element in the configurationat time t shown in Fig. 1 is interpolated using [1,2]
txðr; s; fÞ ¼ txm þ ftxb with txm ¼X4i¼1
hiðr; sÞtxi and
txb ¼ 12
X4i¼1
aihiðr; sÞtVin; ð1Þ
where hiðr; sÞ is the two-dimensional interpolation function of thestandard isoparametric procedure corresponding to node i, txi is
the position vector of node i, and ai and tVin denote the shell thick-
ness and the director vector at the node, see Fig. 1.The following representation of interpolation function hiðr; sÞ is
useful in the element formulation:
hiðr; sÞ ¼ 14ð1þ nirÞð1þ gisÞ with i ¼ 1;2;3;4;
n1 n2 n3 n4½ � ¼ 1 �1 �1 1½ �;g1 g2 g3 g4½ � ¼ 1 1 �1 �1½ �; ð2Þ
in which ni and gi are permuted together.The incremental displacement vector u from the configuration
at time t to the configuration at time t þ Dt is
uðr; s; fÞ ¼ tþDtxðr; s; fÞ � txðr; s; fÞ: ð3Þ
ζ
2 1
4
3
rs
31V
t
32V
t
3n
tV
xi
yizi
z
x
y
Fig. 1. A standard 4-node quadrilateral continuum mechanics based shell finiteelement in the configuration at time t.
Using Eq. (1) in Eq. (3), we obtain
u ¼X4i¼1
hiðr; sÞðtþDtxi � txiÞ þ f2
X4i¼1
aihiðr; sÞðtþDtVin � tVi
nÞ; ð4aÞ
with
tþDtxi � txi ¼ uiix þ v iiy þwiiz; ð4bÞand to quadratic order
tþDtVin� tVi
n ¼ hi� tVinþ
12hi�ðhi� tVi
nÞ; hi ¼ tVi1aiþ tVi
2bi; ð4cÞ
in which ix, iy and iz are the base vectors of the global Cartesiancoordinate system, and at node i, ui, v i and wi are the corresponding
displacement components, tVi1 and tVi
2 are unit vectors orthogonal
to the director vector (tVin) and to each other, and ai and bi denote
the rotations of the director vector about tVi1 and tVi
2, respectively[4].
Substituting from Eqs. (4b) and (4c) into Eq. (4a), the incremen-tal displacement is obtained as
uðr; s; fÞ ¼ um þ fðub1 þ ub2Þ; ð5aÞwhere
um ¼X4i¼1
hiðr; sÞui; ð5bÞ
ub1 ¼12
X4i¼1
aihiðr;sÞð�lVi2aiþ lVi
1biÞ;ub2 ¼�14
X4i¼1
aihiðr;sÞða2i þb2
i ÞlVin:
ð5cÞWe next group the displacement terms in Eq. (5) as
u1 ¼ um þ fub1; u2 ¼ fub2; ð6Þwhere u1 and u2 contain the linear and quadratic terms of unknowndisplacements and rotations, respectively.
2.2. Green-Lagrange strains
The covariant Green-Lagrange strain components in the config-uration at time t with respect to the reference configuration at time0 are defined by
t0e ijðr; s; fÞ ¼
12ðtgi � tgj � 0gi � 0gjÞ; ð7Þ
in which tgi ¼ @tx@ri
are covariant base vectors with r1 ¼ r, r2 ¼ s,
r3 ¼ f.Using Eq. (3) in Eq. (7) applied at time t and t þ Dt, the incre-
mental covariant strain components are
0eijðr;s;fÞ¼ tþDt0 eijðr;s;fÞ� t
0e ijðr;s;fÞ¼12
tgi �u;jþu;i � tgjþu;i �u;j� �
;
ð8Þwith u;i ¼ @u
@ri.
Substituting from Eq. (6) into Eq. (8) and retaining only thestrain terms up to second order of unknowns, the incrementalstrain components can be written as
0eijðr; s; fÞ ¼ 0eijðr; s; fÞ þ 0gijðr; s; fÞ; ð9Þwith
0eijðr; s; fÞ ¼ 12
tgi � u1;j þ u1;i � tgj
� �;
1
1
1r
s
1
.0 constε ζst
.0 constε ζrt
)(0
Aζr
t ε
)(0
Bζr
t ε
)(0
Cζs
t ε)(0
Dζs
t ε
.0 constε ζrt
.0 constε ζst
Fig. 2. Tying positions (A), (B), (C) and (D) for the assumed transverse shear strainfield of the MITC4 shell element. The constant transverse shear strain conditions areimposed along its edges.
1
1
1r
s
1
)(0
Amrr
t ε
)(0
Bmrr
t ε
)(0
Cmss
t ε)(0
Dmss
t ε)(
0Em
rst ε
Y. Ko et al. / Computers and Structures 185 (2017) 1–14 3
0gijðr; s; fÞ ¼12
u1;i � u1;j þ tgi � u2;j þ u2;i � tgj
� �with
u1;i ¼ @u1
@ri; u2;i ¼ @u2
@ri;
where 0eij and 0gij are the linear and nonlinear parts, respectively[4,31].
2.3. Assumed Green-Lagrange strains
To alleviate shear locking, the assumed transverse shear strainsof the classical MITC4 shell element are used [2]
t0~erf ¼
12ð1þ sÞt0e
ðAÞrf þ 1
2ð1� sÞt0e
ðBÞrf ;
t0~esf ¼
12ð1þ rÞt0e
ðCÞsf þ 1
2ð1� rÞt0e
ðDÞsf ; ð10Þ
where the tying positions ðAÞ, ðBÞ, ðCÞ and ðDÞ are shown in Fig. 2[1,2,8].
