EAS-elements For Two-Dimensional, Three-Dimensional
EAS-elements For Two-Dimensional, Three-Dimensional
SUMMARY
The enhanced assumed strain (EAS)method, recently proposed by Simo and Rifai13, is used to develop new
four-node membrane, plate and shell elements and eight-node solid elements. The equivalence of certain
EAS-elements with Hellinger-Reissner (HR) elements is discussed. For instance, the seven-parameter
element EAS-7 with 2 x 2 integration is identical to the HR-element of Pian and Sumihara’. Eight-node
solid elements which are free of volumetric locking and four-node shell elements which have an improved
membrane and bending behaviour, compared to the Bathe-Dvorkin shell element’, are introduced.
Numerical tests for linear elastic problems show an improved performance of the EAS-elements.
1. INTRODUCTION
Due to their efficiency and simple geometry, low-order elements are often preferred in structural
mechanics. Unfortunately, pure displacement models exhibit, in many cases, severe stiffening
known as locking. Shear locking occurs in thin plates or shells when pure bending modes are
spoiled by parasitic shear strains so that the kinematic Kirchhoff condition cannot be repres-
ented. Membrane locking is encountered in thin shells when bending modes cannot be separated
from membrane strains and, thus, not allowing the verification of pure inextensional modes. For
nearly incompressible conditions, volumetric locking may also occur; in this case, deviatoric
modes always come along undesirably with volumetric strains. The locking phenomena have
been explained and discussed in detail in the literature.
To overcome locking, a number of techniques have been proposed and most of them can be
classified as hybrid-mixed methods, i.e. in addition to the displacement approximation, one or
more independent field assumptions for strains, stresses or incompatible displacements are
assumed. In general, the additional fields may be discontinuous from element to element so that
the corresponding unknowns can be eliminated on the element level which finally leads to an
element stiffness matrix.
Among these are formulations based on a Hellinger--Reissner (HR) principle with stress (or
strain) and displacement assumptions,’ ti the so-called assumed natural strain (ANS) ~oncept,~-’’
the enhanced assumed strain (EAS) methodI3 and the method of incompatible displacement
In the ANS concept the strain fields are linked directly to the displacement-
dependent strains at certain sampling points so that no independent degrees of freedom show up.
It was shown by Simo and Rifai13 for the EAS-formulation that the classical method of
incompatible modes is included as a special case.
The present paper focuses on the EAS-method which is applied to develop new four-node
membrane, plate and shell elements and eight-node solid elements. The underlying variational
principle and the basic equations of the EAS-method are given in Section 2. In Section 3 a new
four-node EAS-element is introduced which, for a linear elastic material, turns out to be identical
to the well-known HR-element of Pian and Sumihara.’ Therefore, a general equivalence theorem
for EAS- and HR-elements is discussed. In Sections 4 and 5, the EAS-formulated eight-node solid
and four-node plate and shell elements are developed. The shell formulation is based on the
degenerated solid approachi8 in which the concept of stress resultants is introduced.” Finally,
numerical tests are preformed in the linear elastic range which show the excellent behaviour of the
EAS-elements. For the application to small strain plasticity, see Reference 20.
2. I. Variational basis
The variational basis of the EAS-method is the principle of Hu-Washizu, where the internal
energy of one element can be written as
If stress and additional strain fields are chosen so that they are orthogonal to each other, the last
term in equation (3) vanishes:
ue= I, D
f (BU, -k
D
&)*C(BUh 4- & ) d n
The only variables to be assumed explicitly in equation (5) are the displacements and
the additional strains. For a detailed discussion of the method-in particular, the meaning
of the orthogonality condition in equation (4) in the context of the patch test condition-see
Reference 13.
EAS-ELEMENTS 1313
where
K= BTCBdQ (9)
is the usual stiffness matrix of a displacement model and D and L are defined as
D= 6, MTCMdR
L= 6. M'CBdQ
R is the vector of applied nodal forces used in the displacement method. Condensation of the
strain parameters M yields the element stiffness matrix
k = K - LTD-'L (12)
which is further processed by the conventional assembly procedure.
and its inverse J-’, the compatible strains are calculated as usual according to
( a = 1,2; r,
= t2 = q). The additional strains, defined in the global Cartesian frame, are
interpolated according to equation (7):
g, = MU (16)
where
det Jo
M=-- T ; ~M,
det J
det J denotes the determinant of the Jacobian J, and det Jois its value at the origin Jo= J15=q=o.
