Professional Documents
Culture Documents
Editors
W Wunderlich, E. Stein, and K.-J. Bathe
This work is subject to copyright. All rights are reserved, whether the whole Of part of the material is concemed,,
specifically those of translation, reprinting, re-use of iIIustrations, broadcasting, reproduction by photocopying
machine or similar means, and storage in data banks. U nder § 54 of the German Copyright Law where copies are
made forotherthan private use a fee is payable to 'Verwertungsgesellschaft Wort',Munich.
© Springer-Verlag Berlin, Heidelberg 1981
Originally published by Springer-Verlag Berlin Heidelberg New York in 1981
Softcover reprint ofthe hardcover Ist edition 1981
The use of registered names, trademarks, etc. in this publication does not imply. even in the absence of a specific
statement, that such names are exempt from the relevant protective laws and regulations and therefore free for
general use.
2061/3020 - 543210
Preface
I. BABUSKA
UlliverSlty of Maryland, College Park, USA
Summary
The paper shows how a flnite element solver for linear ellip-
tic equations with an a-posteriori estimator can be extended
to the case of eigenvalue problems and nonlinear elliptic
problems.
Introduction
Lu = f (la)
u = 0 on 3r1 (lb)
( 3)
5
(4)
E(U(lli»
9 (Sb)
for any
Lu = AU (7a)
u = 0 on an (7b)
and in addition
(9 )
Then
LC. 2 1 ,
1.
and of course
Lu + g(u) = f (lla)
u = 0 on 3>1 (llb)
Let now V be the exact solution of (1) with the right hand
side f - g(u(lJ» and v(lJ) be the finite element solution.
Obviously u(lJ) = v(lJ) • Denote by E(u(lJ» the estimator
for Ilv-u(lJ)II E • Nowwehave
1
Proof. Write V = ~+u(lJ) . Then for "Iv E' HO (>I)
We have also
and therefore
I I z I IE < E (u ( )l) ) cr ( 1) as E -* 0 .
Now
~ (l+cr(l))E(u()l))
and because
The FEARS
During recent years the solver FEARS (finite element adaptive
research solver) was developed which has various special fea-
tures which will not be discussed here. For more information,
see [ 2 1, [3 1, [4 1, [5 1, [ 6 J •
DOF 100
*+t E
neTT
20 14.697 1. 015
40 11. 664 1. 095
52 10.168 .981
76 8.136 1. 0002
130 6.332 .9895
173 5.744 .97107
On the sides
we prescribe ul = 0 , u2 = 1 On
References
[lJ Babuska, I.i Rheinboldt, W.C.: A-posteriori error esti-
mates for the finite element method. Int. J. Num. Mech.
Eng. 12 (1978) 1597-1615.
J F BESSELING
UOlverslty of Technology, Delft, The Netherlands
Summary
1. Introduction
for each finite element the necessary and sufficient conditions for
rigid body motion in terms of the nodal velocities, or equivalently
to define the deformation modes by properly chosen generalized
strain measures. If these strains by constitutive equations are
related to the dual stress multipliers, we arrive at the system of
equations by which we can solve a stress- and deformation problem.
This approach, that was indicated in [2J, will be discussed here in
some more detail.
JdB t.ou.dA +
~ ~
JB pf.ou.dV
~ ~
= JB pu.ou.dV
~ ~
(2.1)
doui dOU.)
[ - - + --1. = oe: .. D. (2.2)
dX. dX. ~J
J ~
which now must hold for all kinematically admissible virtual displace-
ments ou .•
~
t.
1.
t .. n. (2.4)
1.J J
and
dt ..
--2:.J. + p f. (2.5)
dX. ~
J
13
the right-hand side of (2.3) may rightly be called the virtual work
of deformation per unit volume.
In the case of elastic material behaviour, or if the inelastic
deformations remain sufficiently small, deformations with respect to
an initial geometry of the body are physically relevant variables.
The deformation tensor may then be considered as the sum of an elastic
and an inelastic part, such that for the elastic components hold
unaltered the elastic~ty relations that are valid in the ~nitial
(2.6)
(2.7)
(2.8)
Using the equality pdV = podVo and again with the aid of a symmetric
tensor field of multipliers, craB' the principle of virtual work is
expressed with volume integrals in terms of the initial geometry B
o
by
J
dB
t.ou.dA +
~ ~
J
B
p f.ou.dV
0 ~ 1 0
JB(pOu.ou.
ll
+ cr BoE B) dV,
a a 0
(2.9)
o o
14
(2. 10)
J
B
t .. OE. ,dV
~J ~J
J
B
o
(2. II)
The principal difference between O£ij and oE aB lies in the fact that
O£ij cannot, and OE aB can be derived from a measure of finite strain
as a first variation. The strains EaB are by their elastic components
, 0
EaB = EaB - EaS related to the stresses GaB' that are the proper duals
of the virtual deformations oE aS '
To allow for a maximum freedom in approximation methods we now write
the virtual work condition (2.9) for an assembly of N finite elements
, h anumb er 0 f mu It'~p I'~er f unc t'~ons, 0aB'
w~t * * and t*,~ *.
ti'
N
L J
e=1 Be
f-p 0
(f,-u:)ou: +
~ ~ ~
o
+ 00*
aB
N
+ L J(lBe [ ( 0u ~ ~
- 0u: ) t ~ + (u ~ - u:) 0 t ~] dA
~ ~ ~ ~ ~ 0
+
e=1
o
+ JdB u [-ou~t~* ~ ~
+ (~, -u~) ot~*]
~ ~ ~
dA= O. (2.12)
Here E~S represents the small inelastic stralns and aCT-To) is the
specific thermal expansion. The prescribed quantities ~. and
1
r. 1
will
usually be given on the external boundary in the deformed state.
e ' ou e ' and
By considering the variations oE aS ~ui
* we derive the
k
following equations
0* (2.14a)
as
3u e
_3_{ 0* (0 + ~)} + pofk (2.14b)
3a S as ka 3aa
e
0*
as
[0 ka
3Uk )
+ 3a nS t*
k
on 3B e
o '
(2.14c)
a
* - t*
t i (!!) =
on element interfaces, (2.14d)
i(-!!)
t~ dA = t. dA on 3B t . (2.14f)
1 0 1
e e e e
3u 3u S 3uk 3uk I
Ee
as 2
1 a
-- +
3a s 3a + 3aa
a
aa; J in Be0 ' (2.ISa)
e
u. u~ on 3B e (2.ISb)
1 1 o '
u~ u. on 3B u . (2.ISc)
1 1
(2.16)
0.1)
The conditions for a rigid body motion are expressed by oEaS '" O.
However, in the derivation of the finite element equations by appli-
cation of the principle of virtual work it suffices to require
locally oEaB '" 0, such that for each finite element a number of
linearly independent conditions is obtained that is equal to the
dimension of its deformation space. In other words, for the finite
element model we may derive the contribution of the subsidiary condi-
tions in the variational statement (2.9),
0.3)
E.=D.(u.). (3.4)
L L Ie
(3.5)
where
(3.8)
p e
o
= p0 e(E a 13) = p 0 e(£.,
1
~
a
), (3.9)
E(£.) =
1
JBe p e(£.,~ ) dV •
0 1 a 0
(3.10)
o
The virtual work condition again assumes the form (3.6), but the
generalized stresses 0i are now determined by
19
dE
a. = (3. II)
l. Cle:i'
0, a*
a8 8
n = t*a . (4. I)
(4.2)
(4.3)
If we let M~S = M:~, V~, M~, and H~i satisfy the equations
(4.4a)
(4.4b)
21
(4. 4c)
then the displacement we disappears from (4.3) and we have for each
element
3
oM* k* dA
as as 0
+ L J [ow*v* + o~*M* +w*ov* + ~*oM* ]ds
n n n n
i=1 s.
~
3
+ L (ow*H*
c~
+ w*oH*.).
c~
(4.5)
i=1
e e- 1 e w * ~ * e w ~
-~ui (Sij OJ + Dikwk + Dii~i) + 0i (Dikow~ + Diio~~). (4.6)
The matrices lS~jJ, [D7kJ , and ID!i J have been given in [5J 1)
The displacements w~ and the rotations ~~ at the boundary are defined
with respect to the basis triangle of the element under consideration.
Continuity of the structure requires that they are expressed in terms
of the displacement and rotation functions defined on the interfaces
of the finite elements. Apart from a rigid translation, which has no
consequences, this implies for the displacements an orthogonal
transformation. The elements of the complete transformation matrix,
as determined by the displacements of the vertices of the basic
triangle in the case of small membrane deformations, are given in the
appendix. However, for the rotations it is a different matter.
Infinitesimal rotations can be decomposed into orthogonal components.
The infinitesimal rotation o~* for a particular element is then
equal to the difference between the tangential component of the
E.
~
(4.7)
(4.8)
23
$*m (4.9)
o.~ (D~~, k -
(4.10)
D~oS •.
~"~J
(D':
J ,m
-D~ B
In nm
)~
m
+ D~ S.
~R, ~J
D~
Jm m
$ = f~
R, .
(4.11 )
by going from the flat plate approximation to the flat plate with
initial curvature, determined by the midside initial displacements
w*o
Finally it can be observed that ~n case of distributed loading
normal to the surface no improvement of accuracy may be expected
going from equivalent nodal forces corresponding to a linear deflec-
tion interpolation to equivalent nodal forces found by the quadratic
interpolation which is possible for the TRIM-6 element. This is due
to the fact that the bending properties of the element are derived
from a moment distribution, that satisfies the homogeneous equilib-
rium equations (4.4a). However, as it was mentioned in [6J, for
pressure loaded shells, that permit a membrane solution of the equi-
librium equations, we can enter the finite element into the structure
in a state of initial membrane deformation with the corresponding
membrane forces acting on the surrounding structure. Thus spurious
bending stresses are avoided.
25
References
Appendix
al x3 - x 2 bl Y2 - Y3 22 a2 + bL
1 1 1
a2 xI - x3 b2 Y3 - YI 22 a2 + b2
2 2 2
a3 x 2 - XI b3 YI - Y2 22 a2 + bL
3 3 3
and
I
bl
I
al
e lUI e hi e
WI
1::\'
d l = 2A b 2 ' d 2 = 2A a2 u U2 v W2
b3 a3 IU 3 = w W3
_ bTw e + veT(baT_ab T) we
_aTw e _ ueT(baT_ab T) we
I + bTu e + a Tv e + u eT (baT-ab T) v e
-w
~z
z
=[ o
-w w
x
y
The tangential unLt vector along the side pq of the basic triangle
is given by
27
x -x +u -u y -y +v -v W -w
q P q P e + qpqpe~e.
L -x \ -y R,i -z
~
S.N.ATLURI, H.MURAKAWA
Georgia Institute of Technology, Atlanta, USA
Abstract:
Introduction:
The topic of rate (incremental), multi-fleld, varlational
principles, in general, and the rate complementary energy
principles, in particular, and the corresponding finite ele-
ment methods, for finlte strain analysls of compressible non-
linear-elastic solids were discussed in detail by Atluri and
Murakawa [1]. Also dlscussed in [1] were the contributions of
Koiter, Zubov, and Fraeijs de Veubeke, dealing with the sub-
29
PRELIMINARIES:
We use a fixed rectangular cartesian coordinate system, and
employ the notation: (_) denotes a second-order tensor; (_)
denotes a fourth-order tensor; (-) 1mplies a vector; ~ = A.b
implies a. = A .. b.; A.b implies a product such that (A.B) ..
1 1)) - - - - 1)
= A .. B. k ; A:B = trace (AT.B) = A .. B .. ; and u.t u.t .. A parti-
1J J - - - - 1J 1J - - 1 1
cle in the undeformed body has a position vector ~ = Xa~a(a=1 .. 3)
where e are unit cartesian bases. The gradient operator Vo
-a -
in the undeformed configuration C is VO = (e a/ax ). The po-
o - -a a
sition vector of the particle in the deformed configuration,
N
1-1.. The gradient operator in CN is -v =
say CN' is Lv = y.e
(~ia/aYi). The deformation gradient tensor is ~ = (ZO~)T,
such that F.1a y.1, a = (ay./ax
1 a ). The nonsingular F has the
polar-decomposition, ~ = 0. (!+9) where the rotation a is or-
thogonal and the stretch h is symmetric and +ve definite. The
Green-Lagrange strain is g T T T
1/2(~ .~-!) 1/2(~+~ +~.~)
o T -
where e = (Z ~) , ~ = ~-~.
r (3)
RATE FORMULATIONS:
Now, we consider the (incremental) rate analysis of finite
strain problems of an inelastic solid with a rate-sensitive
constitutive law. In doing so, one can choose an arbitrary
reference frame. In practice, however, two choices, one the
so-called total-Lagrangean (TL) and the other, the so-called
31
t (5a,b)
o (6a,b)
32
AMB: (7a,b)
or, equivalently,
. N • N . N .T
~.:t: + ~.:t: +~ T • W (7c)
compatlbility:
(Ba,b)
TBC: (9)
Let us suppose, for the moment, that the consitutive law for a
rate-dependent material can be expressed (as shown later in
this paper) in terms of certain rate-potentials, as:
(lla-c)
We consider the Legendre (contact) transformations of the type:
TIHR(~'§) = J: VN
{-s*(:;;) + pN~·~+hN:[(2N~).(~N~)T]
TIC(~) f VN
-E* (t) dv + [
SUN
!.~ ds ( 15)
TIHR(~'~) = f VN
{-E*(t)dv p
N·
~.~
.
+ e: [(~N~) T] }dv + contd.
33
(18)
From -chese, one can establish 5* (§) , E* (!.) and R*(t) as de-
fined ~n Eqs. (12a-c). Thus, we focus attent~on on the poten-
tial V.
It is worth noting that for materials with rate-independent
constitutive laws, such as classical elastic-plastic materials
35
2 (~:S!')S!'
q* 2\1~ + A(~:!)! - l2a\1 (26)
(a' :a') [6\1+2(aF /aW P )]
- - 0
where, A and \1 are Lame' constants, a ' - S!-1/3(S!:~)~ is devi-
ator1c Kirchhoff stress, and the Y1eld-surface is represented
h
by F [3J 2 (a')]2 - F O.
- 0
where < ¢> denotes a specific function, such that < ¢> ¢ (F)
for F>O, and ¢=O for F~O. The parameter y is called the
fluidity parameter and ~a is the general anelast1c strain. It
has been shown by Zienkiewicz and Coworkers [15], and Argyr1s
and Coworkers [16] that classical, rate-1ndependent elasto-
plastic solutions can be obtained from the above theory, when
(i) either y~oo or (ii) a stationary solution of the visco-
plastic flow 1S sought. Various forms of ¢ were reviewed
by Perzyna [14]. For the Hencky-Mises-Huber Y1eld cr1ter1on,
one can define F to be:
.a (1 ).a .a t (38)
E = -6 ~N + B~N+l N~t~tN+~t
when Egs. (35) and (38) are used, Eg. (37) becomes,
~a*
~
= L :~E - ~tL : (Ea+Bv~a*)
~e ~ ~e ~N ~ ~
(39)
From which, upon rearranging terms,
~ a * = M:
~ ::; (~E -~ Ea ) ~ ~
( 40)
wherein the definition of ~ is apparent, and ~~a is known and
is given by: ~Ea ~tE~. From Eg. (40) one can immediately
write
and I themselves.
where E' is TL rate of Gree-strain, and ~', ~', t', and~' are
TL rates of e, s, t, and r respectively. Now a' is subject to
~ ~ NT ~
the constralnt that a .a' is skew-symmetric.
n =
(L-L )L ~ is used. The eigen-value solution for n at bi-
o 0
furcation-necking for the perfect bar, as obtained in [18]
for the presen~ linear-hardening but rigid-plast1c material,
. C
1S n 0.4B.
TOTAL DEGREE OF
FREEDOM 3/2
E =10
1/=03
O"y=40000
12
Mlx BI~TlON H = 50000
LOAD POINT
11
--- FUNDAMENTAL SOLUTION
10 - NECKING SOLUTION
O~~_~ __~__~~~
o 01 02 03 04 05 06
7]-(L-L ol/L o 02 04 067]=LL~o
F1gure 1 Figure 2
z
:;{
1/=0.3 TJ
0:
1-.6 Lo =12# 5931%
If) /;57.61%
Bo =4' 05
-l4 E =10
7
pSI
Ilr5536%
<1:. 5286%
z CTy=40000 pSI 04
~.2 h = 50000 PSI
z
O,~--~--~~-------- or 48~%J I
o <l
I
F~gure 3 Figure 4
The ~ecked profiles of the bar, and the progressive develop-
ment of unloaded regions (shaded) are shown in Fig. 5 at
various values of n. Note that unloading begins at the cen-
ter of the loaded face of the bar at n = nC = 0.482. Final-
ly the distribution of Cauchy stresses, '11 (in the direc-
tion of loading) '22 (in the width loading), and '33 (in
the thickness direction of this plane-strain specimen), at
the neck (y~ = L/2) are shown in Fig. 6. These results are
in excellent qualitative agreement with those of Needleman
[20] who also analyses the necking and post-necking problem
of an initially perfect cylindr~cal bar. It ~s noted that
A \ =59.31%
8: \=56.86%
A
~111
2.0 8
..!<
0 ..!< ;i2 ..!< ;i2 0
I.D
0
I.D 0 0-
eo ::: rI') ~ I.D ~
N
If)
""
If)
If)
If)
I.D
If)
I'-
If)
eo
If)
" "
t=:" "
t=:" "
t=:" "
t=:" "
t=:"
A
Figure 5 Figure 6
41
-Is GO
t.uds (46)
42
d via xa
...
(cS
la
+u~ )e.+N, x3=G
1,a-1-a a
(a=1,2i i=1,2,3) (47)
11.: h e e .
af)-a- 13' ha13 = ha ~x.)
1 [i=1, ••• 3i a, 13=1,2] (48b)
From Eqs. (48a,b) it is seen:
Since, for isotropy, !1, g, and E' are coaxial, Eg;. (3) becomes,
r = t.a (53)
The tensors t and a are assumed to be:
t = t .. e . e .; a =a .. e . e . [ i, J' =1,2,3; 1 (54)
lJ-l-J lJ-l-J
where, t .. t .. (x l ,x 2 ,x 3 ); a .. = a .. (x l ,x 2 ) (55)
lJ lJ lJ lJ
Finally, the external forces distributed on the plate are as-
sumed to be specified per unit area on the mid plane of the
plate to be g.1 = g.1 (x a ) (i=1, .. 3, a=1,2). When the assumptions
in Eqs. (50-55) are substituted in Eq. (4-6), we find through
atraight-forward algebra, that
Is
1T
-f Cu
[!'. (~*-~*) + ~.< (<;<-~) ·~3> ldc -1Co
[,!:.~*+~. (~-;r:) ·~3ldc
(56)
where ~ = M .e e
al-a-l
(a=1,2; i=1, ... 3), and,
M .
al
and (57)
ClW*/Cl!1* = ~.a -
Band ClW*/ClX = M.a - N
-- (58)
and T. + g. = 0 (a=1,2; i=l, ... 3) (59)
al,a 1
-i Co
8· (g-p ·~3dc. (60)
+ sinwcos8~3~2 + cosw~3~3·
In a Von-Karman type plate theory w is assumed to be moderate-
ly large, such that cosw ~ 1-(w 2 /2) and sinw = Wi while the
angle 8 can be assumed to be arbitrary.
......
---ro 'X 2
Efj_
X
1 5 .5
l DIRECTION OF
4 L COMPRESSION
14-.5=.,j -XI .
3 8/2 BOUNDARY CONDITIONS
PRESENT FEM .....,
2 ~
'-..
UI =U2"'W=0. xl=~a/2or x2=±a/2
-- LEVY(NACA REP. 737) X Wx -0 x 2 " ±a/2
ro
0 Px i/Et2 -l Wy=O XI = ~a/2
E =ld
...... 0 8 12 16 20 24
----ro
X
E =107
V-0316
a = 1.0
i ~ PRESENT
• LEVY V=0.316 20
t =0.01
06 a=1.0
16
0.4 t =001
12
P=0.022 • LEVY( NACA REP.737)
02 08
~ PRESENT FEM.
0 Pxa2/Et2 O~ Pa4/Et
0 2 3 4 5 0 80 160 240 320 400
Flgure 7 Flgure 8
46
Acknowledgements:
The research reported herein has been supported in parts by:
NASA under grant NASA-NAG3-38; by ONR under contract number
N00014-78-7636; and by AFOSR under contract number F49620-78-
C-0085. The authors gratefully acknowledge these supports.
The authors express thelr sincere thanks to Ms. Margarete
Eiteman for her untiring efforts in the preparation of this
typescript.
References:
[1) Atluri, S.N., and Murakawa, H., "On Hybrid Finlte Ele-
ment Models in Nonllnear Solid Mechanics", in Finite
Elements in Nonlinear Mechanics, Vol 1 (Ed. P.G. Bergan,
et all Tapir Press, Norway, 1977, pp 3-41.
[2) Murakawa, H., "Incremental Hybrid Finite Element Meth-
ods for Finite Deformation Problems (With Special Em-
phasis on the Complementary Energy Principle) Ph.D.
Thesis, Georgia Institute of Technology, Aug. 1978.
[3) Murakawa, H., and Atluri, S.N., "Finite Elasticity
Solutions Using Hybrid Finite Elements Based on a
Complementary Energy Principle" Journal of Applied
Mechanics Trans. ASME, Vol 45, 1978, pp 539-548.
[4) Murakawa, H., and Atl'Uri, S.N., "Finite Elasticity
Solutions Using Hybrid Finite Elements Based on a
Complementary Energy Principle - II. Incompressible
Materials", Journal of Applied Mechanics, Vol 46, 1979,
pp 71-78.
[5] Atluri, S.N., "On Rate Principles For Finite Strain
Analysls of Elastic and Inelastic Nonlinear SOllds",
ln Recent Research on Mechanlcal BehaVlor of SOllds
(Professor H. Miyamoto's 60th Anniversary Volume) Univ.
of Tokyo Press, Tokyo, Japan 1979, pp 79-107.
[6) Atluri, S.N., "On Some New General and Complementary
Energy ~heorems for the Rate Problems of Finite Strain,
Classical Elastoplasticity" Journal of Structural
Mechanics, Vol. 8, No.1, 1980, pp 61-92.
47
Summary
Introductlon
INCREMENTAL FORMULATION
Varlatl0nal principle for incremental finite element
methods are described in Ref. 1. Modifications of variation-
al principles for relaxing continuity conditions along
interelement boundaries have been used for the derivation
of many hybrid finite element models [2].
For th8 contact problem, consider two bodles A and
B which are divided into finite elements Vn where n refers
to the nth element. The boundary surface aV n of a typical
element may be composed of So n , Su n ' Sand
n Sc n ' which
are respectively, the surface with prescribed tractions,
the surface with prescribed displacements and the inter-
element surface and contact surface between A and B. The
entire contact surface is discretized into contact element
SCm" A modified variational principle which includes the
conditions of continuity and equilibrium on the contact
surface and for which the stresses satisfy the equilibrium
equation is
..
11
mc [[1
n V
flu
"..
dS +
n
stationary (1)
in which
£. = stresses
T = element boundary tractions
51
on interelement
boundary between
Va and Vb
T + 6 ..,
..- T = if + 6 if on Sa
n
J
(T + 6 T)A + A 0
(T + 6 T)B - A 0
-A A
and u + 6 -A
u + x -B
u + 6 ~B + x (2)
from which
R 6 e (5)
52
and
(7)
· 1 acenents ~A-B
Th e d ~sp ~ and Q over each contact
element Sc are then interpolated in terms of the nearby
m
nodal displacements qA and qB of the neighboring elements
of the contacting bodies.
The funct~onal 1Tmc ~s in the form of
1T
mc L[ ~ 8 ~ T ~n 8! -
n
where
(lO)
53
AT JA _
- ( 8q (11)
where
K = [ G T H -1 G
_n _n _n
n
8Q = '[ 8~n
n
R = Ln ~n Re Lm R
e
-m
~ LJ
_m
A JB
l J B
_m
(12)
m m
,... + .....
K 8 q 8 Q R
(13 )
o
"'"
-
t
54
Iterative Procedure
is satisfied.
EXAMPLE SOLUTIONS
R 15 mm
x IT
I
J-V V -
~
.- 1
I-
1 I
I
I
128 mm
- - - ~
I""" -
.....- ~
(~)
153.6 mm
Overall Mesh Pattern ----- (b) Mesh Near Contact
Surface
F~gure 1 Elast~c Half-D~sk and Half-Plane Contact
(E = 21,000 Kg/mm2, v = 0.3)
Non-Sl~ding Contact
Problems are solved for the case wlth both applled loads and
prescrlbed dlsplacements at the top of the disk. In the
problems, the ratio of Young's moduli are varied over a
57
-4
range from 1 to 10 and Sllghtly different mesh patterns near
the contac~ surface are used to accommodate for node-to-node
and node-to-internode contacts. Loads or displacements are
appl1ed by three increments until the length of contact sur-
face becomes about 2.4 mm. For each 1ncrement, the converged
solutions are reached with three or four iterations. For
these solutions, the best results for contact tractions are
obtained when calculated from equivalent nodal forces and
are compared excellently with the Hertz solution in all cases.
The half disk and the semi-infinite half-plane are also used
to demonstrate the capab1lity of this formulation to solve
extensive sliding contact problems. Since the contact be-
tween the two bodies is frictionless, the solution is inde-
pendent of the path, thus a Hertz solution 1S again available
for comparison. Solutions are obtained for prescribed dis-
placements at the top of the disk with both vertical and hori-
zontal components. Stress distributions on the plane of con-
tact, for two prescribed displacements, are plotted 1n Fig.2,
where zero position represents the point of initial contact.
Displacement
at D1Sk Top
Traction
u -0.4 mm Kg/mm <)-_ Present Solut1on
v -0.3 mm (on plane, ~ = 0.2)
Present Solution
(on disk, ~ = 0.2)
Present Solution
(on plane, ~ = 0)
Hertz Solut1on
(~ = 0)
u = -0.2 mm
v -0.15 mm
x (mm)
-2.0 -1. 0 o 1.0
F1gure 2 Normal Traction on Contact Surface
58
Conclusions
References
E.RAMM
UOlversltat Stuttgart, Germany
Abstract
For the prebucklmg range an extenslVe llterature of effectIve
solutlOn techmq ues eXIsts for the numerIcal solutlOn of structural
problems but only a few algorIthms have been proposed to trace
nonllnearresponsefrom the pre-llmlt mto the post-limIt range.
Among these are the sImple method of suppressmg eqUlllbrlUm
IteratlOns, the mtroductlOn of artIflClal sprIngs, the dIsplace-
ment control method and the "constant-arc-Iength method" of
Rlks /Wempner. It IS the purpose of thIs paper to reVIew these
methods and to dISCUSS the modIfIcatIons to a program that are
necessary for theIr ImplementatIon. Selected numerIcal exam-
ples show that a modlfled Rlks /Wempner method can be espe-
cIally recommended.
1. IntroductIon
load
1
1=1,2/3···
'U Ju displacement
FIgure 1: NotatIOn
In VIew of the fact that Iteration takes place m the displacement and
load s pace the load level may change from one Iterate to the other.
J' 1
In thIS caseanmtermediateposltlon J' forthe same load level A = A
IS mtroduced before the final state J IS reached (flgure 1).
(3 a)
(4)
det cK = 0 (5 )
to red to detect negatIve eigenvalues. ThIS IS the pomt when the lImIt
load IS passed and unloadmg should start.
67
load
B C ___ ---OC I
Increm. +
deration
load
IG
If
Ip
problems where the sprlllgs can keep the destabllizlllg structure alIve.
The method cannot be recommended for structures wIth local buckllllg
or when a tendency to bIfurcation IS present.
3. 3 DIsplacement - Control
The most often used method to avoId the slllgularIty at the crItIcal
POlllt IS the mterchange of dependent and mdependent varIables. Here
a smgle dIsplacement component selected as a controllIng parameter
IS prescrIbed and the correspondmg load level IS taken as unknown.
The procedure was llltroduced fIrst by ArgyrIs [9J but III the meantIme
has been modIfIed by several authors.
For SImplICIty let us assume that the stIffness expreSSIOn, eq. (3),
IS reordered so that the prescrIbed component f::, u~J) = U2 IS the last
one m the dIsplacement vector AU(J). Then equatIOn (3) may be de-
(6 )
(7)
+ IR _ IK A
(8)
1 12' U2
(9)
correspondmgtothe two parts of the right hand side of eq. (8). That
is, both solutiOns are obtamed simultaneously using two different
"load" vectors
'K " , - P,
AU W - (10 a)
'K 11 . AU(J)n-
, - 'R , - 'K12 · u"2 (lOb)
2' 2'
Q---------~ O------~~
IU, IU
2
i
The displacement mcrement ;:1"1 J ), eq. (9), is mtroduced mto the
second part of eq. (7). This allows the determination of the load
parameter t::. A(J):
(11 )
71
(12 a)
(12 b)
where the underhned term III eq. (10 b) lS not required to be formed.
Agalll both solutlOns are added:
(13 a)
(13 b)
(14)
t:, X. (1) 1S slmply a scalmg factor prov1dmg the constramt t:, u~l) = u2 .
Batoz and Dhatt [15J even drop th1S first cycle. They update the dis-
placement field only by 1tS component t:, u~l) and start to iterate.
For all further cycles u (J) does not change 1. e. t:, u (J) 1S zero and
2 2
X. (J) 1S
t:,
J =2,3 . (15)
(16)
(17 a)
or In matrIX notahon
(17 b)
J =2,3
'A
new to ngent
'normal plane'
tan gent
'u
FIgure 5: Constant - Arc - Length Method
74
(18 a)
(18 b)
Also here AU{J)I and AU{J)II are obtamed by equatIons (12) usmg
eIther the reference load vector P (D. A = 1) or the out-of-balance forces
IR as rIght hand SIdes. Then eq. (18) lS mserted mto the constramt
eq. (17) and solved for the unknown load mcrement D. A(J)
(19)
GeometrIcally thls lS the mtersectIOn J of the new tangent t{J) wlth the
"normal plane" (flgure 5). Eq. (19) lS equivalent to eq. (15) but con-
tams the mfluence of alldlsplacement components man mtegral sense.
The load mcrement D.A (1) m the denommator, which ObVIOusly has
another dlmensIOn, expresses the different scalIng of the load aXlS
Wlth respect to the dlsplacement space. It may be seen for the one
degree-of-freedom system mflgure 6 a that a low value D. A(I) tends to
a dlsplacement control and a large value to a load control of the ltera-
tion. In many degree-of-freedom systems the value D. A(I) m eq. (19)
does not play an lmPOrtant role and may be suppressed.
i~) Durmg the preparatIOn of thls study the author became aware
of the valuable paper by Crlsfleld [ 17J devoted to the same
subJect.
75
(20)
Instead of Iteratlllg III the " plane " normal to the tangent ~(1)
t It mIght
be useful to deflne a "sphere" wIth a center at m and a radius ds [ 17J
(see appendIx II). Alternatlvely the "normal plane" may be updated
llleverYlteratlOncycle (fIgure 6 b). That IS, III eq. (19) ~u(l) IS re-
(1)
placed by the total lllcrement U . It was found that except for very
large load steps the dIfferences resultlllg from these formulations are
mInor.
IU
oj bJ
FIgure 6: ModlflCatlOn of constant - arc - length method
The algonthms for the displacement control method and the modified
Rlks IWempner method dIffer only m the equation used for the evalua-
tion of D. A(!) The algorIthm IS summarized as follows:
2. In any step:
a) Solve the equihbrlUm equations for P and linearly scale
the load and dIsplacements to produce the length ds. ThIS
determmes D.A(I), ilU(l).
b) AdJust the step slze to the desired number of iteratlOns n.
l'
e. g. Jfl{rl
1 1
.
c) Check the trlangulanzed matrlx for unloadmg.
5. Numencal Examples
5. 1 Shallow Arch
The shallow cIrcular arch under umform pressure (flgure 7) has al-
ready been analysed III [8J applying the artlficlal sprlllg method
(c 11 = 28 lb/m), see also [7J. Ten 8 node Isoparametnc plane stress
elements were used for one half of the arch. The analysIs wIth a basIc
loadolp= 0.3 and usmg the constant-arc-length constraint shows the
tYPIcal step SIze reductIOn m the neIghborhood of the lImIt pOlllt.
ThIrty steps wIth 1 to 2 IteratIOns per step were needed. The analySIS
has been repeated for a basIc load step of p = 1. O. The step SIze has
been adJusted by the factor rn;;;-
I 1
wIth a desired number of IteratIOns
~ = 5. In addItion, the load lllcrement was reduced to 50 % whenever
1
It alternated and the absolute value decreased. Now only 9 steps are
78
14------------~----------------------
~ --03}0
•••• ~ - basIc load step
000 p=1.
12
a.
10.
06
R=100 In I h=2b= 2 In J
02
R
001~----~----~--~----~----~----~
0.00 0.02 004 0.06
w/R
FIgure 7: Shallow cIrcular arch
Sablr and Lock [20J who used a comblllahon of the dIsplacement and
load control techmques. In the present study one quarter of the shell
has been IdealIzed by four 16 node blcublC degenerated shell elements.
As the basIc load step, P = 0.4 kN was chosen. Agalll the load steps
were adJusted wIth ~ and the acceleratIOn scheme descrIbed for
1 1
the arch was applIed. The entlre load deflectIOn dIagram IS obtallled
III one solutIOn wIth 15 steps and 3 to 9 Iteratlons per step as llldlcated
III the fIgure. If the acceleratIOn techmque was not used the number
of IteratIOns Increased consIderably especIally at the mlmmum load.
•• Sablr an d Lock
[201
0.6
fI'----n.....
P I "-
[KNl 1 ' 4
I '\ ",W,
0.4
,,
'\ 5 •
\
"
, 5
0.;:
•
E=3103 KN/mm2,J..l=03
2L
7 I
-02- L I
I
\.
hi n ged
I
R=10 L= 2540 mm
h=635mm
\A
\~
-0.4
0 10 20 WC,W, [m m 1 30
FIgure 8: Shallow cyhndrIcal shell
the free edge. The structure has also been analysed using 36 bIlinear
4 node degenerated elements in combmation with an umform 1 x 1 re-
duced mtegratIOnscheme. ApproxImately the same results have been
obtamed but at about 20 % of the CP-tlme.
-p.. -P
Pcr
06
04
I~ b .~1
----'l .