In order to alleviate membrane locking, we separate the corre-sponding membrane strains from the in-plane strains. The covari-ant in-plane strains in Eq. (7) are expressed as
t0e ij ¼ t
0emij þ f t
0eb1ij þ f2t0e
b2ij with i; j ¼ 1;2; ð11aÞ
3
1
4
2r
s
P rt xs
t xnt
Fig. 3. Characteristic vectors for the element geometry at time t. (a) Two in-plane vec
in which
t0e
mij ¼ 1
2txm;i � txm;j þ txm;j � txm;i� �� 1
20xm;i � 0xm;j þ 0xm;j � 0xm;i� �
;
ð11bÞ
t0e
b1ij ¼ txm;i � txb;j þ txm;j � txb;i
� �� 0xm;i � 0xb;j þ 0xm;j � 0xb;i
� �; ð11cÞ
t0e
b2ij ¼ 1
2txb;i � txb;j þ txb;j � txb;i
� �� 12
0xb;i � 0xb;j þ 0xb;j � 0xb;i
� �;
ð11dÞwith
txm;i ¼ @txm
@ri; txb;i ¼ @txb
@ri:
In Eq. (11a), the term t0e
mij is the covariant in-plane strain at the
shell mid-surface (f ¼ 0), which in general can induce membranelocking.
In the geometric nonlinear formulation of the MITC4+ shell ele-ment, the assumed membrane strain fields are applied based onthe current configuration. Hence the covariant membrane strainin Eq. (11b) is considered
t0e
mij ¼ 1
2tgm
ij �12
0gmij with i; j ¼ 1;2; ð12aÞ
withtgm
ij ¼ txm;i � txm;j þ txm;j � txm;i; ð12bÞ
3
1
4
2
dt x2
tors txr and txs , and the plane P with normal vector tn. (b) Distortion vector txd .
Fig. 4. Tying positions (A), (B), (C), (D) and (E) for the assumed membrane strainfield.
(b)
(a)
0 1 2 3 4 5 6 70
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.7
0.8
0.9
1
reference16x1 MITC416x1 MITC4+
max/ PPtipwLoad
( )
tipu
4 Y. Ko et al. / Computers and Structures 185 (2017) 1–14
0gmij ¼ 0xm;i � 0xm;j þ 0xm;j � 0xm;i: ð12cÞWe next define the three characteristic vectors in the configura-
tion at time t [1]
txr ¼ 14
X4i¼1
nitxi;
txs ¼ 14
X4i¼1
gitxi;
txd ¼ 14
X4i¼1
nigitxi; ð13Þ
in which ni and gi are given in Eq. (2). The geometric representa-tions of the three vectors at time t are shown in Fig. 3.
As shown in Fig. 3(a), the two vectors txr and txs form the planeP with the normal vector tn
tn ¼txr � txs
ktxr � txsk ; ð14Þ
and the dual base vectors tmr and tms on the plane
tmri � txrj ¼ dij;tmri � tn ¼ 0 with i; j ¼ 1;2:
For the MITC4+ shell element, the following assumed field isused for the covariant membrane strain [1]
t0~emij ¼ 1
2t~gm
ij �12
0~gmij with i; j ¼ 1;2 ð15aÞ
where
t~gmrr ¼
12
1� 2taA þ sþ 2taA � s2� �
tgmðAÞrr
þ 12
1� 2taB � sþ 2taB � s2� �
tgmðBÞrr
þ taC �1þ s2� �
tgmðCÞss þ taD �1þ s2
� �tgmðDÞ
ss
þ taE �1þ s2� �
tgmðEÞrs ; ð15bÞ
t~gmss ¼ taA �1þ r2
� �tgmðAÞ
rr þ taB �1þ r2� �
tgmðBÞrr
þ 12
1� 2taC þ r þ 2taC � r2� �
tgmðCÞss
þ 12
1� 2taD � r þ 2taD � r2� �tgmðDÞ
ss
þ taE �1þ r2� �
tgmðEÞrs ; ð15cÞ
t~gmrs ¼
14
r þ 4taA � rs� �
tgmðAÞrr þ 1
4�r þ 4taB � rs� �
tgmðBÞrr
þ 14
sþ 4taC � rs� �
tgmðCÞss þ 1
4�sþ 4taD � rs� �
tgmðDÞss
þ 1þ taE � rs� �
tgmðEÞrs ; ð15dÞ
in which the tying positions ðAÞ, ðBÞ, ðCÞ, ðDÞ and ðEÞ for the corre-sponding strain components are shown in Fig. 4, and the geometriccoefficients are
z
M
b
L
x y
P (b)(a)
Fig. 5. Cantilever problems (16 � 1 mesh, width b ¼ 1:0, thickness a ¼ 0:1,E ¼ 1:2� 106 and m ¼ 0:0). (a) Case of the tip shearing force (L ¼ 10:0). (b) Case ofthe tip moment (L ¼ 12:0).
