a new 8-node element for analysis of three-dimensional...
TRANSCRIPT
Computers and Structures 202 (2018) 85–104
Contents lists available at ScienceDirect
Computers and Structures
journal homepage: www.elsevier .com/locate /compstruc
A new 8-node element for analysis of three-dimensional solids
https://doi.org/10.1016/j.compstruc.2018.02.0150045-7949/� 2018 Elsevier Ltd. All rights reserved.
⇑ Corresponding author.E-mail address: [email protected] (K.J. Bathe).
Yeongbin Ko a, Klaus-Jürgen Bathe b,⇑a 227 Dongil-ro 86, Nowon-gu, Seoul 01618, 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 10 January 2018Accepted 26 February 2018
Keywords:3D finite element for solids8-node elementMITC methodEAS methodIncompatible modesShear and volumetric locking
a b s t r a c t
We propose a new 8-node hexahedral element, the 3D-MITC8 element, for the analysis of three-dimensional solids. We use the MITC method and find the assumed strain field from a thought experi-ment using a truss idealization. For geometric nonlinear analysis, when needed to suppress hour-glassdeformations, the formulation also uses automatically displacement-based contributions to the shearstrains. The element shows a much better predictive capability than the displacement-based element.It is computationally more effective than the 8-node element with incompatible modes, and consideringaccuracy, in linear analysis performs almost as well, and in nonlinear analyses we do not observe spuri-ous instabilities. We show that the new 3D solid element passes all basic tests (the isotropy, zero energymode and patch tests) and present the finite element solutions of various benchmark problems to illus-trate the solution accuracy reached with the new element.
� 2018 Elsevier Ltd. All rights reserved.
1. Introduction
A three-dimensional 8-node hexahedral solid finite element isfrequently employed for the finite element analysis of solids inengineering practice. The element can be used to model manythree-dimensional (3D) solids and performs considerably betterthan the 4-node tetrahedral element. However, the standard 8-node 3D solid element does not satisfy the inf-sup conditions,hence the solution accuracy can severely deteriorate due to shearand volumetric locking [1,2].
To improve the behavior of the standard displacement-basedelement, additional ‘‘incompatible modes” are frequently used[3]. The incompatible modes technique is a special case of theenhanced assumed strain (EAS) method, and the resulting 8-node3D solid element requires, compared to the standard puredisplacement-based element, an additional 9 internal degrees offreedom to represent the conditions of pure bending [3–6]. The ele-ment is quite powerful since it alleviates both shear and volumet-ric locking, but it uses the additional degrees of freedom and canshow a non-physical instability in the analyses of nonlinear prob-lems [5–8].
The instability of the EAS elements has been observed to occurin both small and large strain nonlinear analyses [7–14]. If an ele-ment mesh is subjected to compression, a spurious hour-glassbending mode may occur in elements eventually resulting into
an indefinite stiffness matrix at a certain critical compressive strain[8]. Initially, the hour-glass deformations are small but as theygrow, the incremental analysis leads to a spurious collapse of themodel.
To treat the spurious instability special solution methods andvarious element formulations have been developed. The varia-tional principle for nonlinear analysis has been modified, stabiliza-tion parameters have been proposed, and mixed-enhancedelements have been developed [9–14], see these references andthe references therein. However, further developments are ofmuch interest and, based on our success of developing reliableand efficient shell elements based on the MITC technique [1], webelieve that we can also obtain an effective 3D eight-node MITCelement.
In this paper we propose a new 8-node hexahedral elementbased on the standard displacement interpolations and the MITC(Mixed Interpolation of Tensorial Components) approach[1,2,10,15–20]. To obtain a stable element, we choose the tyingpositions and strain interpolations based on the physical behaviorof a simple truss structure that idealizes the 8-node solid element.For geometric nonlinear analysis to suppress hour-glass deforma-tions, the formulation also uses automatically when needed a sta-bilization scheme based on displacement-based contributions inshear strains. Using these key ideas for the assumed strain field,we find the 3D-MITC8 element to give solution stability, good solu-tion accuracy, and to be computationally efficient. We also extendthis element to obtain the 3D-MITC8/1 element based on a mixeddisplacement-pressure formulation.
86 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
In the next section we present the concepts we use for the tyingand interpolation of the strains in the MITC procedure. Then, inSection 3, we propose the new 3D-MITC8 and 3D-MITC8/1 ele-ments using the total Lagrangian (T.L.) formulation. Of course,the linear behavior corresponds to the first step in the T.L. formu-lation. Further, in Section 4, the stability and accuracy of the ele-ment are assessed using basic tests (the isotropy, zero energymode and patch tests) and the solutions of various benchmarkproblems. Finally, in Section 5, we present our conclusions.
2. Tying and interpolation of strains in the MITC procedure
In this section, we present the concepts we employ in the MITCprocedure for the new element. The tying positions and interpola-tions of the assumed strain components are developed consideringstability in linear and nonlinear analyses.
The geometry of an 8-node hexahedral solid element is shownin Fig. 1. The element domain is given and the strain componentsare defined corresponding to the three natural coordinates, r, sand t. For the 3D solid element, there are three normal (in thedirections of r, s and t) and three shear (on the planes of rs, stand tr) strain components. The six assumed strain componentsare denoted by 0~err , 0~ess, 0~ett , 0~ers, 0~est and 0~etr .
The choice of assumed strain interpolation must be such thatthe solid element is stable corresponding to each strain compo-nent. To obtain insight, we idealize the hexahedral domain as atruss structure with 8 joints that correspond to the nodes, seeFig. 2(a), of the 8-node 3D element. The selected truss structureis shown in Fig. 2(b). This structure is stable and consists of theminimum number of 2-node truss elements. We next considerthe location and direction of each truss element to correspond toan assumed strain component. The location of tying is given bythe truss element but to obtain better accuracy using the 3D ele-ment in analyses we can move these locations to correspondingGauss integration points.
If the truss structure we use is stable with the minimum num-ber of truss elements, we can expect that in linear analysis the 3DMITC element will also be stable and will not lock, because a min-imum number of truss elements is used. The use of the truss struc-ture to idealize the solid element is similar to how the classicaltransverse shear assumption was developed by Dvorkin and Bathefor 4-node shell elements, notably for the MITC4 shell element[15]. Here the 4-node shell element transverse shear behaviorwas idealized by the behavior of four 2-node isoparametric beamelements located along the edges of the shell element, with eachbeam assuming a constant transverse shear strain [1,15].
Fig. 1. A standard 8-node hexahedral 3D solid element.
We place a tying location at the center of each truss, and inter-polate the assumed strain components according to these loca-tions, see Fig. 2(c). The normal strains (0~err , 0~ess and 0~ett) areinterpolated bilinearly over the planes defined by their respectivetying locations and the shear strains (0~ers, 0~est and 0~etr) are interpo-lated linearly between the tying points.
While the resulting assumed strain field yields stability in linearanalysis, there is an instability that can arise in nonlinear analysis.The phenomenon has been widely observed for enhanced assumedstrain elements when initially regular meshes undergo compres-sion [7–14]. Indeed, for the incompatible modes elements, spuri-ous bending deformations or hour-glass modes are seen at acritical state even in small strains [8]. For an 8-node hexahedralelement, possible 2D and 3D hour-glass modes are depicted inFig. 3(a) and (b), respectively. This behavior occurs if an 8-nodeelement has the ability to express pure bending deformationsand the surrounding elements cause mixed behavior of bendingand compression. The behavior is possible for the incompatiblemodes element.
The mechanism of this nonlinear instability was studied bySussman and Bathe [8], where it was found that a spurious bendingdeformation occurs when a critical compressive strain state isreached. For the two different kinds of hour-glass modes we treatthe potential instabilities separately. We suppress the accumula-tion of a 3D hour-glass mode by using the incremental displace-ments to calculate the constant compressive strain. Further, wesuppress the 2D hour-glass modes and their coupling to the 3Dhour-glass mode by interpolating an additional stabilizing shearstrain term bilinearly on the respective planes defined by themid-points on the edges, see Fig. 3(c).
Incorporating these ideas, the assumed strain field is proposedas
0~err ¼ A0rr þ A1
rrsþ A2rrt þ A3
rrst;
0~ess ¼ A0ss þ A1
sst þ A2ssr þ A3
sstr;
0~ett ¼ A0tt þ A1
ttr þ A2ttsþ A3
ttrs;
0~ers ¼ A0rs þ A1
rst þ S1rsr þ S2rssþ S3rsrs;
0~est ¼ A0st þ A1
str þ S1stsþ S2stt þ S3stst;
0~etr ¼ A0tr þ A1
trsþ S1trt þ S2trr þ S3trtr;
ð1Þ
in which the Akij and Skij are the unknown strain coefficients. The con-
stants Akij (k = 0, 1, 2, 3) and corresponding interpolations allow
overall stability in linear and nonlinear analyses. The constants Skij(k = 1, 2, 3) are designed to automatically suppress 2D hour-glassdeformations in nonlinear solutions and are significant only whencompressive strain has been accumulated.
3. Formulation of 3D-MITC8 element
We use the left-superscript t to denote the current configura-tion (or ‘time’) of the element. We employ the total Lagrangian for-mulation with the reference configuration at time 0 indicated bythe left subscript 0.
The geometry and displacement of the standard 8-node hexahe-dral 3D solid element is interpolated by [1]
tx ¼X8i¼1
hiðr; s; tÞtxi
¼ txix þ tyiy þ tziz
¼ tx ty tz½ �T ;
with txi ¼ txi tyitzi½ �T ,
Fig. 2. Identification of tying strains and positions. (a) 8 joints forming a hexahedron. (b) An ideal (minimally stable) truss structure. (c) Tying positions for each assumedstrain component.
Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104 87
hiðr; s; tÞ ¼ 18ð1þ nirÞð1þ gisÞð1þ fitÞ;
n1 n2 n3 n4 n5 n6 n7 n8½ �¼ �1 1 1 �1 �1 1 1 �1½ �;
g1 g2 g3 g4 g5 g6 g7 g8½ �¼ �1 �1 1 1 �1 �1 1 1½ �;
f1 f2 f3 f4 f5 f6 f7 f8½ �¼ �1 �1 �1 �1 1 1 1 1½ �; ð2Þ
in which hiðr; s; tÞ is the standard 8-node interpolation function forthe isoparametric procedure, and ix, iy, iz are base vectors in the glo-bal Cartesian coordinate system, see Fig. 1. The incremental dis-placement vector u from the configuration at time t to theconfiguration at time t þ Dt is
uðr; s; tÞ ¼ tþDtxðr; s; tÞ � txðr; s; tÞ;
u ¼X8i¼1
hiðr; s; tÞui
¼ uix þ viy þwiz¼ ui v i wi½ �T
with ui ¼ ui v i wi½ �T : ð3ÞThe covariant base vectors and displacement derivatives for the 3Dsolid element are given as
tgr ¼@ tx@r
; tgs ¼@ tx@s
; tgt ¼@ tx@t
; u;r ¼ @u@r
;
u;s ¼ @u@s
; u;t ¼ @u@t
; ð4Þ
and we define the following ‘distortion vectors’
txrs ¼ 18
X8i¼1
nigitxi;
txst ¼ 18
X8i¼1
gifitxi;
txtr ¼ 18
X8i¼1
finitxi;
txrst ¼ 18
X8i¼1
nigifitxi; ð5Þ
Fig. 3. Identification of tying strains and positions. (a) Two-dimensional hour-glass modes and corresponding distortion vectors. (b) Three-dimensional hour-glass mode andcorresponding distortion vector. (c) Tying positions for each assumed strain component.