0.2 T
a
..i
LLJLLL-4-LLl...L.LJ- ~
b
~b/2~ P=6h 2
b =4a =1680 mm I h= 6mm
E =210 kN/mm2 JJ=03
OO~----~----~------~----~----~----~
o 10 20 We [mml 30
n 2 ·E·h 3
Pcr = k b
12 (1-fJ 2 ) 0 2
(21 )
The basIc load step chosen was p= 0.25. In fIgure 9 the normalIzed
load IS plotted versus the center lateral dIsplacement. The plate falls
under combllled geometncal and matenal faIlure. The mltlal YIeld
POlllt at a deflectlOn of about 6 mm IS ImmedIately followed by the
lImIt POlllt at about 8. 3 mm. ThIrty steps with 1 or 2 lteratlons per
step were used. The elasto-plastlc analysIs was supplemented by a
purely elastlc solutlOn also shown III the flgure. Here the typlcallll-
creaslllg postbucklIng response of plates IS recogmzed.
The buckllllg analysIs of the closed cylmdncal shell under wmd load
(flgure 10) studIed III [21] has been extended to the postbucklIng range.
~
p
10
TL
1350 18
1t
1 -10
= 0918 E(1/ is =
p
(22)
.h.- f1[ - 0 657
RVh
One quarter of the sheills Idealized by 2 x 18 bicubiC 16 node elements.
Two elements of unequal length are used in the aXIal dIrectIOn, whIle
the 18 elements in the circumferential direction are concentrated near
the stagnation zone. The first load lllcrement defllled the basic step
SIze as p = 0.25. Both the perfect and an imperfect shell have been
analysed. FIgure 11 shows the dIsplacement pattern of one quarter of
the shell near the lImIt pomt. A faIlure mode wIth one half a wave in
the aXial direction and a few bucklIng waves III the circumferentIal
direction, located III the compression zone, is indicated. The post-
bucklIng mimmum of the load-deflection diagram (figure 12) IS about
60 % of the limIt pOlllt. The ImperfectIOn assumed for the second
analysIs corresponds to the failure mode of the perfect structure. The
maXImum Imperfection amplitude is 2. 5 times the wall thickness. The
83
load deflectlOn path (flgure 12) llldlcates a reductlOn of the lImIt load
to 68 %of that for the perfect shell. The postbuckllllg mlllima nearly
cOlllclde. It should be noted that the example IS numerically very
sensitIve because of the extreme slenderness ratIo and the local nature
of the failure mechamsm. In both cases over 60 steps were necessary.
15.-------------------------------------~
P
~I
perfect shell
10
"..--0--__ _.....,.,
/ " ~~
---0------ ~~~--~o_~
~ I mperfect shell
/
/
05 /
I
I
I
I
I
oo+-----------~----------~--------~----~
o 5 10 w/h 15
FIgure 12: Load - deflectIon - dIagram of a wllld loaded shell
6. Conclus lOns
ThIS study on IteratIve techmques for passlllg lImIt pOlllts allows the
followlllg concluslOns:
cessful.
84
Acknowledgement
The author would like to thank Professor D. W. Murray, Umversity
of Edmonton, currently at the University of Stuttgart, for valuable
discussIOns.
85
References
+ (A 1)
(A 2)
1=1,2,3 n (A 3)
'f = C . 'U (A 4)
(A 5)
ell T
C =-2-'P'P (A 6)
P,
The Iterahon equatIon, eq. (3 a), IS modifIed to
The nght hand sIde expresses the out-of-balance forces. After itera-
tlOn (J ..., m + 1) the real loads are determined by eq. (A 2):
(A 8)
(A 9)
The "sphere" wIth the center at m and the radius ds of the Imtlal
tangent vector t(1) (fIgure 13) IS defmed by
r(ll - ds 2 = 0 (A 10)
~
/ I
/ I
T~)/
'sphere'
-- /'
/
1.1/-.= 1
I
I
I / I
J /
O--;........oI:J/----- ____ J
IU
FIgure 13: IteratIon on a "sphere"
89
(A 11)
(A 12)
or m matrix notatlon
(A 13)
+ c = 0 (A 14)
a = 1 +
(A 15)
Parameter Sensitivity of Nonlinear Structures Concerning
Stability Limit
F. G. RAMMERSTORFER, D. F FISCHER
Voest-Alpme AG, Linz, Austria
SurrmaJ:¥
'Ihe stability of nonlinear structures 1S considered, espec1ally hav1ng
in V1ew the pararreter sensitivity of the stability liroit load. I t is
shavn that under special cirCUIllStances, for which cri terions are de-
r1ved, nonlinear structures behave with a sudden, i.e. non-continuous,
<flange of the limit load if a certain system pararreter is varied. Such
phenanena belong to the f1eld of the catastrophe theory which allows
their classification and useful docurrentation.
1. Introduchon
The paper shONS hew structures can undergo effects like a non-continuous
parameter Sens1 tivi ty. This rreans that an infinitesimal variation of a
particular pararreter of the physical system can cause a finite change of
the stability limit. I t is shavn hew those problem:; are anbedded in the
catastrq;he theory.
(1 )
with the reference load vector, !3-re f' and the current load nultiplier,
n A• The linearized 1ncrerrental eqw.libnum equation for the increrrent
n-n+1 follONwg the tangent stiffness concept (e.g. [1]) 1S
m1:
§ 61,;! = Q (4)
has a nontr1v1al solution at the stab1lity lJ.nu.t. Equ. (4) leads to the
condition
det:
m'"
K~
= det K( . m'" A) =0 (5)
for the fundanental state rrr': in wluch the eqw.librium becares lnBtable.
Therefore, the behaviour of the detenrunant of the load dependent tan-
gent shffness matr1x, det ~(A)' can be used as an wmcator for the
stab1hty of the equihbnum. Denotlng the cntical load level by
m'': .;,
A = A , the follONwg relahons hold [4,5]:
a
llIr! aI det ~(A) = - ro, (7)
A-X' -
92
All relations shawn for det K(A) are also valid for the normallzed de-
~
terrninant
As shawn in detail in [4,6,7] the relations (3) and (4) can lead to an
approximative (linearized) elgenvalue problem:
Solving the elgenvalue problem (9) at a certain load level rnA < A;'
renders an approximation of the crltical load level:
( 10)
Wlth rn~. the smallest eigenvalue, and 6U. the corresponding eigenvector.
J -J
In equ. (9) a linear relatlonshlp between the load dependent parts of
the current stiffness matnx and the stiffness matrix at the stabihty
llffil t is assumed. ThlS approximatlve assumption becares rrore and rrore
accurate when the considered funda!rental equilibrium state rn respresen-
ted by rnA, approaches to the stability limlt conflguratlon, represented
by A"i In this case
. rn- rn
llID n·= n= (11 )
rnA_A'" J
with rnn the exact "nonlinear" elgenvalue, for winch
rn rn
A;' = n A ( 12)
is valid.
sical =1 tical load level of the linear structure, and lntersects the
load-dlsplacarent path, A (uP), at the stablhty limit point of the non-
linear structure (see flg. 2 of the follcwlng chapter) .
The curve wIuch lS obtalned by plotting rn~", versus rnA lntersects the line
f (A) = A at rn~",= A': The behavlOur of the ~'" (A) curve allCMS a distlc-
tion between buckling and snap-through, as shcwn in [6]:
Hcwever S (A) cannot reveal the stablhty hrnit MUch is caused by bl-
P
furcatlon, l.e. buckling. Nevertheless the current stiffness pararreter
is ln any case a very useful and easily calculated pararreter which al-
lcws again in, canbination Wlth another stability indicator, e.g. the
determinant (equ. (6)), an investigatlon of the stability behaviour and
the rrechanism of stability loss as shcwn in [11 J •
94
3. A partJ.cular problem
6
h
~*
/'
5 [= 1. I \
r=1. / \
/ \
4 I \
/
V ~'IUP
- \
\
/ \
3
I~. I
hlu~) "Iu~)
dering the eigenvalue functions A* (u) and the det K(A) CIllVe one gets
n~
(the det K(A) CIllVe approaches the A axis and the A'" curves tend
n~
to
taIch the A(u) paths) but a further load increrrentation "stiffens" the
structure until A '" 2.5. Continuation of the load incrementation leads
finally to the snap-through at A "'_3: '!he detn~ curve reaches a limit
with horizontal tangent. When the A1, (u) curves cross the load chsplace-
ment paths, the detenninant of the stiffness matrix behaves correspon-
chnglyto equs. (6) anG (7). Spec~al attention will noN be given to the
detnK(A)
.. mstory. I t is cbvioos that after a certain pararreter variation
the d.et K (A) curve may cross the A ans just before the "re-stiff~g
n ..
pericd". ~s would render a sudden change of the critical load level A1,.
In other words, an infinitesiroal vanation of this certain system para-
meter would cause a finite change of the stability limit load, i.e. a
non-oontirruous pararreter sens~ tivity ooncerning stability limit.
'!he stab~hty behavl.our of these three typl.Cal cases can also be studied
by oons~deration of fig. 4 wluch represents the A1, (A) chagrams (see equs.
(13) and (14».
97
r=1.
Ii
II
1\
\\
\ \ 4
\ \
\
-), (UK)
\ \ '
\
CorresponCill1g to equ. (14) one can pred1.ct that in all the =nsidered
cases the stabllity loss lS due to snap-through although the detnISP' )
curve for the case C = 0.5 (see flg . 3) does not allow thlS pred1.ctlon
(refer to EqU. (7». '!his uncertalnty In the use of the limit =ndltion
(7) (which lS only a sufficient but not a necessary =nditlon) is al-
ready pointed cut 111 [6].
98
OL----L____J -_ _ ~
o 2 3
-
Flg. 4 Elgenvalue functions )., ,', ()., l for three different C parameter
values
3
~*
r= 1.
2
I
I
I
I
I
1(*
0
.0 .5 1. 1.5 2.
C
Fig. 5 Stability limit load as a function of the parameter C (dotted
line is the expected physically not realizable part of the curv~
Flg. 5 shaYS clearly the conslderable Jump of )., ,', (el at C = e'" '" 0.55.
ThlS jump behaviour has of course serious consequences for the design
99
'!he rather young mathematical theory of catastrophes has its origin III
aV = 0 (18)
aX
represents the behaviour surface, e.g. the equ1librium states of the
elastic structure. '!he parameter danain is in this case the plane a - b
and is named the control surface. The system is structurally stable i f a
small var1atlon of the systen parameters does not change the qualltative
100
a2 v I (19)
ax2 (a,b)"'= 0
v = .1.3 x 3 + ax (20)
or
av
ax
= x3 + ax + b =0 (22)
(23)
A simple exaIT\Ple taken fran [17], the v. Mises truss, may dem:>nstrate
the statanents given above. Flg. 6 shows the behaviour surface of this
structure, provided the Euler-bucklmg of the smgle truss elanent dces
not appear before the snap-through takes place, otherwise the system
would be structurally unstable.
101
BEHAVIOUR SURFACE
Fig. 6 '!be behav~oor surface and the bifurcation set of the v. fuses
truss (taken fran [17 ] w~ th penni.ssion)
(24)
x = ex
a = - ex 2 (25)
o
b = plEA
_UK
z
r=1.
BEHAVIOUR SURFACE
(equilibrium states)
it
C
Fig. 7 The behaviOJr surface of the arch with regard to the equil1.brium
states
'!he potential V describing this behaviour is, naturally, the total po-
tent1.al energy . Consider1.ng the behav1.our surface in fig. 7 one ooserves
that it is cut into two different surfaces at a line along C = C". HenCE,
'!han's theor€lll concerning the exJ.stence of a smooth behaviOJr surface
(see chapter 4.2) is only valid f or two separated parameter sub-danains
which are connected along the l1.ne C = C'·' . Thus, the structure has a
structural instabil1.ty at C = C". This fact corresponds to the ooser-
vatl.On that for C < C', the support K snaps through upovards and for C > C',
dGmYards.
I t should be mentioned that the algori tiun used allONS the calculation of
the syst€lll behaviOJr only for load incr€lllentation up to the limit load.
'lherefore the continuation of the behaviOJr surface for >. > >.'" (C), Le.
the post-buckling behaviOJr, has only qualitatively predicted character.
Concerning the fold curve )/' (C) ill the control surface wh1.ch 1.S relevant
for the stability loss due to increased load, one should nonce the cor-
respondence with fig. 5.
103
T'
F1g. 8 'Ihe catastrophic behaV10ur of the cn tical load level A'" (C, r)
In fig. 8 the behav100r surface is shaND for r:;;; 1. The cross sechon
through the behaviour surface A'" (C, r) at r=1 =rresponds W1th the A'" (C)
mrve m f1g. 5. Furthermore, sane pomts on the behav10ur surface for
C = 1.0, l.e. A'" (r ,C=1.) are calculated. Fran these infonnations the
behav10ur surface was drawn qualltatively. There is one ilrqJortant fact
to be noted: Although the behav10ur surface is a srrooth surface, the
folded part has only mathemahcal but no physical rnean1ng.
Always the la.vest critical value A'" 1S the stability limit. Hence, the
relevant part of the behav10ur surface is agaln slotted and only one
branch of the bifurcahon set in the control surface, namely the 0" (r)
curve, contalns the phys1cally relevant JUITp condl hons.
(26)
= o. (27)
i~
C=C (r)
'!he potential function (28) renders with equ. (26) the necessary condi-
tion (5) for the stability limit. Hence, V in equ. (28) yields the beha-
viour surface A = }/( fig. 8). The condibon for the jurrp of A''', equ. (27), leads
in canbination with equ. (28) to that situation in which the det K(;\.)
n~
curve tcuches the ;\. axis without crossing it. This is exactly the case
1f C approaches C;' (see figs. 9 - 11).
10
av
ar= 8
detn~ 6
58
Z
2 3
~
Fig. 9 '!he nonnallzed determinant of the tangent stiffness matrix as
a function of the load level (thin dotted lines are expected
curves)
105
II
o. t 2. 3.
.0
-.2
- .4
-.6
V
-. 8
-1 .
/
/
-
hg. 10 The potentl.al function V (tlun dotted lines are expected
auves)
.6
.4
tv
a,,2=
.2
a
a" detn~ .o I-----.---r--h-Hllt--.--Hr-T-.--,
-.2
- .4
-.6
- .8
-1.
Fig. 10 shCMS that the potenhal V behaves like a polynaninal with or-
der four which is in correspondence with equ. (21). Referring to '!han's
theoran one can conclude that the assumption of a cusp ca.tastrq:he as
shCMl schanatica.lly in fig. 8 is justified. Furthenrore fig. 10 ShONS
clearly for decreasing e that for values e > C' indeed a po~nt of ~nflechon
in the VeAl anveexistsfora>. value srnallerthanX< butnothll e""C<a
second rninlIllUIll of V appears prior the origmal one. '!he appearance of a
second nu.nimJm of V renders the jump of "A* (e).
3QE~: with
(32)
Qi' = Q, (33)
Le. a sudden jump of the crihcal load level "Ai, (Q) •
107
'!he prcx:>f of these theorems is cbvious by considering equs. (6), (8) and
(26 - 28). Equ. (31) rooans that only minima at load levels smaller or
equal to the stability limit load have to be considered because of the
fact that usually a practical physical structure behaves in the unloaded
state, i.e. A = 0, with
detn K(A=O)
::: > ° (34)
5. COnclusions
furcahon set in the control surface (see e.g. hg. 8). 'Ihis procedure
w~ll help to hnd out critical caromations of system parameters at which a
sudden j~ of the stab~li ty limit load takes place. The kno.vledge of
such a j~ is of great importance for the design of supporting struc-
tures especially in the field of licjhtweigt construction.
6. References
F. BREZZI
Universitil di PaVIa, Italy
Summary
1. Introduction
r
Let !(~,A)=(fl(xl, .•. ,Xn,A), ••. ,fn(xl, ... ,Xn,A» be a C -maE
!(~O,AO)=O. (2.2)
(2.5)
or equivalently
a2f~
D fO:
xx-
(:i,,_z) + L ax] aX~ z
r
y
j
(2.8)
r
(2.10)
(2.11)
has to be nonsingular.
{
X' = (x -.... , ... ,xn-l ,0)
~=~'+a1° (2.16)
a = xn
f
n (x'
- (a,A)+a,f,O
.:t.
,A)=O (2.19)
(2.21)
det (2.22)
r-2
the solutions of (2.19) lie on two C branches crossing
transversally at (X~,AO). If on the opposite the determinant
in (2.22) ~s positive, then (X~'AO) is an isolated solut~on
of (2.19). The case det=O corresponds to a higher degeneracy
and w~ll not be studied here.
We set
~
DX!O£.O = -DA!o
(2.23)
(£. o ,1.* ) = 0,
(2.25)
(2.26)
114
(2.27)
F(u,A)=u+TG(u,A); (3.2)
- ~[w,w] in n,
(3.6)
l¢+A¢,w] in n,
w = aw = ¢ = II = 0 on an . (3.7)
an an
(3.8)
G: ((w'¢)'A)+(~LW'WJ,-[¢+A~,wJ) (3.9)
115
T
(3.10)
~~~i = P1 ' ~1 H~(n) i=1,2
and applY1ng T to both sldes of (3.6) we are 1n the frame-
work (3.2) - (3.5) .
\DUFO¢O = 0 II ¢ °Il V = 1
(3.11 )
1(D FO)*¢* = 0 (¢0,¢*)=1
u
(3.12)
(3.14)
and
(3.15)
116
with
jAo=(DUUFO (cpo ,cpo) ,cp*)
lBO=(OU:\FOcpO+OUU FO (cpo ,0°) ,cp*) (3.16)
(3.17)
(4.3)
PEV' (4.5)
(4.8)
(4.9)
For the proof and for more general results we refer to [4J.
Theorem 3 states that the b1furcat1on pattern 1S not (ln
general) stable w1th respect to the f1n1te element discret1
zation. Except10nal cases are for 1nstance the bifurcation
from the tr1v1al branch and the syrnrnetry-break1ng bifurca-
t1on. See [4] for more deta1ls. PreV10US results 1n th1S d1
rect10n can be found for 1nstance 1n [9), [6J .
From the computat1onal p01nt of V1ew two classes of problems
ar1se when we follow a branch: one 1S due to the fact that,
1f we proceed too fast along the branch, we may jump over
the b1furcat1on p01nt without noticing 1t, and keep follow1ng
an unstable branch. Th1S usually results 1n a change in the
slgn of det(DuF h ); 1f the determ1nant 1S computed at each
step we are able to go back with a smaller size step and try
to locate the b1furcat1on point. Th1S 1S easy but not cheap
1n general. Another problem arises when we corne too close to
the b1furcat1on p01nt: then DuFh becomes (numer1cally) sing~
lar and the "d1rect1on of the branch" does not exist anymore;
we say that we are 1n a "black box". In order to go out of
the black box 1n the d1rect1on of the branches we may use
the follow1ng tr1ck: assume that we are at a p01nt (uh,X)
1n the balck box and let v h be the e1genvalue of DuFh Wh1Ch
1S smallest 1n modul~s. Let ¢- and -*
¢ be the corresponding
e1genvectors:
II ~ I =1 I (4.10)
(4.11)
(4.12)
where
~ ~ ~* ~
(4.13)
-(A-A) (DAF h ,¢ )¢.
References
Abstract
This paper represents a progress report of some of the
research that we are conducting in the finite element ana-
lysis of thin shell structures. We consider our isopara-
metric displacement-rotation thin shell element and our
discrete-Kirchhoff-theory (DKT) plate/shell element, which
we are continuously refinlng for accurate and effective
geometric and materially nonlinear analysis. In the paper we
briefly discuss the locking phenomenon of the isopara-
metric element, the use of this element as a transition
element between shell surfaces and in shell-solid transi-
tions, and we give some results using the DKT element.
1. INTRODUCTION
Much research has been conducted during the last two decades
in the development of thin shell finite elements. However,
despite the large amount of research effort there do not
exist as yet what may be considered to be cost-effective,
reliable and general thin shell analysis capabilities.
We believe that an effective thin shell element should
satisfy the following criteria:
1) The element should yield accurate solutions when model-
ing any shell geometry and under all boundary and
loading conditions. In particular, the element should
not contain any spurious zero energy modes, so that
reliable results can always be expected: The theory
of the element formulation must be well-understood and
should not contain any "numerical fudge factors." These
conslderations are most important and many elements
that have been published do not satisfy thlS
criterion. Such element developments can represent
123
/------.y
much attention has been given to the use of the low-order ele-
ments. However, if the stiffness matrices of these elements
are fully integrated, the elements display a "locking phenomenon"
and therefore some reduced or selective integration techniques
have been proposed [1,2,9,10]. These techniques, which can also
be related to mixed formulations [11, 12, 5) , relieve the
"locking behavior" of the elements but because of the various
dlfficulties encountered have not resulted as yet into a shell
element that satisfies our three criteria of an effective prac-
tical shell solution capability. Indeed, based on the experience
available so far it appears that for the development of low-order
shell elements the approach used in the discrete Kirchhoff theory
is more effective [5].
In the following sections we consider two important ingredients
of the isoparametric shell element: the convergence of the
higher-order elements (when fully integrated) to a finite ele-
ment discretization of the Kirchhoff plate theory and the use
of the transition element.
Although the results of a number of investigations have been
published in which the locking phenomenon in conjunction with
reduced integration procedures was considered, there is still
a need for further insight into the problem of whether a mesh will
lock or not. The objective in our work was to address this
question first for the simpler case when using fully integrated
elements. We are currently extending the results obtained to
elements that are not fully integrated.
]]
h
3
""2 [1 AI5..T f.b 15.. dA + a £ yT ~Y J dA
(1)
where
dS X
ax
dW
dy
- Sy
dS
_ ---X
15.. y (2)
dY
dW
dX + Sx
dS X dS y
ay dX
['
1 v 0
~
E
12 (1-v 2 )
v 1
I-v
0 C
-s
Ek
2(1+v) 0 :}31
0 0 -2-
and
W transverse displacement of plate
section rotat~ons about x and y-axes,
respectively
h th~ckness of plate (assumed constant)
k shear correction factor
E,v modulus and Poisson rat~o
L characteristic length
(4)
(5)
o (6)
o ( 7)
where
q total number of nodal points,
129
NO OF
NO OF DO F POLYNOMIAL ADINA
ELEMENT TYPE
(2q- k) TERMS RESULTS
(np)
0 5 8 LOCKS
D 7 9 LOCKS
0 9 12 LOCKS
(a)
D o 0
17 16
NO OF
ELEMENT MESH NO OF 0 OF POLYNOMIAL ADINA
CASE TERMS
CONSIDERED (2q-kl RESULTS
(np)
SIMPLY
·.
o • 78 64 DOES NOT LOCK
o 0 SUPPORTED
··..
o 0
o 0
o 0
SIMPLY
00 00 00 171 144 DOES NOT LOCK
00 o 0 00 SUPPORTED
00 o 0 00
00 o 0 00
00 o 0 00
00 o 0 00
NO OF 00 F NO OF
ELEMENT MESH CONSIDERED POLYNOMIAL ADINA RESUIJ"S
(2q-k) TERMS
( np )
Em
•• LOCKS
97 112 ('2 - NODE ELEMENTS)
ARE LOCKING
••
•• •• • •
•• ••
••
••
•••• •••• 161 144
DOES NOT
LOCK
•• •• • •
•• •• ••
NO OF
ELEMENT MESH NO OF DOF POLYNOMIAL ADINA
CASE (2q- k) TERMS
CONSIDERED RESULTS
(np)
EB
SIMPLY
36 36 DOES NOT LOCK
SUPPORTED
• •
CLAMPED 31 36 DOES NOT LOCK
SIMPLY
78 81 DOES NOT LOCK
SUPPORTED
• • •
• • •
• • • CLAMPED 71 81 DOES NOT LOCK
thickness = .124
E = 30,000, 000
8 = 30°
Flgure 7 (Continued)
135
600
Top Integration POints
(I e Olodlstonce of 0358
-
above the mid - surface)
500
4,4,2 I 3,3,2
l>-------<>
400 ------ 1
..--..
D- -(]
0---<)
b~ +---+
U)
U)
w 300
'"
I-
U)
200
100
025 20 225 25
OISTANCE ALONG FROM THE X-AXI S
- 200
b~
U)
U)
w - 300
'"
I-
U)
Bottom Integration POints
(I e at a distance of 0358
- 400
below the mid - surface)
4x4x2 3x3x2
Constraint Equation ~ l!r----ll.
Average Normal _________ o--~
- 500 TranSition Element _____ 0---0
2-D Fine Mesh +---+
- y
600
Top Int'li!ra'ion POints
(I e 03~8 above the mid-surface)
0--0 Shell Elem.nt Mesh
500 +-+ 2 - 0 Fine Mesh
400
~
w 300
'"
~
'"
200
100
- 100
- 200
b
'"'"<oJ - 300
'"
~
'"
- 400
Bolfom IntegratIOn POlnls
(I e 0358 below the mld- surface)
o---oShel1 Element Mesh
- 500 +-+ 2-D Fine Mesh
x x
Z Z Z
8z 3
y y y
w3
-
v3 3
8Y3 v3
x x 8Y3
u3
!SM Ke K
~KM' I
K
=~~~~=a=
I I ]
-
I I Z
25
- - - - - - - S-S, CONCENTRATED LOAD
-----s-s, DISTRIBUTED LOAD
------CLAMPED, CONCENTRATED LOAD
20 .t' " "L o - - - - CLAMPED, DISTRIBUTED LOAD
• x
u 20
!i
z
o
~ 15
I&J
....J
"-
I&J
o
Z 10
It:
o
It:
It:
I&J
~ 5
ACZIZr A~B Alilf 0....,
............
'o- _ _ _ _ _ ~~_-
c~~121 C~D cello o o------1p===r-~_ 8
6 'r" 4 6
N=2 N=4 N=8 2 N
Figure 14 Element Meshes for the Analysis Figure 15 Percentage Error in the Predicted Centre
of a Square Plate using DKT Deflection of the Square Plate Using
Elements Element Meshes of Fig. 14
U -displacement s and
rotations about Y, Z
axes zero along this
RIGID V-displacements and
DIAPHRAGM r.otatlons about X,Z
P axes zero along this
SUPPORT edge
(V=W=y=O)
B
Y,V
X aXIs zero along
Z ,W T this arc
RII = 100
LlR = 2
RIGID ... = 03
DIAPHRAGM x, Y axes zero along
SUPPORT this edge
(V=W=y =0) Y,V A
D
0 6-0_,
-"'AO_i
- 0........
-50
0 .....
........
0
-200
D C
0 "·o-~-¢-6 -1I-6-¢_ I
itl_~~
"'-'1
o 'lI.~
-..¢
-5 o ~
RNa
p
. .....~
-10 --ANALYTICAL SOLUTION
..
0 4x4 MESH
6x6 MESH
~...
-15
¢
0
axa MESH
10xi0 MESH ~ 6
6 16Xl6 MESH
-20
D C
o
-5
-20
-0.1
I
C
Figure 17 (Continued)
144
50
,,~-.~'\.
Et Wn
-p-
a
B
.-~.~./, 1 1\1 i
C
I
-50 \
\
CI
\
0 4x4 MESH "
• 6X6 MESH
exe MESH
-150 0
o loxia MESH
" 16 Xl6 MESH
-200
B
a
RNa
p
-5
B C
0 ~..."..~-....~~~~c- I
)'.
~
RNji
P
-5
~
0 4K4 MESH
• 6K6 MESH
0 8 K8 MESH
-15
o 10KI0 MESH ...
... 16KI6 MESH
\
-20
...
0.2
--ANALYTICAL SOLUTION
0 4x 4 MESH
• 6x6 MESH
0 8x8 MESH
0
~ o 10XI0 MESH
P ... 16XI6 MESH 0
0.1 ...
Figure 18 (Continued)
146
-ANALYTICAL SOLUTION
o 4114 MESH
-4 " 6116 MESH
o 8118 MESH
o 101110 MESH
• 16 11 16 MESH
-2
Etu
p
O' A
o
0.003
--ANALYTICAL SOLUTION
o 4x4 MESH
0.002 " 6 X6 MESH
o 8x8 MESH
Ma13 o 10xi0 MESH
p
• 16 x l6 MESH
0.001
-0.001
-0.002
4. CONCLUDING REMARKS
Acknowledgment
500
400
I< 300
200
150
100
E =YOUNG'S MODULUS =107 PS I
~ = POISSON RATIO =l{O:i 300~~__~______L-.
h = THICKNESS =0.12 IN 4 12 22 TIME
a = WIDTH = 24 IN
q = UN I FORM APPL lED PRESSURE APPLIED PRESSURE LOAD
ALL EDGES ARE SIMPLY-SUPPORTED
J::.
.....
u PR ESE NT STUDY
~
LEVY
2
~
• ISOPARAMETRIC [3]
SHELL ELEMENT
II::
Z
2
.....
I.)
'"
..J
IL
IIJ
o
II::
'".....2
IU
I.)
References
[161 Bathe, K.J., and Ho, L.W. "A Simple and Effective
Element for Analysis of General Shell Structures, "
paper to be presented at the ADINA Conference, June
1981, M.LT.
4x4 1.00
6x6 1.86
8x8 3.19
lOxIa 5.73
16x16 21.94
Nonlinear Dynamic Finite Element Analysis of Shells
Summary
Dynamlcal aspects of a general nonllnear flnlte element shell
analysls procedure are descrlbed. The work extends preVlOUS
endeavors of the authors on quaslstatlc plate and shell anal-
YS1S. Several sample problems of a two- and three-dlmenslon-
al nature are presented WhlCh demonstrate the appllcablllty
of the methodology.
1. Introductlon
In thlS paper we are prlmarlly concerned wlth the nonllnear
dynamic flnlte element analysls of shells. The work de-
scrlbed hereln lS a sequel to a general nonllnear statlc
formulatlon presented In [20] and [21] and referred to In
thlS paper as "Part I" and "Part II", respectlvely. Due to
the extenslveness of Parts I and II, and In an effort to
keep the present ~ork as brlef as posslble, the detalls of
the statlc aspects of the theory wlll not be repeated here.
However, to enable the reader to better appreclate the "blg
plcture" we wlll brlefly mentlon the sal lent features of
Parts I and II.
Consistent mass
+1
r::ab ~~ oJ -1J NT
-a Nbp
- 0j 0
d~ dO Sb
_
(2.1)
J J
... d~ dll (three d1mensions)
J ... do
-1 -1
(2.2)
+1
o
J ... dll (two d1mensions)
-1
N 0 0 N Z 0 0
a a a
(three
0 N 0 0 N Z 0
a a a dimensions)
0 0 N 0 0 N Z
a a a
N (2.3)
~a
N 0 N Z 0
a a a (two
dimensions)
0 N 0 N Z
a a a
Lumped mass
1. Calculate
+1
ds
Jo
J Jo
-1
(2.4)
rot
m
a J Na2
0
PoJ o do (2.5)
rot
m a = 1,2, ... ,n en (2.6)
a
For heterosis.
(2.7)
o
a = 1,2, ... ,8
155
disp = 0
mg (2.8)
4. Normalizat1on
n
en
rot
Mrot = I m
a (2.10)
a=l
n
en
Mdisp
l mdisp
a
(2.11)
a=l
mrot
a
.;- (M/M rot ) mrot (2.12 )
a
disp
m
a
.;- (M/Mdisp ) mdisp (2.13)
a
+
[za 1 z - z (2.15)
a a
+1
z2 dl;; I <z >2 + 1 [z 12
a
a J
-1
a
2
a 12 a
(2.16)
n
en
h I [za 1 I nen A V/h (2.17 )
a=l
rot rot
m m a (no sum) (2.19 )
a a a
Remarks
3. Transient Algorithm
The transient algorithm is basically the implicit-explicit
predictor/multi-corrector scheme which has been described
previously (Hughes et al. [12,14,18,19,22,23]). However, a
slight reorganization is required to appropriately treat
rotational degrees-of-freedom. Let i be the iteration
counter; n be the time step number; ~t be the time step;
d ,v , and a be the (generalized) displacement, velocity,
~n ~n ~n
implicit
M* (3.3)
explicit
a {oj = 0 (3 .6)
~n+l ~
3. Equation solution:
Remark
4. Numerical Examples
The calculations described herein were performed on the
California Institute of Technology IBM 3032 computer in
double precision (64 bits/floating point word). In three-
dimensional situations 4-node quadrilaterals were employed
and in two-dimensional cases 2-node elements were used.
These elements are described in detail in Parts I and II,
respectively. In cases in which the material was completely
elastic, we used 2-point Gaussian quadrature through the
thickness ("fiber integration"). This has always been suf-
ficient in our experience. However, in elastic-plastic
situations we found it necessary to use 4-point Gaussian
quadrature through the thickness. In the calculations em-
ploying 4-node quadrilaterals, we used both selective (Sl)
and uniform (U1) reduced Gaussian rules for the lamina in-
tegration (see Part I for further details). For the two-
node element we used only the 1-point Gaussian rule in the
lamina direction (U1). The plasticity theory employed was
that due to Krieg and Key [27] which allows any arbitrary
linear combination of isotropic and kinematic hardening. A
plane stress radial return implementation was employed [26].
In all cases the corotationa1 hypothesis [1-3,5,7-9,33] was
adopted in the context of the constitutive algorithm pre-
sented in Part I. Newmark parameters were taken to be
S = 1/4 and y = 1/2 throughout.
4.1 Transient elastic-plastic response of a simply-
supported plate
This problem was solved previously by Liu and Lin [28] who
used small deflection theory and ignored membrane effects.
In our calculations we assumed full geometric as well as
material nonlinearity. One would anticipate some stiffen-
ing in this case due to the tensile stresses developed as
the plate deforms. Since the center displacement is of
the order of half the plate thickness, one would expect
small, but not insignificant differences. As may be seen
from the results of Figure 1, the expected behavior is
produced by our calculations which tend to be slightly
stiffer than L1U and Lln's wlth respect to peak response.
159
2 .8
e lo s. ic - perfeclly
plcsl ic
2.4
L = 10 in
h :I 0.5 in (fh ic k]
E • 107 ps i
11 . 0.3
.8 .. ....... Liu end Lin 1979
p =
30X 10' ps i (y ie ld .Ire .. )
fTy =
2.588xI0-4 Ib-sec 2 /in4 .¥c - - - imp licit , u
l mped mass
• { 0 • <0 Q)
<) _ . - implic i t, consisten t mass
q 300 ps i 1>0 .4
- explicil, lumped moss
6. = { 2.230 x 10-5 sec implici t
5.575 x 10-6 .ec er.pli cil 0
sir. teen 51 elements
0 2 4 6 8 10
t i me ( 10 - 4 sec)
0.8~------------------------------------------~
0.7
,
..--.
c 0,6 /''tf''
, 00'00
,../
+-
/ 0, 0 Ba Imer and Witmer 1964 (exp't.)