taA ¼tcr tcr � 1ð Þ
2td; taB ¼
tcrðtcr þ 1Þ2td
;
taC ¼tcsðtcs � 1Þ
2td; taD ¼
tcsðtcs þ 1Þ2td
;
taE ¼ 2tcrtcstd
; td¼ tc2r þ tc2s � 1; tcr ¼ tmr � txd;tcs ¼ tms � txd:
ð15eÞUsing the assumed membrane strain fields, the in-plane strain
components are constructed
t0~e ij ¼ t
0~emij þ f t
0eb1ij þ f2t0e
b2ij with i; j ¼ 1;2: ð16Þ
The shell-aligned local Cartesian coordinate system in the con-figuration at time 0 is defined using the unit vectors 0Li, i ¼ 1;2;3,
0L3 ¼0g3
k0g3k; 0L1 ¼
0g2 � 0L3
k0g2 � 0L3k ;0L2 ¼ 0L3 � 0L1; ð17Þ
and the corresponding local strain components are
t0�e ij ¼ t
0~eklð0Li � 0gkÞð0Lj � 0glÞ with 0gi � 0g j ¼ d j
i : ð18Þ
Displacements
0 2 4 6 8 10 12 14 16 18 200
0.1
0.2
0.3
0.4
0.5
0.6
analytic solution16x1 MITC416x1 MITC4+
max/MMLoad
( )
tipw tipu
Fig. 6. Load-displacement curves for the cantilever. (a) Case of the tip shearingforce. (b) Case of the tip moment.
x y
z
maxPP
Initial m
esh
max25.0 PP
P
Initial m
esh
z
maxMM
max5.0 MM
max25.0 MM
M
x y
Fig. 7. Deformed shapes of the cantilever. (a) Case of the tip shearing force. (b) Case of the tip moment.
A
B
P
oR iR
x y
z
Fig. 8. Slit annular plate problem (5 � 40 mesh, Ro ¼ 10:0, Ri ¼ 6:0, thicknessa ¼ 0:03, E ¼ 2:1� 107 and m ¼ 0:0).
Y. Ko et al. / Computers and Structures 185 (2017) 1–14 5
2.4. Assumed incremental Green-Lagrange strains
From the assumed Green-Lagrange strains in Section 2.3, weproceed to obtain the corresponding assumed incrementalGreen-Lagrange strains in a consistent manner.
For the transverse shear we use the following assumed incre-mental shear strains
0~erf¼12ð1þsÞ0eðAÞrf þ1
2ð1�sÞ0eðBÞrf ; 0~esf¼1
2ð1þrÞ0eðCÞsf þ1
2ð1�rÞ0eðDÞsf ;
0~grf¼12ð1þsÞ0gðAÞ
rf þ12ð1�sÞ0gðBÞ
rf ; 0~gsf¼12ð1þrÞ0gðCÞ
sf þ12ð1�rÞ0gðDÞ
sf ;
ð19Þwhere the tying positions ðAÞ, ðBÞ, ðCÞ and ðDÞ are shown in Fig. 2[1,2,4,8].
The linear and nonlinear parts of the incremental covariant in-plane strains in Eq. (9) are expressed as
0 2 4 6 8 10 12 14 16 180
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0 2 4 6 8 10 12 14 16 180
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
(b)
Displacements
BwAw
BwAw
)/( io RRPLoad
( )
)/( io RRPLoad
( )
(a)
reference3x24 MITC43x24 MITC4+
reference5x40 MITC45x40 MITC4+
Fig. 9. Load-displacement curves for the slit annular plate problem. (a) 3 � 24mesh. (b) 5 � 40 mesh.
Initial shape
z
Defor
med
mes
h
maxPP
x y
Fig. 10. Deformed shape for the slit annular plate problem.
x
A
B
Free
edge
C
D
P
P
Free
edge
R
y
z
L
β
Fig. 11. Pull-out of a free cylindrical shell structure (12 � 12 mesh, R ¼ 4:953,L ¼ 10:35, thickness a ¼ 0:094, E ¼ 1:05� 107 and m ¼ 0:3125).
6 Y. Ko et al. / Computers and Structures 185 (2017) 1–14
0eij ¼ 0emij þ f 0eb1ij þ f20eb2ij ;
0gij ¼ 0gmij þ f0gb1
ij þ f20gb2ij with i; j ¼ 1;2; ð20aÞ
in which
0emij ¼ 12ðtxm;i � um;j þ txm;j � um;iÞ; ð20bÞ
0eb1ij ¼ 12ðtxm;i � ub1;j þ txm;j � ub1;i þ txb;i � um;j þ txb;j � um;iÞ; ð20cÞ
0eb2ij ¼ 12ðtxb;i � ub1;j þ txb;j � ub1;iÞ; ð20dÞ
0gmij ¼ 1
2um;i � um;j; ð20eÞ
0gb1ij ¼ 1
2ðum;i � ub1;j þ um;j � ub1;i þ txm;i � ub2;j þ txm;j � ub2;iÞ; ð20fÞ
0gb2ij ¼ 1
2ðub1;i � ub1;j þ txb;i � ub2;j þ txb;j � ub2;iÞ; ð20gÞ
with
txm;i ¼ @txm
@ri; txb;i ¼ @txb
@ri; um;i ¼ @um
@ri;
ub1;i ¼ @ub1
@ri; ub2;i ¼ @ub2
@ri:
maxPP
Y. Ko et al. / Computers and Structures 185 (2017) 1–14 7
In Eq. (20a), the 0emij and 0gmij terms are the linear and nonlinear
incremental covariant in-plane strains at the shell mid-surface(f ¼ 0); these in-plane membrane strains can in general inducemembrane locking.
