0263 8231 (84) 90004 1 PDF
0263 8231 (84) 90004 1 PDF
0263 8231 (84) 90004 1 PDF
H. P. Lee
18 Asprey Court, Brampton, Ontario, Canada
P. J. Harris
Department of Civil Engineering and Applied Mechanics, McGill University, Montreal,
Canada
and
ABSTRACT
1 INTRODUCTION
2 D E V E L O P M E N T OF T H E P R O G R A M
approach has been employed together with the theory of plasticity. The
von Mises yield criterion and the associated flow rule of the Prandtl-
Reuss relationship are followed and subsequent yielding is governed by
the isotropic hardening rule. In the pre-yielding stage, nonlinear
equilibrium equations are constructed by applying a variational principle
using the virtual displacement approach. The equations are solved using
the Newton-Raphson process which converges extremely rapidly and
hence very large load steps can be used which, in turn, mean a consider-
able saving in computing time. In the post-yielding stage the incremental
procedure is used to be consistent with the flow theory of plasticity. The
incremental equilibrium equations with unbalanced force corrections are
constructed by applying an incremental principle using the virtual dis-
placement method. The details of the above-mentioned methodology for
nonlinear finite element analysis have already been well documented 3'4
and will not be repeated here. However, the algorithms used for
generating the tangent stiffness matrix and solving the stiffness equations,
which affect computing time vitally and are of particular importance in
nonlinear analysis due to its iterative nature, are described fully in the
following sections.
It is well known that the variational principle derived by using the method
of virtual displacements can be written as
fv BTcrdv = P (1)
where B T, composed of linear (BL) and nonlinear (BN) parts, is the matrix
which transforms nodal displacements to strains at any point in the
element, or is the vector of stresses and P is the vector of applied loads. To
solve these nonlinear equilibrium equations using an iterative process,
such as the Newton-Raphson method, the linear equilibrium equations
to be solved at each iteration can be written as 4
[fv d e , d r + fv Z EB dv + fv CZ rSN + + dq = dR
or, in short
or
KTdq = dR
Equation (3) can be viewed as the same as eqn (2) since the incremental
solution procedure with unbalanced force corrections is essentially the
N e w t o n - R a p h s o n method with one iteration.6 The KT in eqn (3) is the
same tangent stiffness matrix as given in eqn (2), except that now the B
matrix is to transform incremental nodal displacements to incremental
strains but remains in the same form and the E matrix is to transform
incremental strains to incremental stresses and is no longer a constant
matrix. The E matrix should be continuously updated and its formulation
in finite element application using the flow theory of plasticity has already
been established. 4'"~
If the tangent stiffness matrix is generated according to eqn (2), both K~
and KN have to be re-evaluated and only KL can be retained as a constant
in the elastic stage of the analysis. In the post-yielding stage, even KL has
to be re-evaluated due to the necessary updating of the E matrix.
However, the tangent stiffness matrix as given in eqn (2) can be arranged
to make a significant reduction in the number of arithmetic operations
Nonlinear finite element program for thin-walled members 359
Ov + - - + - - - - +---- + - 2z (4)
Exy = O---x ay Oy ax ax Oy ~
• = AG+½GTHG (5)
OWo
1 0 0 0 0 -Z O 0
Ox
014'0
A = 0 0 0 1 0 0 -Z 0 (6)
ay
OWo OWo
0 1 1 0 0 0 -2Z
Oy Ox
= 1, where the subscripts 1,2 and 3 represent the first, second and third H
matrices corresponding to the strain components ~x, ~y and Exy,
respectively. All three matrices are symmetrical and are set to a size of
9 x 9 to make the dimensions compatible. G represents the displacement
gradient matrix and can be expressed in terms of nodal displacements as
G = Cq (7)
Substitution of eqns (8) into the first expression of eqn (2) and, after a
lengthy but simple matrix algebraic manipulation, the tangent stiffness
can be rewritten in the following form
R¢ = f,~,¢dz
Equations (2) and (3) are solved using a modified Cholesky decomposi-
tion m e t h o d t2 which avoids taking square roots. The algorithm is to
decompose the stiffness matrix as follows
K = LD-tL T (12)
LY = AP (14)
362 H. P. Lee, P. J. Harris, Cheng-Tzu Thomas Hsu
where
D-JLVAq = Y (15)
KAq =
where
i j iAqll
BI A2 B~
B2 A3 B~
B3 A4
Aq2
Aq3
Aq4
=
AP2
AP3
AP4 (16)
IA'-1L II II
BI A2 BX2
B2 A3 B~
B3 A 4 ]
z
M! L2
M2L3
M3 L4
D2 I
D:, ]
L2x M T ]
AI = LID~IL~ L x D i ] L l = AI
B1 = M j D t 1 L I MI = BI(L~) IDt
A2 = M , D ~ M ] + L 2 D 2 ~ L ~ " LzD2'L[ = A2-MID~IM T
B2 = M202'L72 " M2 = B2(L2T)-LD2
A 3 ~ M2D21MT-~L3D31L T L3D31L T ~ A3-M2D2IM T
These equations are used to obtain each diagonal block L~ values by the
modified Cholesky decomposition algorithm and each off-diagonal block
M~ values by solving the corresponding equations.