C
0) 0.5 I - -- Belytschko and Marchertos 1974
E I
,0
- - elastic-perfectly plastic
Q)
u , ···········isotropic hardening
a present
0- 0.4 P kinematic hardening study
(/)
I
I
------ combined isotropic-
\J kinematic hardening
!>-
Q)
0.3
+-
c
Q)
u
0.2
p = 2.61 x 1(j4Ib-sec2/in4 Vo = 5200 in/sec
O"y =4.14xI0 4 psi ~t = 10-6 sec
0,1 E= 1.04x107 psi ten UI elements
Ep= .667X 106 psi explicit, lumped mass
v = 0.3
O.OL-______ ~ ________ ~ ______ ~ ________ ~
d1 [t dl
50.---,---,---,---,---,---.---,
___ _c;,t~l~lr;_(J!,.a.ond c
r • 2 In
B ,._. C
:e 40
L . l O in
1"1-- 0 .05 In
°0~--3~--~6---±9--~12~~15~~18~~21
verticat displacement 110- 2 in)
A B c
F lq 3 Static aXlsymmetrtc bucklmg of elostlc cylmdnco\ sh ells
162
"
".. AXIAL MOMENT 'is 1t3 AT • S lr
., q
, or
2
x
_0
,
~N
co
...J
_01~~~~ __ ~~ ____ ~ __ ~~ ____ 1
!z 0 f O 0.4 0.8 1,2 1,6
~~ AXIAL DISTANCE 11.3 (IN))( to- I
...J'
"XOr
Cl
!V)9
g
zgho~~--~O~
. 4----~O~.B----~I~.2~--~~---4 AX I AL STRESS YS .s, AT ~ • w
AXIAL OISTANCE .,IIN) x 10- 1
~ ~ ~
--
r---
k
V
f'-..
t<
,/
h?
~
!>
~~
I ~
--~'
~
~~ §~
g= 0,00 9 = 0 .6 0 g =1.20
Q "'~"
~
,..-,-1-- /
t7
t;po<: ~ - "..:
V-
~ I--'
~ t-
l\---I V F;'"
---
I"- ..... ~-- ......
r--- < <
....... r-
>< i><
>< K ~
~~'"
~~
.--
V~ ~
i=::
9 = 1.80
~~
g=1.82
~ g = 1.90
(START FROM 9 = 1.80) (START FROM 9 = 1.80)
9 - 1.20 g : 1.80
9 - 1.82 g - 1.90
(START FROM g - 1.80) (START FROM 9 - 1.80)
5. Conclusions
In this paper we have presented dynamical aspects of a
general nonlinear finite element shell capability. This
work is a sequel to stat1cal aspects Wh1Ch have been
reported upon in [20,21J. In particular, we have described
the construction of mass matrices and presented details of
the transient algorithms employed. Several sample problems
166
References
1 Argyr1s, J.H.: Recent advances 1n matr1x methods of
structural analysis. Progress in Aeronaut1cal SC1ences
4. Oxford' Pergamon Press, 1964.
2 Argyris, J.H.: Continua and d1scont1nua. Proceed1ngs of
the Conference on Matr1x Methods in Structural Mechanics.
Ohio Wr1ght-Patterson Air Force Base, Oct. 1965.
3 Argyris, J.H., Balmer, H., St. Dolts1n1s, J., Dunne,
P.C., Haase, M.; Kleiber, M.; MaleJannak1s, G.A.;
MleJnek, H.-P., Muller, M., Scharpf, D.W.. F1n1te
element method--the natural approach. Computer Methods
1n App11ed Mechan1cs and Eng1neer1ng 17/18 (1979) 1-106.
4 Argyr1s, J.H.; Dunne, P.C .. Post-buck11ng, f1n1te element
analysis of circular cy11nders under end load. Report
224, Inst1tut fur Stat1k und Dynam1k der Luft und Raum-
fahrtkonstrukt1onen, Un1vers1ty of Stuttgart, Germany,
1977.
5 Argyr1s, J.H., Dunne, P.C., MaleJannak1s, G.A., Scharpf,
D.W.: On large d1splacement--small stra1n analysis of
structures w1th rotat1onal degrees of freedom. Computer
Methods in Applied Mechan1cs and Eng1neer1ng 14 (1978)
401-451, 15 (1978) 99-135.
6 Balmer, H.A., W1tmer, E.A.: Theoret1cal-exper1mental
correlation of large dynam1c and permanent deformat1on
of 1mpuls1vely loaded s1mple structures. Report
FDL-TDR-64-108, Air Force Flight Dynamics Laboratory,
Wr1ght-Patterson A1r Force Base, Oh1o, July 1964.
7 Belytschko, T., Glaum, L.: App11cat10ns of h1gher-order
corotat10nal formulat10ns for non11near fin1te element
analysis. Computers and Structures 10 (1979) 175-182.
8 Belytschko, T.; HS1eh, B.J .. Non11near tranS1ent f1n1te
element analysis w1th convected coord1nates. Interna-
t10nal Journal for Numer1cal Methods 1n Eng1neer1ng 7
(1973) 255-271.
167
AK.NOOR, J.M.PETERS
George Washington University Center at
NASA Langley Research Center, Hampton,USA
Summary
A two-stage glopal-Iocal approach is presented for predict-
ing the collapse behavior of shells. The first stage is
that of spatial discretization wherein the shell is dis-
cretized by using finite elements (or finite differences)
which cover the entire region of the shell. In the second
stage the vector of unknown nodal parameters is expressed
as a linear combination of small number of global functions
(or basis vectors). A Rayleigh-Ritz (or Bubnov-Galerkin)
technique is then used to approximate the nonlinear equa-
tions of the discretized shell by a reduced system of
nonlinear algebraic equations. For the case of loading
applied by means of axial end shortening, a scalar function
is introduced which measures the degree of nonlinearity of
the structure. Also, a quantitative measure for the error
of the reduced system of equations is proposed. The effect-
iveness of the proposed technique for predicting the col-
lapse behavior of shells is demonstrated by means of a
numerical example of the elastic collapse of a cylindrical
shell with a rectangular cutout.
Introduction
In spite of the significant advances made in numerical dis-
cretization procedures for nonlinear problems, the collapse
analysis of complex shell structures, having thousands of
degrees of freedom, is still not economically feasible on
present-day computers. Hence, increasing interest has been
shown in the application of a two-stage, global-local ap-
proach to these problems (see, for example, Refs. I to 7).
Mathematical Formulation
(2)
{X } = A{Z} (3)
c
fG l (Xf,X c ) }
+ o (4)
~G2(Xf'XC)
In the absence of externally applied loading (case of pre-
scribed displacements only), {Qf}=O and {Qc} equals the
vector of constraint forces associated with the prescribed
nonzero displacements {Xc}. The first set of Eqs. 4 can be
used to determine {X f } and the second set is then used to
evaluate the constraint forces {Qc}.
(5)
where
1':1] (6)
{G(lji)} = [r]T{G(lji)} (9 )
and
(10)
Computational Procedure
The computational procedure outlined in Ref. 7 for the ef-
ficient evaluation of the global functions, generation of
the reduced system of equations, automatic selection of the
displacement increments, sensing and controlling the errors
of the reduced equations is adopted in the present study.
However, the scalar S (current stiffness parameter) used to
characterize the nonlinear response (see Ref. 9) and the
error norm used in Ref. 7 are not applicable to the case of
prescribed edge displacements, and therefore, are modified
as described subsequently.
where
I I
aaQ\c = [[Kef] +
(G2iJHaxfj
ax f . aA + [IKeel + [:::il]IZI (131
J J
r dljl j
aQl ael ]
---ax- E [Klj + aljl . d\
(14)
j=l J
and
175
1
(15)
E!UWft No!tm:
To check the accuracy of the solution obtained by the re-
duced system of equations, Eqs. 7, the following error norm
is used:
(16)
where
(17)
Numerical Studies
To test and evaluate the effectiveness of the proposed
global-local approach, a number of nonlinear shell problems
were solved by this approach. Comparisons were made with
solutions based on the full system of equations of the finite
element model. Typical results are presented herein for the
elastic collapse analysis of the cylindrical shell with rec-
tangular cutout shown in Fig. 1. The problem is one of three
problems used in Ref. 10 to assess the capability of various
programs to analyze shell structures. The load is applied to
the cylinder by means of a uniform axial end shortening which
is increased incrementally until the cylinder collapses .
0'0762m~
'<I:: 0 .3
R :: 0.1541 m
I 0 .1143 m
h = 3.556 " 10-4 m
a
Boundary conditions at x 1 :: 0 b
u 1 =). 0.2421 m
u2 = w ='1 ='2 =0 I II I
a
0r"
m
b
1--- -- - 0.1815 m- -- -----i
2.5
2.0
1.5
Current stiffness
parameter S
1.01-----
0.5
o
Total axial load (Newtons)
A=3. 048 x 10- 5 m. at which value the error norm was checked
and was found to exceed the prescribed tolerance (see
Fig. 5). New (updated) set of seven global functions were
generated and used to advance the solution until A=9.398x
10 -5 m. The predicted collapse load of the cylinder was
13 . 656 x 10 3 Newtons corresponding to A=9. 70 x 10 -5 m. The
collapse loads predicted in Refs. 11 and 12 are 10.008 x 10 3
and 12.792 x 10 3 Newtons, respectively.
linear solution
15
12
9
Tota l axial
load (Newtons)
6
3 - - - Fu ll system
+ Reduced bas is (7vectors )
" ' -_ _ _ ...l...-_ _ _--'-_ _ _----L_ _ _ -----I X 10-4
o 7 14 21 28
Normal displacement w (meters)
15
Bending
12 energy
- - - Full system
3
+ Reduced basIs (7 vectors)
o
strain energy (Newton' meters)
The. uoe. 06 ;(;he. g£.obal-£.ocal appfWach J..n ;(;fUJ., PJtob£.e.m Jte!.JuUe.d J..n Jte.-
du~ng ~he numbek on degJtee!.J on 6Jte.e.dom by a 6ac;(;oJt 06 ove.!t 400 (from
180
- - - Forward path
......... Backtracking
.05
.04
. 03
Error
norm
. 02
, 01
\
o
Total axial load (N ewtons )
" 10 3
15.00
11.25
Total b
exlal N1 = 3.996 " 10 3 Newtons
7 .50
load
(Newtons) 2
3 .75
0.0762m~ au o.1143m
b
N4 =11.949 " 10 3 Newtons
Concluding Remarks
A two-stage global-local approach is used to predict the
collapse behavior of shells. The computational algorithm
presented previously by the authors 1S modified to handle
183
References
1 Besseling, J. F.: Nonlinear analysis of structures by
the finite element method as a supplement to a linear
analysis. Computer Methods in Applied Mechanics and
Engineering 3 (1976) 173-194.
W. WUNDERLICH, H. OBRECHT
Ruhr-Umversitat Bochum, Germany
Sununary
1. Introduct1on
system with unit vectors ~ (see Fig. 1). For the description
of curved members it is useful to use convected, curvilinear
coordinates, such that in all configurations a material par-
ticle is identi f ie d by the same set 0 f values e 1 , e 2 , e 3 . The
notation employed here follows that of [7,8].
current,
deformed state
F1gure 1
k
na a.k M NM
a and the following identit1es hold
k k oM
a· M n Na Na NK oa Na NM (2)
a M K J3 J3 K a K
K
A one point rotation tensor w1th components a. M may be defined
K . K K i a
by introduc1ng the shifter g.1 = GK .9:i to g1ve a· M = (g1 na)N W
(3)
i
whera F.A denotes the components of the deformation gradient F.
It is a two point tensor and may be decomposed as follows
F 9: U (4)
( 4a)
(5)
....
deformed base .9:i to the stretched base QA. Recall, that in terms
of the usual displacement grad1ent the base vectors .9:i are given by
(6)
f.~__- -
F1gure 2
T.L. : V.L. :
0
n
-a (9, + -a)N
-a
n
-a (I aiR
- + --a
0 o
r R + u + ~ r + ~
0 A i (7)
dr (~A + ~'A + ~'A)de (51 1 + -u,.)
1
de
0 0
vi (~ + ~) . (!. + g + g) vi = (;~. + ~) . (.! + £l.l
h~ + a~ .
1 1
191
- 0
~+~
[
cos (<p + <p)
0
Sln (<p + <p)
o ] [
1 <p
'] [
cos<p
0
S1n~ ]
0 0
or a ~ + ~ (~ + I)T ~ (9)
o
The corresponding general express10ns for ~ and ~ 1n terms of ~
1 wKL
wM = -"2 EMKL
-w3 2 1
0 w CI.' CP1
0 0
~ W3 0 -w1 ~ w
2 b CP2 b ~ (10)
-w 2 w
1 0 w3 CP3
o 0
Note that b becomes unity in the reference state. The matrix b
is given in the Appendix (equ. A3). A summary of the above rela-
tionships follows:
T.L.: U.L. :
wij u .. =-u
~J ji
u~J e: mji cPm
wi cP~
( 11)
T.L. : U.L. :
C 0C 1 C
(oA + hA + 2" hA)h CB Eik =
h ik +..!.
2 h~1 h jk
hA
B
= .A(oC +oCI + C 1 )+O.A C 1
aC BUB u B a C u B h~ a k.i ( 0.k + u k.i ll
) +u
J J __J_ J
o
AB
The symmetric part of r
~ AB
r (14 )
oir o~
Here L , Lr denote components of Cauchy stress.
oB 1+v {OB v OK
EA ""E sA - 1+v sK oB}
A
(16 )
v K
B
EA = 1+v
E
B
{sA - 1+v sK oB}
A
( 17)
( 18)
£AB
Since the engineering measures r °
a n d hAB are objective a
Legendre transformation
Q*(~)
Q.
= rB
£A hOB _
A
°
Q(g) (19 )
o 0
exists [5J , where Q (g) and Q*(~) are potent~al funct~ons corre-
spond~ng to the strain energy and complementary energy densit~es,
(20)
w* (~) (21)
195
(22)
C
where nAB = u
IA u CI Band W (f) lS agaln the lncremental strain
energy denslty. Substltutlon of (17) glves
(23)
W + W* -_ t·AB u BI A = sA B + 2s A n B
B EA (24)
B A
oA °B 0A 0
6J CB 6{ J [-Q* C~) + r B hA - Po f uAJ dV
Vo
+ boundary terms } 0
T.L. :
6{ I[-w* oL
(.!:) + a.. B r
AB OAB O.L K
YL IA + EMBL r ~ Y !A
M
W
Vo
o . L OK! rAB wM 1 OK! oA o.B M
+ EMBL ~ Y A 2"Y A rB ~ WM w
(26 )
+1 0 K! o.LOAB _P fA]
2 Y A ~ r wL "13 0 YA dV
V
1 oil
<p j ) - pf j yJJ dV
o~
-2 (r i <p.
J
- r <p~ <p l (27a)
The Euler equations result1ng from (27a,b) are the force equili-
brium (28a or 29a) and momentum balance (28b or 29b) equations,
the polar decomposit1on relation (28c or 29c) as well as all
boundary conditions:
U.L. :
m
h .. + (28c)
m1J <P
E
YJli J1
or
t1JI - pf J 0 (29a)
1
olk j
t 1J + <f jk hi + 1:' a· k symmetr1c (29b)
k
h + a (29c)
YJ 11 J1 J1
R (30)
199
c
where 6 and n are the d1stances from the axis, R 1S the pos1t1on
vector of the p01nt on the aX1S, and ~k are the un1t vectors of
the orthogonal reference tr1ad at th1s p01nt (see F1g. 3). Vec-
tor ~1 1S tangent to the reference aX1S. The base vectors ~K
(31 )
where the Chr1stoffel symbol r K11 accounts for the axial curvature
and tW1St of an initially curved rod.
~
M= center of shear
S = center of gravity
F1gure 3
c
~, ~i and ~i respectively. From (5) it follows that 1n general
a. .c(",B+hB)A (32)
-1 ai U c C-B
(34a)
cB B B B ]
[Y +n a · 2 +6a.3 -wwa., ~ (34b)
The last term in (34a) accounts for warping of the cross section
due to torsion and w(n,6 ) depends on the cross-sectional shape.
Thus, the deformation of any point perpendicular to the referen-
ce axis is completely described by that of the respect1ve point
on the aX1S. Moreover, as a consequence of (33) and (32) the vectors a.
1
are seen to coincide with the tr1ad of principal strains and the
angles ~ and ~ appearing in (26,27) may be identified with the
increments of the rotation angles about the local coordinate
vectors a ..
-1
c c
y, y, 0 6 -n <P, wW
c M c
Y2 Y2 + -(6-6) 0 0 <P2 0 (35)
c M c
Y3 Y3 n -n 0 0 <P3 0
201
c c
It is noted that in (35) Z, y, and ~ denote incremental quanti-
ties, and that the sam~ relationshlp is appllcable in the T.L.
descrlption if rotat~ns remain moderate. Furthermore, (35)
differs slightly from (34) in that (35) provides for the fact
that the cent~s of shear and of gravity may not coincide. The
appropriate modifications of (34) have, however, been made in
all general expresslons whlch follow.
The remaining components ~1K are the normal and shear stresses
acting in the directions of the vectors ~ of the dlsplaced and
rotated cross section (see Fig. 3).
0 0 0 0
011 N M3 M2 Mw
r
A F n+ ~ - F
w(n,~) (37)
nn F~~ ww
0 0
M 0
013
r
Q3 M
- [-(n-n) + w'1: ] ~
J
011
+ r (n-n) 1fT (38b)
A T
In (37) A lS the area of the cross-sectlon, Fnn and F~~ are the
202
0000 000
The quantit1es N, M2 , M3 , Mwand Q2' Q3' M1p 1n (37,38) arc 1dent1-
cal with the usual stress resultants. Th1S may be seen from the
var1at1on of the surface 1ntegral of (26). Not1ng that the in1-
t1al normal vector N is 1dentical w1th ~1(F1g. 3) one obta1ns
OAB
Jt NA 6YB dA =
oB
a· K
J rolK 6YB dA (39)
A A
By Subst1tut1ng (34) or (37) 1nto (39), carry1ng out the inte-
grations over the th1ckness, and comparing coeff1c1ents one
f1nds
011 0
r N
r 0
012 o
r dA Q2 2 (40a)
J 013 0
A r Q3
M M 011
0 -(k;-rJ (ll-ll) r
012
)
r k; 0 0 r dA M ( 40b)
A 013
-ll 0 0 r
011 0
Jw r
A
dA Mw (40c)
203
0 0 0
N is thus the resultant aXlal force, M2 and M3 are the bendlng
o
moments about ~3 and ~2' and Mw lS ~he "warplng bi-moment". ; t
lS noted that the torsional moment M1 In (40b) dlffers from M1p
In (38a,b) by terms whlch are due to warplng. Slmllar terms
influencing 8 8
2 , 3 and ~2' ~3 have been omltted In relatlons
(40a,b) but are taken lnto account in the development of the
next chapter. In matrix notation (39) then reads
f °B
t 6YB dA ~'.Q
° 0 c 0 c
° 611'
6y + M 6w + Mw
(41 )
A
0c 0c 0
It is also seen that y , w , and ~ are the klnematic variables
000 0
which are conJugate to ~ 9, ~ and Mw' By further substitutlng
c 0 c c
6~ = 12 .6~ from (10), where ~ are the rotation angle lncrements
about the lnltial reference trlad, one may rewrite (41) as
follows:
0 0 c 0 0 c
° 611'
a .Q 6y + M b 6.!fJ + Mw
A (42)
c
° 6y
0 c 0
V + B 6~ + Mw 611'
o o 0
where V a.Q
°
B = b°T M
° ( 42a)
oc oc
Their conJugate klnematic variables are y and ~ . Equation (42)
o 0
may thus be interpreted as the vlrtual work performed by V, B
o - 0-
and Mw In the lnitlal conflguration. Flguratively speaking,~
o
conslsts of the sum of the components of .Q pOlntlng In the di-
rection of theinitlal reference system. The simple form of (42)
lmplles that the general flcld equatlons take on a partlcularly
slmple form if expressed In terms of these conJugate quantitles.
where
(43a)
I
T. r
c
X l E
: "Irtual war c
c Q !19.Y) jlIU!) 'f ill
If I
J6
Y'"
---~---~- - ----- • --------,--
,, 'Y" ds.
(44)
5
kmemallc I malenal
~Trj.~)
,
~I~) M
t:1 equallons law
I
/til" /til",
c c
r kmeff'allcal molena I 1 E
c relallons law -2 T(j,'l!) -!1/~) c
ill
'!. I if
vr
I
vr
Ix
I
•
I I
------~-r------- --------1-- (45)
¥ I
I .l(
force -
eq~rllbnum f (9,(1) 21j, ~)
M momenl- I M
M., Mw
of (44).
5. Numerical examples
~-c~~__~L~~~_ _ _ _.~
M= ;mEl analytical
2L
~ numerical
(10 elements
160 load-stepsl
F1gure 4
M --Jill.
l
cross-section
R -1 II 1111111101 jJ II I H
"g;" 6:
II I 0
1" om
II H t
1c
~2
"~rf~10cm~
Figure S
--- ---
8
7
6
5
-large rotation theory
4
-- - moderate rotation theory
3
2
2 4 6 8 10 12 14
Figure Sa: Displacements u 2 , u 3 at M
it 209
8
7 ~--
-- --
......., --- <P.
6 1
./
/'
5 /'
/
4 /
/
/ - large }
3 / rotation theory
- - - moderate
2 h
1/
1,01
10° 20° 30° 40°
Figure 5b : Torsion angle <p at M
it
8
7
6
-- ~-
"-
/
/
, M3
/
/'
5 /'
/'
/'
4 /"
~
3
2
Mz,M 3
50 100 150 200 [kNm]
250
it
-- ----
M1 _
8
~
...-
7
5
4
3
2
M1 [kNm]
5 10 15 20 25 Mw [kNm4
Figure 5d:Torsional moment M, and warping bl-moment M
w
at B
210
cross-section
Figure 6
A-
u3 u2
-- -- --- ---
8
-- ---
7
6
./
/'
5 V
4
3
large }
rotation theory
mo d era te
5 10 15 20 25 30 35
A
8 - - Iorg(' }
rotation theory
- - -moderate
7
6
-
'1\
- -
- -- - -
5
4
3 ""
2
A
8
/
"
7
5
4
3
2
~-~--'---r--~--'---r---~ M"M 3
20 40 60 80 100 12 0 [kNm]
Figure 6c : Bending moments H , M3 at A
2
A
8
-- -------- - ---
7
6
-M.
5
4
J
2
M,(kNm]
2 4 6 8 10 12
• 2
M... {kNm J
Figure 6d : Torsional momen t M, and warping
bi- moment Mw at A
212
6. References
AppendlX
0
[ e,e, 8182C3-C183
(A 1)
.9,.= C2 8 3 C1C3+818283
-8 2 8 1 C2
0 o
where C cos <Pl 8i sin <Pl (i 1, 2,3)
1
(C 1 8 2 C3 +8 1 8 3 ) <P1 (C183-8182C3)<P1
+8 1 C2 C3 <P2 +C1C2C3<P2
-(C1C3+818283)<P3 +(81C3-8283C1)<P3
(C18283-81C3)<P1 (-C1C3-818283)<P1
+8183C2<P2 +83C1C2<P2 (A2a)
+(8182C3-83C1)<P3 +(8183+82C1C3)<P3
o C1<P2+81C2<P3
(A2b)
U) = -81<P2+C1C2<P3 -<P 1 +8 2 <P 3
-C1<P2-81C2<P3 o
1 o
o
E. o C1
(A3)
o -8 1
c c c c c
YI YZ >; WI W z W3 vr I MI I M2 I M3
~ ~ ~ I MOl '"
r-- ""
,~
~
1(:) .
I~
o 0
lel "3 -"2
k oA \0)1.. \00
~ 0 ~ IX31Q3 0 0
£A +IX13 EA +ct.?3 EA + IX33i -"3 I~ I ;;
00
00 I of}. o
IXI/OZ 0 0
~" 1 IX3102 0 I
- EA HX 12 - £A -(XZ2 - EA -a32 "2 -vr I~
-1 ,OJ
6~T'~--~--~----T---~--~--~--~--~----+-------~-------r--o--o---+----t----+----t---~ (A~)
-~
EA
0 o 0
o 0 oR0 0 0 oZ
~ _~2_ 0 _ (XI/(XZI _!!i1 _ (X21 IX31
~I EA +IX23.
I EA Ci22
I I EA EA EA
o 0 0 0 02
(X31030 0
I Ci31 0 Z _ &/1&31 ~I !!J.I
~, £A +Ci33 - £A -a32 -EA
EA £A
o o
~I - "3 "2 -1
o 1
f)1 -v, -u.,t
"3
o o _L
-"2 l/I ~I E~~
_--L
0)1 E~",
L..-
215
c-
• •
o 0 j
~"2+ ""21/1
0 0
~~3+}~3Y, M. M•
.b. : M,1/1 ~ 0 0 0 0 d 1!Q2+~Q3 (A4a)
- '1i1/l "3 - Q 2 + fi1.,1/1 "2- 3
} • 0.
° , fi121/1
~ "2+ ° ° -~y,+~Xt'~X;
° °"2°
J ° 0"3-
- fi1.,1/1 .
o 0 02
<52 +M.,"., -QlEA+Q,
0 M3+M.,
lit, "3°+l~ °
2 31/1 -~"+~X2+~~
-;:'2 +t:t., *'3
+1't:tw1/lX2- t! 3
o 0 02 0
0
• +M.,".,-Q)'EA+G,
M •
-1! Q2+ ~ Q;
M.
•
M; + M.,"2
•• a
-M2+ Mo.";
•• GJT+ 1:'0.,
c c c c c
-'"
'"
Y, Y2 Y3 </, 'P2 ~J I 1/1 I VI I l<j I ~ I /vi, I /vI2 I /vI3 I /vi", }
c
- Q3 Q2 1
-
Y, k] -k2 -a ~ EA
J2 -k3 h 1
c
Y3 k2 -h -1
Abstract
1. Introduction
" ct.'Of'fIWCf
CUI'fe"t stol e "
()Uhll
I I
I I
I
I I
I I
I, I
I
I undeformed I,
I in itialslo le
glCiI
I I
I I
I
j'L ____ . _ __ _
~, .'
!,L _______ _
!, xl
(4) ,
221
o (Sa) ,
o (Sb)
"t
I
I
i
I
K.L global nodes
I
L_ -----------
"
Figure 4. Element external
loads
(~*N) 2 *
1~
ITmR - L{ f(N) [- 2EA
+ (E) N ~(~w, x)
2EI
2J dx
n 1
+ [~ * N ~u + ~ *S ~w + ~ * M ~8Jl(N)}
0
I f(N)(~P1~u+~P3~w) dx - L (~N~u+~S~w+~M~8)
n 1 k
+ L
nIx
f(N) [(E)N(~U, +z,
x
~w, )_(E)M~w,
x xx
-P1~u-P3~wJ dx
U,
x
* (E) M
+ ~ M ( EI + w'xx)} dx
stati onary (6 )
* N,x = 0
~ in l(N) (ia) ,
The functlonal ITmR ' eqs. (6), contalns two types of correction
terms, an equlllbrium correction term (2) and a compatibility
correctlon term ~ . These correction terms are due to the
fact that equllibrlum equations for the initial stresses as
well as strain dlsplacement relations for the initial strains
are nonlinear equations which cannot be fulfiled completely
by a sequence of llnearized incremental equations. Durlng the
incremental procedure presented here the correction terms in
the load step (N) are taken into account ln the following step
(N+l). Accordlng to the nomenclature introduced by Pian [18J
and Stricklin et al [19J the geometrically nonlinear procedure
presented ln thlS paper can be thereafter identified as a
self-correctlng lncremental procedure (see Fig. 5).
4. Interpolation functions
(8 )
B1
(9) •
AT
- .12 feT ;E!;nfe + 1 LliC n
IImR - ~
n
{
n n 2 Jt~ LliC n
AT T AT
LliC n LI~'n +
fen ~n Llltn + LliC n ~n - feTn!(,n }
T
- ~ LliCk Lltk stationary ( 10)
k
where 'V
H is the flexib~lity matrix
/fO is the initial stress stiffness matrix
LI~ ~s the incremental element lUIllped load vector
T -T - - -
LI!C k = {Llu k , Llw k , Ll0 k } and LI~ k= {LINk' LlS k , LlM k } denote the
displacement and load vector of the global nodal point k,
respectively.
stationary ( 13) •
where
(16)
(20)
so that
~A(N) ± (21 )
-T A(N)
~ref ,!(;ref
In our numerical studies, carried out by the new incrementation
algorithm the following choices have been made:
~S (N) = I- T A ( 1) I (22)
p ~ref !Cref
(24) .
In order to test and evaluate the beam element resp. the step-
228
a.
.,"0
a
0
-'
70
I aM' I a .., I
60
E
50
40
3D
o,L-~~~-----~-------r------- 20
10
Q1 02
\0
Q. Q35
74=
Q9 I-
I--
;--- - - u
o
.3 025
t----
f- 020
Q2
01
015
010
-]T
- present analysIs .o.S .Q025
- f-- • Sergan tt aI 1221 .o.~ =0 05
60 80 100
0~0---+--~tO~-~'~5--~2~0-~
20
load P Deflechon Parameter w/z.
10
+---4---+-----'..-...-.... --+---4
vf
~
.
K
a.Q35
~
Ii
/"
~ 030 /,
e.
~
-g
.9 025 t - - - - - j \
-~
020t---+
L \
015
02 -
01 t----
r-
r
~ r--
'---
-t±
010 I-
I
005
-01
-02
t----
t----
-; I---
h
~
I
1000 2000 3000 4000 o 005 010 015 020 025 030 035
Axial Thrust N Load Parameter p
0.002
a§ 11
PfM • Eul. , buc.lth n9 lood
ii: l ' _OJ
1.0
'0
8
-' 0.9
0,8
0,7
0,6
O,S
0,4 1.1(XXj
0,2
- Argyris/Ounne (5)
Q1
• present work
0
o
Rotation e
Figure 12. Right-angle frame: geometry and
imperfection sens itivity
~~ \!
~ 1~ -f~'f--t- J -1--+--+-- +--+---1
~12OO~' \-'\ \ ~
1M» ~ / \ \- - t ----l- -t----t--l
or----+----+--!/
400
J I
1\/
\
-" /)
'
--~----1_--~----~__l
eoot---+---+,i I~
( \;/
• / ~I ,! It---t---'---"'------'----j
I.. UO 24
1__ \ /- -
f ..
20 60 eo )0 II)
Deflections u,W
7. Conclus~ons
References
Dupu~s, G.A.; H~b~tt, H.D.; McNamara, S.F.; Marcal, P.V.:
Nonl~near material and geometr~c behav~or of shell
structures. Compo & Struct. 1 (1971) 223-239.
2 Bathe, K.-J'i Bolourchi, S.: Large d~splacement analys~s
of three -dimens~onal beam structures. Int. J.Num. Meth.
Engng. 14 (1979) 961-986.
3 Ramm, E.: Geometrlsch n~chtl~neare Elastostatlk und
f~n~te Elemente. Hab~litat~onsschrift, Un~vers~tat
Stuttgart, 1976.
4 Frey, F.: L'analyse stat~que non lineaire des structures
par la methode des elements f~n~s et son application a
la construct~on metall~que. Ph. D. Thesis, Universite
de Llege, 1978.
5 Argyr~s, J.H.j Dunne, P.C.: On the appl~cation of the
natural mode technique to small stra~n large displacement
problems. Proc. World Congress on Finite Element Methods
ln Structural Mechanlcs, Bournemouth/England, 1975.
6 Pirot~n, S.D.: Incremental large deflection analyses of
elast~c structures. Ph. D. Thes~s, M.I.T., 1971.
234
K. BRINK" W. B. KRATZIG**
Ruhr-Universitat Bochum, Germany
Dr.-Ing., Hochtief AG Essen, formerly Research Group
Leader, Ruhr-University Bochum, Germany
*> Dr.-Ing., Professor of Structural Engineering,
Ruhr-University Bochum, Germany
Summary
Based on an energy-consistent, geometrically nonlinear theory
of rods several curved finlte bar elements are derived. Large
deformations and large rotations are considered in a kine-
matically rlgorous sense. The solution-procedure - a modified
Newton-Ra~hson-method within the modular software-system
FEMAS - demonstrates the excellent convergence of the derived
elements compared wlth standard examples of the literature.
1. Introduction
The state of the art of computer-hardware and computational
techniques enables the analysis of arbitrarily complicated
structural responses. The finite element method as the most
frequently applied approximative solution technique is often
combined with further (sometimes very tough) mechanlcal or
mathematical approximations. In this contribution the authors
intend to demonstrate the advantage and simplicity of the
use of approved mechanical modells and geometrically complete
formulations.
Considering
238
u*
3
u + Z't!' ' Z = Z (2)
undeformed
initial state z
t
I
deformed state
o 0
o
~ a ~
~1 ;::'1 = ~1 (4 )
a z1 a s
a~1 =
(7 )
03 01
a1 0 a1 0
+ (8 )
y-;;;; ~1
y-;;;; ~3
~1
. 0
e3 = (~1 + ~, 1 )
. 0
(~3 + ~)
0
~1
0
~3 0 ( 11 )
240
l03 1 03
a1
l- j'
_ U'l + u°1 b°11
~ arc tan arc tan
01 03 01 01
1 + u '1 - u b 1 a1
~'1= a 11 [- 03 01
(a 1 ) , 1 a 1 +
03
a 1 (a011 ) , 1] (12 )
L L
~ ~ '1 (13 )
1 * * 0* 0*
El'J' = -2 (r ,l, ' r
~ ~
"] - r "l . r, ],), l,j = 1,2,3
~ ~
(14)
(15 )
0 0 2 2 0 0 2
0 1 0 1 0 1
a 11 u'1 - w 2" (u
x +
2" (u, 1)
x)
+
2" (w, 1) +
0 0 2 0 0 0 0 0 0
+
1
"2 (w x) - u'1 w x + U
w'1 x ,
o o o o
X + cos ~ (- X + ~'1 - u'1 + ~'1 u'l
------------------------------------------
o 0 o o
- ~, 1 w x) + sin ~ (w, 1 - ~'1 w'1
o 0
- ~, 1 u x). (16 )
241
- -
0 0 0 02 0 0 0
I3 SR
11 CP'1 u'1 It + w It + CP'1 u'1 CP, 1 w It
0 0 0
,
02
+ cpw'1 It + CPu It (17)
(18 )
D
Dt
(19 )
/0
s
(EF a 11 a 11 + EI 13 11 ~ 11) d~
s
ds
o
(20)
deformed state
undeformed
imflal state
P,P
components of the
difference vectors
l:Y and !¥ for z = 7
~
a 11 [- 0 ~ 0 01
(w'1 + u x) a 1 +
0 0 0 03
(u'1 - w x) a 1 ]'
(j) , 1
a 11 [- 0 ~ 0 01
(w'1 + u x) , 1 a 1 -
0 ~ 0 01
(W'1 + u x) a 1 ' 1
0 ~ 0 03 0 ~ 0 03
+ (u'1 - w x) , 1 a 1 + (u'1 -w x) a 1 ' 1 J
[ ]
0 0 0 01 0 0 o 03
a 11 ' 1
---2 (w'1 + u x) a 1 + (u'1 -w x) a 1
(a 11 )
(23)
o
Again we have introduced the abbreviations
~1
u
0
u, u
~3
w.