From the assumed Green-Lagrange membrane strains in Eq.(15), we derive the assumed incremental linear membrane strain0emij in Eq. (20b)
0~emrr ¼12ð1� 2taA þ sþ 2taA � s2Þ0emðAÞ
rr
þ 12ð1� 2taB � sþ 2taB � s2Þ0emðBÞ
rr
þ taCð�1þ s2Þ0emðCÞss þ taDð�1þ s2Þ0emðDÞ
ss
þ taEð�1þ s2Þ0emðEÞrs ; ð21aÞ
0~emss ¼ taAð�1þ r2Þ0emðAÞrr þ taBð�1þ r2Þ0emðBÞ
rr
þ 12ð1� 2taC þ r þ 2taC � r2Þ0emðCÞ
ss
þ 12ð1� 2taD � r þ 2taD � r2Þ0emðDÞ
ss þ taEð�1þ r2Þ0emðEÞrs ; ð21bÞ
0~emrs ¼14ðr þ 4taA � rsÞ0emðAÞ
rr þ 14ð�r þ 4taB � rsÞ0emðBÞ
rr
þ 14ðsþ 4taC � rsÞ0emðCÞ
ss þ 14ð�sþ 4taD � rsÞ0emðDÞ
ss
þ ð1þ taE � rsÞ0emðEÞrs ; ð21cÞ
with the same geometric coefficients as in Eq. (15e) and the tyingpositions ðAÞ, ðBÞ, ðCÞ, ðDÞ and ðEÞ in Fig. 4.
We employ the same assumed strain field for the incrementalnonlinear membrane strain 0gm
ij , and hence the incremental in-plane strain components are
0~eij ¼ 0~emij þ f 0eb1ij þ f20eb2ij and
0~gij ¼ 0~gmij þ f0gb1
ij þ f20gb2ij with i; j ¼ 1;2: ð22Þ
x104
0
0.5
1
1.5
2
2.5
3
3.5
4
Displacements
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
PLoad( ) AwDw
Bu
Bu Cu
Cu
reference12x12 MITC412x12 MITC4+
Fig. 12. Load-displacement curves for the pull-out of the free cylindrical shellstructure.
Using the shell-aligned local Cartesian coordinate systemdefined in Eq. (17), the linear and nonlinear parts of the incremen-tal local strains are calculated using the following transformations
0�eij ¼ 0~eklð0Li � 0gkÞð0Lj � 0glÞ;0�gij ¼ 0~gklð0Li � 0gkÞð0Lj � 0glÞ with 0gi � 0g j ¼ d j
i : ð23Þ
2.5. Stiffness matrix and internal force vector
Using the standard total Lagrangian formulation [4,18,31], thetangent stiffness matrix (tKe) and internal force vector (t0Fe) ofthe MITC4+ shell element are obtained
tKe ¼Z
0VBTij�CijklBkld
0V þZ
0V
t0�SijNijd
0V ; ð24aÞ
t0Fe ¼
Z0V
BTijt0�Sijd
0V ; ð24bÞ
in which 0V is the volume of the shell element at time 0, and �Cijkl
and t0�Sij denote, respectively, the material law tensor and the second
Piola-Kirchhoff stress measured in the local Cartesian coordinatesystem.
In Eq. (24), the strain-displacement matrices, Bij and Nij, aredefined by
0�eij ¼ BijUe; d0�gij ¼ dUTeNijUe; ð25Þ
where Ue is the incremental nodal displacement vector
Ue ¼ UT1 UT
2 UT3 UT
4
� �T with Ui ¼ ui v i wi ai bi½ �T :
x y
maxPP
z
Fig. 13. Deformed shape for the pull-out of the free cylindrical shell structure.
(b)
B D
Freeed
ge
R
Lβ
x y
z
θ
egdeeerFC
E
M
A
(c)
(a)
1L2L
3L
4L
1L
2L
3L
4L
CB A
E D
Fig. 14. Bending of a cylindrical shell structure (R ¼ 10:0, h ¼ 30� , L ¼ 20:0, thickness a, E ¼ 2:1� 106 and m ¼ 0:0). (a) Problem description (12 � 12 uniform mesh). (b)Distorted mesh pattern (4 � 4 mesh). (c) Distorted mesh pattern used (12 � 12 mesh).
8 Y. Ko et al. / Computers and Structures 185 (2017) 1–14
In the nonlinear solution procedure, the nodal geometry is
updated using Eq. (4b), and the vectors tVi1,
tVi2 and tVi
n at node iare updated using the following equations:
tþDtVin ¼ Q tVi
n;tþDtVi
1 ¼ Q tVi1;
tþDtVi2 ¼ Q tVi
2; ð26Þwith
Q ¼ 2
q20 þ q2
1 � 12 q1q2 � q0q3 q3q1 þ q0q2
q1q2 þ q0q3 q20 þ q2
2 � 12 q2q3 � q0q1
q3q1 � q0q2 q2q3 þ q0q1 q20 þ q2
3 � 12
0BBB@
1CCCA;
q0 ¼ coshi2
� �; q1 q2 q3½ �T ¼ hi
hisin
hi2
� �; hi ¼ khik;
in which a quaternion representation of large rotations is utilized[32].
In the finite element solutions, we use 2� 2� 2 Gauss integra-tion over the element volume for the 4-node shell elementsconsidered. The computational cost of the MITC4+ shell element
is only slightly higher than the cost of the classical MITC4 shellelement due to the use of the assumed covariant membranestrains.