The matrix T i T contains the components Jijoof Jo:
According to tensor calculus T i T maps the polynomial shape functions of M,, defined in the
isoparametric space (therefore, the subscript 0, into the global frame.I3 This transformation is
restricted to the origin so that the components of T i T are constant and the chosen polynomial
EAS-ELEMENTS 1315
order is not increased. An optimal interpolation of M, for the additional strains can be found by
inspecting the polynomial fields of the compatible strains in the isoparametric space (see Table I).
In order to decouple and enhance the compatible strains, the following seven-parameter interpo-
lation was chosen:
Mi=Lo
0
'
0
0 0 0 5 q o
5 q O O 5 q1 (19)
9 10 11 12 13 14 15
With these additional modes, all the three strain fields are completely bilinear. Now the element
has 15 modes in total, three rigid body modes plus 12 deformation modes, which are evaluated by
a 2 x 2 numerical integration (three strain components times four Gauss points = 12).
Equations (1 3)-(19) are introduced into equations (9)-(11). The element stiffness matrix,
equation (12), can be obtained without any problems since the columns in M, are linearly
independent yielding a positive-definite matrix D in equation (10). Furthermore, compatible and
additional strain fields are independent of each other.
For a linear elastic material, the stiffness matrix of an EAS-7 element with 2 x 2 integration is
identical to that of the Pian-Sumihara element (PS) which is based on a HR-principle. Due to this
fact, a general equivalence theorem between EAS- and HR-elements is discussed in the next
section.
P,= 0 1 0 0 (
1 0 0 1 0 01
1 5 1
1316 U. ANDELFINGER A N D E. RAMM
Uh = N d"
ah = Pb = TOP@
G =
Qe
IPTEN dQ = j P T B dQ
Q,
k = CTH-'G
Uh = PH-lCd"
Although only proven by inspection, this equivalence may be generalized for a linear elastic
material in the following way. The stiffness matrix of a 2 x 2 integrated EAS-element is equivalent
to the stiffness matrix of a HR-element, if the polynomials in M, and P, are complementary. For
a four-node element this means they add up to complete bilinear distributions and a polynomial
term used for a strain component in M, is not considered for the corresponding stress component
in P,. If SM is the n,-dimensional space of the polynomial expansion in M, and S p that
corresponding to P,, the following two conditions must hold for this equivalence:
SM SP = scomplete
(21)
S'nSP=$ (22)
For a bilinear element ScomPlete defines the n,-dimensional space with complete bilinear poly-
nomials. Although not proven formally, it can be shown that the equivalence theorem is also valid
for eight-node solid elements with 2 x 2 integration; in that case Scomp'e'e represents the six-
dimensional space with complete trilinear fields. For higher-order elements, the equivalence holds
only if the elements are linearly distorted (straight edges, mid-side nodes in the centre of edges,
etc.).
The stresses of the two models are also identical if the stresses for the EAS-element are
calculated according to the HR-formula
~h = PH-'Gd" (23)
These stresses always satisfy equation (4):
oiZ,dR = jll 1
-1
dnTGTHPT
det Jo
PTTT-T i T M S a d e Jt d l d q
detJ
With the exception of P, and M,, all other terms are constant so that
1
P:M,d< dy = 0 (24)
In Figure 3, the identities of three models are depicted. The first one, the equivalence of EAS-7
and PS, has already been discussed above. The second example describes an eight-parameter
EAS-ELEMENTS 1317
HR
PS
Mf-E 83 l o o q o g ~ o0
[
Pt= 0 1 0 0 5 0 5y 0
0 0 1 0 0 0 061 1
EAS-4 or Qm6 HR-8
MC = [-I
Q4
1 0 0 ~ ~ 0 ~0 0~ 0 ' 0
0 5r)EqOO 0
0 0 1 0 0 0 0 0 0 6 ~ 6
HR-12
0
~ .
HR-element which is identical to the EAS-4 model (which in turn is identical to the element Qm6
of Taylor er al.," see also Reference 13. If no additional strains are assumed, the stiffness matrix of
the displacement model remains unchanged; according to the theorem discussed above, a HR-
model with a complete bilinear stress interpolation leads to the same stiffness matrix. However,
this is well-known from the limitation principle of Fraeijs de Veubeke.2' Besides these three
examples, of course, identities of other models can be constructed as well.
Remarks:
1. The stress interpolation for the PS-element is defined in equation (20). In a computer code,
a slightly modified interpolation matrix P, can be used, which for constant material
properties results in a very efficient one-point integration plus a rank-2 update for the
stiffness matrix.22
2. The stresses may be recovered by two different procedures:
(a) The stresses are calculated pointwise as usual, based on strains, equation (2). Although
the procedure is variationally not consistent and the stress values are, in general, less
accurate; it is the simplest method and proves advantageous in materially non-linear
analysis.