The incremental d~splacement vectors Q, ~ are defined by
the difference of the position vectors -*
!, rand (1):
0 0
r = r + u + u
'V
,
0* * 0*
r* r + u
*
+ u r + u + U + z (~+!) (24)
(25)
.
+
(a ++
11 + a 11 )
s
(a 11
o
+ i~,)J ds , (26)
(27)
.J
+'+ +.:. +'+ + .:.
[EF(Ci. 7? Ci. 77 + Ci. 11 Ci. 77 )+EI((J11 (317 + (311 (311)J ds =
5 o. • • +
J[(p' +/,') 6, + (m + In) ip - EFCi. i1 all - EI (371 (311 Jds
5 - - (, :;: 1 3, U, = U I U3 = w)
o 0 _ 0 03 0 1)
+ Vi,1 Sin tp(x- tp-t) + tp (x -tp'l }(COS tp 01 +S", tp 01
CoeffICIents of transformatIOn
a/ = 1 + (;,1 - WX
0"3
= W'1 x
0 0'
1 + U
_ [all]
-e: =
(3- = e:
-
.,.
II
g = (Q .,. Qv ) Q
Operator matnces.
Q = [+H 1-:, 1~ I ~ I j 0
~
, xO
"3 l 0'I1 -1 X(1-0
" 11 ) 0'31 0 0 ~
Qv = 0 0 0 0 0 0 0 0 01
X rx-cp.,J x-fX-tp,.) x rX-C{J.I) (x-'P.,) (x -ip'l) 0lcos!p-1
sm!p cos!p cos!p-F sm!p (alcos!p.,.a!sm!p) -atsm!p
Qv =
[..,
for small rotatIOns
x
ip
01
ic 2
oJ -1
C{J.,
x(1-0;)
-1fJ•• X
'3
0 1
epic x'
0
01
'3 'I
o ]
0 1 -1
, T'
~
I-f-
I
if
_0
I
ill
I 6(3-
1-0
I '
Qs Qs =
---+-
1-~6(1
I
kt--I'
o-J
I
I =Qo
L -Q
I
I-
I i
0Cf.
Qo . Matrtx for mltlal stresses due to Cl11
Q(1.. Matrtx for mltlal stresses due to (311
o neglected sub- matrix (elemenrs depend on Z2)
Table 3: Non-l~near ~ncremental strain-displacement relations,
related to the initial state
249
o 1 o
w'11 + 2 W o (30)
R
250
ql
s (z 1 ) a3 . (32)
S( 1)
W Z = a, Sin RZI + a2 COS RZI - a3 R Sin RZI
tpS (ZI)
(ZI)3 (Zl)'
• (°7 • 20,1 )RS" + 012 fi6
Combined mterbolatlon functions
for element displacements
u (Zl) = US(ZI) + U£(Z')
4R'
(z't
SR'
1 _L
R
2z'
-7 -1£4-
R 7
z' fz!):!
2i'? 1!t
3R ~
4R
!.Jii'..k't
Rio SRi>
_...L 6z 1
-fj'f
1
qr
z'
fj'f
1ft. ~ ~.1?J:-
R' R' R R' R
COS-fR
5 5 5' 5' 5'
5m- -Rsm :R
4R --:R 16R 64fiT - 256R'
-fR
cp" =
2- 5' 5' 5'
Sin2~ COS -RslO ;R 2R 4R'" 8R' -76fT
cos~
2R -srniR RO-cosiR 5
2R
5'
8R'"
5'
Tt:fiT
5'
6i:iF"
5'
160R 5
P[/b}
40+--~-+
30-t---
20+-~~r---~-----+----~
10 +---H--+- ~=----.-
O.-------~--------~I--------~I--------r-~I
0,0 0,1 0,2 0,3
R = 133.114 in (3 = 73397°
h =0.1875 In E = 107pSI
b= 1 In A = (32 ~ = 1165
w [15i
16+--
own solutlonJ7l.
large rotatIOn theor.
0 I
[In}
0 6 12 18 24wo
0 03 06 09 12 U o
R =100 in =80 In
L
h= 11n=b (J =5313°
H = 40 In E =107 pSI
7. References
O. C. ZIENKIEWICZ
UmverSIty College of Swansea, U K.
Introduct1on
SOlI. and lndeed rock and concrete are porous mater1als wlth the vOlds generally
fllled wlth water WhlCh exerts a pressure on the SOlld partlcles and mlgrates
through the materlal. Any numerlcal computatlon must therefore allow for the
essent1ally two phase behavlour - and on occaSlon when the saturatlon lS not
complete for a posslble three phase sltuatlon.
In thlS paper we shall conf1ne our attentlon to the purely two phase
sltuatlon and dlSCUSS the formulatlon necessary to obtaln the statlc or dynamlc
response of 'solI structures' up to the pOlnt of collapse. Practlcal problems
to WhlCh the numerlcal Solut1on must be appllcable lnvolve such tYPlcal sltuatlons
as the settlement of bUlldlngs. lncremental collapse of offshore platform structures
due to the actlon of sea waves or the posslble earthquake collapse of earth dams or
nuclear power stat10n foundat1ons. For many problems In such categorles qUlte
large d1splacements and local stralns can be tolerated so approprlate cons1derat1on
has to be glven to both materlal and geometrlc nonllnearlty.
(1 )
where the symbol p stands for the fluld pressure 1n the pores of the SOlld skeleton
The ~dea of using the separat~on g~ven ~n (1) stems from Terzhag~ [1]
who observed that un~form pressure changes alone could not deform the sol~d
mater~al apprec~ably or lead to ~ts failure. The only not~ceable effect of such
pressure change~ can be to cause a volumetr~c stra~n change assoc~ated with the
compress~on of solid gra~ns - th~s we can wr~te as
d~=-dp/KS (2)
(3)
where the f~rst term corresponds to the constitut~ve law assoc~ated w~th the Jaumann,
corrotat~onal, stress increment and the second term accounts for effect~ve stress
changes due to 'rig~d body' rotat~on of an element of volume. In above dE~
stands for any '~n~t~al' or creep stra~n ~ncrements and
p _ 1 P
dE'1 -3 dE,i (4)
The stra~n and 'sp~n' ~ncrements, dE;1 , are def~ned ~n the usual
manner as
and
(6)
At this stage we should note that qu~te generally the tangent modulus
D'Jkl w~ll depend on the current effective stress levels g' ,accumulated
plast~c stra~ns or other h~story or damage parameters and ~n the case of an~sotrop~c
Equillbrium relatlons
Following Blot we shall use the vector w, to denote the mean displacement of the
fluld relatlve to the solld skeleton. Thls lS defined by taKlng the total quantlty
of fluld and dlvldlng by the total cross sectlon. Thus the average dlsplacement
in pores lS W,I n where n lS the poroslty.
Wlth thls we can wrlte the total equlllbrlum of the solld-fluld volume as
(8)
- P"
where ~~ are the 'reslstlvlty' coefflclents whlch are the lnverse of the
permeabillty matrlx.
Mass balance
The flnal equatlon coupllng the behavlour of two phase materlals lS that of mass
balance. Clearly durlng strainlng of the solld matrlx the volume of pores changes
and the resultlng change has to be compensated by the outflow and compresslbllity
of the fluid.
. .p
= - P n IK f - Ell (1 - n I (9)
In the above equatlon we have neglected a very small term arlslng due to the change
of pore shapes when the materlal lS subject to pure effectlve stress - assumlng thus
that durlng such actlon the solid gralns are essentlally lncompresslble. The
lncluslon of thls term In a slmllar framework of analysls lS posslble as shown by
Zlenklewicz et al [4] but represents a negllglble effect.
Alternatlve derlvatlons of relatlonshlps governlng the two phase behavlour
are posslble using a mlxture theory. Some examples of such formulatlons are
avallable In llterature [5] - generally for llnear constltutlve relatlons - but
except for ultra-hlgh frequency phenomena In practlce these reduce to the above
format.
262
An approxlmatlon
Before attemptlng a flnlte element solutlon of the equatlons 1-8 It lS convenlent
but not essentlal to lntroduce an approxlmatlon. In thls we take the effect of
the addltwnal fluld acceleratwn wil as negllglble. Such an approxlmatwn allows
us to use ul and p as prlmary varlables rather than ul and WI thus reduclng In
general the number of unknowns. Wlth that approxlmatlon It lS easy to ellminate
W from (8) and (9) glvlng now In thelr place the well known equatlon of flow
In porous medla
(10)
The above equatlon together wlth (1) through (7) In whlch effects of WI have been
omltted governs the solutlon for all statlc and dynamlc problems of sOlI/rock or
concrete behavlour.
Though the above formulatlon dlsregards one of the dynamlc effects It lS
appllcable for most englneerlng dynamlcs sltuatlons assoclated wlth sOlI behavlour
even If permeablllty values are relatlvely hlgh and the frequencles are of the
order of those occurrlng In earthquake analysls.
In Flg. 2 we show results of a study on a Ilnear problem of a sOlI layer In
whlch the zone of analysls In whlch the approxlmatlon lS not valld lS lndlcated [6]
In terms of two non dlmenslonal parameters.
The same flgure lndlcates zones In whlch the effects of permeablllty are
of no lmportance l.e. the zone of undralned behavlour; and a zone In whlch dynamlc
effects can be totally dlsregarded l.e. the zone of consolldatlon behavlour.
.Q.L = 1
a~ -
If dlsplacements are large thls assumptlon lS stlll valld provldlng an updatlng
of the spatlal coordlnates is carr led out In reasonably small lncrements of dlS-
placement ul Computatlon procedures whlch we shall now descrlbe can follow such a
pattern and the changes of domaln are taken care of automatlcally.
where Nand f:! stand for appropr1ate shape funct10ns and'y and E. for nodal
quant1t1es. In above
1/ =( u • v • w )T
where u,v,w are the Carteslan dlsplac8ment components. Eq.7 can now be wr1tten
uS1ng standard Galerk1n methods and Subst1tut1ng [1) as
MIi =9 [13 )
where
lj = k~ ~ )T ~ Y~ dQ
b
~ = Ef:t~Pft:idQ
In all of the above equat10ns Q represents the Current domain and lf
d1splacements are large an updat1ng of coord1nates 1S necessary at each step of
the computat1on.
Equat10ns [12) and [13) are supplemented by the const1tut1ve relat1onsh1p
[3J and an appropr1ate def1n1t1on of the sp1n vector of [6J. In the preVlOUS
express10ns the "standard" stra1n matr1ces were used. Now we have varlOUS
alternat1ves and the wrlter prefers to retaln the notat1on consistent w1th wr1t1ng
the stresses 1n a vector form. We thus deflne follow1ng Nayak [In 3 d1mens1ons)
264
o o o 0 -Wy .Wz
o o o ,Wx 0 - Wz
o o o -Wx .W y 0
0 -~Wz + rWy
-twz 0 -1Wx
-tWy -twx 0
(14 )
(15)
reducing to
( 16)
dQ' (17)
'!IT = (1,1 .1 .0 .0 ,0 l
The above equatlons wlth a speclfled constltut1ve relat10n complete the
d1scret1sed Solut1on Wh1Ch has to be solved now 1n the time doma1n.
algorithm to the dynamlc equatlon (12). Now we have for the n-th step
(18 )
from whlch ~ n.l can be found expllclty if a lumped matrlx form lS used and
Q~ • en .~n , ~-1 etc. are known.
The stress lncrement can be evaluated uSlng eqs. 16 and 17. Thus first
we compute ~B uSlng a difference form and then flnd
(19 )
(Note that In above calculatlons we have used a slmple Euler rule of forward
lntegratlon although wlth llttle addltlonal effort average values of Q~V2,~~1/2
(20)
From this the new set En.l lS easily obtalned and computation can proceed
further.
Stabllity of such staggered algorlthms has recently been lnvestlgated
thoroughly by Park-Felllpa.
In th~present case we note that
(a) as the permeablllty lncreases l.e. ~_CD the equatwn (13) becomes
approxlmate
(b) as the permeab~l~ty decreases i.e. ~- 0 then the seepage equat~on reduces
to • T.
?1!+9~=O
",0_
uY Q
dIlg' dn I
These yield tangent st~ffness matr~ces due to geometr~c (~n~tial stress) effects
and deta~ls of such calculat~ons can be found elsewhere.
It should be noted that here generally one determinat~on of such st~ffness
Const~tut~ve law
The br~ef scope of th~s paper permits only a very brief d~scuss~on of the
const~tut~ve relat~ons appl~cable to so~ls under dynam~c or quas~-stat~c cond~t~ons.
The bas~c models used generally fall into one or another of the two general
classes:
(1) elasto-plast~c or elasto-v~scoplast~c
(2) endochron~c.
d~ =J del) del]
Concludlng remarks
The fleld of problems descrlbed In thlS paper lS one In WhlCh both the formulatlon
as well as physlcal modeillng present some of the greatest challenges. It lS
hoped that this lnterlm statement of the state of the art wlll stlmulate further
contrlbutlons WhlCh wlll lncrease the efflclency and wldth of appllcation of the now
existing methodology.
268
Acknowledgements
The author is ~ndebted to research co-workers and students ~n the var~ous phases
of th~s work. Dr. D. Naylor. Dr. E. Hinton. Dr. G. Pande. Dr. C. T. Chang.
Mr. K. H. Leung and Mr. V. Norris have made part~cularly important contr~but~ons.
Thanks are also due to the Science Research Counc~l (U.K.) for support~ng some
of the members of the team.
References
6. Zienk~ew~cz. O. C. and Bettess. P.l Soil and other saturated porous mater~al
under trans~ent. dynam~c cond~tion. general formulation and validity of
various simpl~fied assumptions. Soils under Cycl~c and Transient Load~ng.
editors D. C. Zienkiewicz and G. N. Pande. John Wiley (to be publ~shed).
11. Z~enkiewicz. O. C.l Constitut~ve laws and numerical analys~s for so~l
foundat~on under stat~c, translent or cyclic loading. Proceed~ngs of Second
Internat~onal Conference on Behavlour of Offshore Structures (BOSS), London 1979.
17. Zlenklewicz, O. C., Leung, K. H., Hlnto~ E. and Chang, C. T.l Earth dam
analysls for earthquakes. Proceedings Conference on 'Oeslgn of Dams to
reslst Earthquakes', Inst. C1V. Eng. London, October 1-2, 1980.
270
12
- - -...-n:,
_ kP V; _ 1.2..L
n, - w l2 - n: f 2
2 2 T ?
n =J!l.l.. =n:? ( - )
V~ T
k= kl p, g k - kinemal ic permeability
T =2L/V, ,(=( 0' K,ln II p
= P K, /P, n
z K,/ p, (speed 01 sound In waler)
n " 0.33
c
101
A "-- -- - f -- - -....-
.0
It I
ib l
£v~O-l
(d )
2 6 8 10 12
N°ol cycles " 16 18 20 100
5'
0 10 20 30 ~o KYO
( b)
o·75 r--------------------------------------------------------,
0·50 ~------------------------------__tt---------------------_I
0.50 ~-------------...:.....----.l--_t-----------_j
0 . 750~-------------------75------------------~1~
0------------------~15
Ti me: s
( b)
[a) Data assumed for the analysis and the mesh of quadratlc
flnite element
[b) Base motlon input
[c) Deformed mesh plot [displacement x 2.0)
[d) Contours of excess pore water pressure bUlld up
273
3s .
6s .
10s .
( c)
3 s.
6 s.
10 s.
Contour unit =kN/m2
Contour SPQClng=25 kN/m2
(d)
Numerical Methods in Elasto-Plasticity -
A Comparative Study
A. SAMUELSSON, M. FROIER
Chalmers Umversity of Technology, G6teborg, Sweden
Summary
Problems involving nonviscid, quasistatic, small strain
elastoplasticity are discussed in different aspects: the
role of an objective rate of stress, the choice of varia-
tional formulation, the choice of time step procedure.
Introduction
The development of the mathematical theory of nonviscid,
quasistatic plasticity started with Tresca, Levy and St.
Venant in the middle of the 19th century. In the first part
of our century v Mises, Prandtl and Reuss established a
theory for small strain plasticity for an isotropic mate-
rial with isotropic hardening. Their work is still the basis
of the mathematical theory of plasticity. After the second
world war the theory was generalized and extended in many
directions by among others Hill, Prager, Drucker, and
Koiter.
Lately, the theory was redefined in the framework of func-
tional analysis, especially the theory of variational in-
equalities by among others Duvaut and Lions [1], and
Johnson [2]. It was, wlthin this theory, possible to prove
the existence of a solution to the cases with elastic -
perfectly plastic and elastic - linear hardening properties.
275
Questions
dO O d0 1 de 1
°1 (1)
dt 1+e 1 dt - 2 dt
(1+e 1 )
°1
e1
.
-°0 + °1
e1
(2)
*1j *ijkl.
T = S nkl (3)
with
1 kl i' kl
+ [vi (1-2v)]g J g - (a/g) s J s }
277
* ..
where ,1J is the Jaumann rate of a contravar1ant component
of the Kirchhoff stress, n
kl 1S the rate of a covariant com-
ponent of the Lagrang1an strain, ~ 1S the metric tensor in
the deformed body and
1J 1J kl/ 3 (5)
, - g gkl'
Further
kl.
a = 1 if s n kl > 0 and (6)
_ 1j kl
J 2 - gikgjlS s /2 = (J 2 )max
a = 0 else,
,* 1 j =,. i j ik, j 1 + g
+ 12( ' g j k, i l + g
i l, j k +
(8)
+ g Jl , 1k ) n. kl
1k sik lk lk
For small strain g ~ 0 ,nkl ~ E kl , , 00 ' the nomi-
nal stress, here denoted by a 1k • Combinatlon of (3) and (8)
and restriction to small strain and plane stress glves the
followlng tangent elastoplastic relation
l SYM
2 2 2 (10 )
A11 = A - Sx (B-SxS z ) /(A-S z ) - O'x g / G
2 (B-S y S Z}2/(A-S;) - O'yg/G
A22 A - Sy
with
A = g(1-v}/(1-2v}, B = gv/(1-2v}
The numerical experiment reported below indicates that the
effect of using the more correct stress rate definition is
significant also for small strain if the tangent modulus
also is small. There seems to be no practical reason not
to include the extra terms in the tangent matr~x.
(11 )
where
F(O'
~J'
~p} = 13/2\s .LJ I - (a Y+HS P ) (12)
and EP = 12/31s P
~J
I ~s proportional to the Euclidean length
of the 9-dimensional plast~c strain rate vector.
It is assumed that during plast~c flow = 0, that is F
~ •• - Hs'P
~v
= 0 ( 13)
O'ij ~J
and that the rate of plastic strain takes place in the di-
rection normal to F in the pOint O'ij:
279
(14 )
( 15)
(16 )
E ••
.p 1 ( 18)
-H~~
[
Now 1aF/aa.·1
1)
= 13/2 and ClF/dE P = -H so
(19 )
1
(a) -OJi,j = Ui (20)
1
(b) -2 (u
E l' J'
1,J. +u.J,l.)
(c)
- . .p on Q
EiJ - AijklOkl + Eij
'1
(d) -EI?(T . . -0 . . ) + HEP(n-E P ) >
lJ lJ lJ
(f) u 1 (t) = 0 on Q
u
(g) T. n.o ..
1 J Jl
T~
u = ¢ u (22)
!¢UdV + (23)
Q
(25)
(26)
(27)
(28)
and
(29 )
(30)
~f ~ T ~f ~ ~p ~ 2
(0 -T) A(o -T) + H(Et_~t-n) , (32)
"(T,n) EP
~f
o (33)
o (34)
f f
2T22 - 2VT 11 + 2(v0 11 -0 22 ) + A(2T 22 -T 11 ) 0
f
4(1+v)T 12 - 4(1+v)012 + 6AT 12 = 0
A.2H(Oy+Hn) = 0
2 2 2
+ T22 + 3T 12 - (Oy+Hn) = 0
MPa 101
56.9
III
~ 5107 II
I~ lum
5~.5
!~~
52.5
3.0m
K )II 'IIDTH·0.2m 1€ 1
750 2250 m~crostra~n
(a) (b)
y
u ; 0 FOR ALL NODES ALONG THIS Ll';E
x
~
J
...
.....
--I COMPUTA~ION~L SLPPORTS
x
(POI~T A
l'IITH Nt:LL REA-:~ION UP TO
I{OUNDOFF EI{I{O'lS
(c)
Numerical experiment
PIPE 1 025 1 050 1 075 1 100 1.125 1 150 1 175 1 200 1. 225 1 250 1.275
OBJECTIVE 782 874 966 1071 1352 2019 2830 4473 12803
STRESS
II
NOMINAL 78~ 874 966 1071 1352 2019 2814 4376 11742
STRESg
OBJECTIVE 781 866 952 1038 1249 1583 2056 2549 3434 5220 11846
STRESS
III
NOMINAL 731 866 952 1038 1249 1583 2056 2540 3403 5045 10687
STRESS
PIPS 1. 025 1. 050 1.075 1.100 1.125 1.1';0 1.175 1.200 1.225 1 250
PIPE 1 021i 1 050 1 075 1 100 1 125 1 150 1 175 1 200 1 22t; 1 250 1 275 1 300
§~
I", ~
o 0250
" "I '" 317 S., 1.!5J i'O]'j ?QOR 104:l' ~OOIJ 6132 7320
(l 0125 4.
I" 221 12S 60' 1253 2038 21:)10 1'146 5 '104 6115 7311
..
o 0500 127 795 7./ 1702 n''J 1811
'8'8
8"f;
,ji~
o 0250 45 127 21. ,"S
". / 1254 17)] 2208 2713 3286 3FJ19
o 012; '5 I" 21. ,., '03 700 1264 1132 2706 2735 3287 3811
E.. .
SYM·y·r.~.--.r---~----.----'--~
D ••• ~
i
C • • I
B • !
2 3 8
PIPf,.
, 025 , 0>0 , 015 , 'DO
, 12'> , 150 , 17' , ,"0 , ,", , 2')0 , '" , JOO
rirst 1A ,0
"
mcthod lA
Jr.
'" " C,'
It
" 7A <B
'"
li'
Second
" lA
'0 '" " <T'
1< 3B
mt·thod
2. 1, S. 7A AA
Concluslons
References
L. CORRADI, G. MAIER
Pohtecmco dl Milano, Italy
Sunnnary
(la,b)
o (~
--e = --e
D e (x)
--e - (2)
(3)
where: ~ is the elastic matrix; ~ and Ee denote the elastic and plas-
tic portion, respectively, of the total straIn. The plasticity condition
(here assumed to be "regular", i.e. to define a smooth yield locus) and
the associated incremental flow rule can be written as follows 17 I:
(4a,b)
~ > 0
e= ~e ~e 0 (4c-e)
(5)
293
() = D e (6)
(7)
(8a,b)
(8c-e)
E
-'e-
(x) = b(x)
---
E (9a,b)
(101 ,b)
with
(lla,b)
(12)
In order to formulate the plastcc portion of the elEment law, let a row
matrix .!l.(~ of interpolation functions be defined such that
(14)
Thus, as soon as matrix ~(~ is chosen, all the element matrices appe-
aring in eqs.(8) can be defined. Dlfferent element laws are obtained
with different ~(~, each involving different approximations 191. The
implications of some assumptions will be discussed subsequently. Howe-
ver, it is worth noting here that some fOYrrlllations earlier proposed
for the element plastic laws can be regarded as particular cases of the
present approach; e.g. the "yield criterion of the mean", successfully
used for limit analysis and shakedown problems I 121, is recovered from
the preceeding equations when matrix ~(!) reduces to the unit scalar.
The element properties must next be expressed in terms of nodal varia-
bles, as required for the assemblage. To this purpose, the equilibrium
and compatibility equations for the element have to be used. They read:
E: = C U (15)
-f=k-uu -u-k
-up n
L (16)
having set:
k Ct D C k = Ct D (17a, b)
-uu -up
These matrices represent the elastic stiffness matrix and the initial
strain matrix, respectively, consistent with the present approach.
(lSa,b)
(20a)
(20b)
Eqs. (20) provide the relations between matrices ~(20 and Q(20, which
must be fulfllled when ~ = i if eqs.(la) and (9a) (lSa) have to represent
the same strain field within the element.
However, the formulation of the element elastic-plastlc behaviour requi-
res also some connections between matnces Q(20 and M20, since they
both contribute to the representation of the strain field within the
element. In fact, on the basis of the relations of the preceeding sec-
tion, one can write both the following expressions for the plastic strain
rate field over the element
D (_x)
"'-€
= {3¢ e--e (x) = {3¢ e--e--
/3G } ~ e- /3G } A(x) ~ (21b)
296
where x· (j 1, •.• ,J) are the strain point coordinates. Then, the fol-
-J
lowing interpolation matrices can be assumed
b(x) = [... -
--
Ieob.Jex)
-
... J (23a,b)
eq.(la) in terms of~. To this purpose, note first that from eqs.(la),
(lSa) and (2Za) one obta1ns
(Z3)
On the other hand, if eqs.(la) and (9a) (lSa) have to provide the same
strain field, the follow1ng equal1ty must hold
(Z4)
Slnce -C-A
p
= 0, from eqs.(lS) and (18) (19) it follows that:
-
E = M 0 (ZSa,b)
where
M=CA
- --0
(26)
(Z7)
matrix ~(~ with the one obtained in tenns of Q(2S) from eq. (24) .
In thIS way, a consistent element is arrived at; however, its elastic
stiffness matrix ~, eq.(17a), has only rank SJ ,instead of D
If the additional rigid modes are not introduced, two different ele-
ment behavIors are obtained from eqs.(l) and (9), the latter beIng
"stiffer" than the fonner. This second type of inconsistency will be
called "artificial stiffening".
The preceeding discussion shows that, at least in principle, consis-
tent element fonnulation can be achieved starting from the assumptions
(22) (23) ,provided SJ ~ D . Note that, while the components of
vectors -E and -A coincide with the relevant field values at the strain
poir.ts, the same is not true, in general, for the components of 2".. and
~ . However, this coincidence can be enforced by selecting as strain
pomt locations -]x the coordinates of the points that would be used
to evaluate numerically by GaussIan quadrature an mtegral over the
element volume. This result can be readily arrived at if eqs.(9b) and
(13b) are numerically integrated, account taken of eqs.(23).
The above special chOIce of x turns out to be particularly conve-
-]
nient. In fact, the element constitutive law, eqs.(6-S), can be shown
to reduce, in this case, to the material law, eqs.(2-l) , enforced at
the strain (Gauss) points 110 I . Thus, the element properties defined
by eqs.(13b) and (14) need not to be evaluated, with a substantial
saving in computational effort. It can also be shown that the element
stiffness matrix ~ coincides with the customary expression, provi-
ded a numerIcal integration based on the J Gauss (strain) points is
used to evaluate It ,10 , . When SJ < D ,a "reduced integration" is
performed/which automatically generates the required additional rigid
modes in the element behaviour.
Thus, the familiar procedure for elastic-plastic finite element ana-
lysis which enforces the material law at the Gauss integration points
,l , ' can be regarded as a special case of the present approach, based
on the assumptions (22) (23) and on a particular choice of the strain
point coordinates.
The consistency requirements previously discussed might explain some
aspects of the numerical experience described in the literature. For
instance, the inaccurate solutions sometimes obtained when using an
excessive number of Gauss points (such that SJ > D) could be justified
by the fictitions redundancy implicitly introduced in the element
299
Assemblage relations
The equations governing the behaviour of the entire system, composed of
N finite elements, are derived below. Subscript n will denote varia-
bles pertaining to element n (n = 1, ... ,N); e.g. vectors ~ and in will
collect the nodal displacement and force components of element n.
Let ~ and f indicate the vectors of the nodal displacement and force
components, respectively, for the assembled system. For each element,
the followlllg customary connectivity relations can be written
u
-n
= -n-
L U (29a,b)
If -n
u and U are defined In the same reference frame, -n
L IS a Boolean
matrlX, containing at most a single unit entry in each row. The compa-
tibilIty and equIlIbrium equations for the discrete system, relating
the nodal variables of the dlscretized structure to the element genera-
lized strains and stresses, can be obtained from eqs.(29) and (15).
They read
~
= -n
C -n-
L U n = 1, ... ,N (30a)
N
F = L -n-n~
Lt Ct (30b)
n=l
Eqs.(30) and the element law, eqs.(6-8) , written for each element n, pro-
vide a complete formulation of the incremental elastic-plastic problem.
It is worth mentioning that, as shown in 191, the present formulation
fulfIlls all the requirements for the validity of some extremum proper-
ties which, in turn, permit the use of numerical methods based on mathe-
matical programming (specifically quadratic programming) for the solu-
tion of the analysis problem 17,141. This aspect will not be discussed
here; attention will be focused on the llffiit analysis problem alone.
Limit analysis
Let the finite element model in point be subjected to a given distribu-
tion of nodal loads fa. For perfectly plastic systems, the "collapse
multiplier" s is defined as the maxlffiUffi load multiplier t/I such that
loads t/lF-{) can be carried by the structure. The static theorem of limit
analysis 1111 states that s is the maximum value of t/I for which in
all elements generalized stresses ~ can be found which are in equili-
301
brium with loads ~Eo and do not violate the plasticity condition:
s = max ~ subject to :
(31)
In general, when -]
x· represent Gauss point coordinates, upper bounds on
the collapse load of the continuous structure are obtained, even if s
is less than the exact value of the kinematically admissible load mu1-
tip1ier associated with the assumed displacement model. This remark ex-
302
plains why accurate estimates of the collapse load are often obtained
even by means of canparatively "stiff" finite elEment models.
Examples
Sane very simple examples are considered below in order to illustrate
sane of the concepts discussed in what preceeds. Consider first a can-
tilever beam of uniform cross section (with flexural rigidity EI and
limit moment M ). The beam is acted upon by a concentrated force F at
p
the free end. The beam is modelled with a single finite element, with
a cubic displac~lent field; thus, the strain (curvature) distributlon
in the beam is linear.
15 r-------r-----~------_.------_,
0.. J-3(con.)
~
...... 1.0 1-----I-+--~---e-k4-o-c-t- - - I
o 1.0
Since the sectional behaviour was assumed as perfectly plastic, the "ex-
act" load-dlsplacement relationship results in a linear plot up to the
value FL = Mp/£; at thlS point, the load cannot further increase and
collapse occurs. This relation is depicted in fig.l, together with the
diagrams obtained with different element laws.
Since I = 4 and R = 2, a consistent element is obtained with the use of
J = 2 strain points. They were located at the Gauss point coordinates
and matrices QC20 and ~(20 were defined according to eqs. (22) (23). The
plot indicates the same behaviour as the exact one, even if the collapse
load is overestimated by about 29%.
303
The two procedures are fully equIvalent and the dashed line in fig.l
shows that the fictitious redundancy has thus disappeared.
--
20
1 5 L1
cubic element:
x l ,2 = £(1 + 13) /2;
Co
:!; 1 0
LJ./ ~ ~ Q1
quartic element:
....
'- V 'quartic x l ,3 £(1 :;: 13) /2,
...-'
/ x2 £/2.
VCUb'l
o 1 00
00 25 50 75
(from ref. lSi), where the collapse loads obtained for consistent
304
-e
£ (r) = {£ s (32)
Q(r) (34)
0 0 - -sine
r- - -2 -3 r
sine
0 0 -2 -6 r
0 sine
Fram eqs.(34) and (11), the expression of ~(r) can be readily derived.
305
Two strain points can be assumed in the element and matrix ~(r) can be
defined accordingly. If the constitutive law were directly imposed at
the strain points, a fictitious redundancy would show up, since SJ = 8
is greater than D. However, the ,inconsistency can be removed through
the use of eqs.(28); if strain points are located at Gauss points, the
element yield function will result in the following two-component vector
(35)
Conclusions
Elastic-plastic and limit analysis are areas of non-linear structural
mechanics where the finite element method turns out to be, once again,
a powerful tool for accurate numerical solutions in complex situations,
but exhibit some aspects and requirements quite distinct and partly
novel with respect to elastic analysis. Some of these peculiarities,
considered in the paper, basically concern non trivial, sometimes over-
looked "consistency" conditions, to be fulfilled in modelling the ele-
ment fields, or, equivalently, in formulating element plastic laws in
(element) generalized variables; it has been shown that these should be
taken into account in order to avoid sistematic and possibly signifi-
cant source of inaccuracy.
306
Aknowledgment
The financial support of C.N.R. (National (Italian) Research Council)
is gratefully aknowledged.
References
1 Nayak, G.C.; Zienkiewicz, O.C.: Elastic-plastic stress analysis. A
Generalization for various constitutive relations including strain
softening. Int. J. Num. Meth. Engrg., 5, (1972), 113-135.
2 Zienkiewicz, O.C.: The finite element method. McGraw-Hill, London,
1977.
3 Besseling, J.F.: The force method and its application in plasticity
problems. Computers and Structures, 8, (1978) 323-330.
4 Besseling, J.F.: Finite element method. 53-78, in: Trends in Solid
Mechanics, Besseling and van der Heijden edts., Sijthoff & Noordhoff,
1979.
5 Argyris, J.H.: Continua and discontinua. 11-189, In: Proc. 1st
Conf. on Matrix Methods in Structural Mechanics, Dayton, Ohio,
AFFDL TR. 66.80, 1966.
6 Scharpf, D.W.: A new method of stress calculation in the matrix dis-
placement analysis. Computers and Structures, 8, (1978) 465-477.
7 Maier, G.: A matrix structural theory of piecewiselinear plasticity
with interacting yield planes. Meccanica, 5 (1970) 54-66.
8 Corradi, L.: On compatible finite element models for elastic plas-
tic analysis. Meccanica, 13 (1978) 133-150.
9 Corradi, L.: On some "consistent" finite element approximations in
non-linear structural analysis. Proc. 5th AIMETA Conference,
Palermo, Italy, Oct. 1980.
10 Corradi, L.; Gioda, G.: On the finite element modelling of elastic-
plastic behaviour with reference to geotechnical problems. 1st. Int.
Conf. on Numerical Methods for Non-Linear Problems, Swansea, U.K.
Sept. 1980.
11 Martin, J.: Plasticity. MIT Press, 1975.
12 Dang Hung, N.; Konig, J.A.: A finite element formulation of the
shakedown problem using a yield criterion of the mean. Compo Meth.
Appl. Mech. Engrg., 8 (1976) 179-192.
13 Hodge, P.G. Jr.: A Consistent finite element model for the two-dimen-
sional continuum. Ingenieur Archiv, 39 (1970) 375-383.
14 Cohn, M.Z.; Maier, G. Eds.: Engineering Plasticity by Mathematical
Programming. Pergamon Press, New York, 1979.
15 Dang Hung, N.; Trapletti, M.; Ransart, D.: Bornes quasi inferieures
et bornes superieures de la pression de ruine de coques de revolution
par la methode des elements finis et par la programmation nonlineare
Int. J. Nonlinear Mech., 13 (1979) 79-102.