3. Numerical examples
In this section, several numerical examples are solved todemonstrate the performance of the MITC4+ shell element in geo-metric nonlinear analysis. The results are compared with thoseobtained using of the classical MITC4 shell element. The referencesolutions are analytical data or are obtained using a fine uniformmesh of the MITC9 shell element, which is known to satisfy theellipticity and consistency conditions and to show goodconvergence behavior in both linear and nonlinear analyses[4–8,17–20].
We show in the example solutions that both the MITC4 andMITC4+ elements work well when uniform meshes are used anddue to the specific physical problem the large displacements ofthe meshes do not induce locking, but in contrast to the MITC4
(c)
(b)
(a)
Load( )0M
Aw
Displacements
Aw
Aw
Load( )0M
Load( )0M
Au
Au
Au
x104
x104
x104
0 1 2 3 4 5 6 7 8 90
0.4
0.8
1.2
1.6
2
2.4
2.8
3.2
3.6
4
0 1 2 3 4 5 6 7 8 90
0.4
0.8
1.2
1.6
2
2.4
2.8
3.2
3.6
4
0 1 2 3 4 5 6 7 8 90
0.4
0.8
1.2
1.6
2
2.4
2.8
3.2
3.6
4
reference12x12 MITC412x12 MITC4+
reference12x12 MITC412x12 MITC4+
reference12x12 MITC412x12 MITC4+
Fig. 15. Load-displacement curves for the bending of the cylindrical shell structurewith the uniform mesh. (a) a=R ¼ 1=100. (b) a=R ¼ 1=1000. (c) a=R ¼ 1=10;000.
(c)
(b)
(a)
Displacements
Aw
Aw
Load( )0M
Load( )0M
Load( )0M
Au
Au
Au
x104
0 1 2 3 4 5 6 7 8 9
Aw
0
0.4
0.8
1.2
1.6
2
2.4
2.8
3.2
3.6
4
x104
x1040 1 2 3 4 5 6 7 8 9
0
0.4
0.8
1.2
1.6
2
2.4
2.8
3.2
3.6
4
0 1 2 3 4 5 6 7 8 90
0.4
0.8
1.2
1.6
2
2.4
2.8
3.2
3.6
4
reference12x12 MITC412x12 MITC4+
reference12x12 MITC412x12 MITC4+
reference12x12 MITC412x12 MITC4+
Fig. 16. Load-displacement curves for the bending of the cylindrical shell structurewith the distorted mesh. (a) a=R ¼ 1=100. (b) a=R ¼ 1=1000. (c) a=R ¼ 1=10;000.
Y. Ko et al. / Computers and Structures 185 (2017) 1–14 9
A
D C
B
PP
P PA B
R
x y
z
0φ
CD
β
(b) (c)
(a)
1L2L
3L
4L
1L
2L
3L
4L
Fig. 17. Hemispherical shell problem (R ¼ 10:0, u0 ¼ 18� , thickness a ¼ 0:04, E ¼ 6:825� 107 and m ¼ 0:3). (a) Problem description (12 � 12 uniform mesh). (b) Distortedmesh pattern (4 � 4 mesh). (c) Distorted mesh pattern applied (12 � 12 mesh).
10 Y. Ko et al. / Computers and Structures 185 (2017) 1–14
shell element, the MITC4+ element also works well when an ini-tially distorted mesh is used for such problem solutions, see Sec-tions 3.1–3.4.
We also show the important point that the MITC4+ element ismore effective than the MITC4 element in response solutions whendue to the physical nature of the shell problem an initially uniformmesh in the large displacement response can induce locking. In thiscase, the MITC4 element locks in membrane actions whereas theMITC4+ element continues to work well, see Sections 3.5–3.7.
3.1. Cantilever problem
We consider the cantilever bending problem in Fig. 5 [11–13,33]. The cantilever fully clamped at one end is subjected toeither a shearing force P or bending moment M at the free tip.The cantilever is modeled with a 16 � 1 mesh for the MITC4 andMITC4 + shell elements.
For the shearing load case, the reference solution is obtainedusing a 32 � 1 mesh of MITC9 shell elements. We consider the
maximum load of Pmax ¼ 4P0 with P0 ¼ EI=L and I ¼ ba3=12 .
For the moment load case, the cantilever develops to form a cir-
cular arc of radius R ¼ EI=M with I ¼ ba3=12. Using this formula,
the following analytical tip displacements are obtained [12,13]
utip
L¼ M0
Msin
MM0
� 1;wtip
L¼ M0
M1� cos
MM0
� �; M0 ¼ EI
L: ð27Þ
The cantilever beam should bend into a complete circle at theapplied tip moment Mmax ¼ 2pM0.
Fig. 6 shows the load-displacement curves of the MITC4 andMITC4+ shell elements. The solutions using both elements agreewell with the reference and analytic solutions. Fig. 7 shows thedeformed shapes at successive load levels P=Pmax ¼ 0:25 and 1.0for the shearing load case and M=Mmax ¼ 0:25, 0.5 and 1.0 for themoment load case.