(b) The stresses are obtained via equation (23). This is variationally optimal and refers to the
equivalence between EAS- and HR-elements; however, the procedure needs elaborate
matrix operations. It may be simplified if the stresses are evaluated at the centre of the
element so that only the constant stress terms are included.
This interpolation requires a 3 x 3-integration scheme. However, even for arbitrary geometries,
the element EAS-11 results in exactly the same stiffness coefficients as the element EAS-7,
provided that a 3 x 3-integration is applied as well. This means that the extra quadratic strain
modes do not influence the stiffness matrix. Therefore, the seven-parameter approximation (19)
can be seen as an optimum for a bilinear element. It should be mentioned, that the four-parameter
r
assumption
M,= 0 Ov 0
O 0 Ol
already leads to a volumetric locking free plane strain element. The bilinear terms in equation (19)
improve only the bending behaviour of a distorted element. The numerical tests show that for
a rectangular or parallelogram geometry, EAS-7 and EAS-4 are identical.
with
where ( C ? < " / ~ X , ) ~are elements of JO'. For the incompatible modes, a slightly modified strain
D
operator instead of B (see equation (15)) has to be used in order to assure exact constant
D
stre~ses.'~ The stiffness matrix is given by equations (9)-(12); only M has to be replaced by B'Q.
If, in addition, the local displacement components u," are transformed into the global dis-
placements u?:
u; = (g) 0 u,"
using the constant components ( 3 t a / a x i ) ,of the inverse Jacobian the similarity between equation
(29) and the EAS-interpolation (17) can be recognized. As stated in Figure 3, element Qm615 is
EAS-ELEMENTS 1319
identical to EAS-4.13 However, it has to be emphasized that it is not easy to find an invariant
displacement model with incompatible modes identical to EAS-7, because these incompatible
modes are defined in global variables u t and require invariance with respect to the x-y
co-ordinate system. The EAS-method turns out to be more flexible, since natural variables are
interpolated which only require invariance with respect to the natural 5-q co-ordinate system.
:
M, =
yo0 ~ 0 0 0 0 0 0 0 0 0 0 0 0 S'q510 0 0 0 hi 0 0
OIllO , 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5qqi 0 0 0 S'qi 0
00; 000000 0 0 0 0 0 0 0 0 0 or;?/[ 0 0 rri
000 ' <qoooo ti qi 0 0 0 0 0 0 0 0 0 0 0 0 0
000 0 o ~ ; O o 0 0 <qq( 0 0 0 0 0 0 0 0 0 0 0
000 oooo~[ 0 0 0 otrs'i 0 0 0 0 0 0 0 0 0
2 5 2 6 2 7 2 8 2 9 3 0 3 1 32.33 3 4 35 36 37 38 39 40 41 42 43 4 4 4 5 4 6 4 7 48 49 5 0 51 52 5 3 54
1320 U. ANDELFINGER AND E. RAMM
'
ill
u E
.
,-I
3
EAS-ELEMENTS 1321
-- __--
t y’v E 5@ #
, W’
V’
U’
The element (named 3D.EAS-30) has 30 additional and 54 modes in total. Following the
arguments of the previous section, it is identical to a HR solid element using the following stress
as~umption:~
l q [ q j O O O 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 l ( j < [ O O O 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 5 q ~ q 0 0 0 0 0 0
(35)
0 0 0 0 0 0 0 0 0 0 0 0 o o o o l ~
Of course, the element 3D.EAS-30 requires the inversion of a 30 x 30 matrix for each element. To
reduce the numerical effort, some less important additional strain modes can be dropped.
Especially the trilinear modes 49-54 show little influence on the overall performance.
However, in view of a volumetric-locking free behaviour, one should include modes 25,26, 27
and 40-45. These nine modes assure that all three normal strains consist of the same polynomial
fields so that the constraint cii = 0 can be fulfilled and no spurious constraint is produced. It is
important to note that the often used incompatible displacement model using nine extra
quadratic displacement modes
t 2 q 2 [ 2 0 0 0 0 0 0
u;=[. 0 0 t2 q2 r2 0 0
0 0 0 0 0 o ( 2 q 2 [ 2
+
(denoted by Hml 1, because the hexahedral element with 8 3 = I1 modes for each displacement
+
component is the three-dimensional equivalent of Qm6 (4 2 = 6 modes for each displacement
component)) is not free of volumetric locking. The element which is identical to an EAS-element
with modes 25-33-recall, the identity between EAS-4 and Qm6-exhibits volumetric locking
for certain distorted element meshes because modes 40-45 are not considered.