A Finite Element Method for Problems in Perfect Plasticity
Using Discontinuous Trial Functions
C. JOHNSON, R. SCOTT
Chalmers UniversIty of Techno}ogy, Goteborg, Sweden
E = Ao (Hooke's law),
take B = {t E IR?:
s
It I < 1} , where 1·1 is the Euclidean norm
,
~n
/R9 , i. e.,
3
Itl= ( L t~j)'.
~, j =1
For simpl~city, let us assume that B is bounded. Together,
309
-diva =f in n , (1.1b)
u = 0 on r , (1.1c)
da ij
(diva) i dX j
3
E
j =1
H = {or = ('[ ~J
.. ):'[ ~J
.. = '[ J~
.. E IJ 2 (n)} ,
Thus
if -dlv T f,
Inf [(T,s(v»-(f,v)] = {O
vEV -co otherwise,
and hence (1.2) can be wrltten
one has
,r 2
Irl < 1,
1
if
g (r) = t
r-! if Irl > 1.
The funct10nal F:V + /R 1S not coerC1ve on V and thus the pro-
Furthermore, one can prove (cf. [3], [4], [5]) that for
< p ~ 3/2,
problem
Inf F(v), (1.7)
vEB~ (n)
admits a solut~on u E rlb(n). Here the domain of F is extended
to B~(n) by defining for v E rlb(n)
2
F(v) = Sup [-,IITII +<T,E(V» - (f,v)], (1 .8)
TEPnC(Q)s
where c(n)s denotes the set of symmetr~c tensors with compo-
nents in C(Q) = {cont~nuous functions on Q} and <','> de-
notes the duality between C(Q)s and M(Q)s extending the scalar
product (T, E (v) ) •
The safe load hypothesis reads as follows:
X E C{~)s' (1.9b)
P {T E P:div T E [L 3 (n)]3},
L (T ,v) = -,II Til 2_ (div T ,v) - (f ,v) ,
and assume that (1.9a) holds and f E [L 3 (n)]3. Then, one
can prove (see [3], [5]) that the functional L:P x ~ ~ ~
that
V(T,V) E P x V, (1 . 10)
or, equivalently,
written
i.e. ,
o = lIE (u) •
C
h
= {K} be a triangulation of ~ and introduce
2
{v E [L 2 (~)] :vl K E iPk(K), K E C h },
f T t: .. (v)dx.
K 1J 1J
that
2.2 Convergence
def1ned by
2
.!t(T,V) -!IITII + [T,s(v)lh - (f,v),
where
(2.4)
(2 • 5b)
(Jh + (J weakly in H,
uh + u weak* in Bb W) I
Green's formula
(2.6)
(2.7)
so that
IloJI ~ C, (2.8)
weakly in H.
o
Further, by (2.9) it follows, using the fact that BD(~) is
weak* in Ifu(~),
in [L (~)]2.
P
It rema~ns to pass to the l~mit in (2.5). Given w E V there
and thus
(a,E(w)) = (f,w) Vw E V,
Le.
-(diva,w) = (f,w) Vw E V.
F~nally, by choosing T ~n (2.5a) smooth we f~nd since then
that
- 2
o~ lim{-,llahll + (Oh,T) + (d~vT,uh) + (f,u h )}
3. A numerical experiment
Here l'
=
dT
dx and ( ., .) denotes the scalar product in L2 (I) .
fIx) = Af(x) , x E I,
where
,
for 0 < x < 1
:2 ,
fIx)
-
{ :-1 for < x < 1.
tric around x ~) :
A 3
u(x) '6(x - ~)
4
for 0 < A < 12,
1
~(x2 - TI)
o (x)
2
3 A
AX
u(x) -6- + ( 1 - "8)x,
for 12 < A < 16,
AX 2 A
o (x) -2- + 1 - "8 ,
and that no solution exists for A > 16. We note that the
A = 15:
N =9 N = 19 N = 39
°0 °1 <5
2 °3 °0 °1 °2 °3
°0 °1 °2
°3
jUlll' 0.25 0.18 0.11 .0 0.25 0.22 0.18 0 0.25 0.23 0.21 .0
in~
ciently close to !.
323
u
->.--___ 1
x=1 x
Fig. 1
<5 (3) ,
0 3 (3)
as follows:
where
References
M. A CRISFIELD
Transport and Road Research Laboratory, Crowthorne, Berkshire, U. K
Introductl.on
For a number of years, the author has used the finite ele-
ment method to investigate the collapse strength of thin
plated steel structures [1-3]. The work has been directed
primarily towards steel bridges which are usually fabricated
from engineering steel for which the stress-strain curve
exhibits a significant plateau. The collapse behaviour
usually involves an interaction between material and geo-
metric non-ll.nearities and is influenced by initial geo-
metric imperfections and residual welding stresses. The
present communication descrl.bes a number of numerical
techniques that the author has developed in order to analyse
such structures. The topics covered l.nclude approximate
yield criteria, accelerated iteratl.ve methods and incremen-
tal solutions using a 'length constraint'.
Using the von Mises yield criter10n and assuming that the
equivalent stress is at yield throughout the depth of the
plate or shell, Ilyushin [7] derived a yield surface involv-
ing the six stress resultants M ,M,M ,N,N and N To
x Y xy x Y xy
this end, he defined the follow1ng non-dimensional quadratic
stress intensities.
1 (N2 + N2 - N N + 3 N 2) N
Qt X Y X Y xy
N2 N2
0 0
l:..(M 2 + - M M + 3 M 2)
M
Qm H2
xy
M2 X Y X Y M2
0 0 ••• (1)
1 - 10M N
Qtm N"11(Mx Nx + My NY 2 x Y
o 0
102MYNx + 3 M N ) MN
xy xy MN
o 0
e i2 e iO
<P , 11 ... (3)
eil eil
327
1 (4)
Robinson [8] has shown that this is the best of the various
approximations that are llnear in Qt' Qm' Qtm space. The
author [1] invoked normality and treated the yield function
in a flow sense to derive elasto-plastic modular matrices
relating the increments of the stress resultants to the
increments of the total 'strain'. The matrices were incor-
porated in an incremental/iterative finite element formu-
lation and appl1ed to a number of problems involving the
collapse analysis of imperfect steel plates. Reasonable
agreement was obtained with both alternative solutions
using a layered von Mises approach and with experimental
resul ts [ 1] .
1 ( 5)
~x~s =(3EO:Y
= (3EO:r
E is Young's modulus and ~ '~X and ~Xpxy are the incre-
px py
mental plastic curvatures.
As X +00, a + 1 and F2 (equation (5)) coincides with the
ps
original Ilysushin approximation of equation (4). When
Xps = 0, a = 0.6 and equation (5) gives
1 (8)
(9)
- ~ 1 ... (10)
_K.- l
where /).
-~
_~ 2'i ... (12)
v :: free
10, ~ ~ 10rl----------------------------------------~
09 09
08 08
"C
] 07 E 07
.t:: .t::
~ 06 ~06
0- .g
~ "C
"C
.3 05 !!l05
...J
04
02
01 01
OLr__-L__-L__ ~ __ __'L-~L- __L-__ ~ __ ~ __ ~ __J
o·
o 02 04 06 08 10 12 14 16 18 20 o 02 04 06 08 10 12 14 16 18 20
Shortening/squash shortening Shortening/squash shortening
la) IMPERFECTION CJ. o= 0001b (b) IMPERFECTION CJ.o = 0005b
FIg 1 NON-DIMENSIONAL LOAD SHORTENING RELATIONSHIPS FOR UNIAXIALLY COMPRESSED SIMPLY SUPPORTED PLATES
331
-1
~i-l according to
-1 T
K.
-1 -1 ~i-l ~i-l Yi ~i-l
-l ~i-l T
y. o T
~i-l _l -l-l Yi
+~ +
h
T
II i-l
-1
~i-l y.
~i-l
-l
T
Y
Y'
' 1T
l -l-
~l-l
~i-l
y.
_l
T
.. (13)
where y.
_l g.
-l
- 2i-l
... (14)
*
o. -1
• .. (15)
_l - ~i-l 2i
o. * *
-l hl~i + filli-l~i-l + (1 - hi)~i-l • •• (16)
*
where ~i-l is given by
* -1
- K
-1
• •• (17)
~i-l ~i-l 'h-l -a
a.
l
h.
l ~
l
... (18)
d. t.
f.
l
- ~
l
+ (h.l - 1) (1 - -.2:)
b.
l l
ai' b i , d.l and t.l are scalars glven by the inner products
332
a.
~
b.
~
• •• (19)
d;
*
=(0. -
* T
0'1) g .
.,. -~ -~- -~
*
(~1'
* 1) Ty .
o.
.-
-
-~- ~
••• (20)
* . .. (21)
~i = h~~i + elni-l~i-l
c.
e.
~
hi(l - ~)
b. - 1 ... (22)
~
and c. * T
o. . .. (23)
~ -1 Yi
The two methods (equations (16) and (21)) have been applied
to a number of elastic and elasto-plastic large-deflection
problems [13]. Both techniques have been found to require
substantially less iterations than the standard mNR method.
The extra computation per iteration is almost negligable,
the only penalty being the extra storage. In general,
333
0020
Fixed load
0015
N
E
E
Z Fixed length Increments
0-
,j
..... I = no of accelerated Iterations
~ 0010 ii, JI J = no of standard m N·R Iterations
"a.
E
~c
:::)
0050
R = 2540mm, L = B = 508mm
E = 3105 N/mm2, v = 0 3, t = 3 175mm
OL-----------------~--------------~~--------------~
o 10 20 30
o (24)
(25)
(26)
O.
-1
-K
-1
g(A. + OA i)
- 1 _o.1 (A.)
1 + OA.1 ~T ... (28)
where o (A.)
-1 1
-K
-1
g. (A .)
-1 1
... (29)
-1
~T K g • •• (30)
lip. + O. (31)
-1 -1
o (32)
(33)
Acknowledgements
The work descrlbed in this Paper forms part of the programme
of the Transport and Road Research Laboratory and the Paper
is published by permission of the Director.
References
J PAULUN, E. STEIN'
Techmsche Umversltat Hannover, Germany
Summary
The deformation process of stiffened plates (e.g. box gir-
ders for bridges) with increasing stat1c compression loads
until critical states is investigated in the frame of geo-
metrical nonl1near theory. Instead of geometrical imper-
fections a constant normal load is applied on the plate.
The elastic-plastic mater1al is described accord1ng to v.
Mises with an isotropic hardening. Partly plastifY1ng over
the thickness of the plate 1S considered in the FE-descript-
ion by using elements with the full plate thickness. So,
neither layered elements nor smeared elements for plate and
stiffeners are used. Two methods for the numerical treatment
of incremental elast1c-plastic compression and bend1ng are
developed and compared, the first follow1ng a proposal of
Pfluger and the second improving the first by an equilibrium
iteration 1n order to reduce the errors with respect to the
Prandtl-Reuss equations. Up to now a Tresca-material was
mostly used for plate bending. Concering the FEM, rectangu-
lar elements (12-parametric for bending and 8-parametric for
plane stress) are used for the plate and beam elements for
the stiffeners. Calculat10ns w1th various st1ffness para-
meters give insight into the load carry1ng behaviour near
critical deformation states.
Introduction
(2.2)
• •[ 1"3
~p = A ~ - tr (~)!
] • (2.4)
F R U (2.5)
U = 1 + E~ , (2.6)
F R det F (2.7)
b -1 • T
T := (det F) F S F (2.8)
- -
has the relation with the Jaumann stress flux ~
b 0
- --
T = T - T D - D T + tr(D)T (2.9)
0
b
T = 2G~ v
+ 1-2V tr (!?) 2J = T. (2.11)
(2.13)
we finally get
(2.14 )
o (3.4)
T
with oE N = sym {oF N FN }, the variation of the displacements
by the-variation ;f ~he increments, oyN + o~y, and corre-
spondingly o~N + sym {o~FTFN}. From (3.4) follows the sim-
plification in (3.3) as was shown by Klee, Paulun [9]
L L
~E o~E (3.6)
345
L L T N
J tr(o~E ~S)dV + J tr(o~P ~p S )dV (3.8)
Vo Vo
Por purely elastic deformations also the second term in
(3.8) vanishes. In the case of elasto-plastic behaviour
this second term in (3.8) can only be neglected if the
achieved yield stress Y is small compared with the plastic
tangent modulus ~
(3.10)
N
- x (3. 11 )
+ + T
~_ = ~N-1 + v~2 + w~3 RN Lu. (3.12 )
+ dt at.
:= l N. (\/) N.
For the local dlsplacement gradlent H dX axJ -~ -J
i t follows
+
LF := FN+ 1 - FN RN H (3.13)
347
(3.15)
RN
~k ® ~k ::
N
Rjk N.®
-]
N
~k w1th Rjk N.
-]
. ~k·
One gets
N + + N
~Uj Rjk uk or uk Rkj ~Uj . (3.16)
n • +
~a(a= 1,2).(3.17)
-3
348
--7-- -.,.
T( +
!J2
n. local base vectors
_1 in state N
+ + +
w,ac =U3,0I. = !J3 • !Joe
+
For the local rotational tensor R it yields
+ +
<P -w ,1
+ + + + +
+ H -<P -w :::: R (n. ·n.)N.®N
,2 -1 -J _1 -J
+ +
w,1 w,2
N.® N.
-1 -J
. + 1 + +
with the average inplane rotat1on <p 2(u 1 ,2 - u 2 ,1) :::: O. =
For the increments with respect to global directions we get
with lIa 1 = ~1
+.
(~2 -
~2) , lIa 2 = ~2
+
(~3 ~3) and lIa 3 =
. -
~3
.+
(~1 -
~1) . These rotational increments are calculated
with the transformation
RN +
H. (3.18)
349
N + +
La 1 R11 ¢ + R 13 w, 2
N + N + (3.19 )
La 2 -R 21 w , 1 - R22 w,2
N + N +
La 3 -R 32 ¢ + R33 w , 1
·D (4. 1 )
nE
with
:= E -"31 tr(E)1.
--
*+ + *+ ---- + +
.-
.-
.-
.- / "- / , ..-
..-
directions of main
stresses before
--
.-
buckl1ng
-
.- directions of main
---- --
stresses due to
* +
*
buckling
r- I-
r-
e.... / "- / , l-
~
(4.2)
cos a (4.3)
1/2
3 (E - T) (4.4)
3E - (1-2v)T
351
Yleld surface f a
·pl
= o. The relations (4.2 - 4.4) lead after some trans-
-
E
-
format10ns to mod1fied elasticity constants, compare Pfluger
[15] ,
lI/ 3 (1-n)
E ":::"2-;-(-::-1+:-v-;)~+""':"""(71:--n"""")'-:-:(1'-"2'--v') E , (4.5)
T
n = no = 1 - E". (4.7)
10
, 110
1,0 T
E
T
fig. 5. no in dependence of E and v
*
The values E and illv are variable over the thickness and the
coordinates of the middle surface as well.
2
I
I SP middle surface
-+-.- of the plate
I
I n
GP
process
(5.1)
., !!1
tD ~I: @ +
v
+ +
1:13 w,1 w,2
+'T+' +'
The element oriented inner virtual works ov 1 kl v 1 can be
developed very similar to the geometrically linear elastic
case, see Stein, Wunderlich [1~, e.g. Puttlng all kine-
+
matic variables together in vector h we get the lntegral
- +T* +
of the inner virtual work for an element i f o~ ~ ~ dA,
;\- + + + + + +
with n {u,1 v,2 u,2 v,1 w,11 w,22 w,12 w,21}
ill
li
[<
tf1
t3
Q 12 f2
Q
*
j *
f1
ill
f2 *fp + *fG (5.2)
* E* *\I 0 0
* N
0
N
0
fp ~ ~ °11 °12
1 -\I
*
\I 0 0
N N
0
1-~ ° 11 0 °21
0 0 -2- 0
N
0 0 0
1-~ °21
N 0
°22 0
-2-
N N
0
°12 0 °22
(5.3)
+i (5.4)
v
follows from (3.16) and (3. 19). For a node k the transfor-
mation matrix has the form, see (3.16) and (3.19)
+
u flu 1
+
v flu 2
+
w flu 3
+ (5.5)
lIa 1
w,1
lIa2
+
w,2 l',a 3
with N
(5.6)
+
o 0 0 X3k 0
~P
000 (5.7)
o o
[~ - (5.8)
A S!. A
(5.9)
(5.10)
(5.12)
(5.13 )
358
6. Examples
In the f1rst example, f1g. 8, membrane stresses and hori-
zontal displacements of the m1ddle surface of a slmply
supported square plate under uniform load are calculated.
144 square elements in a quarter of the plate were applied.
These elements have 4 nodes with 5 DOF (degrees of freedom)
in each node. The compression stresses in a ringlike outer
reg10n effect the supporting of the inner tens10ned plate
1n states of large deflections. Comb1ned w1th this behav10ur
we see hor1zontal displacements of the support lines which
1S known as the pillow effect.
E = 21 000 kN/cm 2
v = 0,3
t = 0,.5 em
00076 load
[cm[QOt;07;/;;6--'----'---'---'---'--'---'---L-'-'-+---------'~
q = 1 N/cn,2
r
120cm
load step
6q = 0,2 q
girder
s
(
~7
I
w{",,,.'
/;. Poin'
D
A} Glrkmann
Point 8
";::===::::::::j E • 21 000 kN 2
ern
v = 0.3
y = 2 1. kN 2
o ern
(. C" 100. kN 2
ern
8 load i nc rements
--....;;::j APi = 0 . 08 ~P 1
(1 = 2, . . . , 8)
~YER
POINTS _ . - , -~y-=- ._.-.-armlr SECTION I-I
7. Conclusions
References
O. M. EIDSHEIM, P. K. LARSEN
The NorwegIan Institute of Techno!ogy, Trondheim, Norway
Abstract
Constitutive laws for elasto-plastic shells based on strains
and curvatures of the middle surface and the corresponding
stress resultants and stress couples are investigated. The
stress-strain relationships are obtained using the classical
theory of plasticity for work-hardening material, expressed
in generalized variables. Two expressions for the consequ-
tive Yleld - condi tion are studied, and modified to lmprove
the behaVlour. The models are verlfied against results ob-
tained by numerical integration through the thickness for
the special case of bending and stretching of beams.
Introductlon
- f ~u
T-
~Tds
S
an
(1)
(2)
where
l:: ::1
Note that Egs. (1) and (2) are based on Love's first order
shell theory, and that no explicite dependence on the trans-
verse shear forces is indicated in Eg. (2).
Elastic materials
{::} +e ]{: }
by Hooke's law, which on matrix form is written as
wlth
367
.! 0
h 0
and
-v
1
o
o (a. 1,2)
dA (4 a,b)
( 5)
+ 2 NM)~ 1 (6)
NoMo
2
N N2 2 - N
+ N22 N + 3N 2 (7a)
11 I I 22 12
_2
M M2 + M2 - M M + 3M 2 (7b)
I I 22 I I 22 I 2
NM N
II
M
I I
+ N M
22 22
- ~(N
I I
M
22
+ N M ) + 3N M
22 I 1 I 2 I 2
(7c)
and
1
N ha M h 2a (8)
0 0 0 "6 0
The initial yield condition 1S g1ven by Eq. (6) when the posi-
tive value of ± NM is chosen
f F
o
- 1 o ( 9)
N
L
= ho
0
M
L
= "23 M
1 2
4" h 0 0
(ll)
1
f F - 1 (12)
c c
13
(13)
M (14)
370
with
-P
K
J_2_(dKP2
o 13 11
- + _1_ INMI) ~
(~ + -M
F*
c N2
L
a 2 M2
L 3
r.:
bNLM L
1 (16)
08+-----~--~~--~----~--~
02+---~-----+-----r~~+-~~
o2 o4 o6 o8 1 0
N
"N..
Fig.l - Initial and limit yield
surfaces for beam-column
Horizontal plateau
Hardening regime
--~-
Elastic - perfectly plastic
Model
-------1
Crlsfleld's Yleld surface, Eq. (12), may be rewrltten as
2 2
(N + M 4s NM )!:>
16 + - 0 (20)
h 2 a 2h 4 13 ah 3 0
where
s = INMI
NM
ao dOC
c
~
dN + ~ dM as
as
a.S as
dA (22)
dO c dH
dO c __ dO C
__ aa dK
-P
dH
P + -P- -P
dE dN dK dM da dK aA
as as as as
373
whlch Ylelds
H' • 1
( 23a)
ho
c
1
H' ( 23b)
ho
c
da
-p
The equlvalent plastlc straln lncrement dE may be expressed
in terms of the invarlants of the plastlc strain and curva-
ture lncrements. Alternatlvely, the following expressions
are appllcable /5/
or
- -P 1
dE (25)
he;:
c
Model 2
o (26)
where
2 _ 2
N M*
( - + 36 (27)
h2 h4
374
daB daB
NaS dN aS + dM aS
d>"
~S
(28)
dH
p
daB
+
dH
p
daB
- --
daB
dM*
daB
. B
d£aS dN aS dK aS dM aS
as
dM aS
B (29)
Numerlcal studles
1,0 ~ ~~
-, =
M ~
ML 0,8 /
I~
II
0,6
Thickn . Integration
L
0,4 ---- Model 1 -
.. .. . .. Mode l 2
V
0, 2
K
K
0 .0
0,0 0, 4 0,8 1,2 1,6 2 ,0 2, 4 2 ,8 3, 2 L
1,0 -- .. ..
M
ML .-/ ~
b' ,, '
0,8
V ····
0,6
L '
;/
0,d
;: /
. >'
----
Thickn . Integrati on
Model 1
./
.. ... ........
;/ Model 2
V
0, ?
K
K
0,u L
0,0 0 ,4 0,8 1,2 1 ,6 2 ,0 2, 4 2, 8 3, 2
M 1,0 -- .=
M
./ I~
.' ..
q.~
0,8
.....
'
L
~
0,6 /'
/. .'
I
I-
~
0,4
;/ Thickn . Integra tion
-
---- Model l'
V
... .... .. ....
/ Model 2-
V
0,2
0,0
:) , 0 0 ,4 0,8 1, 2 1 ,6 2, a 2,4 2, 8 3,2
"
1,0
-...........~ "'''-'....~
0,8 0,8 ~,
.,
1\,,
~
0,6
~.,-, '\ ',
0,6
\ 1\\,'. ' \\1\\
v.
\; '\
0,4 0, 4
0, 2 0, 2
0 ,0 0, 0
0,0 0, 2 0 ,4 0 ,6 0,8 1, 0 0, 0 0 ,2 0,4 0,6 0,8 1 ,0
N N
Thickn . Integrati "" N Th ic kn . Integrat i on
-L
Mode l Mod e l I"
1,0
N
NL //
<,"~
~
0,8
//V
,-
0,6
/
/
V I
1/
.I
Thic kn . Integrati on
0,4 I
---- Model I
,V . . . ... . Mode l 2
1/
V
0, 2
0, 0
0,0 0,4 0,8 1, 2 1, 6 2,0 2, 4 2 ,8 3, 2
1,0
-
M
ML
0,8
/~~ ..:::- :: :::::: ::::::::::: ::: ::: ::~~:;~.~.~ ..
/ r-FL'I
/
0,6
Thi ckn. Inlegroli on
0,4 Model -
0, 2 / Model 2
0,0 ~1/---+----~--4---~---4----+---4---~
0,0 0 ,4 0,8 1,2 1,6 2,0 2, 4 2,8 3,2
0, 2 / Mode l 2
/
0 , 0 ¥----+----l---I----l--~
0,0 0 ,4 0,8 1,2 1,6 2,0
0 ,4
-
0,2
0,0
0, 0 0,4 0,8 1, 2 1, 6 1,b 2, 0
0,0 0,4 0,8 1, 2
K
,.
Thickn _ Integrotion Thickn _ IntegroHon
KL
Model ---- Mod,,1
Model 2 ... .. .. ...... Mod,,1 2"
1 , 0 ~~~r----'-----r----'----.
0,8+-----r-~~~---r----+_--~
Thickn Integroti on
Model
0 , 4+----.r---~~---r--~~--~
Model 2
o,o ~--~~---+-----r----+---~
Work-hardening materials
0=0
o
20
o
For the case of pure bend~ng the accuracy of the two models
~s comparable with that of the non-hardening case. The re-
sults of bending or stretch~ng of a prestrained section are
given ~n F~gs. 13 and 14. From F~g. 13 ~t ~s clear that the
correct l~m~t case M = 0 and N = NL ~s not reached for model
2*, which ~s due to the fact that the quant~ty M* ~n the ex-
pression for the equ~valent stress goes asymptot~cally to-
wards the value M* = 0.25 ML ~nstead of the correct value
M* = O. For the proport~onal load~ng path K/KL = 2 E/EL
both models reproduce the correct moment-curvature relation-
sh~p, wh~le the relationsh~p between Nand E shows greater
deviations, F~g. 15. For the case K/KL = 0.5 E/E L , however,
the opposite ~s the case, and the moment-curvature relation-
ships display s~gnif~cant deviation from the correct one,
F~g. 16. F~nally, the interaction diagram for the two cases
are depicted in F~g. 17. The l~m~t surface FL =1 ~s here
exceeded for both models. Th~s is due to the fact that f~nite
M \ ,.
r--r-'--'--'~~~-'~~-.--r-.--.
.1 , 2 -1 . 0 -0,8 1. 0 1, 2
Thickn . Integrati on
·O, B Model
~ ---- Model 2
1,0
N
NL 0,8
0,6
Thickn. In tegra ti on
0, 2
0,0
0,0 0,4 0,8 1, 2 1, 6 2, 0 2 , 4 2,8 3, 2
1,0 ~~
....
~
ML / ~ r"....-
V.
0, 8
//
0,6
;/" Thickn , In tegration
0, 4 / ' :I ----
V
Model I'
... . ... .. , .. . Model 2-
V
,/
0, 2
0 ,0
! I
0, 0 0,4 0,8 1, 2 1 ,6 2 ,0 2 ,4 2 ,8 3, 2
0 , 8~----.---~-----r----.----.
0,6+----+----~--~--_+--__1
V
0,2+--F~r---_+----~----t---~
Model 2*
O , O f---~r---_+----4-----t---~
0,0 0,4 0,8 1,2 1,6 2,0 £
L
:::~±0~:::
Thickn . Integralion
Model . "
-~-----==--
Mode l 2'
0,0 ~---;-----r----+-----~--_;
K
0,0 0,4 0 ,8 1 ,2 1,6 2,0 K
L
M
ML
1,0 '-
-- - 1
0,8
0 , 6+----+-+.~~--_P~~----1
° + ----,f-\-
Th ickn . In te gration
,4 Mode l ,"
Model 2 ·
N
0,0 ~----~---+~~~----+---~ NL
0,0 0, 2 0 ,4 0,6 0,8 1,0
References
J BANOVEC
Umverslty Edvard Kardel] ofL]ubl]ana, YugoslavIa
Summary
A general rrethod for the elastic - plastic analysis of plane frames in-
cluding nonlinear gec:rcetric effects (large displacarentc;, large lquadra-
ticl strains) is presented. A mixed type element with three degrees of
freedan d. o. f. at earn end, and with n (n = J ,2,3, ... ) internal d.o.f.
aloog the element is proposed. The mixed approach W1th the approximation
of the axial force and the curvature presents better the elastic -
plastic state than the well kna,.m displacement approach.
Introduction
Plane frarres are alrrost the rrnst frequent structures in the civil engi-
neering and the nonlinear analysis of these structures has attained coo-
siderable interest. Many publicatioos treat ooly material nonlinearity
(elasto - plastic frarre) or only geanetrical noolinearity (large dis-
placarents and quadratic strains; large displacerrents and small strains;
the secood order effects). cnly a few papers analyse both ncnlinearities,
e. g. [Ref. 3) to [5). Mas t of these rrethods separate the frarre colunns
and beams into large nunber of elements. Usually element unifonn loads
IlU.lSt be replaced by adequate nndal forces and m:xrents.
I
I
I
I
t
Kinematics
'ltle derivation starts with the expression for the extensiooal strain £
(1)
387
or
D_
~ =
1/10+2~ex
V·· dU
dx
+ dudu
dx dx
-1.0 (2)
du
(1 + ~)CX)S <P = cos tjJ + dx
(3)
dw
(1 + ~)sin <P = - sin tjJ - dx
du du ~ dw ~
- = -a + -a (4)
dx dx x dx z
du
(1 + ~) (oos <p - 1) = cos tjJ + dx - 1 - ~
and assuming that the reference systan is chosen so that the local rota-
tion <p is a small order quantity (Fig. 1):
(5)
the left sides of (3*) are expanded into Taylor's series aboot the point
<p = 0.0:
ClF ;/F co2
(6)
F = (1 + ~) (oos (!> - 1) ~ F[ + Cl<P (D + -Cl<p-2 2
(peO.O
<0=0.0 <[)=O.O
D = du +cosljJ-1+ ~<p2
N dx 2
(8)
dw . d,
~ <p = - dx - Sl.n '"
~ = (1 + ~)I (9 )
<1>=0.0
(10)
If <p is small (5), the rigid local rotatioo ljJL is also a small order
quantity (Fig. 1):
(12)
Assuming that:
L IlL _ L
Lo
oos L. = (13)
LO
~ is given by:
L
(14)
~ = LO
du ~
~= dx +cosljJ-1+ 2" <p2
dw
'N<P=- dx -sinljJ
(15)
L
~ = LO
389
kinematic relations are d:>tained. '!hey are taken into acca.mt in deriva-
tion of equilibriun equations by the energy principle. '!he influence of
~ will be discussed later.
e. g. lit = 0.0 and it is assumed that the deformOO length of the element
L is equal to L0 (~= 1. 0) •
Stress resultants
dA
y
a
x
(16)
r= A
f tXdA=Na +Qa
x z
(17)
where:
N=
A
f adA normal force
A
390
M = J iX
A
ax z dA = AJ a z dA (18)
Constitutive equations
a =a (E) (19)
a=
dU
de: (20)
u= J UdA (21)
A
it can be concluded fran (1) and (16) to (21) that the energy becanes
function of stretching and bending defonnatioos ~ and ~:
(22)
M= au
aDM
391
N =f a dA =N (~, ~)
A (23)
M =f a z dA =M (~, ~)
A
(24)
da
A =
t f dadA
de: St =f de:
zdA
A A
I
t
= Af da
de:
z2 dA
where e. g. t:N presents the nonnal force inCrE!l'lent. '!he Eqs. (24) are
used in the displacenent awroach.
Selecting the nonnal force N and the bending deformation ~ for inde-
pendant variables a roodified fom of (24) is obtained:
(25)
'!he Eq. (24) will be used in the mixed energy principle. I t is coovenient
to introduce mixed defonnatioo energy per tmit initial length, (22):
(26)
:~ = - ~
* au* = M
aOM
aM (27)
aN
392
Equilibrium equations
where Q is the potential of the elerrent loads and Q presents the poten-
tial of the end actions (Fig. 1):
(29)
0% = o( ~ ) + % <P 6<P
dw
% o<p = - 0 ( dx ) (30)
Using (22), (28) and (29) the first variation of II can be stated in the
form:
LO LO
c5II = J (N 6D.~ + ~1 6DM) dx - J (p-x 6u + p z C'M)dx-
o I
o
6
-l'&6v=0 (32)
k=l -k k
Considering (30) and (31), and using the integration by parts, approxi-
mate differential equations:
(33)
d (.!..... dH _ N (0) 0
dx ~ dx + Pz =
S3 = - M(O) S6 = M(Lo)
(34)
S2 = - ( .!..... dM - N <O)x = 0
'1~ dx
S5 = (~:: - N (0) x = L°
are obtained. '!he Eqs. (33) and (34) are quite simple in canparison with
the "exact" ones [1]. I t will be shown that relatively gCXJd results can
be obtained.
U -l EI
n
Bending rigidity
Number of integral
intervals
w M LO
w=
ET
!f-
o- - - - - L °_____...,.1
One integral interval
"Exact" rrethcx:1 Simplified rrethcx:1s
~,l'\., - % error for u,w
a (38)
- 51 00(0) - 54 oU(L )
(39)
51 = - N(O)
Solving thE!l1:
396
(40)
the above equatioos can be used as the coostraints; thus the elenent
mixed defonnation energy fl.IDctiooal is defined:
LO
11* (w, No) = J [U* (N,~) + Q(u,w)]dx + Q(vk ) +
aL °
+
a
J N(N0 ) (du
dx
+ oos ljJ - 1 + ~
2
c(2)dx (41)
Ws = 12 v2 + Is Vs + c N L (1 3 v3 + Is Vs) -°
- 0.25 LO(Cu ljJ - sin ljJ) (1; - E;3) (42)
n
wb = k~7 1k vk n ~ 7
~ cp = - dx
dw
- sin ljJ D = dcp
M dx
where
x = 0.5 LO(l + E;)
12 = 0.25 (2 - 3 1; + 1;3)
(43)
IS = 0.25 (2 + 3 1; - 1;3)
397
I3 = 0.125 ( - 1 + ~ + E2 - E3)
(43)
I6 = 0.125 (1 + E - E2 - ~3)
and for k) 7:
(44)
dIk dIk
~ (- 1) = Ik (1) = dr (- 1) = dr (1) =0
(43) are recognised allro3t in all oublications, which analyse beam ele-
ment. '!he above polynanials satisfy the ccnditioo for the shape fl.U1ctioo.
The eZement equations. n* is nClftl dependent 00 vk (k:l 1,4) and 00 NO.
Equating to zero the first variatioo of n* , oonsidering Eqs. (26), (29),
(40), (41) and (42) follClftl1ng expressioos are ootained:
L ° a~
s. =
-lc
J0 (M -aV dx + c._N N cP ~
aV
- p
z
aw)dx
aVk
k k
1
-u LO 0 ~ dx 2
LO
=.L
LO
J (- ~ + 0 G)dx
0
Q-
(46)
J
where:
~= 0
(47)
~ =0 for7~k~n
vazying the shape fl.U1ctioo Ws the expressioos for end actions 52' 53'
5s, and S6 are ootained. '!he bubble fl.U1ctioo wb does not displace the
ends of the elenent and the correspooding generalized force ~ nust be
zero (47). Similarly NO' Fq. (40), does not influence the glOOal elanent
equilibrium and the correspooding generalized stretching defonnatioo %
must be zen> (47). '!he integrand in (46) can be treated as a difference
between tlNo stretching defonnatioos, the first ooe ~ is defined by
398
lISk
n
= i:2
f aM a~
t.Vi ~ ( a~ aVi + N ~ aVi aVk )dx
a<p a<p
i;&4
LO aM a~
+ llNo J(
o
aN av.k + ~ (j) ~
k
) dx
(48)
LO a~
+ t.v,+ - t.vl - m
D 0 aN
dx J-
'!he increrrents (48) can be expressed in more cam:>act fom, i f (24) and
(26) are taken into acoount:
(49)
servative. '!he Egs. (50) and (51) are used for the statical ooodensa-
tion of oorresponding increments NO and v k (k ~ 7); thus the total
frane matrix presents Sate mcxlified stiffness matrix and it is of the
sane order as in the case of the classical beam analysis.