Fig. 4. Characteristic geometry vectors. (a) Three covariant base vectors at thecenter of the element. (b) The three covariant base vectors characterizing a regularmesh.
88 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
which measure hour-glass deformations. The three ‘2D distortionvectors’ txrs, txst and txtr correspond to 2D deformations, and the‘3D distortion vector’ txrst corresponds to a 3D deformation, seeFig. 3(a) and (b).
We define the following covariant base vectors at the elementcenter,
tg_
i ¼ tgið0;0;0Þ; with tg_
1 ¼ tg_
r ;tg_
2 ¼ tg_
s;tg_
3 ¼ tg_
t: ð6Þ
These three vectors characterize the regular geometry (tV_
e) of theelement domain (tVe), see Fig. 4.
The seven vectors tg_
r , tg_
s, tg_
t , txrs, txst , txtr and txrst establish thegeometry of the 8-node hexahedral element and we define the cor-responding incremental displacements from time t to t þ Dt using
tXC ¼ tg_
rtg_
stg_
ttxrs
txsttxtr
txrst
h i;
UC ¼ tþDtXC � tXC ¼ ur us ut urs ust utr urst½ �; ð7Þ
where we note that the displacement vectors ur , us, ut , urs, ust , utr
and urst are independent of each other.The contravariant base vectors are obtained from the covariant
base vectors using the standard relationship [1]
tgi � tg j ¼ d ji ; with tg1 ¼ tgr ;
tg2 ¼ tgs;tg3 ¼ tgt ;
tg1 ¼ tgr; tg2 ¼ tgs; tg3 ¼ tgt : ð8Þ
The covariant Green-Lagrange strain components in the configura-tion at time t with respect to the reference configuration at time0 are
Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104 89
t0eijðr; s; tÞ ¼
12ðtgi � tgj � 0gi � 0gjÞ: ð9Þ
Using Eq. (3) in Eq. (9) applied at time t and t þ Dt, the incrementalcovariant strain components are
0eijðr; s; tÞ ¼ tþDt0eijðr; s; tÞ � t
0eijðr; s; tÞ
¼ 12ðtgi � u;j þ u;i � tgj þ u;i � u;jÞ; ð10Þ
which can be separated into a linear part (0eij) and a nonlinear part(0gij)
0eij ¼ 12ðtgi � u;j þ u;i � tgjÞ; 0gij ¼
12ðu;i � u;jÞ; ð11Þ
in which tgi ¼ @tx@ri
and u;i ¼ @u@ri
with r1 ¼ r, r2 ¼ s, r3 ¼ t. Note that
0e11 ¼ 0err , 0e22 ¼ 0ess, 0e33 ¼ 0ett , 0e12 ¼ 0ers, 0e23 ¼ 0est and0e31 ¼ 0etr , and similarly for 0gij.
We use the strain components transformed to a fixed covariantcoordinate system defined at the element center in the referenceconfiguration
0 e_
ij ¼ 0eklgki g
lj; 0g
_
ij ¼ 0gklgki g
lj with g j
i ¼ 0g_
i � 0g j; ð12Þand
0kðr; s; tÞ ¼0jð0;0;0Þ0jðr; s; tÞ with 0jðr; s; tÞ ¼ det 0gr
0gs0gt
� �; ð13Þ
as done in previous works [4–6,10,14,16].We also use the following geometric measure in geometric non-
linear analysis
tc ¼ �X8i¼1
nitxi
!�X8i¼1
nitxi
!�
X8i¼1
gitxi
!�X8i¼1
gitxi
!
�X8i¼1
fitxi
!�X8i¼1
fitxi
!;
t0c ¼ tc � 0c; ð14Þwhere tc measures the normal compressive deformation of the hex-ahedral element and t
0c accumulates the history of the measure.In order to correctly account for the element domain (0Ve), we
employ the volume-averaged constant incremental strains [16]
0 e_conij ¼ 1
0V
Z0Ve
0 e_
ijd0Ve; 0g_con
ij ¼ 10V
Z0Ve
0g_
ijd0Ve;
0V ¼Z
0Ve
d0Ve; ð15Þ
which results in the same constant strain part as given by the stan-dard 8-node displacement-based element.
We express the 3D distortion vector as
txrst ¼ tdrtg_
r þ tdstg_
s þ tdttg_
t ; ð16Þwith
tdr ¼ txrst � tg_r ; tds ¼ txrst � tg
_s; tdt ¼ txrst � tg_t with tg
_
i � tg_ j ¼ d j
i ;
and obtain likewise the corresponding displacement vector
urst ¼ tdrur þ tdsus þ tdtut : ð17ÞIn Eq. (17) the 3D distortions are expressed using the regular defor-mations. To obtain overall stability in nonlinear analysis, the 3Dhour-glass deformations need to be suppressed using the constantstrain parts. Hence, we use Eq. (17) in Eq. (15) and denote the result-ing ‘constant strain’ as 0~econij and 0~gcon
ij , see Appendix A for details.
Based on the above considerations, we thus use for the 3D-MITC8 element the assumed incremental linear strains
0~err ¼ 0~econrr
þffiffiffi3
p
4 0 e_
�0; 1ffiffi
3p ; 1ffiffi
3p�
rr0kðr; s; tÞ sþ t þ
ffiffiffi3
pst
� �
þffiffiffi3
p
4 0 e_
�0;� 1ffiffi
3p ;� 1ffiffi
3p�
rr0kðr; s; tÞ �s� t þ
ffiffiffi3
pst
� �
þffiffiffi3
p
4 0 e_
�0; 1ffiffi
3p ;� 1ffiffi
3p�
rr0kðr; s; tÞ s� t �
ffiffiffi3
pst
� �
þffiffiffi3
p
4 0 e_
�0;� 1ffiffi
3p ; 1ffiffi
3p�
rr0kðr; s; tÞ �sþ t �
ffiffiffi3
pst
� �;
0~ess ¼ 0~econss
þffiffiffi3
p
4 0 e_
�1ffiffi3
p ;0; 1ffiffi3
p�
ss0kðr; s; tÞ t þ r þ
ffiffiffi3
ptr
� �
þffiffiffi3
p
4 0 e_
�� 1ffiffi
3p ;0;� 1ffiffi
3p�
ss0kðr; s; tÞ �t � r þ
ffiffiffi3
ptr
� �
þffiffiffi3
p
4 0 e_
�1ffiffi3
p ;0;� 1ffiffi3
p�
ss0kðr; s; tÞ �t þ r �
ffiffiffi3
ptr
� �
þffiffiffi3
p
4 0 e_
�� 1ffiffi
3p ;0; 1ffiffi
3p�
ss0kðr; s; tÞ t � r �
ffiffiffi3
ptr
� �;
0~ett ¼ 0~econtt
þffiffiffi3
p
4 0 e_
�1ffiffi3
p ; 1ffiffi3
p ;0�
tt0kðr; s; tÞ r þ sþ
ffiffiffi3
prs
� �
þffiffiffi3
p
4 0 e_
�� 1ffiffi
3p ;� 1ffiffi
3p ;0�
tt0kðr; s; tÞ �r � sþ
ffiffiffi3
prs
� �
þffiffiffi3
p
4 0 e_
�1ffiffi3
p ;� 1ffiffi3
p ;0�
tt0kðr; s; tÞ r � s�
ffiffiffi3
prs
� �
þffiffiffi3
p
4 0 e_
�� 1ffiffi
3p ; 1ffiffi
3p ;0�
tt0kðr; s; tÞ �r þ s�
ffiffiffi3
prs
� �;
0~ers ¼ 0~econrs þ 12 0 e
_ð0;0;1Þrs � 0 e
_ð0;0;�1Þrs
� �0kðr; s; tÞt
þ tanhðt0cÞ0 e_stabrs ðr; sÞ0kðr; s; tÞ;
0~est ¼ 0~econst þ 12 0 e
_ð1;0;0Þst � 0 e
_ð�1;0;0Þst
� �0kðr; s; tÞr
þ tanhðt0cÞ0 e_stabst ðs; tÞ0kðr; s; tÞ;
0~etr ¼ 0~econtr þ 12ð0 e
_ð0;1;0Þtr � 0 e
_ð0;�1;0Þtr Þ0kðr; s; tÞs
þ tanhðt0cÞ0 e_stabtr ðt; rÞ0kðr; s; tÞ; ð18aÞ
0 e_stabrs ¼
ffiffiffi3
p
8
�0 e_
�1ffiffi3
p ; 1ffiffi3
p ;1�
rs þ 0 e_
�1ffiffi3
p ; 1ffiffi3
p ;�1�
rs
�r þ sþ
ffiffiffi3
prs
� �
þffiffiffi3
p
8
�0 e_
�� 1ffiffi
3p ;� 1ffiffi
3p ;1�
rs þ 0 e_
�� 1ffiffi
3p ;� 1ffiffi
3p ;�1
�rs
��r � sþ
ffiffiffi3
prs
� �
þffiffiffi3
p
8
�0 e_
�1ffiffi3
p ;� 1ffiffi3
p ;1�
rs þ 0 e_
�1ffiffi3
p ;� 1ffiffi3
p ;�1�
rs
�r � s�
ffiffiffi3
prs
� �
þffiffiffi3
p
8
�0 e_
�� 1ffiffi
3p ; 1ffiffi
3p ;1�
rs þ 0 e_
�� 1ffiffi
3p ; 1ffiffi
3p ;�1
�rs
��r þ s�
ffiffiffi3
prs
� �;
90 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
0 e_stabst ¼
ffiffiffi3
p
8
�0 e_
�1; 1ffiffi
3p ; 1ffiffi
3p�
st þ 0 e_
��1; 1ffiffi
3p ; 1ffiffi
3p�
st
�sþ t þ
ffiffiffi3
pst
� �
þffiffiffi3
p
8
�0 e_
�1;� 1ffiffi
3p ;� 1ffiffi
3p�
st þ 0 e_
��1;� 1ffiffi
3p ;� 1ffiffi
3p�
st
��s� t þ
ffiffiffi3
pst
� �
þffiffiffi3
p
8
�0 e_
�1; 1ffiffi
3p ;� 1ffiffi
3p�
st þ 0 e_
��1; 1ffiffi
3p ;� 1ffiffi
3p�
st
�s� t �
ffiffiffi3
pst
� �
þffiffiffi3
p
8
�0 e_
�1;� 1ffiffi
3p ; 1ffiffi
3p�
st þ 0 e_
��1;� 1ffiffi
3p ; 1ffiffi
3p�
st
��sþ t �
ffiffiffi3
pst
� �;
Fig. 6. Cantilever problem (1 � 4 mesh, E ¼ 3:0� 104 and m ¼ 0:0). (a) Regularmesh. (b) Distorted mesh.