In the numerical simulations, the following EAS-models are compared:
3D.EAS-9: Modes 25-33 are considered; the element is identical to H m l l which is based on
equation (36).
1322 U. ANDELFINGER AND E. RAMM
3D.EAS-15: Modes 25-39 are considered; the element has an improved bending behaviour
compared to 3D.EAS-9.
3D.EAS-21: Modes 25-45 are considered; the element is free of volumetric locking.
3D.EAS-30: Strains are complete up to trilinear fields; the element is identical to the HR-
element of Pian and Tong3 here named 3D.HR-18.
As in Section 3, equivalent HR-models can be constructed for all 3D.EAS-models.
5.1. Introduction
In this section the EAS-method is applied to four-node shell elements. The shell formulation is
based upon the degenerated solid approach' which takes into account transverse shear effects.
The stress state across the thickness is represented by stress r e s ~ l t a n t s In
. ~ order
~ to improve the
element performance, the EAS-method is used for the membrane and bending component, while
the transverse shear component is formulated according to the so-called A N S - m e t h ~ d .This ~
combination was selected because the ANS-concept is ideally suited for the transverse shear
component. The EAS-formulation for the transverse shear part does not lead to shear-locking
free elements for arbitrarily distorted geometries unless additional projection methods' are used.
On the other side, the ANS-concept is not able to decouple both directions of the membrane and
bending components because of the $1-influence.
f - x + 2ihp
~
(37)
x is the vector of the reference plane, p is the normalized director, the thickness co-ordinate 5 takes
values from - 1 to + 1 and h denotes the thickness of the shell (Figure 5).
Equation (37) is discretized
2, = Nx" + -Nh"p"
i
2
A typical four-node shell element is shown in Figure 6.
X J
The displacement 6 of an arbitrary point within the shell can be written in terms of the
displacement u of the mid-surface and the rotation of the director
i
G = u + 7 hAp (39)
1
or in discretized form
i
6,, = NU"+ NR"Ap" (40)
2
The rotation of the director Ap" can be expressed by two independent rotational degrees of
freedom.
e, = es (42)
e, = e, x e, (43)
The orientation of the orthogonal co-ordinate system within the mid-surface is arbitrary; for
simplicity, e, is chosen in the direction of e,.
5.3.I . Membrane and bending components. In a geometrically linear analysis, membrane and
bending strains are:24
1324 U. ANDELFINGER AND E. RAMM
(a, b = 1,2; r1 = r, r2 = s). The three translational and the two rotational degrees of freedom are
separated so that the matrices for the kinematic operators can be decomposed into
ax _a
_ F
as as
ax a ax - a
- -+-
Br as as ar
In equation (47b), F is a 3 x 2 matrix linking the two rotations to the three global direction^.'^
5.3.2. Transverse shear component. In order to alleviate shear locking, the assumed natural
strain concept7x9will be introduced. The shear strains are referred to the co-ordinate system t-q
with base vectors er = (ax/a<)and e, = (dx/dq). The linear equations for the components of the
transverse shear strain are
ax
auT
=-p
atz
+7
a<
Ap
identical formulation for the two parts assures that a membrane-bending coupling can be
handled easily. This coupling arises when eccentric stiffeners are present or a non-symmetrically
layered material is used. The transverse shear part will be approximated according to the
ANS-method. Alternative hybrid-mixed formulations for the different parts will be discussed at
the end of this chapter.
5.4.1. The EAS-formulation of the membrane component. The EAS-formulation of the mem-
brane stiffness matrix is determined by equations (9)-(12). Hence, the matrices B, C and M need
to be specified. With
(50)
a x a_
_ x T_ _
axax’
__ ar as
the derivatives of the bilinear shape functions ” ( I = I,. . . ,4)with respect to the local ortho-
gonal co-ordinate system r, s can be computed:
Nf,
Equations (51) and (46) lead to the operator matrix El,. However, in order to assure a constant
stress state, at least for a flat element (patch test condition), constant base vectors ( a x / a r ) , and
(ax/ds),, calculated at the origin 4 = v = 0, are inserted into equation (46). If the element is
warped, the base vectors are computed individually at every integration point.
The usual material matrix is given by
C,=--- Eh
1 - v2
Eh = MU (53)
with
det J,,
M=-- TG’M, (54)
det J
In equation (18) J i j o are replaced by Aijc),where the elements A,, of the matrix A, equation (50),
are again evaluated at the origin C = q = 0. Alternatively, the seven- or the four-parameter
interpolations (19) or (26),respectively, are used for M, which leads to two different shell elements.