Comparison of different order bubbZe funations. '!he problan can be
solved analytically only in Sate sinple cases whim can be used for
testing the el.ements with different order bubble functions.
~
PI P3 P"
Number of % Error % Error % Error
elements
1 6.40 0.06
2 1.32 O. O.
4 0.75
6 0.16
8 0.05
'!he results srow that the relative error for critical normal force is
negligible even in the case when cne P" element presents the beam. To
get nearly the SaIre results 8 equal length PI elerrents have to be used.
Figure 5 shCMS the elasto - plastic cantilever which was analised using
f. o. t. Nurrerical results are calculated by nurrerical integration alonq
the elerrent. '!he j-th order polynanial can be represented by j+l values
of ~, which is approximated as the j-th order polynanial; thus the IlOSt
suitable number of subdivision points for P. elerrent is j+1. Using great-
J
er muTter of points a strong divergence of results may occur. '!he results
in Fig. 5 shaN that me elerrent P" gives a quite good approximation of
cantilever up to 97 % full plastic force P fp'
~p}
-----------r----------__ E
L
"I
W=
% Error for w
W Ole elerrent - Lcbatto integration
PLj Ps P6 Pa
pi + p2 = -0 ; L (p2 e ) = 53 + 56 (53)
z
Transforrred form (ex = a)
x of the simplified equations (33), (34):
-
pi + pi = 0; ~ LO(p2 9Z) = 53 + 56 (54)
where (Fig. 1 )
(58)
When using (57) and (58) the correct tangent matrix is obtained. Many
authors use rrodified Egs. (54) and (58) with ~ ~al to 1 ; this way a
ncnsyrretric tangent matrix is derived and usually the term 5 5 l'.lj! is ne-
glected. Another way of obtaining an approximate tangent matrix is to
introduce L = LO into Fqs. (53) and (58). In the method proposed here
the global element ~librium is satisfied exactlv, while neglecting
6~ leads to an approximate tangent matrix. In the case of zero bending
deformation (truss) this matrix berorres the rigorous tangent matrix.
402
Conclusion
References
J. C. F. TELLES, C. A. BREBBIA
Southampton UnIVersIty, U. K., UnIversIty of CalIfornIa, Irvme, USA
Swnmary
1. Introductlon
2. Governing Equat~ons
~ (u. ~,J
+ U. . ) = ~':"
J,~ ~J
+ ~J:.;
,I
a
~J,~
+ b] =0
~n the ~nter~or ~ of the body, and
p~ - a~JnJ =0
on the boundary r, being b J the body fv I: . . , '
per un~t volume, a~J the st.ress rate CLdl' )l1t::
tract~on rate components per un~ t a.re:i a,_
ward normal to the boundary of t.he be ctr '
ar~ses,
406
(5)
h
were a• P
ij represen t s th e componen t s 0 f th e "1.'n1.'t1.'al
stresses" given by,
.p _ .p + 2Gv .p
a ij - 2G £ij 1-2v £kk °ij (6)
and
(10)
.p .p
b,
J
b,
J
- 2G(£ .. , + _v_ £kk, j)
1.J,1. 1-2v
(11)
and
. .p v .p
I\ Pi + 2G(£ij n,J + 1-2v £kk nil (12)
* -1 [ ] ar
Pij = 4CY.'lT (l-v)rCY. { (1-2v) 0ij + Br,i r,j an
* -- c jkrs E*rsi - 1
°jki 4CY.'lT(1-v)rU {(1-2v) (r,k 0ij
(17)
I:
~ .q~~r0ptc tensor of elastic constants
. ~, I Y wr 1 tten as,
'LJ
" PJ' dr - dr + i u* b J d>l
>l 1J
r
~ I (19)
, >l
i
,F " '''''11-1 ')'1
-I, ax
*
ap iJ
m
u. dr +
J >l
au *
~ b
ax
m
J
d>l
(21)
I II
409
·L
a a<1 *jki
'p dfl } 'p
aXm
{
L * £jk
<1jki aX m £jk dfl
(8-10v) 'p
+ 15 (l-v) £im (22)
<1 ..
1J
L (-<1~jk) Pk dr -
L * uk dr
Pijk
+
L * ) b. dfl +
(-<1 ijk k
L':jk' 'p
£kQ, drl
Cij Uj ~ L *
u ij Pj dr -
L *
Pij u j dr +
L * b, dO
u ij
J
+L A*
0' jki
'p
Ejk do (24)
where
. .
O'ij
L *
(- O'ijk) Pk dr
-ir *
Pijk uk dr
+ i
0(- *
O'ijk) b• k dO + iA*
oO'ijkR.
'p
EkR. dO
(26)
in which
*
O'ijkR. G r2 [ 4v r ,i r ,j 0kR. - 2\1 0ij 0kR. ]
+ 2 7I(l-v)
(27)
(29)
C'k
J rs CrSJ'k (30)
moreover
Thus,
dO = L* .p
Ejki (Jjk dn (32)
(33)
(Jij
L *
(- (JiJk) Pk dr -
L*
PiJk uk dr
+
L *
(- (Jijk) b k dO + L *
EijkR.
.p
(JkR. dn
(34)
-L {G(u~,
2Gv
~J,
k + u~k
~ ,J
,) +
.p
1-2v uiR..,R.. 0jk} E Jk dn (35)
which after integrating by parts gives the identity,
Jr 1<
u ij
2G (.p
&jk n k
+ _v_·p
l-2v &.Q,.Q, nj
) d
r
Jo u ij1< 2 (.p
G Ejk,k
+ _v_·p
1-2v E.Q,.Q"j
) d
n (36)
(37)
where OJ and Pj were given in (11) and (12) respectively.
Therefore, we have arrived at a plasticity formulation in
which tractions and body force rates are fictitious (depend
on the plastic strains), but the displacements are correct.
In order to apply equation (37) one has to be aware that
although it looks like the elastic application of the
boundary element method, the internal stresses still have
to be computed by use of equations (1) and (4) , i.e.
L
..:.
°ij
1<
(- °ijk) Pk dr - Jr Pijk Uk dr
1< •
.
+L
.p
(- °ijk) bk dO - Cijk.Q, Ek.Q, (38)
413
H u = ~ p + Q aP (39)
(41)
(42)
(43)
i· j.
where u l and u l are the displacement rate values at nodes
i and J and t is the element size. For plane stress v is
replaced by v
5. Numerical Integration
The process of integrat10n over boundary elements is well-
known and the procedure presented in [lJ was followed.
For the initial stress rate integrals the semi-analytical
1ntegration scheme fully described in [9] was adopted. The
technique is formulated with reference to a cylindr1cal
coordinate system (r,~) based at the singular point. Due
to the very nature of the fundamental solutio~ analytical
integration with respect to r turns to be very simple and
removes the singularities. Integration with respect to the
angle ~ is then performed using one dimensional Gaussian
quadrature formulae, five integration pOints being suffic-
1ent.
Notice that the procedure is also app11ed to compute the
principal value of the domain 1ntegral presented in (34).
But if higher order interpolation functions as well as
different cell shapes were chosen, the princ1pal value could
415
F(a.
1J
,k) = 0 (44 )
(45)
Jl a ..
11
1 (46 )
J2 2' Sij S 1J
..
1
J3
3" S.1J Sjk Ski
where
416
S.
1J 1J 3 J 1 us:
a .. - 1:.
ij
(47)
) -< ~
'IT
- "6 :5.. 8 6
(48)
Tresca:
(49)
Von Mises:
~ - a (k)
2 0
o (50)
Mohr-Coulomb:
Jl
:r sin¢ + 1J2 (COS8 - -=1
/3
sin8 sln¢) - c cos¢ o (51)
Drucker-Prager:
(52)
where,
2 sinp 6c cosp
a. = K (53)
13 (3-sin¢) 13 (3-sin¢)
a tan<jl 3c
K (54)
(55)
o .. dE~ = dk (56)
°e 1J 1J
o . (57)
1J
.p
E. (58)
1J
a}' af
in which ak £ (60)
ao k £ ao k £
F a .. a . - d1jJ k
dk
o (61)
1J 1J
o (62)
a .. a .. ~ a .. a .. dA o (63)
1J 1J dk 1J 1J
dA (64)
where
y a
ij
C
ijk~
a ko +
N
~ddk a ij a ij (65)
(66)
y a C a +H' (67)
1j 1Jk~ k~
a . (68)
1J
419
in wh~ch
1
CiJk~ - y C1Jrnn a mn a op Copk~ (69)
·e
0, , (70)
~J
·e
where stands for the components of the elastic stress
0,
1J
rate tensor (1.e. these represent the stress values as if a
pure elastic problem were being solved).
·e 1 ·e
0, , 0, , C, , a a ok 0 (71)
~J ~J y ~Jmn mn k ~ Iv
which means that the true stresses can be computed from the
correspond~ng elast~c stresses. In addition to this, the
~n~t~al stress rates presented in (6) can also be calculated
by the relat~on,
.p ·e
o , o 0, (72)
~J ~J 1J
E E' + I (73)
This g~ves,
·e
0 G' p - H' u + Q* aP (74)
where,
Q* Q' + E (75)
420
Ax = f + Q oP (76)
and slml1arly,
·e
a A' x+ i' + Q* aP (77)
.
where vector x contain the unknown tractions and boundary
displacements.
From the multiplication of (76) by A-I we get
x R aP + m (78)
where
R A-I Q (79)
and
m A-I f (80)
a·e S oP + n (81)
in which
421
s Q* - A' R (82)
and
n £1 - A' m (83)
(84)
x = R(a P + ~p) + A. m
~ - (85)
and
e (86)
a
x =R aP + S m (87)
and
a·e = S aP + Sn (88)
8. Examples
Following the discussion presented in the end of the last
section, the results for a series of examples solved by the
boundary element technique are now compared with analytical
solutions where such solutions are available and with finite
element results.
Notched Tensile Specimen:
This first example is one of the very early plasticity prob-
lems solved by using the finite element technique. Plane
stress and plane strain results have been presented in
several papers, creating a good opportunity to compare the
425
2C7.
14 eTo
(8 EM) limit load achieved 1 21
-------123
12
----119 FE limit loads
10
08
06
04
02
20.
00"
""
2 .0 F. E.col_pse 1.85
B.E.colapse 1.6 4
1. 5
1.0 IV
~c
1---1
0 .5
- - F.E.M
- - - B.E.M .
Uc
1
0.005 0 .0 1 0 .015
~ --
I ,
I "
~--..;'+--.. - I',
~.:__ ~ __~~L __~,1 __ _
o I 1 ksi
.2.4 ks i
,,
,,
• ,,
/ ._.J._.__ C£
a
C
- 7. 5
/aker(F.E.M .)
B EM
/'
, Reyes ( F.EMJ
\./,".
v·,".
-5 .0 ,".
':;,....
0v :"0.';.: •••••••
Bake r (F.E.M.J
- 2.5
1BEM
~ Reyes (FEM)
!Lx 3 4
x
r
5
Rough Punch:
In the third example the elastic-plastic behaviour of a
square block compressed by two opposite perfectly rough
rig~d punches is stud~ed. The problem is analysed under
plane strain condition and material is considered to be per-
fectly plast~c obeying von Mises yield criterion.
9. Conclusions
p =mea n pressure
W= 0 .5 on
·tt
® Fml te Element mesh @ OiscreUz8t ion used for B.E results
(no bounda ry elemenlS along
symmetry axes)
!It
v3p
200 • fEM .
• B.E.M.
1 1'"4-- - -----,
6 T,~-.,
,,
,,
,
,,
E = 10 7 pS'
00 = 13000 psi ,
L______ ____ _
v = 0 .33
H' = 0
0 .0 1 0 .02 0 .03
References
Appendix A
G
{ SClr
" [ (1-2v)0 .. r k +
2a1T(1-v)r S on 1J ,
+ v(o'k r . + o'k r .) - yr .r .r k J
1,J J,l ,1 ,J ,
+ Sv(n.r .r k + n.r .r k) +
1 ,J, J ,1 ,
(Al)
--(;--Q
2a1T (l-v) rf.'
{s (1-2v) (0 .. r k r
1J"
R, + 0kQ,r . r
,1 ,J
)
+ SV(Oa. r r or . + O'k r or
N1 , Jr , k + O'k
J,N ,l 1,N,J
+ O'a r .r k) - Syr .r .r k r a +
IN ,1 , ,1 ,J, ,N
+ 01 k r ,N,J
or . + o'or
I N ,.r
l, k J
+ 20 ko r .r . - 8r .r .r k r 0 } (A3)
N ,1 ,J ,1 , J , ,N
B. H. KROPLIN, H. DUDDECK
Technische UnivefSltat BraunschweIg, Germany
Sununary
A Y1eld funct1on, which descr1bes the plastic flow of a
plane cross sect10n of a thin plate approx1mately is
d1scussed and compared to the layer approach. Further an
art1fic1al (viscous) damp1ng 1S suggested to overcome
1nstab1lit1es due to snap throughs w1th or w1thout b1fur-
cations.
Introduct1on
(1)
iii = 1
ijj = z
(2)
437
J 4J
z
dz =0 !4J ZdZ =1.
z
(4)
(5)
(6)
where
An = na.~ Ja.~gA n 9A
( 1oa)
asz :st
f1g. 2: Shape funct10ns for stresses 1n the plastic
llm1t state
( lob)
m
- reference cross section
"'---'---'_L--'---I_ n=1 fut
Sy2 o yield function (11)
n 1
0,75 ~-~~--i1----t-Y
o
a
11
9-
O,5L-----~----~-----L
I-----L----~------L-----L-~e:-·p-
I-O,5~
S = Soy H d £ p. (13)
eP
441
(14 )
m::fyi A 22
2
10"T-~_ A: A12 :: n m Sy
B: A12 : 0
2
3"
"-.
Alz= t nQ./3J a /3QA mQA
1..-_ _ _ _ _ _ _ _ _ _ _ _ _---..:...____ n =i-~
to Y
-p
fig. 4: Change of the Yleld surface wlth A12 and E
. .~ £P ay· ay .!. P
WM = WM - 25 ( as 5 +2at P E ) = stat .. (15 )
442
Application
The Y1eld funct10n model was applied to post buckl1ng
443
= [ no.
a.
n t3
~l
• N 1
~2 = 25
[~E~.a.p~]
~ =[o,mJ
tL =U;~~]
F Flexibllity of the cross sectlon
b boundary integrals.
Examples
G G,U
Gy
1.0 ~__~?/Yt=3'175mm
m" _ N/mrrf
w S' n
'0 Gr2SO
E-206500 N/mm2
t ~"0,3
w~ b/1000
alb-O,875
",'=Without yielding
-~-:~-=:=-)(
.5
_ ,~~I
,..x~
first yielding
- .: 7 /
4:;6~Z ,. displ.
5 layers
mixed
3. yield f.
.1
Wm
1.0 2.0 t
G
Gy
1.
.t alb =1
t
I :
2x
a IG y i
0 100 t 24
G
Gy
1. .t . 2x
-:
t
a/b=1I2
Gb
D Ib
.......
t . a
SLyGY
0 100 t 24
G
[b_
Gy a/b=O
1. ,
.!., . -Gb
t a
t, ~VJhi
0 100 t 24
• experiments (11)
+ mixed FEMJyield function
Conclusion 1:
~(x)~ + 9 ~ = I! . (18)
447
(19 )
D_*n:: °
£M- (20)
Examples
flQ +alflQI
0.=02 (22 )
1+0
QJ
t - 7 mm
<.l 1.0 I =-1000 mm
....
.....0 E = 2.06 10 5 N/mm2
.S? ~= 0 , 0
il
c
E
L.
0
y
.A.
.m y§ ~,n
Wm
U1 =0 /021 mm
W" a sm"!!.!.-+ a sm ~
1 I 2 I
c t al a2
.1 .2
A 0 ,1 0
BO 0,01 0,0707
C 0 0,0707
number of iteratIons
A*, B", c*: with damping
A,B,C incremental
c 0 without damping
....u
0
.!!!
Qj 20 K l
"0
C
....
L.
C
III
u
.1 Wm
1°
t
Conclusion 2
Acknowledgement
Literatur:
G. MEHLHORN, D. KLEIN
Technische Hochschule Darmstadt, Germany
1 . Introduction
4
{I _J()/AII/m 1
'l.- J()OOOMII/m t
v.-o.z Fig . 1 Crack distribution
Ti. rod. of Shear Wall WT7 of the
1 ~SJ·l,ocm: tests by Leonhardt/Walther
~ :..A$l·~'cm, [1]
-A~, • 1.0 em
I--~ SSt .10/500
64()kN - -
CfGOO up 10 610 kN - -
1qro k N --
* I
1
, " I J '"
J
lood
(kNfin2)
50
{ lin. elastic ~fallure load
(uncracked)
./ {Yield lme Theotg}
/ /
V I
~ I'
30 ~working lood (DIN 101fS) Fig. 3. Load-mid-
j pOint deflection
L A{
20 curve of a simply
dis - and re,dodlng supported rectan-
1.1 of the slob gular slab tested
//;
10
by Franz [2]
1
0,5 ~o l,S 2,0 .2,5 40 Defledlon [em}
.0:n
7'd
1,0
][ I
I- bwnd oti~3 fl)('
0.5 F1g . 4. Fallure criterion of
l-
I-
>-
e CfOeXs in O~
di~ction
crode3 in two
concrete in the biaxial state
-<Tn of stresses by the formulatlon
I- ~ directions
of L1nk [7] based on Kupfers
AI tests (6)
-0.5 -1,0
,?/~ - o.SI5
-1,0 l.,{%o}
Under unloading the steel obeys the same law as under first
loading. At cyclic loading with many cycles, however, the
strength of the steel is diminished. Temperature does in-
fluence the steel properties very much. So the strength is
reduced at high temperature.
I~-
conu.nfric pull- out
f
.
<
specimen
Q ·'SlotN/m'
INN/m'}
~
l ~::
-
excenlric pUlI- out
specimen
10
t
lilil -
Q · Ia/~ral
Compr.
-~
lapp~d splice
f??
speCimen
.d{mm]
., .,.
?,
C>1-'
Cl
~ :!".
fti3f crock
lolerol compression
DirecJion of
.~.-----
A B A Sec/ion A Section B
I I I 1
~ 0:
I
I
J.--_ ___.-1
l
I ',,--r ..,
I
I
II Flg. 13. Cross-
section through a
I cracks I reinforced concrete
I lateral I element under
I reinforctm~nt I bending moment
I I
A B
The most versatile but also most expensive method for the
modelling of reinforced concrete is the separate idealization
of concrete and reinforcement by elements (see F1g. 14) .
aJ Example for planat element under inplane farces '
b) Example for la!lered bending element.
concrete logers
anedimensianal linle
element far bond- 51ip contact element
for bond-Slip
onedimensianal eleme'" t.:::===;z::::=~ smeared reinforce -
for re;n(or«menf bo, ment layer
no rension st;f(ening
a) Singlt cracks along the ~emt!nf boundorit3 :
stepped respOl/3e
~r crocking
l\
'\ /' \
nodlll PQinfs eraclc
elemel/f wilh
two crockdirecfiol/.r
al arbitrary ol/gle
p p
L-__________~u U U
u
incff!menlal melhod+ in&l'rmen/of mil/hod ~ inuemen/af me/hod #
N_Im-Rophson iler. n",,/if. Newton -Raph- modif. Ntwfon - Roph -
(wn» longenfsnflness) son ltero/ion Jon iluo/ion
(wilh iniliol stiffness) (new sHffnes$ 0/ any
load step)
3. Numer1cal Examples
3.1 Shear Wall WT7 of the Tests of Leonhardt/Walther f1J
The system and the discretisation 1nto fin1te elements used
by D6rr [1~ 1S shown in F1g. 18. The wall 1S character1zed
by a concentrated tie rod at the lower side and strong
vert1cal re1nforcement bars. After cracking of the concrete
1n the tension zone, the vertical reinforcement must trans-
mit the load applied to the lower side of the panel upward
into the uncracked compression zone, where it is transmitted
by bond to the concrete and carried by arching action to the
supports. To study the effect of bond-sl1p between the
reinforcement and the concrete, a matter of main importance
1n this example, Dorr [191 idealized the reinforcement by
separated bar elements and represented the bond-sl1p by
realistic shear stress-slip relations including the influence
of lateral compression. F1g. 19 shows the crack-development
of the solution of Dorr [1~.
f\ /\ lif\ Vi ~ V11\
V\1 \V NV1NV
1
i
-
1-16 I· s+cm-----j
- - - "infofUmenf
!I
Fig. 18. Finite element idealization
of Dorr [19} for the recalculat~on
of the Shear Wall WT7 (see Fig. 1)
Load
p~~~--~~--~~~~~=+----+---~
Miiller [18}
q8~----+---'~~~~~-----+----~----~
Geisfefeld [21}
~6~--~~~--+----r---~---r---~
o Hoshino [11]
t:. Stouder (10] ---!----1
- test results
1 2 J If Oeflection [mm)
F~g . 20. Comparison of calculated load-midspan deflection
relations with the test result of Leonhardt/Walther [1]
469
SlM '
SSt +1O/5QQ
MI soo/sso
L,;, I
Load Load
{kN;In'l step,
50 V Ii,. :=-- - --
--:;;; ~ I-- --
18
(uncrocked) ~
"0
J ~ 16
/
f+
R ~ 12
JO
'\
/d
1 of.1Schafer
calc. results . 1 [13 J fO
20
~~
calc. resu!f; of Maurer [16J 6
10
t "
2
lowpr side
fe3f
// j \\... ""-
// \" urper side
ealeulalian
"\ //
'" \
/ /
I /
471
lDod
P{IrN} -
20
Ih"knes.
e m
d- ~u till
eHecIdepih. At =~33 em
ptrr:enlDf (tmt 6 t· 0.85%
1
5
~
--+_T...__geom ,mear
, } •
geom nDn/m fixed corners
I
15~~71:~~~
'failure load
(Yield fine Theory)
- - - frat results
qs 1,0
load
p~~
~~-------r--------+-
¥ r-H-----r------
qz
relnforcement
relnf. : =-u
bond: onedun. llnk
element
Hoshlno [11] panels under
In-plane forces
sep. ldeal. of
concrete and
relnforcernent
concrete:
relnf. :
L:4/:
=-u
v
il{U:
panels: two-dlm.llnk
element
Schafer [13] slabs, lncludlng comb. plate panel u , u'n, v: v'-dw'n
In-plane forces element (;rnt. pOlnts
V, jWi w, r
\l,x: u, , v: v'x:
W, i
w,~x; W'x;i W'~y
Mehlhorn/Krlst- slabs, lncludlng comb. layered plate
]ansson/Kleln In-plane forces panel element L!{:u, v: w'n
Ui Vi w: W'xi W'y
[14], [15], [23J w'xx' W'xyi W'yy
Maurer [16] folded plate comb. layered plate concrete } [ > { u : v, w,
structures panel element +
relnf. bars ~~~~f.W~~~hs 9 x i9y iBz
relnf. bars. ~"'u
t>t
Zeltler/Dletrlch/ slabs, lncludlng comb. layered plate u'x i U'y:
Mehlhorn [17] In-plane forces panel element v, v'x' V'yi
W: w'x i W'y
Muller [18J panels under
In-plane forces
sep. ldeal. of
concrete and relnf.
concrete +
relnf. mesh
relnf. bars: =
t7 u: v
__ u
bond: contact element
Dorr [19] panels under sep. ldeal. of concrete comb. ~; v
In-plane forces concrete and relnf. wlth relnf. mesh
bars relnf. bars: =<-u
bond: onedlm. cant. element
475
unlaxlal hypo of Ml.ses elastlc- elastlc- along fully st~ffen~ng pure lterat.
nanllnear 1.n compr., ~deal- ~deal- element of bond- method w~th
(DIN 1045) max cr 1n plast~c plastlc boundar Ilnks secant
tensl0n stlffness
unlaxlal hypo of Mlses elastl.c- elastl.c- along the nonllnear --- pure lterat.
elastl.c- (no reduct. of ~deal ~deal- ]ol.nts of normal method
~deal tenslle plast~c plast~c prefabr. shear w~th
plashc strength) panels force secant
lnteract. st~ffness
blaXl.al bl.axl.al nonll.near perfect crack held reduct of stral.n soft l.ncrernental+
nonll.near (L~nk [7] ) w~th~n the shear of concrete mod. Newton/
(L~nk [7] ) element st~ffness Raphson ~ ter
b~ax~al b~ax~al nonl~near perfect crack held reduct. stra~n soft ~ncremental+
nonllnear (L~nk [7] ) or w~th~n the of shear of concrete mod. Newton/
(L~nk [7J ) b~l~near element st~ffness Raphson 1 ter
un~ax~al unrestr. comp.J l~near perfect crack f~eld fully no gradual cons
llnear max er ~n elastlc w~th~n the of cracks by
elast~c tensl0n element new st~ffness
References
Abstract
Introductlon
(1)
.
cr C£ - CAn (2)
485
A = H-1/ IT')
0 = \n C£ /(H+n T Cn)
\n T') (3 )
f(o,£P) (4)
2 2 2 1/2
2 (r -1) cos9+ (2-r) [4 (r -1) cos 9+5-4r]
g(9) , (5)
4(r 2 -1)cos 2 9+(r-2)2
9 (6 )
(7 )
(8 )
Fig. 2 Biaxial
behaviour of
ELASTIC LIMIT concrete
HARDENING
REGION
_ _ _-SOFTENING
(9)
o (11)
487
E 1%0
-1 -2
o/fc
.8
.6
fc= 32.17 N/mm2
.4
E =32500 N/mm2
.2
I1v %0
-.2 -.4 -.6 -.8 -1 v
Cracking
Reinforcement
For reinforcing steel, the uniaxial bilinear elastic plastic
material model is used.
Reinforced concrete
Complete compatibility between concrete and reinforcement
is assumed, so that the deformation is continuous over the
adjoining boundaries of concrete and steel. In slabs and
panels, the reinforcement net is treated as smeared layers
of thin wires. The effect of separate reinforcing bars is
transferred to the nodes of the finite element mesh through
the shape functions. The shear resistance, which stiil exists
in the cracked region due to aggregate interlocking and
dowel action, is taken into account in reduced shear modulus
of concrete [27]
0.1 G (12)
Equations of equilibrium
In total Lagrangian approach, the principle of virtual work
can be presented in the form
R(q) + M q Q (14)
~9
where
(15)
are the internal force vector, the mass matrix, and the
external force vector, respectively. The incremental form of
the equation (14) is
(16)
*
8 CE (17)
(18)
Solution techniques
(19)
~E (20)
(21)
'+1 Hl i
where ~q1 qn+l - qn+l. For the first iteration cycle,
ql+l = q is taken. The iteration is continued until
n -i+l n i+l ,
II Aq II < e: II qn+ 1 - qn II where e: 1S a tolerance parameter.
Consistent mass matrix is used. In order to avoid too drast1c
or artif1cial changes due to cracking, the tangent stiffness
was updated in the first iteration cycle but held constant
thereafter. This also reduces computing time.
~U
~T (24)
Applications
au/ax + (aw/ax)21z+za<p/Clx
Clu/Cly + Clv/ax + (aw/ax) (aw/ay) + z(a<p/
(25 )
lay + a'li/ax)
Yxz = aw/ax + <p
etc. where <p and 'Ii represent the rotations of the normal
with respect to x- and y-axes. The expressions of von
Karman follow when the constraints Y
xz
= Y
yz
= 0 are
~mposed.
" I
? ... ffi:
I
I
I
---+--- 0.85%
I
I A z
I P
I I
I I
--j--,--
I I
1 2 3 E
--~-....j.-- 0.2 0.4 0.6 %0
I I
I I
~
36 IN
,/ 7
P[klbfl
3.6
2.8
2.0
.4
def
4 8 12 16 20 24 28 32 36 40
2 8 ~[O]I(203.2mm)
(2 cp2S:mm)", 8" I 12"
(304.8mm)
~ P[kN ]
(203.2mm)
40
CYCLE 1 CYCLE 2
2 v[mm ] 4 6 v [mm)
- 20
-40
- -- EXPERIMENTAL
--0--0-0- CALCULATED
P[kN] CYCLE 3
CYCL E 4
P[kN]
90
60
v [mm ]
40 v [mm ]
Fig. 5. Continued
v [mml
123.9
295
k:L I
P[kNl
t
,IE
.. 2
i~-----~.....,........J
I
6066 [msl
50" lIffIr
I
1D
(1270mm)
20
(254mm)
10"
2" 00 2 #6
(50.8mm) ~ (2ct>19.05mm)
6"
(152.4mm)
- - EXPERIMENTAL
- - - CALCULATED
10
t[msl
10 20 30 40 50 60
12000mm PIMNl
I 10
I 8
I
----1- - - -
E I
E I
<:)
<:) I P
:
<:)
<:)
I
I
--"1--T--
I I
----i--t---
: I
I
5ms 100ms
def. [mm)
12
11 0-0 MINDLIN THEORY
2 x 2 el.
10 ~ KIRCHHOFF THEORY
2x2 el.
9 3 x 3 el.
8
7
6
5
4
3
2
1
5 10 15 20 25 30 t [ms)
p 2
1+ =2000 N/m2 A
P = 3.4 MN/m
I = 300 N/m2
+
t = 2.7 ms
t- = 40 ms
t t
Mlndlm's theory
experiment [6 ]
central
deflection
100 [mm] FEM [6]
so P uniformly distributed
over plate area
t [ms]
10 20 30 40
Fig. 8. Central deflection vs. time of one-way slab under uniform
pressure. Data. reinforcement 0.17% and 0.085% fc = 36.6 MPa,
f t = 4.90 MPa, Ec = 28.4 GPa, Vo = 0.15, fs = 450 MPa, Es = 210
GPa, span 2280 rom, width 1230 rom depth 130 rom.
A rectangular slab, simply supported on two opposite edges and
free on the remaining edges, was considered by Nilsson & Johansson
[6] . The slab was subjected to a uniformly dlstributed pressure
load varying with time. In addition to the test, the authors made
finite element calculations which included cracking and plastic
deformations. The central deflection-time history of thlS study
was computed using four elements for a slab quadrant (Fig. 8).
Varying time step h = 0.025 ... 0.5 ms was used for the Kirchhoff
theory with 2 by 2 mesh and h = 0.01 ... 0.2 ms for the Kirchhoff
theory with 3 by 3 mesh and for the Mindlln theory.
Discussion
The numerical results obtained for the problems of beams and slabs
indicate that the present method is capable to predict satisfac-
torily the response of relnforced concrete structures subjected
to transient loading. More numerical results and experimental
data are still needed. Modelling of rate dependent behaviour should
be developed. More realistic description of aggregate interlocking,
dowel action and bond slip is also necessary.
499
References
Summary
The stability of nonlinear step by step computing schemes
for solid mechanics problems can be insured by either re-
quiring that the energy of the approximation be conserved or
by insisting that the energy be bounded for all time. While
techniques exist for analyzing conserving procedures, few
analysis methods exist which are capable of establishing the
boundedness of the energy of nonconserving computing schemes.
In this paper an analysis procedure for demonstrating the
stability and convergence of nonconserving step by step solu-
tion algorithms is introduced. As a typical example this
technique is used to analyze a Crank-Nicholson-Galerkin al-
gorithm for nonlinear elastodynamics problems.
Introduction
Energy analysis procedures have been used by several investi-
gators [1-4] to study the stability of step by step solution
algorithms for nonlinear structural dynamics problems. This
previous work has been principally directed to developing
algorithms which, either implicitly or by constraint, con-
serve some scalar or energy like quantity from time step to
time step. various unconditionally stable algorithms have
been constructed using these procedures. In this paper a
method of analysis for nonconserving algorithms is developed.
This analysis is based on the premise that, in the approxi-
mation, energy should rema1n bounded for all t1me. This meth-
od of analysis should be useful in analyzing various noncon-
serving implicit algorithms, load extrapolation algorithms,
and explicit algorithms.
503
• P.m-l
UEL",,(O,ooiW2 (I)).
Regularization Procedures
Stability and convergence results, of the type discussed in
this paper, are extremely difficult to derive using the second
order equation (2). Thus (2) is represented as a first order
system and certain regularization (artificial viscosity) terms
are added to obtain the following form:
pv - D(t(l + Du)) + £v = f
.
u - v + au o (£,a>O) (3)
it - v + au = 0 (4)
U - V + aU = 0 (8)
(EVD(v-V) ,DW) o ( 9)
eu E
u + &'U
E
u = u - U eu U U
e E +&' E v - V cf V V (10)
v v v v v
where
SUP 3· -
X = t ~ [0,00] [2v(DU(t) + DU(t))]
XEI
In addition total energy errors can be defined by P(w,r) =
~(w) + U(r) qnd P*(w,r) = ~*(w) + U*(r). The errors 8 u and
8 v can be established through the following Theorem:
Thm. 1. If the exact solutions u and v are sufficiently regu-
lar and the mesh discretization parameter h is sufficiently
restricted, there exist positive constants gl' g2' and g3'
independent of hand t, such that
(12)
3
ED8u 2
(2 ,D8v ) - '
(pE v , 8 v t
(13)
d
dt U(8u ) - (EvD8v,D~u) + 2aU(8u )
+ (-2
3E08 2,08 ) + 1> (15)
u v
- (E'J [DE
u
- DE
v
+ aOE ) ,08 )
u u
- (EV[OE
u
+ aOE u ) ,08 )
u
where
v *2 Eh \ 1
3EC*2 y 1
2p + Y2 + 2p
3EIIDEuilL
00
(I)C*\4(~)
+ 2ph
2
+ 0: I I DEU I I L2 (I) )
The constants Yl' ... 'Y6 are adjustable. They can be chosen
3
such that d l = dS/h , d 2 = X, and d 3 = d 6 h where d S and d 6
are independent of h. Then by making minimal restrictions
on the regularity of the solutions, it is possible to show
that, because of (7), d 4 ~ d 7 h 2k - l Then from (15) and (16)
~
dt
F*(8 8) + ~ ~*(8 ) +
v' u p v
(20: - d 6 h)U*(8u )
(17)
F*(t) S. fS (
1+
[~-.::fr _fAt)
4 2 e -
f4 + f2
f
4
= J
g2
2
-
4 g XTh2k-4
gl 3e
n+l n) k
(p V ~t- V ,W + (EDu n + 2,DW) + 3 (E(DUn+~)2,DW)
2"
+ E(Vn+~,W) = (fn+~,W)
u n + l _ Un n+~ n+~
~t - V + aU o (19)
Un(W)
~n(W)
-~ln
where
(20)
[: +~
2
and
512
g2F*n+~(~v'~U) + g3ex(n+l)~th-3(F*n+~(~v'~u))2
• (21)
wn + l + wn 2
+ ~ta3 ( 2 ) (22)
( 23)
where
where
ilta 3
A - -4-
ilt n
B 1 + T(a 4 - a 3 w* )
ilt a3 n n
C = -(1 - T(a 4 - _w* »w* •
2
By reduc1ng ilt, it 1S poss1ble to make the term 4AC/B 2 in
(24) small. Thus the square root term in (14) can be evalu-
ated using a binomial series. The following expression is
obtained:
C AC 2 (1 + 2AC
2
+ a) . (25)
- B
B3 B
where Sis a term which approaches zero as '" t .... O. The cor-
·
respon d 1ng . ,1n terms 0 f wn+ 1 an d wn.1S d e f'1ne d as
express10n
follows:
(26)
where
where
"'t
1 - T(a 4 - T
a3 n
w* )
1 = 1 "'t
+ T(a 4 - a 3 w* n )
1 { ",ta 3 1/!w*n }
y = -2 1 + .2. + ex
1 21
for N=O,l, ... ,oo. The size of the term on the right hand
side of (28) can be fairly easily est1mated. In fact
N
1 - IT ].1i -<1 N+l
-].1 (29)
1=0
where
].1 (30)
From (30) and (31), it can be seen that for ~t small enough
-2 (a 4+a 3 a S ) (N+l)~t N+l -(a 4+a 3 a S ) (N+l)~t
e :::'].1 :::. e (32)
and so on.