3.2. Slit annular plate problem
We next consider the slit annular plate problem shown in Fig. 8[12,13,15,34]. The shearing force P is applied at one end of the slitwhile the other end is fully clamped. We use 3 � 24 and 5 � 40meshes for the MITC4 and MITC4 + shell elements. The maximumload per unit length Pmax=ðRo � RiÞ ¼ 0:8 is considered. The refer-ence solution is obtained using a 10 � 80 mesh of MITC9 shell ele-ments. Fig. 9 shows the load-displacement curves. The solutionobtained using the MITC4+ shell element is slightly better thanthe solution using the MITC4 shell element. The final deformedshape of the structure calculated using the MITC4+ shell elementis presented in Fig. 10.
3.3. Pull-out of a free cylindrical shell
We consider a pull-out of the free cylindrical shell structureshown in Fig. 11 [12,13,34]. The shell structure is subjected to a
Fig. 18. Load-displacement curves for the hemispherical shell problem with theuniform mesh shown in Fig. 8(a). (a) 8 � 8 mesh. (b) 12 � 12 mesh.
0
50
100
150
200
250
300
350
400
0 1 2 3 4 5 6 7 8 9
0 1 2 3 4 5 6 7 8 90
50
100
150
200
250
300
350
400
Displacements
PLoad( )
PLoad( )
(b)
(a)
Au
Au
Bv
Bv
reference8x8 MITC48x8 MITC4+
reference12x12 MITC412x12 MITC4+
Fig. 19. Load-displacement curves for the hemispherical shell problem with thedistorted mesh shown in Fig. 8(c). (a) 8 � 8 mesh. (b) 12 � 12 mesh.
maxPP
maxPP
Y. Ko et al. / Computers and Structures 185 (2017) 1–14 11
pair of pull-out loads (P) at its center. Due to symmetry, only one-eighth of the structure corresponding to the shaded region ABCD inFig. 11 is modeled using a 12 � 12 mesh of the 4-node shell ele-ments. We use the following boundary conditions:w ¼ b ¼ 0 alongBC, u ¼ b ¼ 0 along AD, and v ¼ a ¼ 0 along AB. The analysis is per-formed up to Pmax ¼ 4� 104. The reference solutions are obtainedusing a 32 � 32 mesh of MITC9 shell elements. Fig. 12 shows theresulting load-displacement curves. The MITC4 and MITC4+ shellelements perform very well. Fig. 13 presents the final deformedshape of the structure obtained using the MITC4+ shell element.
x y
z maxPP
maxPP
Fig. 20. Deformed shape of the hemispherical shell (12 � 12 uniform mesh used).
3.4. Bending of a cylindrical shell structure
We solve the bending problem of a cylindrical shell structureshown in Fig. 14(a) [11,16–18]. The structure is subjected to uni-form bending momentM along BC. Three thickness to radius ratios,a=R ¼ 1=100, a=R ¼ 1=1000 and a=R ¼ 1=10;000, are considered.The applied moment varies with the thickness a considered
PIn-plane
POut-of-plane
zb
L
x y
A
Fig. 21. Twisted cantilever beam problems (4 � 24 mesh, L ¼ 12:0, b ¼ 1:1,thickness a ¼ 0:0032, E ¼ 2:9� 107 and m ¼ 0:22).
0 1 2 3 4 5 6 70
0.5
1
1.5
2
2.5
3
3.5
4
0 1 2 3 4 5 6 7 8 9 10
8 9
11 120
0.5
1
1.5
2
2.5
3
3.5
4
(b)
Displacements
(a)x10-2
x10-2
PLoad( )
PLoad( )
Au
Au
Aw
Aw
reference4x24 MITC44x24 MITC4+
reference4x24 MITC44x24 MITC4+
Fig. 22. Load-displacement curves for the twisted cantilever beam problem. (a)Case of the in-plane load. (b) Case of the out-of-plane load.
12 Y. Ko et al. / Computers and Structures 185 (2017) 1–14
according to M ¼ M0a3. A fully clamped boundary condition isapplied, u ¼ v ¼ w ¼ a ¼ b ¼ 0 along DE. For each thickness, weconsider the load level up to ðM0Þmax ¼ 4:0� 104.
In addition to the uniform mesh in Fig. 14(a), we also considerthe distorted mesh pattern shown in Fig. 14(b). For an N � N ele-ment mesh, a pair of edges are discretized in the following ratio:L1: L2:L3: . . . LN = 1: 2: 3: . . . N. The distorted mesh is shown inFig. 14(c). For the 4-node shell elements a 12 � 12 mesh is used.A 32 � 32 mesh of MITC9 shell elements is employed to obtainthe reference solution.
Fig. 15 shows the load-displacement curves for the uniformmesh, where the solutions obtained using the MITC4 and MITC4+shell elements agree well with the reference solution. Fig. 16shows the load-displacement curves when the distorted mesh isused. As the shell thickness decreases, the solutions obtained usingthe MITC4 shell element depart from the reference solution. How-ever, using the MITC4+ shell element good response predictionsare always obtained.
3.5. Hemispherical shell problem
We solve the hemispherical shell problem shown in Fig. 17(a)[13–15,17,18,34]. The hemispherical shell with an 18� cutout atits pole is subjected to alternating radial point forces (P) at itsequator. In this bending problem, the shell structure undergoesalmost inextensional deformations and thus we test whethermembrane locking occurs. Due to symmetry, only one quarter ofthe structure corresponding to the shaded region ABCD in Fig. 17(a) is modeled using uniform meshes of 8 � 8 and 12 � 124-node shell elements. We use the following boundary conditions:u ¼ b ¼ 0 along BC, v ¼ b ¼ 0 along AD, and w ¼ 0 at A. The max-imum load of Pmax ¼ 400 is considered.