With equations (50)-(54),the membrane stiffness matrix is derived from equations (9)-(12).
5.4.2. The EAS-formuEation of the bending component. The EAS-formulation of the bending
part is analogous to that of the membrane component. Again, equations (50) and (51) are applied
to derive the B-matrices from equation (47). Also here, for flat elements, J x / d r and ax/as are
1326 U. ANDELFINGER AND E. RAMM
replaced by (ax/&), and (dx/ds), in equation (47). Now the material matrix is given by
h2
Ch = -cc, (55)
12
Again, for the additional bending strains either the seven- and alternatively the four-parameter
model is used.
5.4.3. The ANS-formulation ,for the transverse shear component. The concept of the ANS-
technique, for example, as applied to the four-node shell element,’ is well-established in the finite
element literature and, therefore, is only summarized here. The keypoint of the method is the
reduction of the bilinear transverse shear fields to constant-linear fields ye; and yq defined on the
basis of two sampling points for each of the two shear strain components:
Table IIJ. Four-node shell elements with different membrane, bending and transverse shear formulations
As mentioned before, the membrane and bending stiffness matrices of the element HR-ANS
can be evaluated by a one-point integration plus an additional rank-2 update. This is not only
advantageous from a computational point of view but satisfies automatically the patch test;
furthermore, it i s not necessary anymore to use constant base vectors in equation (46) or (47b).
6. NUMERICAL EXAMPLES
6.2.1. Eigenvalue analysis. In order to test the behaviour of solid elements in the (nearly)
incompressible range, an eigenvalue analysis for one square element with a side length 1.0 and
E = 1.0 is performed. The material is assumed to be incompressible, i.e. in the numerical
calculation L' = 0.4999 was used, rendering eigewalues of 103-105instead of cc . Table IV shows
the eigenvalues for 18 modes; the six zero-eigenvalues for the six rigid body modes are not shown.
The elements 3D.EAS-21,3D.EAS-30 and 3D.HR-I 8 give the correct eigenvalues. For a volumet-
ric-locking free behaviour, it is important that the elements contain only one incompressible
mode; i.e. its eigenvalue, is infinite. For the displacement model DISP, six deviatoric modes are
always mixed up by parasitic volumetric strains which for v = 0.5 leads to six unrealistic infinite
eigenvalues. The elements 3D.EAS-9 and 3D.EAS-15 still have three undesired infinite modes
which also may lead to volumetric locking in some cases.
6.2.2. Regular block with almost incompressible material. A regular block is considered with
side lengths of 100 and a height of 50 (Figure 9). The structure is fixed at the bottom and loaded at
the top by a uniform pressure load of q = 250/unit area, acting on an area of 20 x 20 at the centre.
The material parameters are chosen to model an almost incompressible material.
1328 U. ANDELFINGER AND E. RAMM
t"
-F E = 3000
-
j-y7-3.;,2] ---cF v =0,0
dL F == 0.1
h 1.0
I -
10.0
I \
-... --
... ...-. .. -... -__.__..-... I
E = 210000
v = 0,1999
q 250
Figure 9. Regular block loaded by a uniform pressure load of q = 250/unite area acting on an area of 20 x 20 in the centre
Figure 10. Distorted mesh for regular block (only the contours at the surfaces are shown)
One quadrant of the structure is idealized. First with a uniform 5 x 5 x 5 mesh, as shown in
Figure 9, and then with a non-regular mesh, Figure 10, to test the elements in a distorted
configuration.
In Table V the vertical centre displacement, u, is given for the different finite element models.
The element 3D.EAS-30 is identical to 3D.HR-18 so that the same displacements are obtained.
The elements DISP and 3D.EAS-9 are too stiff while the elements 3D.EAS-21, 3D.EAS-30 and
3D.HR-18 are free of locking.
In Reference 20 it is shown for small strain plasticity that the elements 3D.EAS-21 and
3D.EAS-30 are capable of avoiding volumetric locking.
1330 U.ANDELFINGER AND E. RAMM
L = 100
h = l
E = 10"
v = 0.3
P = 16.367
6.2.3. Solid elements in plate bending. A square plate with clamped edges is loaded by
a concentrated load P at the centre (Figure 11). One quadrant of the structure is idealized with
one layer of 2 x 2 and 4 x 4 solid elements. Table VI shows the centre deflection for different finite
elements. The analytical value according to the Kirchhoff theory26is w = 0.0611 PL2/Eh3 = 1.0.
Except for the displacement model DISP and 3D.EAS-9, the solid elements show an excellent
bending behaviour. Again, since 3D.EAS-21, 3D.EAS-30 and 3D.WR- t 8 are identical for regular
meshes, the same displacements are obtained.