•
From the results (21) and (23) the energy of 8 v and 8 u can be
estimated. Initially it is convenient to assume that, as in
n
the semi-discrete case, F* (#v,8u )=O when n=O. The energy
error growth condition (21) contains certain time dependent
functions. To make the equations agree, certain restrictions
must be employed. Suppose that the calculations are to be
carried out until t=(N+l)6t. Then for any time step n=O,···,N,
equation (21) gives
(34)
( 3S)
where
517
*n+l 2k-l
F (~ ,~ ) < c(h + 6t 4 )
v u
•
(36)
References
1. Belytschko, T., and Schoeberle, D.F., "On the Uncond-
ltional Stabillty of an Impllcit Algorithm for Nonllnear
Structural Dynamlcs", Journal of Applied Mechanics,42,1975,
pp. 865-869.
2. Hughes, T.J.R., "Stability, Convergence and Growth and
Decay of Energy of the Average Acceleration Method in
Nonllnear Structural Dynamlcs", Computers and Structures,
6, 1976, pp. 313-324.
3. Hughes, T.J.R., Caughey, T.K., and Liu, W.K., "Pinlte
Element Methods for Elastodynamics which Conserve Energy",
Journal of Applled Mechanics,45,1978, pp. 366-370.
4. Wellford, L.C., and Hamdan, S.M., "An Analysls of an
Impllclt Flnlte Element Algorithm for Geometrically Non-
linear Problems of Structural Dynamics, Part 1: Stablllty",
Computer Methods ln Applled Mechanics and Engineerlng,14,
1978, pp. 377-390.
5. Oden, J.T., and Reddy, J.N., Mathematical Theory of
Flnite Elements, Wiley Interscience, New York, 1976.
6. Wellford, L.C., "On the Theoretlcal Basis of Flnite
Element Methods for Geometrlcally Nonlinear Problems of
Structural Analysis", Formulatlons and Computational
Algorlthms in Flnite Element Analysis, Ed.,K.J. Bathe, et.
al., MIT Press, Cambrldge, Massachusetts, 1977.
Part V:
Solution Methods
Direct Solution of Equations by Frontal and Variable Band,
Active Column Methods
ABSTRACT
form
523
Ku f (1 )
K 2: ~e (2 )
e
and
f
2: 1e
e
The connectIon propertIes of each element will define the
non-zero structure of the sparse matrIx K. In variable band
methods the non-zero envelope of K is called the profile of
the matrIx. The profile wIll be defined in terms of equa-
tion numbers assigned to each degree-of-freedom at each node
and the list of nodal numbers attached to each element. In
most implementations of the variable band method the equa-
tion numbers are assIgned independently of the nodal
numbers. Often profIle optImization schemes are employed to
obtaIn "good" ordering of the equation numbers. We shall
524
f (5 )
x (6 )
and
U u = 1..
525
solving (1)
very little cost. With one vector in LCM and one in SCM no
SUBSTRUCTURING
zeroes within the profile may not fill during the triangular
completed.
540
CLOSURE
REFERENCES
I
1 Table 1.
1
1 1
: Mruumum Front Width for 3-d Meshes :
1 1
1 1
: wi th M-nodes on fuch Edge :
1 1
1 1
+--1
1
Element Type 11
1 1
1 1
M : &-node 2O-node 27-node I
I 1
3 i
1
14 22 27 j
1
1 1
5:
1
32 36 47 I1
7:
1 1
1
~ ~ ~ I1
1 1 1
I 9:1 92 e6 111 I1
1
1 1 1
: 11 : 134 120 155 :
I
1 I i
1
Table 2. 1
1
1
1
3-d Element Types Wl th 3-dof/ node 1
1
1
1
: &-node (1) 20-node (S) 1
1 27-node (1) 1
1
1 1 1
1
M 1iFront Core Front Core
1
1
1 Front core I1
1 1 1
1 1 1 1
1 1 1
1 3 1
1 42 9J3 66 2211 1
1 81 3402 1
1
1 1 1
1 1 1
1
1 5 1
1 96 4656 108 5886 I 153 11781
1 1 1
1 1 1
1
1 7 1
1 174 15225 174 15225 I 237 28203
1 1 1
I I I
I
I 9 I
I 276 38226 I 258 33411 I
I 345 59685
I I
I I I
I
: 11 1
I 402 81003 : 360 64~ : 477 114003
I I I I
_I_L I 1
544
i Table 3. i
i i
iI Triangular DecomIXlsi tion Operations (in millions) iI
i II
Element Type
I
I
Mesh 8-node I 2O-node 27-node I
I I
~----~--------~---------+--------~I
I
Table 4. I
I
I
I
27 node lBgrangian Element I
I
I
I
Band Numbering Frontal Numbering I
I
-
I
Table 5.
20 node Serendipity Element
Table 6.
Forward Solution or Backsubstitution
Operations (in thousands)
Element Type
Mesh 8-node 2O-node 27-node
Table 7. 1
1 Vectorizatlon Timings in Fortran and Assembly language :
: for CRAY-1 and COC-7600 Computers :
: Time In nano-seconds to produce a single element for an n-vector. :
: REGOL REGOLD COLBAC:
1 Time I CAL Time I CAL Time ICAL 1
1
: n=1
: CAL 3Xl.0 1.00 3000.0 1.00 600.0 1.00
! CFT(ON=V) 2200.0 2.75 9000.0 3.00 1400.0 2.33
: CFT(OFF=V) 2100.0 2.63 4000.0 1.33 1300.0 2.17
: COMPASS 2300.0 2.&3 2000.0 0.67 1000.0 1.67
: FTN(OPT=2) 3500.0 4.38 6000.0 2.00 1300.0 2.17
1
:n=B
: CAL 362.5 1.00 375·0 1.00 175.0 1.00
: CFT(ON=V) 937.5 2.59 1125.0 3.00 337.5 1.93
! CFT(OFF=V) 850.0 2.34 1000.0 2.67 662.5 3.79
: COMPASS 1012.5 2.79 1000.0 2.67 1050.0 6.00
: FTN(OPT=2) 1487.5 4.10 1500.0 4.00 1425.0 8.14
1
: n=64
: CAL 70.3 1.00 1 109·4 1.00 56.3 1.00
: CFT(ON=V) 167.2 2.38 203·2 1.86 75.0 1.33
! CFT(OFF=V) 576.6 8.20 734.4 6.71 531.3 9.44
: COMPASS 446.9 6.36 562.5 5.14 478.1 8.50
: FTN(OPT=2) 806.3 11.47 5015.6* 45.86 817.2 14.53 :
1 1
1
: n=1000 1
1
:1 CAL 30.7 1.00 57.0 1.00 46.6 1.00 :
: CFT(ON=V) 69.8 2.27 97.0 1.70 49.2 1.06 :
: CFT (OFF=V ) 540.0 17.59 666.0 11.68 513· 7 11.02 :
1 COMPASS 365.3 11.90 442.0 7.75 386.4 8.29 1
1 FTN(OPT=2) 1113.9* 36.28 872.0 15.30 1482.9* 31.82 1
[-4 5 2 ;-7 8 ] = LG
Kll K12 K13 K14 K15
K= 4 Pivot location
JJ= 7 Global equation number
[-4 5 2 8 J= LG
Block # 1 2 3
3 cals 2 cals 2 eals
2 6
3 4 7
5 8 10 13
9 11 14
12 15
16 17
18
1 2 3 4 5
Block 1 llJ11314151 K-Array
11 13 15 I Pointer to Diagonals
1 2 3 4 5 6 7
Block 2 1 6 17 1s19110 111 112 I K -Array
[ill] Pointer
1 234 5 6
B1 aek 3 113 114 115 116 117 Ils I f-Array
I 4 I 6 Pointer
21
i7
46 941 36 .31 26
0
71 66 61 56 51
..
81•
96
•
91 86
76
11
17
29
61
3 27 # 23 fJ 7 64
7 75 71 55
111
81 7) 73 tE7 114
1 5 12 105
129
"
, .. :-,
! ' FISSlE:7800
! FACTORIZATION
i
l00~--~----~~~~_----
I
1 ________~-+----~~~~----~
"'-li--~;-'-'--':-I
-~.~- -:"~l,~l , .~ '. ' i
, ! :I: ' '1,
I I
: 'r ,"- ~-
-" : I : ::
I I
10r-~=-~~~--~~----r-~----~~
., LINE SOR: 7800 - +
(456 ITER., 10~ ' : •
lr-------~-7~--~~.-
_4~:::~~_
--_-_--~~----~.-----~
_-,_..._-J-_______
.+-_. ---
_____.. --_ ... _- ---+--:;:-...:.:_-: _.
.
-:::::-7. ~~: ~~ ,
.-
, .~-=-: . t
- - - 1---
. --.----1---....L-_____ __._
; i
1--7""---="":-, --:-;- ,
: !! t ! !:
COLSOl : 7800 ....:
R.H.S. REDUCTION
, + BACKSUB. :
: j! I ~ •
I:
1~~~~
• . -. ,. -~
-, ~
-:~-.-_~
--- - - r-...L- f -7" - ~- FISSlE : CRAY
. .... .. . . .• -
,. ,' +
--..,-......;-y--+-~:+ R.H.S. REDUCTION
BACKSUBSTITUTION
-'-;~ 'TI -----'---"-~-"' -
: ' !I :f I: ;
i
; !' 4, 1!
1T"Tf
I : I
'I i i i!1
.Ol~
! I . __~__~~~~
__~~~~~~____~~~~~~ : I I
...o 8 ...§
AVG. COL. HGT. -
552
2 5
12 13
Substructure 1.
1 2 3 4 5 6 7 8 9--11--13 1 2 3 410 9
XXX X XX XX
XX X XXX
XX XX XXXX
X XX XXX
X X X X0 0 III III
X X X0 0 III
X X X0 X Substructure 2.
X X0 X 5 6 7 8 911
XXX
XXXX xxxx
XX X xxx
XX xxxx
X xxx
QQ
iii
Substructure 3.
x - Non-zero in Triangular 9--11--13
Factor. QQQ
o - Zero ln Triangular Factor. Qxx x
Qx x
a - Substructure Coefficients. xx
x
P.G.BERGAN
The Norwegian Institute of Techno!ogy, Trondhelm, Norway
Summary
A method of adjustlng the load lntensity in order to minimize
the residual (unbalanced) forces from the equilibrlum equa-
tions is suggested. This technique may be used as a means
of controlling the steps for load or displacement pattern
incrementatlon schemes. It may also be used for obtaining
improved convergence during equllibrium iterations of initi-
al stress (strain), quasi-Newton or true Newton-Raphson type.
Introduction
( 1)
and
( 2)
555
where
R
unbal,j
= R - R
int,J
(4 )
R
1
r
I
R= p Rref (5)
prescrlbed
load path
minimum
dlrection
This is the condition for finding the best suited load level
The "s~ze" of the unbalanced force should be measured
in terms of some norm, e.g. a weighted Euclidean norm defin-
ed by
N
II X II (E (7)
k=l
N
E W2 R R
k=l k ref,k int,k
(8)
N
L
k=l
R = R(p,r) (9)
Cl P-Pa
Cl~ I I Runball I -;;-11
op
R
a
+ - - ( R -R)
Pb-Pa b a -
R
1nt
II o (10)
N 2
Z Wk2 (R b ,k - R k)
k=l a,
secant path
R
n
true
load
path
(12)
Incrementation of displacements
(13)
561
II ~r ref II
y (14)
IIMil1
Discussion of performance
diverging direction
r
2
converging direction
t ideal direction
displacement space
R21~1 3 pr
1
b2 bl b
~:pace
Rl
Flg. 5 - Convergence propertles when using constant gradlent
563
~rk,j-l + ~rk,j
(15)
~rk,J-l
Numerical studies
sp (16)
where index 1 refers to the in1tial state and the first step.
lIPi II MIl I
S (17 )
P lIP l ll llr i l1
0.5
a.
O.O+-____~______~__~~----~~~--~_
-0.5
o2
1.0
b. 0.5
3.5
o1
s
P
1.0
A
c. 0.5
-0.5
V2
a.
sp
-0.5
-1.0
!t
1.0
A
B
O.S
'
•••••
.
/.
...... / C
'.
a.
'. '.
" . ....
°2
loS
1.0
~ ..............................
O.S
1.0
Results from some more test cases are described in Ref. /8/.
References
v2
H ,
Ho v
L 0 L
Spring coefficients
a+bv 2 , a , bL
k1 = const. , k2 m
0 m1
k1 k1
Auxi11iary quantities
f
f 1 + H2 + v 2 f2 (....2)~ - 1
0 0 0 f1
2 2 1+f2
f1 1 + (H o -0 1 ) + (v + 2)
0 f3
~
Nondimensiona1 potential energy
Equlllbrlum equatlons
P1 2f 2 (H o - 01)
P2 mo 0 2 + m10~ - 2f 2 (v o +0 2 )
571
1: BELYTSCHKO
Northwestern University,Evanston, USA
Swnmary
Introduction
and implic~t methods. They have shown that the max~mum fre-
quency in the explic~t partition governs the t~e step, so
that if the stiff portions are ~ntegrated ~mplic~tly, the
t~me step can be much larger. Hughes and Liu [3] have pro-
posed an expl~c~t-~mpl~c~t part~t~on by elements w~th s~mi
lar stab~l~ty properties; the element bas~s of the~r part~
Governing Equations
f (1)
where
f (2)
( 3)
.
E B ~ = ~ ~E u
. (4)
K u (5)
Mu + K ~ (6)
n+l n
u u + I1tn+~ un+~ (8)
+ (10)
n+l
o (11)
K u ( 13)
2c
w (14)
max T
(15)
wn+l,ext
(17a)
577
wn,~nt
E
+
I
2" l'It J ~n+~{~n + ~n+l) dV (17b)
VE
w~nt
L w~nt
E
(17c)
E
I .n+~ T
(~ ) M un+~ (18)
2"
where w~nt and wext are the ~nternal energy and external
work, and T is the k~net~c energy. Energy balance then re-
quires that
(19)
Part~tioned Algorithm
Table 1
Partitioned Explic1t Algor1thm
o
o. in1tial conditions: ~ ,
579
1. tM -+- tM + Li tM
2. compute !ext, set !int = 0, LitN = 0
4. if tG>t M, go to 7
9. ~f tN > t M, go to 11
12. go to 1
Adapt~ve Algor~thm
step litE' and each node is advanced by its own t~me step
Li~. The nodal time step ~s the min~murn of the element
580
(20)
Table 2
Adaptive Explicit Algorithm
o ._!:z ·lnt
o. condltlons: u , u , f = 0
o for all elements and nodes, tM
1. loop over elements: E = 1 to NELE
2. If tE > tM - threshold, go to 7
9. If tN > t M, go to 14
15. tM + tM + t,tM
16. go to 1
The time step for adjacent nodes and elements cannot be com-
pletely arbitrary. One ~mportant restr~ction is that the
element time step must be smaller than the nodal time steps
of its nodes. Th~s restrict~on detracts from the method
substant~ally, for ~t would be des~rable to be able to
increment the element at the largest possible t~me step,
s~nce updating an element is far more costly than updating
a node.
Example
The adapt~ve method was also tr~ed for other problems of the
same type. At t~mes the solution was observed to drift
from the correct solution. It was not a violent Neumann
type of instab~l~ty, although energy balance was v~olated.
10 Elements (~=O.l)
-0.1
Ul
Ul
Q)
'"
~ -0.2
10 20 30 40 50
Tlme
Analytlc
Ul
Ul
Q)
'"
.j.J
rJl
-0.2
x = 4.5
10 20 30 40 50
Tlme
Acknowledgement
Foundation.
References
ABSTRACT
1. INTRODUCTION
2. MECHANICS
k
an = pE aX k dv (1)
1V ~l
- f (tkm,m k
+ pf ) ,~xk dv + f km
(t km _ t_
+ )nmox k da
r 7°
_ s k )ox da
+ ,f (tkmn
m k 0 (2)
7 1
t km fm o in r ( 3)
o on g> ° (4 )
(5)
589
FII 1 The body 'If WIth surface tractloru I' on the boundary ~ I and a prescnbed motion on the boundary Jl An Intenor
boundary cI 0 with a umt nonnal vector ,./r IS PictUred
a
on y> (6)
ayk Q
.:.A- (x"',t) (7)
axel
(8)
d km 1
2 (vk,m + vm,k) (10)
1
wkm 2 (vk,m - vm,k) (11)
3. CONSTITUTIVE THEORIES
(13)
V'rs
t (14)
' l d '~ng ~s
Wh en no y~e ' crsmn ~s
, occurr~ng, , the ~sotrop~c
, , tensor
Agrsg mn + 2~grmgsn, where A and ~ are the Lame parameters.
When plasti; yielding is occurring, that is, when! ~~,~~ij _
2 c~ ~,r~ 2 lJ
k = 0 and ~rs ~ > 0, the moduli are replaced by
(15)
593
where
/2k S ~ P
3 Hld rs I , k (O) (2/3) 1/2 t
0
\/
a 2 P
(1 - S)
rs 3" H d rs ars(O) 0
where
n - 1
-2-
l2 0 (l
,2
t:1J,t "i j ) exp(-Q/R8)t"
rnn
i i
(1 - a)x n + aX n + l (17)
h .. u .. (18)
1J 1,J
e .. (a) (19)
1J
1
e .. sym (20 )
1J 2"
596
1 i -1 1
R ..
1J
(oi
k
- l1t 2' w k) (gkj + ~t 2' wkj ) (21)
-n+~
a R~i R~j a n.. (23)
rs r s 1J
5. SPATIAL DISCRETIZATION
L-------------------------~_r
y (24)
When all the elements around node 1 are considered, the basis
function obtained from the interpolation functions in each
element can be pictured. Flgure 4 shows the result for this
case. As can be seen in Fig. 4, the basis functions obtalned
600
(26)
a vector
601
k~a
s ~k a vector.
The integrals for [T] over each element are performed numeri-
cally using Gaussian quadrature rules. A 2x2 integration is
used to evaluate the area integral. In the process a con-
stant bulk strain is used. This may be viewed as a 1 point
integration of the volumetric strain energy and a 2x2 inte-
gration of the deviatoric strain energy, [28-30].
6. DYNAMIC RELAXATION
(27)
(28 )
where
q a vector
(29)
(29) become
(30)
(31)
605
tn (32)
c
(33)
2 c/h (34)
(W ) (35)
,0
(36)
7. APPLICATIONS
DEGRADED ANHYDRITE OR
SALT SALT POLYHALITE ~g~~~~~i~'!SALT
Young's Modulus (Pa) 2.5 x 10 10 2.5 x 10 10 7.3 x 10 10 5.28 x 10 10
Creep Constants
n 5. 5. ------- 5.
Coeffl.cl.ent of Frlctl.On)J O.
610
T
11.
~~r
1.~~ .j
P = 1.41 x 10 7 Pa
Material Proj2erties
9
G = 9.93 x 10 Pa
K = 1. 65 x 10 10 Pa
EC = Aa
n
.036
.032
.028
.024
100. x IItcr
.
~
"'-
.020
..~
-
,
Q)
.016
-..-:..
--........
o
.-I
U
~~ ....
Q)
lit 10. x lit _~
-~
.-I
o cr _
..,.,.".'" ..,;"
~
::r:: .012
__ -.:-':;:""-;; = II t
__ ~.... cr
/ ......
.008
/:/
.,//
.004
V-;/'
/
/
o.
o. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
Time (Years)
l !III
19.32
I
g
5.03
104.62
Lx
Figure 8. Dimensions and Boundary Conditions
for the Two-Dimensional Analysis of
an Underground Opening
613
POL YHALITEI
ANHYDRITE CLAY SEAM
-
=
~
601.7
606.6
609.9
611.4
622.1
(MODEL BOUNDARY)
=
VVVVV\
~ 626.7 ~
..i:ivViM
~
634.6
639.5
-- --
--
VI/VVII\
647.1
- 649.5
~
663.6
664.5
670.6
683.4
~ 685.8 ~
693.7
-
=-==-
-:.-:..
~ 698.0
708.0
~ 710.5 (MODEL BOUNDARY) 'VVINV\
711.7
-==~
~-=- 717.8
~~
2000
.14
1800
/
.....-
....-
-- -
....-
....- ....-
.12 1600
s
-- --- - --
Iterations
'"--
1400
,,,
~
Q)
1-1
.10
::I
rn
0
I 1200 rn
r-i
I::
0
U •.-1
.08 .j..I
.j..I
'1-1 1000 III
1-1
•.-1 Q)
1-1 .j..I
C H
r-i .06 800 r-i
III III
U .j..I
•.-1
0
.j..I
1-1 600 Eo!
Q)
:> .04
400
.02
200
o. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.
Time (Years)
GEOMETRY
WA = 3. 0 in. ( . 0762
-----------l
6L = 7 . 0 in. (.1778 m) L
IIIIIIIIII!I~ ~
I" L + 6L
PROPERTIES 11
E = 30. x 10 6 psi (2.07 x 10 pal H =5 x 10 4 psi (3.447 x 10 8 pal
0y = 4. x 10 4 psi (2.758 x 10 8 pal v = .3
- O.OS
Cooverqanc.
- 0.10 Tolerance ttl
- 0 . 15
- 0.20
- 0.25 L--.........---"_~--'-~'---o---'---I
O. . 1 .2 .3 •• .5 ., .7 ••
\! .3
Theory
t ---] a=l. b=2.
8 6 Finite Element Results
cry = 3.1 x 10 Pa
Pb j k = cr
y 113
o. ;p 1.0
0.9
0.8
0.7
.><:
N
0.6 ~
N'><: ~
81'" b "-<D
b
0.5
0.4
1.
-0.7 0.3
REFERENCES
H. D MITTELMANN
Ulllversltat Dortmund, Germany
Summary
We cons1der var1ational problems resp. var1at1onal in-
equal1tles which correspond to quasilinear elliptic boun-
dary value problems without resp. with llnear constraints.
Finite element discret1zat1on leads to nonlinear algebraic
systems resp. llnearly constrained nonlinear programm1ng
problems. Eff1c1ent algor1thms are given for the solution
of these problems. Convergence 1S shown and some numer1cal
results are presented. Applications are for the unrestric-
ted case all p,-oblems Wh1Ch are glven by a m1n1mizat1on
problem and for problems w1th constraints e. g. elast1c-
plast1c torsion or elast1c contact problems.
1. Introduct1on
In the following we will consider optimization problems of
the form
f(x) = m1n
S (1. 1)
M
S = {x e.JR , g(x) G'l'x + g < O}
o -
where f : JRM -+ JR .
1S a str1ct 1 y convex not necessar1'1 y
(1 .5)
p <
M
S = {x E JR , c < x < d} (1 .6)
(k) M (k+1)
Let x e. S = {x E JR , x .::. o} be given. Compute x accor-
ding to
O.~, k = f ~~
.. (x.~, k) (1 .8)
f(x) min
S (2.1)
M T
S {xEJR , G x + go > O}
h
were G = ( EM,-E ) EJRM, 2M , EM t h e MxM 1
d entlty matrlx
' and
M
T T T i
go = (-c ,d ) . We denote the columns of G by g ,i=1, ... ,2M,
T
and deflne I(x) {lE{1, ... ,2M}, glx + goi = O}.
k
tor on the subspace of the actlve constraints. Finally let
r =
k
S = {
O, if j =
T T
° or Ak = 1 or llk
k zkrk/zk_1rk_1 otherwise.
j 1, i f Ak = 1
j + 1 otherwise,
k k + 1 and go to Step 1.
(2.3)
(k) f. (x. k)
(k+1 ) ( . (d 1 1, ) )
xi max c i ' mln i'x i - wi,k a (x )
ii i,k
(2.4)
i = 1, ••• ,M, a ii diagonal element of A in (2.2)
628
3. Numerical Results
r 2 2 1/2
a(l-r Ib)
.
,1f x = 0, -b ~ Y ~ b
¢ (x,y)
l 0 otherwise
Operat.
15000
10000
5000
cg-SMOR-N
---
residual
0
10
MOR-N
10
-2 "'----
-...
"""" .......... .......
-4
10
phase I phase I I
-6
10
10 20 0
relaxation steps
Fig:. 2 Residual for relaxation and 2-phase method
act. constr. I
17
/
.,P" cg-SMOR-N
10 I
MOR-N I
"
5
I
10 20 30 40
relaxation s:eps
Fig:. 3 Number of active constraints
these reduces to
Au = f. (4. 1 )
For a sequence of stepsizes {h,q,} with
h
0
> h1 > h2 > ... > h,q, > . .. > 0
(4.2)
1,2, ..•
(4.3)
(4.4)
T
if A~ D~ + L~ + L~ . Then a simple computation shows that
where S
o
5. Conclusions
References
Summary
In the present paper, distlnction lS made between frictional
and frictionless contact problems. After their formulation,
the epitome of frlctlonal problems, V1Z. the problem of
frictional roillng contact is considered, and the two most
recent and successful methods for deallng wlth lt are pre-
sented. Then the frictlonless problems lS considered. Two
palrs of methods are presented, each palr conslsting of a
non-varlatlonal method, wlth ltS variational counterpart
authored by us. The flrst palr, V1Z. Paul & Hashemi I Kalker
& Allaert, deal wlth the general half-space frictlonless
contact problem. The second palr, V1Z. Reusner I Kalker & De
Mul, deal with the contact of bodles of revolutlon, WhlCh
may be approximated by half-spaces. The codes implementlng
the frictionless problems suffer from long operatlng tlmes.
Research lS stlll ln progress to speed up the codes.
1. Introductlon
outer norffial on av a (I )
638
Pn ~ 0 in contact o outside it (2 )
a a
p u (3 )
min!
u,p together
II z=o
Pn(X'Y){~Un(X'Y) - h(x,y)}dxdy
sub PT(x,y) 0
u + 0 as (x,y,z) + 00 (5)
u
n
(x, y) - 1 - cr
- 21i'G
II z=o
Pn (x' , y' )
I
dx'dy'
2
-
2
(x-x') + (y-y')
(x,y) : Cartes~an coordinates on the surface of the
min!
u,p together
II z=O
Pn(x,y,t){~un(x,y,t)-
- h(x,y,t)}dxdy + II K(t-T)
PT(x,y,t){~UT(x,y,t) -
min!
u,p together
II z=O
Pn(x,y,t){~un(x,y,t)-
- h(x,y,t)}dxdy + fI PT(x,y,t){~uT(x,y,t)-
J z=O
640
- UT(X,y,t-T) - p(x,y,t;T)}dxdy
2 2
sub Pn ~ 0 , IpTI ~ g (g = - fPn ' known) (8)
PS(x,y,t)g(x,y,t)dxdy =
o= II z=O
OPn{un(t) - h(t) + D}dxdy + II z=O
OPT{uT(t)-
u. (x,y) =
1
ff K
p (x' ,y' )H .. (x-x' , y-y' )dx'dy',
J 1J
E u (x,y)p. (x,y)dxdy
1 1
N
~ I I
I,J=l 1,j=1,2
M2I + 1 ,2J+]' P2I+i P2J+]'
642
min!
{Pi}
(Y L
I,J=l i,j=1,2
(~M2I+~,2J+J P2I+i P2J+j) +
Y
I=l i=1,2
L (u 2I + i (t - T) + P2I+i(t;T))P2I+~ QI)
2 2 2
sub P2I+i + P2I+2 gI ' I = 1, ... ,N ; QI: area of
2 2 < 2
restrictions P2I+l + P2I+2 gI· In the present program, p
is represented by the polar representation (P2I+l,P2I+2)
= (r I cos ¢I ' r I sin ¢I) when r I = gI' so that the restric-
tion reduces to the bound r I ~ gI.
The rectangular representat10n 1S mainta1ned when r I < gI'
because the polar representation becomes slngular when
r I = 0, and this introduces spur10us Kuhn-Tucker points. As
a consequence, we shift back and forth between represent-
ations, whereby the adaptation of the gradient and Hessian
is most consuming.
(12)
ap'l'(x,y)
w(x,y) = slip p(x,y) - L , x: rolling
ax
direction (a)
IpH I
L(l - -g---)PT = - APT
IpH I
A (-g---) - l)L ~ 0 if IpT I g
which is Coulomb's law.
The algorithm must be provided with a boundary condition,
for wh~ch reference is made to [10J. It is extremely fast, a
reduction factor of 100 in speed of calculat~on is achieved
with respect to the algorithm of sec. 3.1, with errors of an
estimated maximum of 15% of the maximum value of the total
tangential force. Such high errors are acceptable in rail
vehicle dynamics, where owing to pollution of wheels and
ra~ls, unpredictable deviations of up to 40% from the theory
have been observed.
(14)
a2F
Let f(x,y) be such that dxdy = f(x,y) ; then
-1
Now, if f(x,y) = I x 2 + y2 9 F(x,y) =
-1
= x sh- 1 (Y/ixi) + y Sh- 1 (x/iyi) ; sh z = inverse
so that
(( F( ) I -x J -1
Jxx=x JY I -Y J -k "close" (17a)
x,y -x +1 y=y -y +k
I J I J
4lk
"far"
/ (XI-X J ) 2 + (YI- YJ)2
2 2
if (xI - x J ) + (Y I - YJ )
e.g. 4 (17b)
( 18)
truly fast.
Whereas ~n the railway application, congruent rectangles
seem appropriate, this is not so in the bearing application,
where stress concentrations occur at the transition to the
sharp rounding of the roller. The large gradients of the
stress necessitate a locally very fine mesh, so f~ne that
the number of equal elements would run into many hundreds or
even thousands, which would lead to a very slow algorithm.
When the discretization is no longer uniform, the ~nfluence
_ 1/6 / x 2 + y2 3
fJ Q h(x,y)dxdy
I
2
- kJ ) } ; (19)
with C » 1 , e.g. 2:
MIJ ~ 16
I (xI - x J )
2
+ (YI - YJ)
2 -1
lK1Jklk J (19b)
5. Conclusions
L1terature
Captions of figures
Common
normal
(b)
negative distance h
(b)
n~o~c~o~n~ta~c~t~~c~o~n~ta~c~t_______ x
Boundary
pOint
Exterior Penalty Methods for Contact Problems in Elasticity
J.1:0DEN
The Umverslty of Texas, Austill, USA
Summary
This paper reviews recent results of aden, Kikuchi, and
Song [4] on the use of exterior penalty methods as a basis
for finite element approximations of contact problems in
linear elasticity.
Introduction
Th1s note reviews several results recently obtained by the
author and his colleagues, N. Kikuchi and Y. J. Song, on
the use of exterior penalty methods for unilateral problems
in elasticity theory. The physical problem considered is
that of a l1near elastic body deflected by the applicat10n
of external forces so that it comes in contact with a rigid
fr1ct10nless foundation. An appropr1ate var1ational state-
ment of this problem is as follows:
Find (u,o) E V x M such that
B{u,v) - <o,y n (v»
_
_ - s> > 0
<T - o,y n (u)
= f{v) V
V
v EV
TE M
} (l)
Here,
o = contact pressure
656
D
B(u,v) 2fJ Jrl E:.(u) E: D•• (v) dx , where
1J - 1J-
1
(u) = -2 (u. . + u. .)
E: ••
1J - 1,J J,l
<.,.> denotes duality pairing on W' x W , where
v. .v. . dx ;
1,J 1,J
}
There exists B > 0 such that
(3)
I <T,y n (v)
_ >I
B IITllw I < sup Ii T € W'
v€V
Moreover, set
a
E
- !E j (y (u ) -s)
n -E +
(4)
Approx~mations
L
E G
J(f} =
e=l
Je(f} , J
e
(f)
L
j=l
(S)
'" frC f ds •
659
h h 1 h h h
L (v ,T ) 2" B (~ ,~ ) - f (v )
£ -
(7)
h h
then its saddle p01nts (u ,0 ) are characterized by the
-£ -£
system
h h J[o h£ y n (v_ h ) j V ~ h E: Vh
B(u ,v )
-£ -
- f(v h )
} (8)
h h h £Ohj} > 0 h
J{ (T -0£) [(Y n (~£) -s) + £ - V T € Mh
(9)
660
_ h }]
IJ[Thy n {v
h
}
8hllTllo< sup
h
~ E Vh
lI~hlll {lO}
for every Th € Wh
1
+ "D
iJh
IIyn (u-v
- -
h ) 110 r ]
, C
h
+ inf [II a_1 h Ilw' + <1 -a,y (u)
n -
1hE~
_ s>11/2] + inf I<T-a h ,y n (u)
-
-s> I
1/2
1eM
1/2
+ ~ II a h II 0 r + EJ } (11)
iJh ' C
Discussion of Results
1. G3 FAILS
2. G2 FAILS
3. Gl FAILS
4. S CONSTANT
(stable)
5. T CONSTANT
(stable)
6. G3 FAILS
7. S CONSTANT
(stable)
8. T CONSTANT
(stable)
9. G2 FAILS
10. T CONSTANT
(stable)
ll. G3 FAILS
12,. Q8 G2 FAILS
13. Q8 Gl FAILS
14. Q8 S CONSTANT
(stable)
15. Q8 T CONSTANT
(stable)
I
-Llhearly Elastlc Founda tlon-
I
1000
1 4
J
0.3
I. 8 .1
Flg. 2 Model for numerical example of a
Signorini problem.
664
100
200
100
200
Hertz' F.E.M.
Solution Solution
300
a 3.873 3.6
p 258 275
400
References
1. INTRODUCTION
The main goal of this paper is to describe some numerical
methods for solving nonlinear variational problems in incom-
pressible finite elasticity.
In Sec. 2 we discuss a decomposition princlple for a large
class of variational problems and then derive several itera-
tive methods of solution from this principle. We give in
Sec. 3, the formulation of elastostatics two-dimensional and
axisymmetric problems for incompressible Mooney-Rlvlln mate-
rials.