In addition to the uniform mesh in Fig. 17(a), we consider thedistorted mesh pattern shown in Fig. 17(b) in which we use anN � N element mesh, each edge is discretized in the followingratio: L1: L2: L3: . . . LN = 1: 2: 3: . . . N. The distorted mesh in onequarter of the hemisphere is shown in Fig. 17(c). To obtain the ref-erence solution, a 32 � 32 uniformmesh of MITC9 shell elements isemployed.
Figs. 18 and 19 present the load-displacement curves for theuniform and distorted meshes, respectively. As the mesh is refined,the solutions obtained using the MITC4+ shell element converge tothe reference solution more rapidly than those obtained with theMITC4 shell element. For the distorted mesh cases, the MITC4 shellelement gives a response with a large error from the referencesolution. However, the MITC4+ shell element still shows a goodbehavior. Fig. 20 gives the final deformed shape of the hemispher-ical shell calculated using the MITC4+ shell element.
3.6. Twisted cantilever beam problems
We consider the twisted cantilever beam problems shown inFig. 21 [26,34,35]. The initially twisted beam is fully clamped atone end and is loaded by a point load P at the center of the freetip. Two load cases are considered: an in-plane and anout-of-plane load as shown in Fig. 21. We use a 4 � 24 mesh ofthe 4-node elements while an 8 � 48 mesh of MITC9 shell ele-ments is used for the reference solutions. The maximum load levelis Pmax ¼ 4� 10�2 for both load cases.
Fig. 22 gives the load-displacement curves for both the in-planeand out-of-plane loads. No severe locking is present for the in-plane load case, where both 4-node shell elements perform well.When the out-of-plane load is applied, the response predictedusing the MITC4 shell element deviates significantly from the ref-erence solution due to membrane locking. However, the MITC4+
Fig. 23. Deformed shapes of the cantilever beam. (a) Case of the in-plane load. (b)Case of the out-of-plane load.
A
xy1R
2R
1θ
2θ
b
PA
x y
z8 elements
12 elements
Fig. 24. Hook problem (4 � 20 mesh, R1 ¼ 14, h1 ¼ 60� , R2 ¼ 46, h2 ¼ 150� , b ¼ 20,thickness a ¼ 0:02, E ¼ 3:3� 103 and m ¼ 0:3).
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1x10-4
PLoad( )
reference4x20 MITC44x20 MITC4+
Au Aw
Y. Ko et al. / Computers and Structures 185 (2017) 1–14 13
shell element shows a good predictive capability. The finaldeformed shapes of the cantilever beam obtained using theMITC4+ shell element are presented in Fig. 23.
0 10 20 30 40 50 60 70 800
0.1
Displacements
Fig. 25. Load-displacement curves for the hook problem.
3.7. Hook problem
Finally, we consider the hook problem shown in Fig. 24, referredto in linear analysis as the Raasch challenge, see Ref. [36]. Thestructure is fully clamped at one end and is loaded by a shear loadP applied as a uniformly distributed traction at the free tip. For thesolution, we use a 4 � 20 mesh with the MITC4 and MITC4+ shellelements and an 8 � 40 mesh of MITC9 shell elements to obtainthe reference solution. The load is applied up toPmax ¼ 1:0� 10�4. Fig. 25 shows the resulting load-displacementcurves. Using the MITC4+ shell element produces a significantly
more accurate solution than using the MITC4 shell element. Thefinal deformed shape of the hook obtained using the MITC4+ shellelement is shown in Fig. 26.
x y
z
maxPP
Initial shape
Deformed mesh
Fig. 26. Deformed shape of the hook.
14 Y. Ko et al. / Computers and Structures 185 (2017) 1–14
4. Concluding remarks
We presented the geometric nonlinear formulation of theMITC4+ continuum mechanics-based shell element which is for-mulated using the MITC approach to alleviate shear and membranelocking. The assumed shear and membrane strain fields used in lin-ear analysis are developed for geometric nonlinear analysis in aconsistent manner. The nonlinear performance of the MITC4+ shellelement is numerically tested through the solutions of variousexamples. The computational cost of the element is only slightlyhigher than the cost of the classical MITC4 shell element.
We can conclude that the MITC4+ shell element provides reli-able and efficient solutions in large displacement problems. Com-pared to the original MITC4 element, the MITC4+ shell elementshows improved performance when distorted meshes in the initialconfiguration are used. Moreover, the MITC4+ element performsmuch better than the MITC4 shell element when due to the natureof the shell problem, the large displacements of the mesh caninduce locking. Hence, we can conclude that the MITC4+ shell ele-ment shows excellent behavior in both linear and nonlinearanalyses.
Acknowledgments
This work was supported by the Basic Science Research Pro-gram through the National Research Foundation of Korea fundedby the Ministry of Science, ICT, and Future Planning (No.2014R1A1A1A05007219), and a grant (MPSS-CG-2016-04) throughthe Disaster and Safety Management Institute funded by Ministryof Public Safety and Security of Korean government.
References
[1] Ko Y, Lee PS, Bathe KJ. A new MITC4+ shell element. Comput Struct2017;182:404–18.
[2] Dvorkin EN, Bathe KJ. A continuum mechanics based four-node shell elementfor general nonlinear analysis. Eng Comput 1984;1(1):77–88.
[3] Bathe KJ. The finite element method. In: Wah B, editor. Encyclopedia ofcomputer science and engineering. John Wiley and Sons; 2009. p. 1253–64.