For comparison, the results of the shell elements BD and EAS7-ANS are listed as well.
Although these values are close to the Kirchhoff solution, they must be viewed with caution since
a plate theory including transverse shear effects renders an infinite displacement under the point
load.
6.3.1. Eigenoalue analysis of d i e r e n t four-node plate elements. For the 2 x 2-integrated dis-
placement model DISP, the reduced integrated models URI and SR1 as well as for the elements
BD and EAS7-ANS, the stiffnesses matrix of a square-plate element is investigated by an
eigenvalue analysis. For the rectangular geometry the elements EAS4-ANS and HR -ANS are
identical to EAS'VANS. The tested element has a side length of 1.0 and material constants of
E = 1.0 and v = 00.
EAS-ELEMENTS 1331
m m m m m
0 0 0 0 0 0 0 22g g
m m m o o
gggzz
303030"
1332 U. ANDELFINGER AND E. RAMM
The 12 eigenvalues and the corresponding strains are shown in Table VTI. The values of the
fully integrated displacement model DISP are given in the second column. Due to parasitic shear
strains y';'" and y?, the eigenvaiues of the curvature modes ICE< and JC& are too large (0.118056
instead of 0.083333). The reduced integrated models URI and SRI render the exact value SO that
the constant bending modes are locking free. However, due to underintegration, the element URT
has four and SRI two additional zero-energy modes.
The element BD has no zero-energy modes and passes the patch test; i.e. the bending modes
K& and K;,, are exact as well. The element EAS7-ANS has linear bending modes dinwhich have
improved when compared to BD or SRI (0.027777 instead of 0041667) because K~~ is assumed to
be constant.
A complete decoupling of the linear shear modes and the constant drilling mode KE,, is not
possible. However, this coupling only leads to locking when the elements are warped (see Section
6.3.4). It does not show up in reduced integrated elements URI and SRT because they do not
contain linear shear modes.
6.3.2. Morley's 30" skew plate. In this example Morley's 30" skew plate (length L = 100,
thickness h = 1, E = lo5,v = 0.3, uniform pressure load q = 1) is analysed. The boundary condi-
tions of the simply-supported plate (Figure 12) are idealized as the so-called soft support (SS1). In
Table VIII, the centrepoint deflection is given for different element meshes. Due to the parallelo-
gram geometry of all elements, EAS4-ANS and EAS7-ANS are again identical. For all meshes
they show a clear improvement over the element BD.
In the literature, the Kirchhoff solution 4-455 obtained by Morley2' is often used as a reference
value. However, even for the length to thickness ratio of only 100, the shear deformations cannot
be neglected so that a correct value of about 4.64 is anticipated. In the case that the thickness of
the plate is further reduced, all elements come close to the results of Morley.
6.3.3. Scordelis-Lo roof: The geometry of the Scordelis-Lo roof, a cylindrical shell under dead
load, is given in Figure 13. The length of the structure, which is supported through rigid end
diaphragms, is L = 50, the radius is R = 25 and the thickness is h = 0.25. Due to symmetry
conditions only one quarter of the structure is idealized. In Table IX the vertical deflection wB is
reported for different finite element meshes. A 'deep shell solution' w, = 03008 is given in
Reference 6.
For the 2 x 2 mesh the elements BD and SRI show seemingly better results than the elements
EAS4--ANS and EAS7-ANS. The reason for this is that the too stiff behaviour of the BD and
SRI elements is counterbalanced by a convergence from above in this example. However, after
a mesh refinement the elements EAS4-ANS and EAS7-ANS show the anticipated excellent
performance.
L = 100
h = l
E = lo5
V = 0.3
q =I
L
I 4
Figure 12. Morley's skew plate
EAS-ELEMENTS 1333
E= 4 , 3 2 . 108
v = 0.0
h = 0.25
P = 360
The advantage of the elements EAS4-ANS or EAS7-ANS becomes even more apparent if one
looks at the stress distribution. For instance, if the membrane shear forces nrs are considered and
computed for an 8 x 8 mesh over a quarter of the structure, the BD-element (Figure 14) exhibits
distinct oscillations, especially at point B, where the forces oscillate between - 17400 and
+ 17400. On the other hand, the element EAS7-ANS leads to a smooth and qualitatively correct
solution (Figure 15);according to equations (23)and (20) the stresses are constant in each element.
Apparently, a one-point stress calculation at i:= q = 0 avoids the oscillations as well.