In Sec. 4 we descrlbe the application of the algorithms of
Sec. 2 to the iterative solution of the mechanlcal problems
of Sec. 3. In Sec. 5 we give some brief indicatlon on the fi-
nite element approximation of the problems of Sec. 3
Finally some numerical results obtained applying the methods
of Sec. 4,5 are presented and discussed in Sec. 6
where J : V + IR is defined by
It appears from (2.~) that J(.) and therefore (P) have very
special structures and taking this into account it is then
quite natural to design special methods of solution for (P).
(2.3)
Remark 2.3 : The new problem (TI) has clearly a mixed formula-
tion "flavor" since the linear relation Bv-q 0 suggests the
introduction of a Lagrange multiplier.
Remark 2.4 : Problems (P) and (TI) are equivalent, but consi-
dering (TI) we have in some sense simplified the nonlinear
structure of (P), at the expense however of the new variable
q and of the relation
(2.7) Bv-q o .
(2.8)
{2. 9}
669
n n n ,n+1
then if n ~O, A given, we compute u ,p ,A by
~ Find {un ,pn} E VxH such that
(2.11)
t.er(Un,pn,A n } $£r(v,q,>,n} TV {v,q} EVXH,
(2.12) An +1
n .... u
(2.14) u strongly in V,
n ....
(2.15) p P = Bu strongly in H,
-1 0
(2.17) {u ,A} E VxH, given,
then for n ~ 0, u
n-l ,A n given, we compute p n , u,/\
n, n+l £y
(2.22)
671
rl r (n
u ,pn ,/\,n+]/2) -<rlr ( v,p n ,II.,n+l/2) Vv E V,
(2.24)
1+ 15
(2.26) o < p <--2- r.
-1
(2.28) u given,
then for n ~ 0,
n-l n-l
ru A2 (u ),
(2.30)
n n
ru - A2 (U ),
n+l/2 A ( n+l/2)
ru - 1 u
(2.35)
~~ + A(u) 0
where A
gor~thms of Sec. 2.5 and ADI methods have been observed for
the first time by CHAN-GLOWINSKI [23]. In Sec. 2 we have
basically followed the presentation of FORTIN-GLOWINSKI [5]
and GLOWINSKI [6].
- = Jnp0 Jan
(3.1) n(v) (o(v)-f'v)dx - S·V df ,
- - - - 0-
2
where n is a bounded domain of lRN corresponding to the re-
ference configuration ; an(= an1 u an 2 ) is the boundary of n,
the body being fixed on anI' We have denoted by o(~) the
stored energy functional (per mass unit). For a Mooney-Riv-
lin material we have
(3.2) a (v)
(3.3)
(3.4) F Ilv+I
(3.6) S
-0
K 1
(3.8)
a. e. ,
d
(3.10) [~,~] = ~ (det F(u»v . . ,
~,J --- ~ - ~,J
(3.11)
2 NxN .
Find {~,;, ~} E: W = XxYx (L W» , stat~onary point
over W of the augmented lagrangian
(3.12)
.eR(~'~'~) = n(v) + ~11~~+;-~1I~2
- Jrl~· (~~-;-~) d~ ,
implles
)
o
(3.14) I+Vu o
e (u)
u
where e(u) = 1+ ~
r
is the extenslon ratio in the circumfe-
rential direction, and where
dU.
_J
dZ
678
2
e + (0 .. +u. )
2
1J 1, J
det(~+?~) = e(1+ul,lu2,2-ul,2u2,1+ul,1+u2,2).
For incompressible materials we have
det(~+?~) = 1 a.e.
implying in turn
2 2 2
I 2 (tl) = e (o .. +u . . ) + lie.
- 1J 1, J
a) Spaces
X {VE (lfl)2, v=O on 3~1}'
and with 1 s i, j s 2
124
z {ll
o
, ll· . } Eol
1J
x (.e) }
b) Augmented lagrangian
olR(V,G,ll) R
= n(v) + -2 f (0 .. +v . . -g .. ) 2 dx
- ::: '" - ~ 1J 1, J 1J -
(3.15)
-J ( R
ll .. (o .. +v .. -g .. )dx + -2 f (g -e(v)) 2 dx-
~ 1J 1J 1J 1J - ~ 0 - -
679
-f S"l
~o(e(v)-g
-
)dx.
0-
'"
{Un,F n } XxY and V {v,G} Xxy we have
-
E E
'"
(4.2)
JJR(~n'F'n,:n)
L__ .~ S
JJ
L
( G ,n) ,
R ~, '" ' ~
nnn, n n n
(4.4) .eR(~ ,!
,~ )
s .... R(~ '~'0) VG E Y, ;' E Y,
n
(4.5) d~R (~ ,! n ' ~ n ).~ = 0 "Iv E X, U
n
E X,
(4.6)
(4.7)
(4.8)
(4.9)
G ....
'"
J!G 2
[RG. . - 2 (R (u .
1J 1,J
. +0
1J
) - A .. ) G.
1J 1J
] dx
-
681
w~th
1 a.e. on ~}.
Find {F .. }
-- ~J
E c:. and minimizing over C) the functional
(4.13)
where
(4.14)
(4.15)
where
(4.17)
(4.18)
682
solution of
(4.19)
2 2
We suppose that zl+z4 ~ 0 ; then it can be shown that (4.19)
has a unique solution in ]-R,+R[ ; moreover, using the im-
plicit function theorem (see [24], [25] for more details) one
can show that this solution p of (4.19) between -R and R is
prec~se1y the one associated with the global minimum of
c + RC~-2z.c. over £.c~
~ ~ ~ ~ ~
= 2, to which, therefore, corresponds
a unique global minimizer given from p in (4.16), (4.17) (ac-
tually there is no other minimizer (local or global) than the
above global minimizer).
5.2.1. Quadrangulation of ~.
(5.1)
k
UIh
E
~k·
"
X2 A4
AI
AS "Xl
AZ A3
(5.2)
(5.3)
(5.4)
685
1
(5.6) Xh Hh n X,
(5.7) Zh (P ) 4
h
(5.8) Yh Y n Zh'
Remark 5.1 : It follows from (5.2), (5.4) that the above fi-
nite element approximat1on is of the isoparametric type . •
Either
or
( 5 . 9) 2 sh (
~h) \k = L 2 -proJec
. t '10n 0 f "~~h on Zh'
det (I+\lv) 1
at GT only.
Notice that because the nonsymmetrlc structure of the AQL
element we have to make some restriction on the triangula-
tion upon which the space of the approximate displacements
is to be deflned ; one suggests in [30) the following pro-
cess that generates nearly as general a mesh as any other
finite element triangulation :
We first partltion Q into convex quadrilaterals arbltrarily
then each quadrilateral is subdivided into two triangles by
one of its two diagonals (any of them can be used). The
edges over which the restrictlons of the second order poly-
nomials of P 4 / 3 are allowed to be quadratlc are preclsely
those dlagonals.
Hl o -
(5.12) {Yh E C W), YhIT EP 4j3 'IT E 'eh } ,
h
1
(5.14) Xh X n Hh
4
(5.15) Zh (Ph) ,
6. - NUMERICAL EXPERIMENTS.
All the numerical experiments displayed in this sect10n deal
w1th Mooney-Riv1in materials and either a plane strain or
aX1symmetric situation ; moreover the numer1ca1 results pre-
sented here have been obtained using the quadrilateral fini-
te element approxim~tion discussed in Sec. 5.2. Other numeri-
cal experiments are done in [24] ,[25] (including the numeri-
cal treatment of plane stress problems) ; numerical exper1-
ments using th1s Ruas' element of Sec. 5.3 (and some of its
generalizations are discussed 1n [30] ,[31].
-4
match the applied tension with a 10 precis1on.
CAACK u,. 0
C1 =1. PSI
HEIGTH =1.75IM
WIDTH =1.95IM
CRACK LENGTH =0.50IM
STRAIN ENERGY =2. 212FT L8
Figure 6.1
I,
I,
I,
I,
I,
anI an2
I,
,I
I,
I,
I,
I,
r
I 1'-.
V/////////////// '/ '/'//// '/'/ '/////"
F:'gure 6.2
The numerlcal values glven in Table 6.1, both for the ana-
lytical and numerical solution correspond to
The computed Cauchy stresses 0ee and orr in the mid cross-
section are indicated (as functions of r) on Fig. 6.3, for
different values of the parameters Nand Q ; these computed
values (indicated by crosses on Fig. 6.3) are exactly loca-
ted on the curves obtained analytically in [32] and repro-
duced on Fig. 6.3 ; this confirms the validity of our compu-
tations.
Table 6.1
Cl =0 . 8750pSI
C2 =0. 1250pSI
HEIGT H =1.2IN
3. 0 a ee
0.5 ____
. .. . . .
~_~ _ _ _ _ _ __
~
O. 1
0.2
0.3
O. 4
O. 5
0.6
0.7
0.8 0
rr
Figure 6.3
693
REFERENCES
D.S.MALKUS
IllmOis Institute of Technology, ChIcago, USA
Summary
The basic stability condition for mixed/Lagrange multiplier
variational principles in incompressible media problems is
the LBB condition. It plays an essential role in determin-
ing whether or not the problem is well-posed and governs the
choice of finite elements in the discretization. This paper
examines the discrete eigenstructure of a well-known La-
grange multiplier formulation for linear elasticity or
Stokes-flow. It shows how the weak incompressibility con-
straint is reflected in the elementary-divisor structure of
the eigenproblem whose solution determines the finite ele-
ment approximation to the natural modes. In this context
the discrete LBB condition can be seen to be a condition
determining the limiting disposition, as the mesh parameter
decreases, of a matrix pencil with infinite eigenvalues.
Pure pressure modes and the load vectors required to trans-
mit them are paired in cyclic subspaces of the infinite
eigenspace. This pairing can be related to a well-known
heuristic interpretation of the LBB condition. The rela-
tionship between the natural mode eigenproblem and the
eigenproblems which determine the norm of the inverse of
the discrete operator of a static or steady-flow problem is
described. Finally, because of the equivalence between
classes of mixed and penalty formulations, it is shown that
these results apply to penalty/reduced integration finite
element methods.
Introduction
This paper discusses certain aspects of the discrete
stability condition for finite element models in incompress-
ible elasticity or fluid flow. The term stability is used
to denote the uniform invertibility of the discrete operator
in the variational statement of a static or steady-flow
problem [1-4J. The approach taken here is to view the dis-
crete problem from an algebraic point of view. This approach
does have its limitations because, by nature, a stability
697
{ Find U E H';" ¥ V E: H (1 )
B(U,V) = F(V)
F E: H* is a standard load integral. H is the Hilbert space
HIo X HO• U = (B,p) ~ H is a displacement or velocity/
pressure pair, and the subscript "0" denotes that B may
satisfy appropriate essential boundary conditions. For the
purpose at hand, there is no loss of generality in assuming
that these conditions are homogeneous, so that H is a linear
space. H* is the dual of H. The form of B(U,V) is assumed
to be:
B[(B,p),(~,q)]= l[2YE: ij f ij -E-uq- fup]dD. (2)
where E: ••
1J
=~ (~Ui
uX.
+ ~Uj)
uX.
, f ..
1J
=~ (~Vi
uX.
+ :v j )
uX.
and repeated
J 1 J 1
indices imply tensorial contraction; p is shear modulus, and
-pip is the hydrostatic pressure function for elasticity;
p is a viscosity with p the pressure in Stokes-flow.
It can be shown that B(U,V) satisfies three important
conditions which guarantee that the problem of (1) is well-
posed, given appropriate conditions on the domain [3J and
698
ii.
(4)
sup
h
I ( E: I:.
J.J.
,JIll ~ ch3
Ilu'" II H1=1
Inequality (6) is the continuous LBB condition and inequality
(7) together with the requirement
lim inf c~ ~
h-+o
c; > 0 (8)
(9)
uT ~2 v = 2(1)
.n.
E-~j~jda ¥Bh ,r E: Sh
T (10)
u ~l P
T
v F=
pT .9q =
.n.
1
phqh dll >fph,qh E ~ (11)
(14)
l[y
Define the weakly constrained Rayleigh quotient
E ij E- ij - EiiPJdIl.
R(u,p) - (16)
~u (17)
or
(18)
701
B. Convergence
Assuming the consistency of the finite element approxima-
tion, that is assuming that Sh x Th can accurately approximate
arbitrary members of H [3J, what is needed to establish the
convergence of the finite element solutions to (9) towards
the exact solution is verification of condition ii* (equation
4). Using the norm given in (13), the satisfaction of ii*
can be determined from the following eigenproblem:
(19)
The following fact is surely well-known by most, and is
stated here as the first theorem for the sake of completeness.
The proof of this result and the proofs of all other theorems
stated here may be found in an extended version of this paper
[6 J.
THEOREM 1: I f I A 11 .s. I A 21 .s. ... .s. I ). N I are the eigen-
values of (19), then
inf sup IB(Uh,Vh)1 1/..11
II vhll=l Iluhll=l
C. LBB Eigenproblem
The eigenvalues of the following eigenproblem determine
the satisfaction of the discrete LBB condition:
(20)
where
(21)
m { 0
_ ~Ol
I
___ +- ----
I
[~ ~J
I
I
1
2(k-m) I • (23)
t~ ~]I ___ _
I
1- ;..
I n-k+m-A
n-k+m
{
This is a block diagonal matrix. Un-
specified entries are zero.
It is useful here to introduce an auxilliary eigenproblem:
*
B. Adjoint LBB eigenproblem
- T -1 -
lSI Jj lSI P = A Qp (24)
Let ql' ... ,qn be the null eigenvectors of (24) and PI'··· ,Pk-m
be the eigenvectors of (24) with nonzero /I.. Let" 1" refer
to orthogonality in the norm of ~ (given via Q):
Lemma 3.2.1: The qi span the kernel of lSI' Ker lSI' and
Pi span the coimage, Colm RI = {Ker ih~ L. For each Pi'
the vector z.l = M- I rvKl p.l is dilatational, i. e., KIT z l. /.L 0,
t"V I"V
x. -- [__C2._]
1 Pi
[-~~-] 1
[-;~-l 1
b) JS [-~~-l = 0 (27)
1
up" in a nonlinear divisor. The fact that for any pure pres-
sure mode [ 0:. p T]T
[-~~-l
1
[-~~-l
707
-1 - T
Q
~
Kl u.l
~
1 + vll + 4cri
with A = A~ = 2
-1 - T
Q
~
Kl u.l
~
1 - vll + 4cri
wi th /,. = 1\ i = 2
Penalty Methods
The work of this author and T. J. R. Hughes [lOJ has shown
that the penalty/reduced-selective integration technique pro-
duces a static or steady-flow problem
(~2 + z ~l)u = F (29)
~2 is the same matrix as in (9), and z is a large penalty
parameter. When ~ is constructed based on the integration
points of Sh used to evaluate ~l as described in [lOJ, then
relation (21) holds, and (29) produces the same solutions as
a slightly perturbed version of (9) [4,10J. More recently
Hughes and Malkus [12J have established the correspondence
between penalty methods and Lagrange multiplier methods,
which allows a much broader choice of pressure trial spaces
~, yet yields relation (21) relating penalty methods to
Lagrange multiplier methods. So in what follows assume that
~l in the penalty method is given by (21) for Rl and g de-
fined for an associated Lagrange multiplier method which also
has the same ~2 and F.
The penalty method (29) has the following eigenproblem
which determines the norm of the inverse of the discrete
operator:
709
(30)
This rearranges to
,lS1 u = cr,lS2 u (31)
where
A -1
z (32)
References
1. Eabu~ka, I.; Oden, J. T.; Lee, J. K.: Mixed-hybrid
finite element approximations of second order elliptic
boundary-value problems. Compo Meth. !2E!.. Mech. ~.
11 (1977) 175-206.
2. Erezzi, F.: On the existence, uniqueness and approxi-
mation of saddle point problems arising from Lagrangian
multipliers. R.A.I.R.O. 8 (1974) 129-151.
3. Eabu~ka, I.; Aziz, A. K.: Mathematical Foundations of
the Finite Element Method. New York: Academic Press 1972.
4. Eercovier, M.: Perturbation of mixed variational pro-
blems. Application to mixed finite element methods.
R.A.I.R.O. 12 (1978) 211-236.
5. Hlava~ek, I.; Ne6as, J.: On inequalities of the Korn
type. Arch. Rat. Mech. Anal. 36 (1970) 305-334.
6. Malkus, D. S.: Preprint, available from author.
7. Gantmacher, F. R.: Matrix Theory II. New York, Chelsea
1971.
8. Moler, C.; Stewart, G. W.: An algorithm for generalized
matrix eigenvalue problems. S.I.A.M. ~ Numer. Anal.
10 (1973) 241-256.
711
P.l. PAHL
Techmsche Umversitiit Berlin, Germany
Summary
The global data for finite element analyses are conven~ently
stored in matrices and table trees. In order to optim~ze the
use of the available storage capacity, the matrix and table
tree data structures are ~mplemented on the basis of a common
storage structure. The paper describes the properties and
functions of storage, matrix and table tree managements for
fin~te element programs.
1. 1 Introduct~on
as follows:
ThlS list shows that tabular and matrix data structures are of
equal importance ln finite element programs.
The following table shows that the required worklng set chan-
ges signlficantly during the analysls. In particular, the re-
latlve volume of tables and matrlces in the working set chan-
ges. The use of the available prlmary memory for out-of-core
solutions is therefore optimized if the table and matrix data
structures are implemented on the basis of a common array sto-
rage structure. Space released by matrices can then be used
for tables, and vice versa. This concept is the main feature
of the data management proposed in this paper.
717
2. STORAGE MANAGEMENT
The flnlte element program and its data are stored in a pro-
gram area on prlmary memory and in data sets on secondary
memory, which are asslgned through the operatlng system.
If the pool contains neither empty nor free arrays, the exe-
cution ~s term~nated due to lack of storage space.
DEFINE POOL
An empty pool block table ~s set up.
STORE POOL
All arrays ~n the blocks of the pool are stored ~n the~r files.
DEFINE FILE
The adm~nistrat~ve data for a specif~ed f~le are entered in
the f~le table.
RELEASE FILE
All arrays of the specified f~le are released in the pool. The
file ~s set to empty.
DELETE FILE
The specified f~le is released. Its definition ~n the file
table ~s nullified.
STATUS FILE
The parameters describing the file ~n the file table are re-
turned.
720
ASSIGN BLOCK
A block is assigned in the pool by the algorithm described in
section 2.3. Its number is returned.
FIX BLOCK
The contents of the specifled block is fixed ln the pool: the
block lS not available for reassignment.
FREE BLOCK
The specified block is avallable for reassignment.
RELEASE BLOCK
The status of the speclfled block lS set to the value empty.
STATUS BLOCK
The parameters describlng the block in the pool table are re-
turned.
ASSIGN ARRAY
A new array lS assigned in the specified flle by the algorlthm
of section 2.3. Its number is returned.
GET ARRAY
The specifled array lS transferred from the flle to a specl-
fled block ln the pool.
PUT ARRAY
The array stored in the speclfled block lS transferred to the
flle reglstered in the pool block table.
STATUS ARRAY
If the array resides in the pool, the number of the block ln
which it is stored is returned, otherwise the value zero.
3. MATRIX MANAGEMENT
The hypermatrix, the submatrix table and the block table are
usually stored in fixed blocks of the pool. Between runs, the
hypermatr~x and the submatrix table are stored ~n the f~le
3.5.1 Conventions
DEFINE MATRIX
The propertles of the matrlx are entered in the matrlx table.
RELEASE MATRIX
All arrays of the matrlx are deleted. The flle is set to empty.
DELETE MATRIX
The matrlx lS released. Its deflnltlon In the matrlx table lS
nullifled.
724
STORE MATRIX
The arrays of the matrix are stored 1n the associated f1le. All
blocks assigned to the matr1x are released.
PRINT MATRIX
The hypermatr1x and adm1n1strat1ve tables of the matr1x are
pr1nted.
STATUS MATRIX
The contents of the assoc1ated row in the matrix table is re-
turned.
INDEX MATRIX
The pool indices of the first element of the hypermatr1x and
its profiles are returned.
DEFINE SUBMATRIX
The spec1f1ed submatr1x 1S def1ned and recorded.
SET SUBMATRIX
All coeff1c1ents of the submatr1x are set to a g1ven value.
The profiles are set to zero.
PRINT SUBMATRIX
The coeffic1ents and the prof1les of the spec1fied submatr1x
are printed.
INDEX SUBMATRIX
The pool 1nd1ces of the f1rst element of the submatr1x and of
its prof1les are returned.
SET GROUP
The coeff1c1ents of a specified group are set to g1ven values.
ADD GROUP
The coeff1cients of a specified group are mod1fied by adding
given values.
GET GROUP
The values of the coefficents of a specified group are re-
turned.
4. TABLE MANAGEMENT
table.
trated in Fig. 9.
The backward search starts w~th the specified table. The num-
ber of the preceeding table ~s read ~n column 7 of the adm~n
DEFINE TABLE
The propert1es of the table are entered in the adm1nistrative
table.
729
PUT LINE
The data are wrltten from the FORTRAN-vector lnto the specl-
fled llne. If necessary, space for pOlnters and for the data
lS automatlcally asslgned ln the table tree.
GET LINE
The contents of the speclfled llne is read ln the table and
stored ln the FORTRAN-vector. If the llne lS deflned, the
code is set to 1, otherwlse to 2.
STATUS LINE
The code is set to the status of the speclfied llne (defined
or undeflned). The functlon lS used to check the eXlstence of
speciflc data, ego a node wlth a partlcular number.
Acknowledgement
1
Data Set
File 4
Data Set 2 Q)
C)
co
11 0
+-'
en
~
co
"'C
c
Data Set N 0
u
E
IIIIII
Q)
Q)
+-' en
en
>
en
I
i
File 6
C)
c
.;::;
e
Q)
0-
o
>
.0
Virtual Memory
o
+-'
0- Q)
en C)
c co
co 0
.= Admlnlstr Pool +-'
en
S A
~
co
E
~
Real Memory
j
FIGURE 1 . Storage areas of a finite element system
2 3 4 5 2 3 4 5
K U p
I 1 3 5 7 9 11 13 15 3 1 3 I
1 1
I 1
3
'" 3
'"
5 5
7 7
*
'" 9 9 '"
11 11
'" 13 13 '"
]15 15 I
I
t--- s s s -- -f- L - - L - f-
1 2
Structure of matrix K 3 4
5
1 2 31 4 5
I I I I I
1 1 1 1
3 3 3 3 3
,
5 5 5 5 5
3 5 3 5 3 5 3 5 3 5
-.!.
2-
2+ 35
~ lg1 ~
1 3
1 1 I 1 ' 4
1 1 1 2 3 · 4
3 3 3 3~ 5 ~ ~
5 5 5
3 3 3 Base
array
submatrix n1
t submatrix n2 submatrix n3
coefficients profiles
2 17 2
M~
K = 13 J JHYP
l
J =2_t--t--HI--II--IH-t-t-t-t--+-f' i3'ROb"P·...-+-+-+-+-+-.\--I~-I
1-:"", -. 3
5
~-+~+-~~~+-~-+~~~~~~-+'~,-
,,'~~
:}' 6
b'' "
1M " .~ SUBMATR I X ¥, ,?S' 6 2
., :LI' .,
i.= k ,d. 10
t--+-+-+-~~~~-r-r-r-+-+-+-+-+-f-4-4-4-4~~~~ 11
K 1 3 5 7 11 13 15 19 21 23
1
KHYP 2 3
Scale: X = c * A X = c * D
Add. X = A1 ± A2 X=A±D
Multiply' X = A1 * A2 X=D*A
X=A*D
Solve X = R- 1 * A X = D- 1 * A
X = A * R- 1 X = A * D- 1
L * R = Q determine L, R
N
r---
.- -
r-
r- -----
--
- r---
- r--- - -
- r-
- r--- Loadin g cond ition table Load in condition table
- -
Coordinate table
I- U1 U2 U3
- -- F1 F2 F3
I- -
I- - ~
I- -
I- - FIGURE 8 :
I- - Table tree for nodal data (simplified)
r- -
r- -
'-- -
Node number table
739
Result The line of table 65 assigned to node 815 IS supplied In the pool
1 2 3 4 5 6 7 8 9 10 11 12 20
13
- 4 2 4 .256 27 1 42 28
-
-
27 4 1 2 512 - - 17 72 38 75
65 9 4 3 192 13 3 18 33
Chaining of arrays
2 2 3 4 2 3
Table 27 13 65
Type' number I * 4 R * 8
File 4 4 9
Array 38 42 18
Block 75 28 33
Line, 303 2 5
Column. 3
L~terature
H. A. KAMEL, J. M. TAN
Umversity of Anzona, Tucson, USA
1. In~roductlon
3. Definition of Primitives
OEFINE,device,filename,filetype,hypersize,recordsize
ATTACH,device,filename,filetype,hypersize,recordsize
TRANSFER,devicel,filel,recordl,device2,file2,record2
MULT,filel,recordl,file2,record2,file3,record3
SUBT,arrayl,recordl,array2,record2,array3,record3
MINV,arrayl,recordl,array2,record2
TPOS,arrayl,recordl,array2,record2
As llEIltioned before, an algori tlun will be chosen, and a coder and simulator
will be developed to predict its performance on specific hardware con-
figuration. The following gives this device:
ATTACH,DK,STIF,2D,M,M,S,S
DEFINE,MC,WS,1D,5,S
DEFINE,AP,BUF,1D,5,8
D=1
WHILE D<M DO
TRAN,DK,STIF,D,D,MC,WS,1
TRAN,MC,WS,1,AP,BUF,1
MINV,BUF,1,BUF,2
I=D+1
WHILE I <=M DO
TRAN,DK,STIF,D,I,MC,WS,2
TPOS,WS,2,WS,3
TRAN,MC,WS,3,AP,BUF,3
J = I
WHILE J <= M DO
TRAN,DK,STIF,I,J,MC,WS,4
TRAN,MC,WS,4,AP,BUF,4
TRAN,DK,STIF,D,J,MC,WS,5
TRAN,MC,WS,5,AP,BUF,5
MULT,BUF,1,BUF,5,BUF,3
SUBT,BUF,4,BUF,3,BUF,4
TRAN,AP,BUF,4,MC,WS,4
TRAN,MC,WS,4,DK)STIF,I)J
J = J+1
ENDWHILE J
I = 1+1
ENDWHILE I
D = D+1
ENDWHILE D
Figure 1. Program to Triangulate a Hypermatrix
On the other hand,we must state that the approach here may be
generalized to include many of the effects described above.
751
DIJ ## PRE OPCODE DEV DEVI FILE REC DEV2 FILE REC DEV3 FILE REC
1-- 1 o TRAN D-M DK STIF 1,1 MC WS 1
1-- 2 1 TRAN M-A MC WS 1 AP BUF 1
1-- 3 2 INV AP AP BUF 1 AP BUF 2
12- 4 o TRAN D-M DK STIF 1,2 MC WS 2
12- 5 4 TPOS MC MC WS 2 MC WS 3
12- 6 5 TRAN M-A MC WS 3 AP BUF 3
12- 7 6, 3, 2 MULT AP AP BUF 3 AP BUF 2 AP BUF 1
122 8 o TRAN D-M DK STIF 2,2 MC WS 4
122 9 8 TRAN M-A MC WS 4 AP BUF 4
122 10 0 TRAN D-M DK STIF 1,2 MC WS 5
122 11 10 TRAN M-A MC WS 5 AP BUF 5
122 12 2,11, 6 MULT AP AP BUF 1 AP BUF 5 AP BUF 3
122 13 9, 6,12 SUBT AP AP BUF 4 AP BUF 3 AP BUF 4
122 14 13, 8 TRAN M-A AP BUF 4 MC WS 4
122 15 14, 8 TRAN D-M MC WS 4 DK STIF 2,2
123 16 15 TRAN D-M DK STIF 2,3 MC WS 4
123 17 16,14 TRAN M-A MC WS 4 AP BUF 4
123 18 11 TRAN D-M DK STIF 1,3 MC WS 5
123 19 18,12 TRAN M-A MC WS 5 AP BUF 5
123 20 12,19,13 MULT AP AP BUF 1 AP BUF 5 AP BUF 3
123 21 17,20 SUBT AP AP BUF 4 AP BUF 3 AP BUF 4
123 22 21,17 TRAN M-A AP BUF 4 MC WS 4
123 23 22,16 TRAN D-M MC WS 4 DK STIF 2,3
133 24 18, 5 TRAN D-M DK STIF 1,3 HC WS 2
133 25 24, 6 TPOS MC MC WS 2 MC WS 3
133 26 25,20 TRAN M-A MC WS 3 AP BUF 3
133 27 26, 7,20 MULT AP AP BUF 3 AP BUF 2 AP BUF 1
133 28 23 TRAN D-M DK STIF 3,3 MC WS 4
133 29 28,22 TRAN M-A MC WS 4 AP BUF 4
133 30 24,19 TRAN D-M DK STIF 1,3 MC WS 5
KEY INFORMATION:
DIJ Current va~ues of loop control parameters
## Instruction sequence number
PRE Prerequisites
OPCODE Operational code TRAN=transfer, INV=invert
MULT=multiply, SUBT=subtract, TPOS=trans-
pose
DEV Device on which instruction is executed,
MC=main computer, DK=d1sk, AP=array
processor, D-M=disk/computer bus, M-A=
computer/AP bus.
DEVl Device used with first operand
FILE File where first operand resides
REC Record address for first operand
(DEV2,FILE,REC), (DEV3,FILE,REC)
Device, file and record for second and
third operands
Figure 2. List of F1rst 30 Instructions Generated by the
Code Generator for Simulation Purposes
752
AP @ I 0· I@
..
1
OK- MC - AP CD jl® 0 ® I@ I® ® I® I® I@ 1\
Me @ ·1 ~
.2 .3 .4 5 .6 .7 (sec r
® 10
~ ..
AP
MC - AP
10 ® I® I@ I
I@
I""
®7
MC ® ® 1\
MC - OK (2) 10 ® ® I T@ I@· \
.1 .2 .3 .6 .7 (sec)
trol the MAP-200 was done via direct memory access (DMA).
lNumber of 688 40
nodes
lNumber of 575 64
~lements
Half-bandwidth
Maximum 318 43
r.m.s 210 32
7. Acknowledgements
8. References
G.F.CAREY
UmVerSlty of Texas, Austin, USA
Summary
Introduct~on
Special Techniques
There are a few special techniques used in flnlte element and
finite difference methods that merit particular consideratlon
in the context of parallel and vector processors. Here we
llmit the treatment to substructuring and splittlng strategles.
(1) Substructuring
Substructuring was introduced In the 1960's to solve large
scale structural analysls problems, such as aircraft "wing-
body" interaction, wlthout heavy dependence on slow out-of-
core solution algorlthms. The basic idea is to consider the
entire structure S as made up of a set of substructures
S., i
l
= 1, • • . ,n. For example, the wing and body may be
treated as separate structures in the "wing-body" problem.
Treating each substructure as a separate domain, the element
contributions can be calculated and assembled to the sub-
structure stiffness K
_l. .
Using statlc condensation (pre-
ellmlnatlon) the substructure stiffness can be reduced In
766
(2) Splitting
Splitting techniques have been more extensively utllized in
finite difference computations, but they can also be applied
in conjunction with finite element methods and other approxi-
mation techniques. Splltting can be defined in a general
manner to reduce the number of dimensions and/or to separate
individual contributions ln the governing equation. [3 - 5].
In the former instance, the splitting can be carried out on
the differential equation by replacing the differential
operator by a sequence of one dlmensional dlfferential opera-
tors. For example, let the governing dlfferential equation
of motion have the general form
767
au + aF + aG 0
at ax ay
o o
lit
u 1' J' - -x-
uy (G ..
1J - G1J-
.. 1)
(n+l)
u ..
1J
A A
~x
+ A
~y
and
n+l -1 n
u (I - D A)u + k
-1
where k = D b The components of the right-hand side can
be computed independently in parallel or as blocks in parallel.
The calculation involves formation of the rows of the itera-
tion matrix on the right, followed by a matrix x vector
product and a vector addition. We can use the vector product
form discussed previously for direct solvers. Thus the in-
efficient scalar Jacobi scheme can be easily implemented as
a vectorized parallel scheme in an array processor. However,
the Jacobi iteration is inherently so slow that it would re-
quire unusual circumstances to make it really practlcal. SOR,
which is superior for scalar processors, lS sequential by
nature and cannot be so readily adapted to parallel processing.
Conjugate gradient and accelerated conjugate gradient
771
Non11near Solution
As indicated ear11er, straightforward Newton-Raphson itera-
tion leads to a sparse linear Jacobian system to be solved at
772
F(u) 0
and componentwise as
... , u
~-
l'u.,u.+ l , . • • . , u )
~ ~ n
o
for ~ = 1, • . .
, n.
Mot~vated by linear iterative schemes, let u~k) be the
J
estimate of component u. at iterate k • Now consider the
J
i'th equation and solve
. .. , (k)
u i _l ' u i '
u (k)
~+l'
{aF (u(k+l,m-l»/au(k+l,m-l)}
~ ~ ~
773
(k+l 0) (k)
where we have chosen as startlng lterate ul ' = ul
the current estimate of ul a. Note that ln thlS formula
the dependence on u, j I 1 has been suppressed for nota-
J
tional convenience.
In this Jacobl-Newton scheme, each equatlon of the
outer (Jacobi) iteratlon can be assigned to a processor and
inner Newton iteratlons computed independently. Thus the
scheme is very suitable for parallel array processing. If we
accelerate the scheme and use most re ently calculated values,
convergence rates improve but parallellsm is lost. In [ 11 )
we dlSCUSS the convergence and other properties of thl~ f?mily
of iterative methods.
k
W: L n(mk (2FE + 2C))
k=l
W : 2Kn(FE + C)
k
W mk ~(2FE + C)
k=l P
and for 1
W _ 2k ~(FE + C)
p
Conc1udlng Remarks
The difficulties concernlng design of advanced computers for
scientlflc computing go signiflcant1y beyond the design of
processors, data bases and support systems. The numerlca1
methodology as well as algorithms are still evolving and to
achieve the required processlng speed there must be a strong
lnterdependence between these and the actual processor design.
Parallel array processors and vector processors with some
parallelism provide a means to achleve the desired improve-
ment In processor speed. Thls lmp1les that the deslgn must
775
Acknowledgment
This study has been supported 1n part by NASA grant NSG-2338
and by ONR contract N00014-78-C-0550. I wish to express my
apprec1at1on to Dav1d Young and L1nda Hayes for their helpful
d1SCUSS1ons.
References
1 Mat1soo, J.: The Superconduct1ng Computer, SC1ent1f1c
Amer1can, 242, 5, 50-65, 1980.
Additional References
B1rkhoff, G. and A. George: Elimination by Nested Dissection,
Complexity of Sequential and Parallel Numerical Algor1thms,
J. F. Traub, ed., Academic Press, New York, 1973, pp. 221-269.