[4] Bathe KJ. Finite element procedures. Prentice Hall; 1996, 2nd edition KJ Bathe,Watertown, MA; 2014 and Higher Education Press, China, 2016.
[5] Bathe KJ. The inf–sup condition and its evaluation for mixed finite elementmethods. Comput Struct 2001;79(2):243–52.
[6] Bathe KJ, Iosilevich A, Chapelle D. An inf-sup test for shell finite elements.Comput Struct 2000;75(5):439–56.
[7] Bathe KJ, Lee PS. Measuring the convergence behavior of shell analysisschemes. Comput Struct 2011;89(3):285–301.
[8] Chapelle D, Bathe KJ. The finite element analysis of shells – fundamentals. 2nded. Berlin: Springer; 2003. p. 2011.
[9] Chapelle D, Bathe KJ. Fundamental considerations for the finite elementanalysis of shell structures. Comput Struct 1998;66(1):19–36. 711–2.
[10] Lee PS, Bathe KJ. On the asymptotic behavior of shell structures and theevaluation in finite element solutions. Comput Struct 2002;80(3):235–55.
[11] Bathe KJ, Dvorkin E, Ho LW. Our discrete-Kirchhoff and isoparametric shellelements for nonlinear analysis - an assessment. Comput Struct 1983;16(1):89–98.
[12] Arciniega RA, Reddy JN. Tensor-based finite element formulation forgeometrically nonlinear analysis of shell structures. Comput Meth ApplMech Eng 2007;196(4):1048–73.
[13] Sze KY, Liu XH, Lo SH. Popular benchmark problems for geometric nonlinearanalysis of shells. Finite Elem Anal Des 2004;40(11):1551–69.
[14] Klinkel S, Gruttmann F, Wagner W. A robust non-linear solid shell elementbased on a mixed variational formulation. Comp Meth App Mech Eng2006;195(1):179–201.
[15] Jeon HM, Lee Y, Lee PS, Bathe KJ. The MITC3+ shell element in geometricnonlinear analysis. Comput Struct 2015;146:91–104.
[16] Bathe KJ, Dvorkin EN. A formulation of general shell elements - the use ofmixed formulation of tensorial components. Int J Numer Meth Eng 1986;22(3):697–722.
[17] Bucalem ML, Bathe KJ. Higher-order MITC general shell elements. Int J NumerMeth Eng 1993;36(21):3729–54.
[18] BucalemML, Bathe KJ. Finite element analysis of shell structures. Arch ComputMeth Eng 1997;4(1):3–61.
[19] Bathe KJ, Lee PS, Hiller JF. Towards improving the MITC9 shell element.Comput Struct 2003;81(8):477–89.
[20] Lee PS, Bathe KJ. Insight into finite element shell discretizations by use of the‘‘basic shell mathematical model”. Comput Struct 2005;83(1):69–90.
[21] Lee PS, Bathe KJ. Development of MITC isotropic triangular shell finiteelements. Comput Struct 2004;82(11):945–62.
[22] Lee Y, Yoon K, Lee PS. Improving the MITC3 shell finite element by using theHellinger-Reissner principle. Comput Struct 2012;110–111:93–106.
[23] Lee Y, Lee PS, Bathe KJ. The MITC3+ shell element and its performance. ComputStruct 2014;138:12–23.
[24] Lee PS, Bathe KJ. The quadratic MITC plate and MITC shell elements in platebending. Adv Eng Soft 2010;41(5):712–28.
[25] Vampa V. Analysis of in-layer strains in the low order MITC shell element.Latin Am J Solid Struct 2007;4(2):87–102.
[26] Belytschko T, Leviathan I. Projection schemes for one-point quadrature shellelements. Comp Meth Appl Mech Eng 1994;115(3):277–86.
[27] Choi CK, Paik JG. An efficient four node degenerated shell element based on theassumed covariant strain. Struct Eng Mech 1994;2(1):17–34.
[28] Kulikov GM, Plotnikova SV. A family of ANS four-node exact geometry shellelements in general convected curvilinear coordinates. Int J Numer Meth Eng2010;83(10):1376–406.
[29] Wall WA, Bischoff M, Ramm E. A deformation dependent stabilizationtechnique, exemplified by EAS elements at large strains. Comput Meth ApplMech Eng 2000;188(4):859–71.
[30] Sussman T, Bathe KJ. Spurious modes in geometrically nonlinear smalldisplacement finite elements with incompatible modes. Comput Struct2014;140:14–22.
[31] Yoon K, Lee PS. Nonlinear performance of continuum mechanics based beamelements focusing on large twisting behaviors. Comp Meth Appl Mech Eng2014;281:106–30.
[32] Greenwood DT. Advanced dynamics. Cambridge University Press; 2006.[33] Ko Y, Lee PS. A 6-node triangular solid-shell element for linear and nonlinear
analysis. Int J Numer Meth Eng 2017. http://dx.doi.org/10.1002/nme.5498.[34] Jung WY, Han SC. An 8-node shell element for nonlinear analysis of shells
using the refined combination of membrane and shear interpolation functions.Math Prob Eng 2013;2013.
[35] Belytschko T, Wong BL, Stolarski H. Assumed strain stabilization procedure forthe 9-node Lagrange shell element. Int J Numer Meth Eng 1989;28(2):385–414.
[36] Knight NF. Raasch challenge for shell elements. AIAA J 1997;35(2):375–81.