6.3.4. Locking test: a quarter of an in$nitely long cylinder. In Figure 16, a section of an
infinitely long curved panel is shown which is defined by a quarter of a cylinder. It is clamped on
one side and loaded by a constant moment on the other side so that the problem is in a pure
bending mode. To test the locking of the finite elements, the thickness of the structure was
reduced successively by a factor of 10 and, hence, increased the displacements each time by
a factor of lo3.
For a regular mesh of 3 x 3 elements, all models render the correct displacements and exact
stress resultants: constant moments, zero membrane and shear forces. However, if the mesh is
1334 U. ANDELFINGER AND E. RAMM
bigure 14. Membrane shear rorces nrs of the Scordells-Lo roof, using the element BD (half of the structure is shown)
Figure 15. Membrane shear forces n,, of the Scordelis-Lo roof, using the element EAS7TANS (half of the structure is
shown)
distorted in such a way that the elements are warped (Figure 17), the elements HR-ANS,
EAS7-ANS and BD exhibit severe shear locking. The displacements increased only by a factor of
10 instead of the expected factor of lo3.Furthermore, distinct oscillations for the transverse shear
stresses occurred.
In order to exclude membrane locking, the element 1 x 1-AN§ was tested as well. Here the
membrane and bending parts are 1 x 1-integrated and the transverse shear part is formulated by
the ANS-concept. Since the only elements that do no1 lock are URI and SRI, the locking can be
identified as shear locking.
It can be argued that a coarse distorted mesh is a rather rough geometrical representation of
a cylindrical shell and the resulting kinks might change the response of the structure. However,
EAS-ELEMENTS 1335
a = 10
L 10
i
v = 0.3
E = 1000
URI
SR I
10Z
1 2
*
1 10 100 1000 10000 100000 R'h
this argument is wrong because of two reasons: Firstly, the reduced integrated elements show
results close to the original shell behaviour. Secondly, a mesh refinement based on the original
coarse 3 x 3-mesh including the kinks also converges to the results of the smooth structure.
The example demonstrates the sensitivity of four-node shell elements when they are warped.
Also, the constant drilling curvature, utq,always comes along with linear transverse shear strains.
It has been found in this case that the locking depends on the boundary conditions. For instance,
no locking is observed when the symmetry boundary conditions are released and the shell has
three free edges.
7. CONCLUSIONS
It is shown that the enhanced assumed strain method is well-suited to improve and optimize
continuum and structural finite elements. If for linear elements, the additional strains are assumed
1336 U. ANDELFINGER AND E. RAMM
so that the total strain fields are of the same order as the displacement fields, the EAS-formulation
leads to elements which are identical to optimally interpolated HR-elements. However, in order
to keep the numerical effort as low as possible the number of additional strain parameters should
be kept to a minimum. For instance, it is not advisable to use complete trilinear strain modes for
the eight-node solid element.
In shell analysis, the BD-element is improved by the EAS-formulation for the membrane and
bending components. Especially the stress resultants are more accurate and free of oscillations.
For the transverse shear components, the ANS-method is applied. However, the last test problem
indicates that elemcnts with an ANS-formulated transverse shear part may still exhibit shear
locking when they are warped; as a remedy the shear stiffnesses may be scaled.
The application of the EAS-concept to small strain plasticity is described in References 13 and
20. Further studies are necessary
- to base the equivalence theorem which has been found primarily by inspection on a more
rigorous theoretical foundation,
- to extend the EAS-formulation into the large deformation range where the additional
variables have to be defined in terms of displacement gradients instead of strains,
- to clarify the kind of stress recovery which has to be variationally consistent but still efficient,
in particular, for material non-linearities.
REFERENCES
1. S. W. Lee and J. J. Rhiu, ‘A new efficient approach to the formulation of mixed finite element models for structural
analysis’, Int. j . numer. methods eng., 21, 1629-1641 (1986).
2. T. H. H. Pian and K. Sumihara, ‘Rational approach for assumed stress finite elements’, Int.j. numer. methods eng., 20,
1685-1695 (1984).
3. T. H. H. Pian and P. Tong, ‘Relations between incompatible displacement model and hybrid stress model’, Int.
j . numer. methods eny., 22, 173-181 (1986).
4. ’ IH.
.H. Pian and C. C . Wu, ‘A rational approach for choosing stress terms for hybrid finite element formulations’, Int.
j . numer. methods my., 26, 2331-2343 (1988).
5. J. J. Rhiu and S . W. Lee, ‘A new efficient mixed formulation for thin shell finite element models’, Int. j . nunter. methods
eng., 24, 581-604 (1987).