Fig. 5. Mesh used for the patch test.
Table 1Predicted y-displacement at the tip (point A) in the cantilever problem.
Elements Regular mesh Distorted mesh
H8 0.235611 0.204659H8I9 0.347833 0.3453073D-MITC8 0.347833 0.304211Exact solution 0.347833
0 e_stabtr ¼
ffiffiffi3
p
8
�0 e_
�1ffiffi3
p ;1; 1ffiffi3
p�
tr þ 0 e_
�1ffiffi3
p ;�1; 1ffiffi3
p�
tr
�t þ r þ
ffiffiffi3
ptr
� �
þffiffiffi3
p
8
�0 e_
�� 1ffiffi
3p ;1;� 1ffiffi
3p�
tr þ 0 e_
�� 1ffiffi
3p ;�1;� 1ffiffi
3p�
tr
��t � r þ
ffiffiffi3
ptr
� �
þffiffiffi3
p
8
�0 e_
�1ffiffi3
p ;1;� 1ffiffi3
p�
tr þ 0 e_
�1ffiffi3
p ;�1;� 1ffiffi3
p�
tr
��t þ r �
ffiffiffi3
ptr
� �
þffiffiffi3
p
8
�0 e_
�� 1ffiffi
3p ;1; 1ffiffi
3p�
tr þ 0 e_
�� 1ffiffi
3p ;�1; 1ffiffi
3p�
tr
�t � r �
ffiffiffi3
ptr
� �;
ð18bÞ
in which the superscript ‘stab’ denotes the strain terms obtaineddirectly from the displacements for stabilizing the spurious hour-glass modes. These strain terms are multiplied by the function tanh(t0c) and hence the effect of stabilization is maximum at large com-pressive deformations. In linear analysis there is no accumulated
Fig. 7. Cook’s problem (4 � 4 mesh, E ¼ 1:0, plane stress conditions with m ¼ 1=3;plane strain conditions with m ¼ 0:3, 0.4 or 0.499).
Table 3Predicted y-displacement at point A in Cook’s problem with plane stress conditions.
Elements Mesh
2 � 2 4 � 4 8 � 8 16 � 16 32 � 32
H8 10.9771 17.3332 21.5484 23.2055 23.7209H8I9 20.3889 22.6184 23.4766 23.7740 23.88403D-MITC8 19.0821 22.2906 23.4092 23.7695 23.8893Reference solution 23.9642
Table 2Predicted stress-xx (�102) at the support (point B) in the cantilever problem.
Elements Regular mesh Distorted mesh
H8 0.466667 0.351417H8I9 0.700000 0.6182223D-MITC8 0.700000 0.580764Exact solution 0.700000
Fig. 8. Convergence curves for Cook’s problem with plane strain conditions. The bold line represents the optimal convergence rate.
Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104 91
deformation, t0c in Eq. (14) is zero, and thus the strain terms in Eq.
(18b) are not activated.We use as tying positions the Gauss numerical integration
points 12 a ¼ 1ffiffi
3p
� �for the in-plane bilinear interpolations of the nor-
mal and shear strains in Figs. 2 and 3 and the mid-points of thefaces 1
2 a ¼ 1� �
for the linear interpolations of the shear strains inFig. 2. These positions of tying the normal and shear strainsdecrease the solution errors.
We employ the same assumed strain field for the incremental
nonlinear strain 0g_
ij, and hence the incremental strain componentsin the global Cartesian coordinate system are
0easij ¼ 0~eklðii � 0g_kÞðij � 0g
_lÞ; 0edilii ¼ 0~econkl ðii � 0g_kÞðii � 0g
_lÞ;
d0gasij ¼ d0~gklðii � 0g
_kÞðij � 0g_lÞ; d0gdil
ii ¼ d0~gconkl ðii � 0g
_kÞðii � 0g_lÞ; ð19Þ
in which superscripts ‘as’ and ‘dil’ denote assumed and dilata-tional strain parts. The dilatational strains are constructed onlyfrom the constant part of the assumed strains to alleviate volu-metric locking.
The constant volumetric strain is obtained as [1,21]
0evol ¼ 0edilxx þ 0edilyy þ 0edilzz ¼ Bvolue;
d0gvol ¼ d0gdilxx þ d0gdil
yy þ d0gdilzz ¼ duT
eNvolue; ð20Þ
and the deviatoric strain components are obtained from
0edevij ¼ 0easij � 13 0evoldij ¼ Bdev
ij ue;
Table 4CPU time used for evaluating the element stiffness matrices of Cook’s problem (320 �320 element mesh).
Elements CPU time (s) used Normalized CPU time
H8 16.1 1.0H8I9 24.2 1.503D-MITC8 18.3 1.143D-MITC8/1 19.0 1.17A laptop with dual core Intel 3.50 GHz CPU and 8.0 GB RAM is used
d0gdevij ¼ d0gas
ij � 13d0gvoldij ¼ duT
eNdevij ue; ð21Þ
with the nodal displacement vector ue ¼ uT1 . . . uT
8
� �T . The effi-cient computation of the strain-displacement matrices given inEqs. (20) and (21) is presented in Appendix B.
Fig. 9. Clamped square plate subjected to in-plane moment (plane strain condi-tions, L ¼ 1:0, E ¼ 1:0, M ¼ 1:0, and m ¼ 0:3, 0.4 or 0.499). (a) Problem descriptionwith regular mesh (4 � 4 mesh). (b) Distorted mesh (4 � 4 mesh).
92 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
The stiffness matrix (tK) and internal force vector (t0F) are [1]
tK ¼Z
0Ve
jBvolTBvold0Ve þ
Z0Ve
BdevT
ij Cdevij Bdev
ij d0Ve
þZ
0Ve
13ð0t Sxx þ 0
t Syy þ 0t SzzÞNvold0Ve þ
Z0Ve
0t SijNdev
ij d0Ve;
t0F ¼
Z0Ve
BvolT 13
t0Sxx þ t
0Syy þ t0Szz
� �d0Ve þ
Z0Ve
BdevT
ijt0Sijd
0Ve; ð22Þ
in which t0Sij denotes the second Piola-Kirchhoff stress measured in
the global Cartesian coordinate system, and j ¼ E3ð1�2mÞ is the bulk
modulus with Young’s modulus E and Poisson’s ratio m. Forthe linear isotropic material considered in this study,
Cdevxx ¼ Cdev
yy ¼ Cdevzz ¼ E
1þm and Cdevxy ¼ Cdev
yz ¼ Cdevzx ¼ E
2ð1þmÞ.
The incremental Green-Lagrange strains in Eq. (10) areexpressed using the linear and nonlinear strains in Eqs. (20) and(21), and the global incremental strain values are then obtained as
0evol ¼ Bvolue þ 12uTeN
volue; 0edevij ¼ Bdevij ue þ 1
2uTeN
devij ue: ð23Þ
The second Piola-Kirchhoff stress is updated as
tþDt0Sij ¼ t
0Sij þ 0Sij; 0Sij ¼ j0evoldij þ Cdevij 0edevij ; ð24Þ
and the nodal geometry is updated using Eq. (3).
Fig. 10. Convergence curves for the clamped square plate subjected to in-plane momenoptimal convergence rate.
Alternatively, we can construct an 8-node element using theassumed strain field in Eq. (18) to obtain a mixed displacement-pressure (u/p) formulation for alleviating volumetric locking.
The stiffness matrix (tK) and internal force vector (t0F) of the3D-MITC8/1 element are
tK ¼ tKuu � ð1=KppÞtKuptKT
up;
tKuu ¼Z
0Ve
BdevT
ij Cdevij Bdev
ij d0Ve þZ
0Ve
13
t0Sxx þ t
0Syy þ t0Szz
� �Nvold0Ve
þZ
0Ve
t0SijN
devij d0Ve;
tKup ¼ �Z
0Ve
BvolTd0Ve; Kpp ¼ �
Z0Ve
1jd0Ve ¼ �
0Vj
;
t0F ¼
Z0Ve
BvolT 13
t0Sxx þ t
0Syy þ t0Szz
� �d0Ve þ
Z0Ve
BdevT
ijt0Sijd
0Ve: ð25Þ
The single pressure variable (pe) is
pe ¼ �ð1=KppÞKTupue; ð26Þ
and the second Piola-Kirchhoff stress is updated as
tþDt0Sij ¼ t
0Sij þ 0Sij; 0Sij ¼ �pedij þ Cdevij 0edevij : ð27Þ
t using (a) uniform meshes and (b) distorted meshes. The bold line represents the
Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104 93
When compared with the 3D-MITC8 element, the 3D-MITC8/1element shows the same performance in linear analyses and cangive slightly different solutions in nonlinear analyses. For thenumerical solutions below, the results are the same as for the3D-MITC8 element unless otherwise noted.
In the finite element solutions, we use 2 � 2 � 2 Gauss integra-tion over the element volume considered.
Fig. 11. Pressurized cavity problem (plane strain conditions, L ¼ 1:0, E ¼ 1:0,p ¼ 1:0 for uniform and p0 ¼ 1:0 for smoothly varying loads, and m ¼ 0:3, 0.4 or0.499). (a) Problem description with uniform pressure (4 � 8 mesh). (b) Smoothlyvarying pressure.
4. Illustrative solutions
In this section, we present the performance of the new 3D solidelement through the solutions of various 3D examples. Compar-isons are made with the standard displacement based 8-node ele-ment referred to as ‘‘H8” as well as the standard 8-nodeincompatible modes element referred to as ‘‘H8I9”, see Refs.[1,3–5].
First, the new 3D solid element is assessed considering the basicnumerical tests. Then the performance of the new 3D solid elementis investigated in linear plane stress and plane strain analyses,namely, a cantilever beam problem, Cook’s problem, a clampedsquare plate problem, a pressurized cavity problem, and a clampedplate subjected to a uniform pressure. The geometric nonlinearsolutions are included to test the occurrence of instability: a can-tilever subjected to tip forces, a rubber block problem, and a panelproblem. We use ADINA to obtain the solutions of the geometricnonlinear problems with the incompatible modes element [22].In all analyses we assume that the material response is describedby Young’s modulus E and Poisson’s ratio m, where we recognizethat there is a strain limit that such material can be subjected tobefore a material model instability occurs [1].
Using the derivations in Appendix B, we have found that the cal-culation of the 3D-MITC8 element stiffness matrix is not muchmore expensive than the calculation of the H8 element stiffnessmatrix commonly used (see Section 4.3 for an example).
4.1. Basic numerical tests
We consider the isotropy, zero energy mode and patch tests inlinear analysis.
The spatially isotropic behavior and invariance to the nodalnumbering is important for any element [1,17,19]. The 3D-MITC8element passes the isotropy test.
In the spurious zero energy mode test, the stiffness matrix of asingle unsupported element should show only zero eigenvaluescorresponding to the correct rigid body modes. Six rigid bodymodes corresponding to the 3 translations and 3 rotations shouldbe present. The 3D-MITC8 element passes this test.
In the patch tests [1,15,23], the patch of elements shown inFig. 5 is subjected to the minimum number of constraints to pre-vent rigid body motions, and forces are applied at the boundarycorresponding to the constant stress states. The predicted normaland shearing stresses should be the analytically correct values atany location within the patch of elements. The 3D-MITC8 elementpasses the patch tests.
4.2. Cantilever problem
We solve the cantilever problem shown in Fig. 6. The structureis subjected to a shearing force at its tip. The cantilever is modeledusing regular and distorted meshes with four elements as in Refs.[20,24,25].
Tables 1 and 2 give the tip vertical displacement at point A andthe xx-component of stress at point B with reference to the refer-ence solutions. As when using the H8I9 element, the use of the3D-MITC8 element gives the exact solution with a regular mesh.
For the distorted mesh case, the solution using the 3D-MITC8 ele-ment is much more accurate than when using the H8 element, andnot much less accurate than obtained using the H8I9 element.
4.3. Cook’s cantilever problem
We consider Cook’s problem shown in Fig. 7 [4,20,24,26]. Thecantilever is clamped at one end and is subjected to a distributedshearing force of total magnitude P at its tip. We consider the solu-tions using plane stress conditions with Poisson’s ratio m ¼ 1=3,plane strain conditions with m ¼ 0:3, 0.4 and 0.499, and meshesof N � N elements with N ¼ 2, 4, 8, 16 and 32. The plane strain con-dition is modeled by imposing w ¼ 0 for the entire mesh.
Table 3 gives the vertical displacement at point A for the planestress case. The performance of the 3D-MITC8 element is similar tothat of the H8I9 element. For the plane strain case, we measure thesolution error using the s-norm [27,28] with reference solutionsobtained using a 72 � 72 mesh of the standard 27-nodedisplacement-based element for the cases m ¼ 0:3, 0.4 and the27/4 element for the case m ¼ 0:499 [1].
The s-norm used is given by
kuref �uhk2s ¼ZXref
DeTDsdXref ; with De¼ eref � eh; Ds¼ sref �sh;
ð28Þand then the relative error is
94 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
Eh ¼ kuref � uhk2skuref k2s
; ð29Þ
where uref is the reference solution, uh is the solution of the finiteelement discretization, the subscript h corresponds to the elementsize used, e and s are the strain and stress vectors, and Eh is the rel-ative error.
The behavior of an element is optimal if the convergence for theelement in a sequence of meshes is given by
Eh ffi Ch2; ð30Þ
in which C is a constant independent of the material properties andh is the element size.
Fig. 8 shows the convergence of the relative error as a functionof the element size h ¼ 1=N. For m ¼ 0:499, the convergence isseverely degraded when using the H8 element. The 3D-MITC8and H8I9 elements behave nearly optimally and show comparableperformance.
To indicate the computational effectiveness of the new element,Table 4 gives the CPU time used to calculate the element stiffnessmatrices for a fine mesh with N ¼ 320, that is, 102,400 elements.We employ the standard formulations for the H8 and H8I9 ele-ments, with the internal degrees of freedom of the H8I9 elementcondensed out [1,3–5]. The time used for the 3D-MITC8 elementis larger than for the standard H8 element, but not by much.
Fig. 12. Convergence curves for the pressurized cavity problem using (a) the uniform and
4.4. Clamped square plate subjected to an in-plane moment
We solve the clamped square plate problem shown in Fig. 9 [20]and assume plane strain conditions with Poisson’s ratios m ¼ 0:3,0.4 and 0.499. For the solutions, we use regular and distortedmeshes with N � N elements and N ¼ 2, 4, 8, 16 and 32, seeFig. 9(a) and (b), where the element edges are discretized in theratio L1:L2:L3:. . .:LN = 1:2:3:. . .:N. The following boundaryconditions are imposed: u ¼ v ¼ w ¼ 0 along AD, and w ¼ 0 forthe entire mesh.
The relative error is measured using the s-norm with thereference solutions obtained using a 72 � 72 mesh of the standard27-node displacement-based element for the cases m ¼ 0:3, 0.4 andthe 27/4 element for the case m ¼ 0:499. Fig. 10 shows a compara-ble performance of the 3D-MITC8 and H8I9 elements, where theconvergence is nearly optimal for the uniform and distortedmeshes.
4.5. Pressurized cavity problem
We consider a square structural domain with an internally pres-surized circular cavity as shown in Fig. 11. Plane strain conditionsare assumed with Poisson’s ratios m ¼ 0:3, 0.4 and 0.499. Because ofsymmetry, only one-quarter of the structure is modeled usingN � 2N elements with N ¼ 2, 4, 8, 16 and 32. The following bound-
(b) smoothly varying loads. The bold line represents the optimal convergence rate.
Fig. 13. Clamped square plate with uniform pressure (L ¼ 1:0, E ¼ 1:7472� 107,p ¼ 1:0 and m ¼ 0:3). (a) Problem description with regular mesh (4 � 4 mesh). (b)Distorted mesh (4 � 4 mesh).
Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104 95
ary conditions are imposed: u ¼ 0 along BC, v ¼ 0 along DE, andw ¼ 0 for the entire mesh.
The structure undergoes in-plane compression or bendingaccording to the type of internal pressure p used. We test the casesof uniform internal pressure, see Fig. 11(a), and smoothly varyingpressure, see Fig. 11(b).
Fig. 12 shows the convergence of the solutions measured usingthe s-norm, for which the reference solution is obtained with a 36� 72 mesh of the standard 27-node displacement-based elementfor the cases m ¼ 0:3, 0.4 and the 27/4 element for the casem ¼ 0:499. Unlike the H8 element which severely locks near theincompressible limit, the 3D-MITC8 and H8I9 elements behaveoptimally.
4.6. Clamped square plate subjected to a uniform pressure
We consider a clamped square plate transversely loaded by auniform pressure, see Fig. 13(a) [17,19,20]. Utilizing symmetry,only one quarter of the plate is modeled using N � N elements withN ¼ 4, 8, 16, 32 and 64. We vary the ratio of the shell thickness tothe overall dimension of the structure, t=L ¼ 1=10, 1/100 and1/1000. The element size is h ¼ L=N. The boundary conditions areu ¼ hy ¼ 0 along BC, v ¼ hx ¼ 0 along DC andu ¼ v ¼ w ¼ hx ¼ hy ¼ 0 along AB and AD. We use only one layerof elements in the thickness direction.
We also perform the convergence studies with the distortedmesh pattern shown in Fig. 13(b), where each edge is discretizedby the elements in the following ratio: L1:L2:L3:. . .:LN =1:2:3:. . .:N for the N � N element mesh.
Fig. 14 shows the convergence measured using the s-norm; thereference solution is obtained using a 72 � 72 element mesh of theMITC9 shell element [28,29]. The element behavior is optimal if theconvergence curves have the optimal slope and the curves areindependent of the plate thickness and material properties. Forthe 3D solid elements, we can hardly expect to see as good a per-formance as when using shell or plate elements. However, the 3D-MITC8 element shows the optimal convergence behavior whenusing the regular mesh. The convergence behavior deteriorateswhen a distorted mesh is used, but the performance of the 3D-MITC8 element is better than the performance of the H8I9 element.
4.7. Cantilever in large displacements
We solve the cantilever problem shown in Fig. 15; the structureis subjected to tip forces of magnitude F. To test the mixed effect ofcompression and bending, two tip forces are applied at corners ofthe structure. The increment in load per step is 2.0. The fullyclamped boundary condition of u ¼ v ¼ w ¼ 0 is applied at thebottom.
Figs. 16 and 17 present the load-displacement curves and thedeformed shapes, respectively. The reference solution is obtainedusing the standard 27-node displacement-based element. The3D-MITC8 and incompatible modes elements follow the referencebehavior, however, using the H8I9 element, the cantilever is pre-dicted to already collapse at F ¼ 68:0 due to the propagation ofan element instability near the bottom of the model.
4.8. Rubber blocks in large displacements and large strains
We solve the rubber block problems shown in Fig. 18 [12,13,20].The structures are supported at the bottom, loaded with the pres-sure p in plane strain conditions with m ¼ 0:49. The increment inload per step is 1.2.
We analyze the rectangular and trapezoidal structures usingcoarse meshes and, of course, do not claim to have reached solu-
tion convergence of the continuum problems. The purpose of thesolutions is to investigate whether an instability arises when usingthe 3D-MITC8 element.
Fig. 19 shows the calculated load-displacement curves. SincePoisson’s ratio is not very close to 0.5, we obtain the reference solu-tions using the standard 27-node displacement-based element.The standard H8 element shows a too stiff behavior. Using theH8I9 element, the analysis stops at the spurious collapse loadsp ¼ 80:4 and p ¼ 85:2 for the rectangular and trapezoidal struc-tures, respectively. The solutions using the 3D-MITC8 elementare as stable as when using the 2D-MITC4/1 or the standard 27-node elements. Fig. 20 shows deformed shapes obtained usingthe 3D-MITC8 element; these shapes are acceptable in contrastto those obtained when using the H8I9 element which show spu-rious hour-glass patterns.
4.9. Rubber panel in large displacements and large strains
We consider the panel shown in Fig. 21. The pressure load p isapplied asymmetrically on the top in order to also have significantbending of the structure. The increment in load per step is 1.6.Since the mesh is coarse, we do not claim to have reached conver-gence in the solution of the continuum problem. The purpose ofthis problem is to investigate whether a spurious instability arises.