6. A. F. Saleeb, T. Y. Chang and W. Graf, ‘A quadrilateral shell element using a mixed formulation’, Comp. Struct., 26,
757-803 (1987).
7. E. N. Dvorkin and K. J. Bathe, ‘A continuum mechanics based four-node shell element for general nonlinear analysis’,
Eng. Comp., 1, 77-88 (1984).
8. H. C. Huang and E. Hinton, ‘A new nine node degenerated shell element with enhanced membrane and shear
interpolation’, Int. j . numer. methods eny., 22, 73-92 (1986).
9. T. J. R. Hughes and T. E. Tezduyar, ‘Finite elements based upon Mindlin plate theory with particular reference to the
four-node isoparametric element’, J . Appl. Mech., 48, 587-596 (1981).
10. R. H. MacNeal, ‘Derivation of element stiffness matrices by assumed strain distributions’, Nucl. Eng. Des., 70, 3-12
(1982).
11. K. C. Parks and G. M. Stanley, ‘A curved C“ shell element based on assumed natural-coordinate strains’, J . Appl.
Mech., 53, 278-290 (1986).
12. J. C. Simo and T. J. R. Hughes, ‘On variational foundations of assumed strain methods’, J . Appl. Mech., 53, 51-54
(1986).
13. J. C. Simo and M. S . Rifai, ‘A class of mixed assumed strain methods and the method of incompatible models’, Int.
j . numer. methods eng., 29, 1595-1638 (1990).
14. G. P. Bazeley, Y.K. Cheung, B. M. Irons and 0.C. Zienkiewicz, ‘Triangular elements in plate bending-conforming
and nonconforming solutions’, Prac. 1st Con$ Matrix Methods in Structural Mechanics, Wright-Patterson ATBFB,
Ohio. 1965.
15. R. L. Taylor, P. J. Beresford and E. L. Wilson, ‘A non-conforming element for stress analysis’, Int. j . numer. methods
eng., 10, 1211-1220 (1976).
16. E. L. Wilson, R. L. Taylor, W. P. Doherty and J. Ghaboussi, ‘Incompatible displacement models’, in: S . T. Fenves
et al., (eds.), Numerical and Computational Methods in Structural Mechanics Academic Press, London, pp. 43-57,
1973.
EAS-ELEMENTS 1337
17. Ch.-Ch. Wu, M.-G. Huang and T. H. H. Pian, ‘Consistency condition and convergence criteria of incompatible
elements-general formulation of incompatible functions and its application’, Comp. Struct., 27, 639-644 (1987).
18. S. Ahmad, B. M. Irons and 0.C. Zienkiewicz, ‘Analysis of thick and thin shell structures by curved finite elements’,
Int. j . numer. methods eng., 2, 419-451 (1970).
19. W. K. Liu, S. E. Law, D. Lam and T. Belytschko, ‘Resultant-stress degenerated-shell element’, Comp. Methods Appl.
Mech. Eng., 55, 259-300 (1986).
20. U. Andelfinger, E. Ramm and D. Roehl, ‘2D- and 3D-enhanced assumed strain elements and their application in
plasticity’, in: E. Oiiate and D. R. J. Owen (eds.), Proc. 3rd Int. Conjerence on Computational Plasticity, Fundamentals
and Applications, Barcelona, April 1992, Pineridge Press, Swansea, 1992.
21. B. Fraeijs de Veubeke, ‘Displacement and equilibrium models in the finite element method‘, in: 0. C. Zienkiewicz and
G. C . Holister (eds.), Stress Analysis Wiley, London, 1965.
22. 0. C. Zienkiewin and R. L. Taylor, The Finite Element Method, 4th edn., Vol. 1, McGraw-Hill, London, 1989.
23. X. J. Wang and T. Belytschko, ‘A study of stabilization and projection in the 4-node Mindlin plate element’, Int.
j . numer. methods eng., 28, 2223-2238 (1989).
24. J. C. Sirno, D. D. Fox and M. S. Rifai, ‘On a stress resultant geometrically exact shell model. Part 11. The linear theory;
computational aspects’, Cornp. Methods Appl. Mech. Eng., 73, 53-92 (1989).
25. U. Andelfinger, ‘Untersuchungen zur Zuverlassigkeit hybrid-gemischter Finiter Elemente fur Flachentragwerke’,
Dissertation, Institut fur Baustatik, Universitat Stuttgart, 1991.
26. S. P. Timoshenko and S. Woinowsky-Krieger, Theory 01Plates und Shells, 2nd edn., McGraw-Hill, New York, 1970.
27. L. S. D. Morley, ‘Skew Pfutes and Structures’; Pergamon Press, Oxford, 1963.