Fig. 22 shows the load-displacement curves. The reference solu-tion is obtained using the standard 27-node displacement-basedelement. While the analysis using the H8I9 element predicts a spu-rious path of solution, the prediction using the 3D-MITC8 elementclosely follows the reference load-displacement curve. Thedeformed shapes are compared in Fig. 23. Unlike the H8I9 element
Fig. 14. Convergence curves for the clamped square plate problem using (a) regular and (b) distorted meshes. The bold line represents the optimal convergence rate.
96 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
mesh which shows strong hour-glass shapes, the 3D-MITC8 ele-ment mesh deforms smoothly as in the reference solution.
5. Concluding remarks
We proposed a new 3D 8-node solid element based on the MITCapproach for linear and nonlinear analyses. We established theassumed strain field from physical considerations using a trussstructure to idealize the solid element and previously obtainedknowledge regarding MITC element formulations, specifically the2D-MITC4, the 2D-MITC4/1 and the MITC4 shell elements[15,20]. However, unlike in previous MITC element formulations,we needed to include stabilization terms of displacement-basedshear strains to suppress spurious modes in geometrically nonlin-ear solutions. These stabilization strain terms are only, and thenautomatically, activated in geometrically nonlinear analyses whenneeded.
The new 3D-MITC8 passes all basic element tests and we pre-sented various problem solutions to illustrate the predictive capa-bilities of the element. In linear analyses, the element compareswell in the accuracy of solutions reached with the H8I9 elementbut is computationally more efficient because no incompatiblemodes are used. In geometrically nonlinear analyses, the elementdid not show instabilities as observed when using the H8I9 ele-ment. However, it would be valuable to further study the element
and in particular to obtain deeper insight into the element formu-lation through a mathematical analysis.
Appendix A. Representation of assumed strain field usingcharacteristic vectors
Here we decompose the covariant strain components using thecharacteristic vectors and use the decomposition to represent theconstant assumed strain in Eq. (18).
Note that the incremental linear and nonlinear strain parts inEq. (11) can be expressed as
0eijðr; s; tÞ ¼ ðtxIuJ þ tyIv J þ tzIwJÞnijIJðr; s; tÞ;
0gijðr; s; tÞ ¼ ðuIuJ þ v Iv J þwIwJÞnijIJðr; s; tÞ; ðA:1Þin which I; J ¼ 1; . . . ;8 are node numbers. Note that any part ofincremental strains can be expressed using the corresponding nodalconstants (nijIJ with nijIJ ¼ njiIJ ¼ nijJI ¼ njiJI). The geometry isdescribed and corresponding incremental displacements are
tXC ¼ tx½1� tx½2� tx½3� tx½4� tx½5� tx½6� tx½7�� �¼ tg
_
rtg_
stg_
ttxrs
txsttxtr
txrst
h i;
UC ¼ tþDtXC � tXC ¼ u½1� u½2� u½3� u½4� u½5� u½6� u½7�� �;
Fig. 17. Deformed shapes for the cantilever.
Fig. 15. Cantilever subjected to tip forces (2 � 1�30 mesh, E ¼ 1:0� 103 andm ¼ 0:0).
Fig. 16. Load-displacement curves for the cantilever.
Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104 97
u½K� ¼ u½K� v ½K� w½K�� �T with K ¼ 1; . . . ;7: ðA:2ÞUsing Eq. (A.2), we define following representative incremental
strains (e½KjL� and g½KjL�) and the corresponding nodal constants (n½KjL�IJ ),
e½KjL� ¼ 12
tx½K� � u½L� þ u½K� � tx½L�� � ¼ txIuJ þ tyIv J þ tzIwJ� �
n½KjL�IJ ;
g½KjL� ¼ 12
u½K� � u½L� þ u½K� � u½L�� � ¼ ðuIuJ þ v Iv J þwIwJÞn½KjL�IJ ;
n½KjL�IJ ¼ 1
128ð1½K�I 1½L�J þ 1½K�J 1½L�I Þ with I; J ¼ 1; . . . ;8 and K; L¼ 1; . . . ;7;
1½1�I 1½2�I 1½3�I 1½4�I 1½5�I 1½6�I 1½7�I
h i¼ nI gI fI nIgI gIfI fInI nIgIfI½ �:
ðA:3ÞUsing Eq. (A.2) in Eq. (4),
tgr ¼@ tx@r
¼ tx½1� þ stx½4� þ ttx½6� þ sttx½7�;
u;r ¼ @u@r
¼ u½1� þ su½4� þ tu½6� þ stu½7�;
tgs ¼@ tx@s
¼ tx½2� þ ttx½5� þ rtx½4� þ trtx½7�;
u;s ¼ @u@s
¼ u½2� þ tu½5� þ ru½4� þ tru½7�;
tgt ¼@ tx@t
¼ tx½3� þ rtx½6� þ stx½5� þ rstx½7�;
u;t ¼ @u@t
¼ u½3� þ ru½6� þ su½5� þ rsu½7�: ðA:4Þ
Using Eqs. (A.1), (A.3) and (A.4) with Eq. (11) we obtain the nodalconstants for the covariant strain
nrrIJðr; s; tÞ ¼ n½1j1�IJ þ 2sn½1j4�
IJ þ 2tn½1j6�IJ þ 2stðn½1j7�
IJ þ n½4j6�IJ Þ
þ s2n½4j4�IJ þ 2s2tn½4j7�
IJ þ t2n½6j6�IJ þ 2st2n½6j7�
IJ þ s2t2n½7j7�IJ ;
Fig. 18. Rubber blocks subjected to pressure (6 � 1 � 15 mesh, E ¼ 1:0� 103 andm ¼ 0:49). (a) Rectangular block. (b) Trapezoidal block.
Fig. 19. Load-displacement curves for the rubber blocks. (a) Rectangular block. (b)Trapezoidal block.
98 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
nssIJðr; s; tÞ ¼ n½2j2�IJ þ 2tn½2j5�
IJ þ 2rn½2j4�IJ þ 2trðn½2j7�
IJ þ n½4j5�IJ Þ
þ t2n½5j5�IJ þ 2t2rn½5j7�
IJ þ r2n½4j4�IJ þ 2tr2n½4j7�
IJ þ t2r2n½7j7�IJ ;
nttIJðr; s; tÞ ¼ n½3j3�IJ þ 2rn½3j6�
IJ þ 2sn½3j5�IJ þ 2rsðn½3j7�
IJ þ n½5j6�IJ Þ
þ r2n½6j6�IJ þ 2r2sn½6j7�
IJ þ s2n½5j5�IJ þ 2rs2n½5j7�
IJ þ r2s2n½7j7�IJ ;
nrsIJðr; s; tÞ ¼ n½1j2�IJ þ tðn½1j5�
IJ þ n½2j6�IJ Þ þ rn½1j4�
IJ þ trðn½4j6�IJ þ n½1j7�
IJ Þþ sn½2j4�
IJ þ stðn½4j5�IJ þ n½2j7�
IJ Þ þ rsn½4j4�IJ þ 2rstn½4j7�
IJ
þ t2n½5j6�IJ þ t2rn½6j7�
IJ þ st2n½5j7�IJ þ rst2n½7j7�
IJ ;
nstIJðr; s; tÞ ¼ n½2j3�IJ þ rðn½2j6�
IJ þ n½3j4�IJ Þ þ sn½2j5�
IJ þ rsðn½4j5�IJ þ n½2j7�
IJ Þþ tn½3j5�
IJ þ trðn½5j6�IJ þ n½3j7�
IJ Þ þ stn½5j5�IJ þ 2rstn½5j7�
IJ
þ r2n½4j6�IJ þ r2sn½4j7�
IJ þ tr2n½6j7�IJ þ r2stn½7j7�
IJ ;
ntrIJðr; s; tÞ ¼ n½1j3�IJ þ sðn½3j4�
IJ þ n½1j5�IJ Þ þ tn½3j6�
IJ þ stðn½5j6�IJ þ n½3j7�
IJ Þþ rn½1j6�
IJ þ rsðn½4j6�IJ þ n½1j7�
IJ Þ þ trn½6j6�IJ þ 2rstn½6j7�
IJ
þ s2n½4j5�IJ þ s2tn½5j7�
IJ þ rs2n½4j7�IJ þ rs2tn½7j7�
IJ ; ðA:5Þwith nrrIJ ¼ n11IJ , nssIJ ¼ n22IJ , nttIJ ¼ n33IJ , nrsIJ ¼ n12IJ , nstIJ ¼ n23IJ andntrIJ ¼ n31IJ . Using Eq. (17) with Eq. (A.3),
~n½Kj7�IJ ¼ tdrn
½Kj1�IJ þ tdsn
½Kj2�IJ þ tdtn
½Kj3�IJ with K ¼ 1; . . . ;6;
~n½7j7�IJ ¼ td2
r n½1j1�IJ þ td2
s n½2j2�IJ þ td2
t n½3j3�IJ þ 2tdr
tdsn½1j2�IJ
þ 2tdstdtn
½2j3�IJ þ 2tdt
tdrn½1j3�IJ : ðA:6Þ
Substituting ~n½Kj7�IJ for n½Kj7�
IJ (K = 1, . . . , 7) in Eq. (A.5), we obtain thenodal constants for the assumed covariant strain~nIJðr; s; tÞ ¼ ~n11IJ ~n22IJ ~n33IJ ~n12IJ ~n23IJ ~n31IJ
� �.
Appendix B. Calculation of strain-displacement matrices
We aim to obtain good computational efficiency in the evalua-tion of the element matrices.
Rearranging the strain transformation in Eq. (12),
n_
IJðr; s; tÞT ¼ G_
nIJðr; s; tÞT ;
n_
IJðr; s; tÞ ¼ n_
11IJ n_
22IJ n_
33IJ n_
12IJ n_
23IJ n_
31IJ
h i;
nIJðr; s; tÞ ¼ n11IJ n22IJ n33IJ n12IJ n23IJ n31IJ½ �;
Fig. 20. Deformed shapes for the rubber blocks. (a) Rectangular block. (b)Trapezoidal block.
Fig. 22. Load-displacement curves for the rubber panel.
Fig. 21. A rubber panel subjected to surface loading (3 � 3 � 9 mesh, E ¼ 1:0� 103
and m ¼ 0:49).
Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104 99
G_
¼
G_
1
G_
2
G_
3
G_
4
G_
5
G_
6
0BBBBBBBBBBBBBBB@
1CCCCCCCCCCCCCCCA
¼
g11g
11 g2
1g21 g3
1g31 2g1
1g21 2g2
1g31 2g3
1g11
g12g
12 g2
2g22 g3
2g32 2g1
2g22 2g2
2g32 2g3
2g12
g13g
13 g2
3g23 g3
3g33 2g1
3g23 2g2
3g33 2g3
3g13
g11g
12 g2
1g22 g3
1g32 g1
1g22 þ g2
1g12 g2
1g32 þ g3
1g22 g3
1g12 þ g1
1g32
g12g
13 g2
2g23 g3
2g33 g1
2g23 þ g2
2g13 g2
2g33 þ g3
2g23 g3
2g13 þ g1
2g33
g13g
11 g2
3g21 g3
3g31 g1
3g21 þ g2
3g11 g2
3g31 þ g3
3g21 g3
3g11 þ g1
3g31
0BBBBBBBBBBBB@
1CCCCCCCCCCCCA;
Fig. 23. Deformed shapes for the rubber panel.
0jG_ ¼ C
_
1 C_
2 C_
3 C_
4 C_
5 C_
6
� �: ðB:1Þ
We define the following representative constant vectors evaluatedat the tying positions of the assumed strain components in Eq. (18)
100 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
C10
C20
C30
C40
C50
C60
C70
C80
C90
0BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB@
1CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCA
¼ 10V
X8L¼1
1 gL fL gLfL½ �TC_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
1
1 fL nL fLnL½ �TC_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
2
1 nL gL nLgL½ �TC_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
3
1 nL gL nLgL½ �TC_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
4
1 gL fL gLfL½ �TC_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
5
1 fL nL fLnL½ �TC_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
6
1 nL gL nLgL½ �TfLC_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
4
1 gL fL gLfL½ �TnLC_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
5
1 fL nL fLnL½ �TgLC_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
6
0BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB@
1CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCA
;
0V ¼X8L¼1
0j�nL 1ffiffi
3p ;gL 1ffiffi
3p ;fL
1ffiffi3
p�;
G11 G1
2 G13
G21 G2
2 G23
G31 G3
2 G33
G41 G4
2 G43
0BBBB@
1CCCCA ¼
X4L¼1
1nLgL
nLgL
0BBB@
1CCCA G
_�0;nL 1ffiffi
3p ;gL 1ffiffi
3p�
1 G_�gL 1ffiffi
3p ;0;nL 1ffiffi
3p�
2 G_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;0�
3
� �;
n10
��TIJ
n20
��TIJ
n30
��TIJ
0BBBB@
1CCCCA ¼ 1
8
X8L¼1
1 nL gL nLgL½ �~n�0;nL 1ffiffi
3p ;gL 1ffiffi
3p�
rrIJ
1 nL gL nLgL½ �~n�gL 1ffiffi
3p ;0;nL 1ffiffi
3p�
ssIJ
1 nL gL nLgL½ �~n�nL
1ffiffi3
p ;gL 1ffiffi3
p ;0�
ttIJ
0BBBBBBB@
1CCCCCCCA
¼
n½1j1�IJ þ 1
3n½4j4�IJ þ 1
3n½6j6�IJ þ 1
9~n½7j7�IJ
2ffiffi3
p n½1j4�IJ þ 1
3~n½6j7�IJ
� �2ffi3
p
n½2j2�IJ þ 1
3n½5j5�IJ þ 1
3n½4j4�IJ þ 1
9~n½7j7�IJ
2ffiffi3
p n½2j5�IJ þ 1
3~n½4j7�IJ
� �2ffi3
p
n½3j3�IJ þ 1
3n½6j6�IJ þ 1
3n½5j5�IJ þ 1
9~n½7j7�IJ
2ffiffi3
p n½3j6�IJ þ 1
3~n½5j7�IJ
� �2ffi3
p
0BBBBB@
n40
��TIJ
n50
��TIJ
n60
��TIJ
0BBBB@
1CCCCA ¼ 1
8
X8L¼1
1 nL gL nLgL½ �~n�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
rsIJ
1 gL fL gLfL½ �~n�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
stIJ
1 fL nL fLnL½ �~n�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
trIJ
0BBBBBBB@
1CCCCCCCA
¼
n½1j2�IJ þ 1
3n½5j6�IJ
1ffiffi3
p n½1j4�IJ þ 1
3~n½6j7�IJ
� �1ffiffi3
p n½2j4�IJ þ 1
3~n½5j7�IJ
� �n½2j3�IJ þ 1
3n½4j6�IJ
1ffiffi3
p n½2j5�IJ þ 1
3~n½4j7�IJ
� �1ffiffi3
p n½3j5�IJ þ 1
3~n½6j7�IJ
� �n½1j3�IJ þ 1
3n½4j5�IJ
1ffiffi3
p n½3j6�IJ þ 1
3~n½5j7�IJ
� �1ffiffi3
p n½1j6�IJ þ 1
3~n½4j7�IJ
� �
0BBBBB@
G14 G1
5 G16
G24 G2
5 G26
!¼X2L¼1
1fL
G_ð0;0;fLÞ
4 G_ðfL ;0;0Þ
5 G_ð0;fL ;0Þ
6
� �;
H14 H1
5 H16
H24 H2
5 H26
H34 H3
5 H36
H44 H4
5 H46
H54 H5
5 H56
H64 H6
5 H66
H74 H7
5 H76
H84 H8
5 H86
0BBBBBBBBBBBBBBB@
1CCCCCCCCCCCCCCCA
¼X4L¼1
X2M¼1
1nLgL
nLgL
fMnLfMgLfM
nLgLfM
0BBBBBBBBBBBBBB@
1CCCCCCCCCCCCCCA
G_�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fM
�4 G
_�fM ;nL
1ffiffi3
p ;gL 1ffiffi3
p�
5 G_�gL 1ffiffi
3p ;fM ;nL
1ffiffi3
p�
6
� �;
ðB:2Þin which the signs are f1 f2
� � ¼ 1 �1½ �, n1 n2 n3 n4� � ¼
�1 1 1 �1½ � and g1 g2 g3 g4½ � ¼ �1 �1 1 1½ �. Wenote that the constants in Eq. (B.2) are calculated only once for eachelement.
We define the following representative nodal constants
n00
��IJ ¼ n½1j1�
IJ n½2j2�IJ n½3j3�
IJ n½1j2�IJ n½2j3�
IJ n½1j3�IJ
h i;
ffi n½1j6�IJ þ 1
3~n½4j7�IJ
� �23
~n½1j7�IJ þ n½4j6�
IJ
� �ffi n½2j4�
IJ þ 13~n½5j7�IJ
� �23
~n½2j7�IJ þ n½4j5�
IJ
� �ffi n½3j5�
IJ þ 13~n½6j7�IJ
� �23
~n½3j7�IJ þ n½5j6�
IJ
� �
1CCCCCA;
13 n½4j4�
IJ þ 13~n½7j7�IJ
� �13 n½5j5�
IJ þ 13~n½7j7�IJ
� �13 n½6j6�
IJ þ 13~n½7j7�IJ
� �
1CCCCCA;
n70
��TIJ
n80
��TIJ
n90
��TIJ
0BBB@
1CCCA ¼ 1
8
X8L¼1
1 nL gL nLgL½ �fL~n�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
rsIJ
1 gL fL gLfL½ �nL~n�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
stIJ
1 fL nL fLnL½ �gL~n
�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fL1ffiffi3
p�
trIJ
0BBBBB@
1CCCCCA
¼
1ffiffi3
p n½1j5�IJ þ n½2j6�
IJ
� �13 n½4j6�
IJ þ ~n½1j7�IJ
� �13 n½4j5�
IJ þ ~n½2j7�IJ
� �2
3ffiffi3
p ~n½4j7�IJ
1ffiffi3
p n½2j6�IJ þ n½3j4�
IJ
� �13 n½4j5�
IJ þ ~n½2j7�IJ
� �13 n½5j6�
IJ þ ~n½3j7�IJ
� �2
3ffiffi3
p ~n½5j7�IJ
1ffiffi3
p n½3j4�IJ þ n½1j5�
IJ
� �13 n½5j6�
IJ þ ~n½3j7�IJ
� �13 n½4j6�
IJ þ ~n½1j7�IJ
� �2
3ffiffi3
p ~n½6j7�IJ
0BBBB@
1CCCCA;
Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104 101
E1jIJ E2jIJ E3jIJ� � ¼ 1
4
X4L¼1
1nLgL
nLgL
0BBB@
1CCCA n
�0;nL 1ffiffi
3p ;gL 1ffiffi
3p�T
IJ n
�gL 1ffiffi
3p ;0;nL 1ffiffi
3p�T
IJ n
�nL
1ffiffi3
p ;gL 1ffiffi3
p ;0�T
IJ
;
E1jIJ ¼
n00
��TIJ þ 1
3 n½4j4�IJ þ n½6j6�
IJ þ 13n
½7j7�IJ n½5j5�
IJ n½5j5�IJ n½5j6�
IJ 0 n½4j5�IJ
h iT1ffiffi3
p 2n½1j4�IJ þ 2
3n½6j7�IJ 0 2n½3j5�
IJ n½2j4�IJ þ 1
3n½5j7�IJ n½2j5�
IJ n½3j4�IJ þ n½1j5�
IJ
h iT1ffiffi3
p 2n½1j6�IJ þ 2
3n½4j7�IJ 2n½2j5�
IJ 0 n½1j5�IJ þ n½2j6�
IJ n½3j5�IJ n½3j6�
IJ þ 13n
½5j7�IJ
h iT13 2n½1j7�
IJ þ 2n½4j6�IJ 0 0 n½4j5�
IJ þ n½2j7�IJ n½5j5�
IJ n½5j6�IJ þ n½3j7�
IJ
h iT
0BBBBBBBBB@
1CCCCCCCCCA;
E2jIJ ¼
n00
��TIJ þ 1
3 n½6j6�IJ n½5j5�
IJ þ n½4j4�IJ þ 1
3n½7j7�IJ n½6j6�
IJ n½5j6�IJ n½4j6�
IJ 0h iT
1ffiffi3
p 2n½1j6�IJ 2n½2j5�
IJ þ 23n
½4j7�IJ 0 n½1j5�
IJ þ n½2j6�IJ n½3j5�
IJ þ 13n
½6j7�IJ n½3j6�
IJ
h iT1ffiffi3
p 0 2n½2j4�IJ þ 2
3n½5j7�IJ 2n½3j6�
IJ n½1j4�IJ þ 1
3n½6j7�IJ n½2j6�
IJ þ n½3j4�IJ n½1j6�
IJ
h iT13 0 2n½2j7�
IJ þ 2n½4j5�IJ 0 n½4j6�
IJ þ n½1j7�IJ n½5j6�
IJ þ n½3j7�IJ n½6j6�
IJ
h iT
0BBBBBBBBB@
1CCCCCCCCCA;
E3jIJ ¼
n00
��TIJ þ 1
3 n½4j4�IJ n½4j4�
IJ n½6j6�IJ þ n½5j5�
IJ þ 13n
½7j7�IJ 0 n½4j6�
IJ n½4j5�IJ
h iT1ffiffi3
p 0 2n½2j4�IJ 2n½3j6�
IJ þ 23n
½5j7�IJ n½1j4�
IJ n½2j6�IJ þ n½3j4�
IJ n½1j6�IJ þ 1
3n½4j7�IJ
h iT1ffiffi3
p 2n½1j4�IJ 0 2n½3j5�
IJ þ 23n
½6j7�IJ n½2j4�
IJ n½2j5�IJ þ 1
3n½4j7�IJ n½3j4�
IJ þ n½1j5�IJ
h iT13 0 0 2n½3j7�
IJ þ 2n½5j6�IJ n½4j4�
IJ n½4j5�IJ þ n½2j7�
IJ n½4j6�IJ þ n½1j7�
IJ
h iT
0BBBBBBBBB@
1CCCCCCCCCA;
n14
��IJ
n24
��IJ
!¼ 1
2
X2L¼1
1fL
nð0;0;fLÞ
T
IJ ¼n0
0
��TIJ þ n½6j6�
IJ n½5j5�IJ 0 n½5j6�
IJ 0 0h iT
2n½1j6�IJ 2n½2j5�
IJ 0 n½1j5�IJ þ n½2j6�
IJ n½3j5�IJ n½3j6�
IJ
h iT0B@
1CA;
n15
��IJ
n25
��IJ
!¼ 1
2
X2L¼1
1fL
nðfL ;0;0Þ
T
IJ ¼n0
0
��TIJ þ 0 n½4j4�
IJ n½6j6�IJ 0 n½4j6�
IJ 0h iT
0 2n½2j4�IJ 2n½3j6�
IJ n½1j4�IJ n½2j6�
IJ þ n½3j4�IJ n½1j6�
IJ
h iT0B@
1CA;
n16
��IJ
n26
��IJ
!¼ 1
2
X2L¼1
1fL
nð0;fL ;0Þ
T
IJ ¼n0
0
��TIJ þ n½4j4�
IJ 0 n½5j5�IJ 0 0 n½4j5�
IJ
h iT2n½1j4�
IJ 0 2n½3j5�IJ n½2j4�
IJ n½2j5�IJ n½3j4�
IJ þ n½1j5�IJ
h iT0B@
1CA;
s00��IJ ¼ n0
0
��IJ þ
13
n½4j4�IJ þ n½6j6�
IJ n½5j5�IJ þ n½4j4�
IJ n½6j6�IJ þ n½5j5�
IJ n½5j6�IJ n½4j6�
IJ n½4j5�IJ
h i;
102 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
S4jIJ S5jIJ S6jIJ� � ¼ 1
8
X4L¼1
X2M¼1
1nLgL
nLgL
fMnLfMgLfMnLgLfM
0BBBBBBBBBBBBB@
1CCCCCCCCCCCCCA
n
�nL
1ffiffi3
p ;gL 1ffiffi3
p ;fM
�TIJ n
�fM ;nL
1ffiffi3
p ;gL 1ffiffi3
p�T
IJ n
�gL 1ffiffi
3p ;fM ;nL
1ffiffi3
p�T
IJ
;
S4jIJ ¼
s00��TIJþ 1
3 2n½6j6�IJ þ n½7j7�
IJ 2n½5j5�IJ þ n½7j7�
IJ13n
½7j7�IJ 2n½5j6�
IJ 0 0h iT
1ffiffi3
p 0 2 n½2j4�IJ þ n½5j7�
IJ
� �2 n½3j6�
IJ þ 13n
½5j7�IJ
� �n½1j4�IJ þ n½6j7�
IJ n½2j6�IJ þ n½3j4�
IJ n½1j6�IJ þ 1
3n½4j7�IJ
h iT1ffiffi3
p 2 n½1j4�IJ þ n½6j7�
IJ
� �0 2 n½3j5�
IJ þ 13n
½6j7�IJ
� �n½2j4�IJ þ n½5j7�
IJ n½2j5�IJ þ 1
3n½4j7�IJ n½3j4�
IJ þ n½1j5�IJ
h iT13 0 0 2 n½3j7�
IJ þ 13n
½5j6�IJ
� �n½4j4�IJ þ n½7j7�
IJ n½4j5�IJ þ n½2j7�
IJ n½4j6�IJ þ n½1j7�
IJ
h iT2 n½1j6�
IJ þ 13n
½4j7�IJ
� �2 n½2j5�
IJ þ 13n
½4j7�IJ
� �0 n½1j5�
IJ þ n½2j6�IJ n½3j5�
IJ þ 13n
½6j7�IJ n½3j6�
IJ þ 13n
½5j7�IJ
h iT1ffiffi3
p 0 2 n½2j7�IJ þ n½4j5�
IJ
� �0 n½4j6�
IJ þ n½1j7�IJ n½5j6�
IJ þ n½3j7�IJ n½6j6�
IJ þ 13n
½7j7�IJ
h iT1ffiffi3
p 2 n½1j7�IJ þ n½4j6�
IJ
� �0 0 n½4j5�
IJ þ n½2j7�IJ n½5j5�
IJ þ 13n
½7j7�IJ n½5j6�
IJ þ n½3j7�IJ
h iT23 0 0 0 n½4j7�
IJ n½5j7�IJ n½6j7�
IJ
h iT
0BBBBBBBBBBBBBBBBBBBBBBBBBB@
1CCCCCCCCCCCCCCCCCCCCCCCCCCA
;
S5jIJ ¼
s00��TIJþ 1
313n
½7j7�IJ 2n½4j4�
IJ þ n½7j7�IJ 2n½6j6�
IJ þ n½7j7�IJ 0 2n½4j6�
IJ 0h iT
1ffiffi3
p 2 n½1j4�IJ þ 1
3n½6j7�IJ
� �0 2 n½3j5�
IJ þ n½6j7�IJ
� �n½2j4�IJ þ 1
3n½5j7�IJ n½2j5�
IJ þ n½4j7�IJ n½3j4�
IJ þ n½1j5�IJ
h iT1ffiffi3
p 2 n½1j6�IJ þ 1
3n½4j7�IJ
� �2 n½2j5�
IJ þ n½4j7�IJ
� �0 n½1j5�
IJ þ n½2j6�IJ n½3j5�
IJ þ n½6j7�IJ n½3j6�
IJ þ 13n
½5j7�IJ
h iT13 2
�n½1j7�IJ þ n½4j6�
IJ
�0 0 n½4j5�
IJ þ n½2j7�IJ n½5j5�
IJ þ n½7j7�IJ n½5j6�
IJ þ n½3j7�IJ
h iT0 2 n½2j4�
IJ þ 13n
½5j7�IJ
� �2 n½3j6�
IJ þ 13n
½5j7�IJ
� �n½1j4�IJ þ 1
3n½6j7�IJ n½2j6�
IJ þ n½3j4�IJ n½1j6�
IJ þ 13n
½4j7�IJ
h iT1ffiffi3
p 0 0 2�n½3j7�IJ þ n½5j6�
IJ
�n½4j4�IJ þ 1
3n½7j7�IJ n½4j5�
IJ þ n½2j7�IJ n½4j6�
IJ þ n½1j7�IJ
h iT1ffiffi3
p 0 2�n½2j7�IJ þ n½4j5�
IJ
�0 n½4j6�
IJ þ n½1j7�IJ n½5j6�
IJ þ n½3j7�IJ n½6j6�
IJ þ 13n
½7j7�IJ
h iT23 0 0 0 n½4j7�
IJ n½5j7�IJ n½6j7�
IJ
h iT
0BBBBBBBBBBBBBBBBBBBBBBBBBB@
1CCCCCCCCCCCCCCCCCCCCCCCCCCA
;
S6jIJ ¼
s00��TIJ þ 1
3 2n½4j4�IJ þ n½7j7�
IJ13n
½7j7�IJ 2n½5j5�
IJ þ n½7j7�IJ 0 0 2n½4j5�
IJ
h iT1ffiffi3
p 2 n½1j6�IJ þ n½4j7�
IJ
� �2 n½2j5�
IJ þ 13n
½4j7�IJ
� �0 n½1j5�
IJ þ n½2j6�IJ n½3j5�
IJ þ 13n
½6j7�IJ n½3j6�
IJ þ n½5j7�IJ
h iT1ffiffi3
p 0 2 n½2j4�IJ þ 1
3n½5j7�IJ
� �2 n½3j6�
IJ þ n½5j7�IJ
� �n½1j4�IJ þ 1
3n½6j7�IJ n½2j6�
IJ þ n½3j4�IJ n½1j6�
IJ þ n½4j7�IJ
h iT13 0 2 n½2j7�
IJ þ n½4j5�IJ
� �0 n½4j6�
IJ þ n½1j7�IJ n½5j6�
IJ þ n½3j7�IJ n½6j6�
IJ þ n½7j7�IJ
h iT2 n½1j4�
IJ þ 13n
½6j7�IJ
� �0 2 n½3j5�
IJ þ 13n
½6j7�IJ
� �n½2j4�IJ þ 1
3n½5j7�IJ n½2j5�
IJ þ 13n
½4j7�IJ n½3j4�
IJ þ n½1j5�IJ
h iT1ffiffi3
p 2 n½1j7�IJ þ n½4j6�
IJ
� �0 0 n½4j5�
IJ þ n½2j7�IJ n½5j5�
IJ þ 13n
½7j7�IJ n½5j6�
IJ þ n½3j7�IJ
h iT1ffiffi3
p 0 0 2 n½3j7�IJ þ n½5j6�
IJ
� �n½4j4�IJ þ 1
3n½7j7�IJ n½4j5�
IJ þ n½2j7�IJ n½4j6�
IJ þ n½1j7�IJ
h iT23 0 0 0 n½4j7�
IJ n½5j7�IJ n½6j7�
IJ
h iT
0BBBBBBBBBBBBBBBBBBBBBBBBBB@
1CCCCCCCCCCCCCCCCCCCCCCCCCCA
: ðB:3Þ
Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104 103
Here, the part of nM0
��IJ related to ~n½Kj7�
IJ is updated using Eq. (A.6) once
for each element per each increment of nonlinear solution. The
remaining parts of nM0
��IJ , and the constants Ej
��IJ (j ¼ 1,2,3), nk
j
���IJ
and Sj
��IJ(j ¼ 4, 5, 6) are independent of solution increments, loca-
tions within elements and are the same for all elements and henceare only calculated once.
It is possible to represent the assumed strain field in Eq. (18) interms of nodal constants as
~nasrrIJ ¼ ~ncon
rrIJ þffiffiffi3
p
40kðr; s; tÞsn_s
rrIJ þffiffiffi3
p
40kðr; s; tÞtn_t
rrIJ
þ 34
0kðr; s; tÞstn_strrIJ;
~nasssIJ ¼ ~ncon
ssIJ þffiffiffi3
p
40kðr; s; tÞtn_t
ssIJ þffiffiffi3
p
40kðr; s; tÞrn_r
ssIJ
þ 34
0kðr; s; tÞtrn_trssIJ;
~nasttIJ ¼ ~ncon
ttIJ þffiffiffi3
p
40kðr; s; tÞrn_r
ttIJ þffiffiffi3
p
40kðr; s; tÞsn_s
ttIJ
þ 34
0kðr; s; tÞrsn_rsttIJ;
~nasrsIJ ¼ ~ncon
rsIJ þ12
0kðr; s; tÞtn_trsIJ þ
ffiffiffi3
p
8tanhðt0cÞðn
_rrsIJr þ n
_srsIJs
þffiffiffi3
pn_rs
rsIJrsÞ0kðr; s; tÞ;
~nasstIJ ¼ ~ncon
stIJ þ12
0kðr; s; tÞrn_rstIJ þ
ffiffiffi3
p
8tanhðt0cÞðn
_sstIJsþ n
_tstIJt
þffiffiffi3
pn_st
stIJstÞ0kðr; s; tÞ;
~nastrIJ ¼ ~ncon
trIJ þ12
0kðr; s; tÞsn_strIJ þ
ffiffiffi3
p
8tanhðt0cÞðn
_ttrIJ t þ n
_rtrIJr
þffiffiffi3
pn_tr
trIJtrÞ0kðr; s; tÞ; ðB:4Þ
~nconrrIJ
~nconssIJ
~nconttIJ
~nconrsIJ
~nconstIJ
~ncontrIJ
� �T ¼X9M¼1
CMT
0 nM0
��IJ;
n_srrIJ
n_trrIJ
n_strrIJ
26664
37775¼
G21 G1
1 G41 G3
1
G31 G4
1 G11 G2
1
G41 G3
1 G21 G1
1
2664
3775E1jIJ ;
n_t
ssIJ
n_r
ssIJ
n_tr
ssIJ
26664
37775¼
G22 G1
2 G42 G3
2
G32 G4
2 G12 G2
2
G42 G3
2 G22 G1
2
2664
3775E2jIJ ;
n_rttIJ
n_sttIJ
n_rsttIJ
26664
37775 ¼
G23 G1
3 G43 G3
3
G33 G4
3 G13 G2
3
G43 G3
3 G23 G1
3
2664
3775E3jIJ; n
_trsIJ ¼ G2
4n14
��IJþ G1
4n24
��IJ;
n_r
rsIJ
n_s
rsIJ
n_rs
rsIJ
26664
37775¼
H24 H1
4 H44 H3
4 H64 H5
4 H84 H7
4
H34 H4
4 H14 H2
4 H74 H8
4 H54 H6
4
H44 H3
4 H24 H1
4 H84 H7
4 H64 H5
4
264
375S4jIJ ; n
_rstIJ ¼ G2
5n15
��IJ þG1
5n25
��IJ ;
n_s
stIJ
n_t
stIJ
n_st
stIJ
26664
37775¼
H25 H1
5 H45 H3
5 H65 H5
5 H85 H7
5
H35 H4
5 H15 H2
5 H75 H8
5 H55 H6
5
H45 H3
5 H25 H1
5 H85 H7
5 H65 H5
5
264
375S5jIJ ; n_s
trIJ ¼ G26n
16
��IJ þG1
6n26
��IJ ;
n_t
trIJ
n_r
trIJ
n_tr
trIJ
26664
37775 ¼
H26 H1
6 H46 H3
6 H66 H5
6 H86 H7
6
H35 H4
6 H16 H2
6 H76 H8
6 H56 H6
6
H45 H3
6 H26 H1
6 H86 H7
6 H66 H5
6
264
375S6jIJ: ðB:5Þ
In Eq. (B.4), we carry out the computations that depend on the nat-ural coordinates (r, s, t) (for each Gauss point), once for each differ-ent nodal coupling (nodes I and J). Using Eqs. (B.2) and (B.3) to Eq.
(B.5), we calculate (once per each element) the constants n_i
kkIJ , n_ jkkIJ ,
n_ijkkIJ , n
_kijIJ , n
_iijIJ , n
_ jijIJ and n
_ijijIJ , and update ~ncon
ijIJ in each increment.The nodal constants for the global volumetric and deviatoric
strains (nvolIJ and ndevijIJ ) are obtained from
nvolIJ ¼ ndilxxIJ þ ndil
yyIJ þ ndilzzIJ ;n
devijIJ ¼ nas
ijIJ �13nvolIJ dij;
nasijIJ ¼ ~nas
klIJðii � 0g_kÞðij � 0g
_lÞ;ndiliiIJ ¼ ~ncon
klIJ ðii � 0g_kÞðij � 0g
_lÞ: ðB:6Þ
The deviatoric strain-displacement matrices (Bdevij and Ndev
ij ) areobtained by
Bdevij ¼ Bdev
ij1 � � � BdevijI � � � Bdev
ij8
� �;Bdev
ijI ¼X8J¼1
ndevijIJ
txJ 0 0
0 ndevijIJ
tyJ 0
0 0 ndevijIJ
tzJ
0BB@
1CCA;
Ndevij ¼
Ndevij11 � � � Ndev
ij18
..
.Ndev
ijIJ...
Ndevij81 � � � Ndev
ij88
0BBB@
1CCCA;Ndev
ijIJ ¼ ndevijIJ
1 0 00 1 00 0 1
0B@
1CA; ðB:7Þ
where similar relation also holds for volumetric strain-displacement matrices, Bvol and Nvol with nvolIJ .
References
[1] Bathe KJ. Finite element procedures. Prentice Hall; 1996, 2nd ed. K.J. Bathe,Watertown, MA; 2014 and Higher Education Press, China; 2016.
[2] Bathe KJ. The inf–sup condition and its evaluation for mixed finite elementmethods. Comput Struct 2001;79(2):243–52.
[3] Wilson EL, Ibrahimbegovic A. Use of incompatible displacement modes for thecalculation of element stiffnesses or stresses. Finit Elem Anal Des 1990;7(3):229–41.
[4] Simo JC, Rifai MS. A class of mixed assumed strain methods and the method ofincompatible modes. Int J Numer Meth Eng 1990;29(8):1595–638.
[5] Kasper EP, Taylor RL. A mixed-enhanced strain method: part I: geometricallylinear problems. Comput Struct 2000;75(3):237–50.
[6] Alves de Sousa RJ, Natal Jorge RM, Fontes Valente RA, César de Sá JMA. A newvolumetric and shear locking-free 3D enhanced strain element. Eng Comput2003;20(7):896–925.
[7] Wriggers P, Reese S. A note on enhanced strain methods for largedeformations. Comp Meth Appl Mech Eng 1996;135(3–4):201–9.
[8] Sussman T, Bathe KJ. Spurious modes in geometrically nonlinear smalldisplacement finite elements with incompatible modes. Comput Struct2014;140:14–22.
[9] Korelc J, Wriggers P. Consistent gradient formulation for a stable enhancedstrain method for large deformations. Eng Comput 1996;13(1):103–23.
[10] Pantuso D, Bathe KJ. On the stability of mixed finite elements in large strainanalysis of incompressible solids. Finit Elem Anal Des 1997;28(2):83–104.
[11] Reese S, Küssner M, Reddy BD. A new stabilization technique for finiteelements in non-linear elasticity. Int J Numer Meth Eng 1999;44(11):1617–52.
[12] Wall WA, Bischoff M, Ramm E. A deformation dependent stabilizationtechnique, exemplified by EAS elements at large strains. Comp Meth ApplMech Eng 2000;188(4):859–71.
[13] Bischoff M, Bletzinger KU. Interaction of locking and element stability at largestrains. In: VIII international conference on computational plasticity.Barcelona; 2005.
[14] Korelc J, Šolinc U, Wriggers P. An improved EAS brick element for finitedeformation. Comp Mech 2010;46(4):641–59.
[15] Dvorkin EN, Bathe KJ. A continuum mechanics based four-node shell elementfor general non-linear analysis. Eng Comput 1984;1(1):77–88.
[16] Radovitzky RA, Dvorkin EN. A 3D element for non-linear analysis of solids. Int JNumer Meth Bio Eng 1994;10(3):183–94.
104 Y. Ko, K.J. Bathe / Computers and Structures 202 (2018) 85–104
[17] Lee PS, Bathe KJ. Development of MITC isotropic triangular shell finiteelements. Comput Struct 2004;82(11):945–62.
[18] Bucalem ML, Bathe KJ. The mechanics of solids and structures – hierarchicalmodeling and the finite element solution. Springer; 2011.
[19] Lee Y, Lee PS, Bathe KJ. The MITC3+ shell element and its performance. ComputStruct 2014;138:12–23.
[20] Ko Y, Lee PS, Bathe KJ. A new 4-node MITC element for analysis of two-dimensional solids and its formulation in a shell element. Comput Struct2017;192:34–49.
[21] Sussman T, Bathe KJ. A finite element formulation for nonlinearincompressible elastic and inelastic analysis. Comput Struct 1987;26(1–2):357–409.
[22] ADINA R & D, Inc. ADINA Theory and Modeling Guide; 2017 <http://www.adina.com>.
[23] MacNeal RH, Harder RL. A proposed standard set of problems to test finiteelement accuracy. Finit Elem Anal Des 1985;1(1):3–20.
[24] Chen XM, Cen S, Long YQ, Yao ZH. Membrane elements insensitive todistortion using the quadrilateral area coordinate method. Comput Struct2004;82(1):35–54.
[25] Cen S, Chen XM, Fu XR. Quadrilateral membrane element family formulated bythe quadrilateral area coordinate method. Comp Meth Appl Mech Eng2007;196(41):4337–53.
[26] Li HG, Cen S, Cen ZZ. Hexahedral volume coordinate method (HVCM) andimprovements on 3D Wilson hexahedral element. Comp Meth Appl Mech Eng2008;197(51):4531–48.
[27] Hiller JF, Bathe KJ. Measuring convergence of mixed finite elementdiscretizations: an application to shell structures. Comput Struct 2003;81(8):639–54.
[28] Chapelle D, Bathe KJ. The finite element analysis of shells – fundamentals, 1sted. Springer; 2003; 2nd ed. Springer; 2011.
[29] Bathe KJ, Lee PS, Hiller JF. Towards improving the MITC9 shell element.Comput Struct 2003;81(8):477–89.