The 'forward' solution for eqn (14) can be expressed as
Mi L2 Y2 = AP2~
ELI l(Y,
M2 L3
which leads to
M3 L4
Y3
Y4
AP3/
AP4 : (is)
L1YI = APt
L2Y2 "- A P 2 - M I Y I
L3Y3 = A P 3 - M 2 Y 2 etc. (19)
These equations are solved for Y . Y2, Y3, etc.
The 'backward' solution for eqn (15) is now expressed as
021 L] M] / ~ Aq2 Y2
/ Aq3 = Y~
D41L'~--J \ A q 4 1 Y4 (21)
which leads to
(D2IL~)Aq4 = Y4
(D3~ L~)Aq3 = Y 3 - D3'M~Aq4
(D;IL~)Aq2 = Y 2 - D~M]Aq3 etc. (22)
Solving eqn (22), the displacements Aqi are obtained in reverse order.
364 H. P. Lee, P. J. Harris, Cheng-Tzu Thomas Hsu
3 E X A M P L E PROBLEMS
L / 2 : 14 in. I=
D 17 18 33 41 4g 87
i ~ t I I I
M _ 0 . 6 7 ~ ~ ~ ~ ~ ~ ~ ~ 1 ~
0.05"
o.Ts-j// ~" \- 0.1"
o.8z'//
0.88 " /
0.94
J
t--y--/7]- T
YT"-/V,I~= 4 i°.
,. o.o~ i°. /-/A~//
O" 16
(7"= 2 7 . 7 6 ksi
Y
t
~
I
Y
MATERIAL PROPERTY
0.9ot
08
0.6
0.5
0.4
0
T~II
0.4
I
0.6 0.8
! I
1.0 1.2
i I
1.4
i
1.6
i
1.8
]Fig. 2. Moment versus curvature.
M/My
I.O
0.9
0.8 ./My • 0 7 )
0.7
---~ RESENT STUDY
0.6 (M¢./My = 0.67)
0,5
0.4 (B/t)
I I I I I I
O 0.4 0.8 1.2 1.6 2.0 2.4
¢_
0.94
FULLY ELASTIC: [ I
P A R T I A L L Y YIELDED:
FULLY YIELDED: ~]
Fig. 4. Yielded zones.
Nonlinear finite element program for thin-walled members 367
~-~ INTERIOR
AVERAGE STRESS
AT TOP FLANGE
i /i
/"EXTERIOR I /
./"~---FACE I /
/ /
I
-~ NTERIOR ~/ /
FACE / /
EXTERIOR
/ / FACE
/ /
/ /
M/M, = 0.94 /
/
/
/
/
ii /
I
//I///
II 3 0 KSI
I/ I 4
II
II
//
// *---WEB
//
/
and outer faces, and for the average stress at the crest of the buckles.
Stresses are distributed nonlinearly as expected.
¢_
79~ a l
Z5
MESH IDEALIZATION
/
Z
o.682" ~----NOD E NO.
7 15 19 2,5 31 57 43 49 ,55 61 67 73
0
M= 2,276 LB'IN
O.I
(INITIALBUCKLING}
0.2
M= 3,913 0.3
0.4
0.,5
0.6
M= 8,456
0.7
0.8
0.9
M = 9,465 1.0
i.I
1.2
M = 9,780 1.3
1.4
(IN.)
,,,J
I0,000
8,000
6,000
4,000
2,000
~c(IN.)
I I I i i i I
AVERAGE (MEMBRANE)
STRESS
TON/IN" "' it"
8' =E
0
COMP. 4, o
12 8 4
Z
0
Z
lud
I--
TENSION SIDE
| 12 16 OF BUCKLE
TENSION
AND HARVEY
Fig. 8. Longitudinal stress distribution across lipped channel section. - - - , Present study
(M = 6720 lb in); - - - , Rhodes and Harvey t5 (M = 6060 lb in).
370 H. P. Lee, P. J. Harris, Cheng-Tzu Thomas Hsu
T-
M = 9,780
I I FULLY ELASTIC
PARTIALLY YIELDED
stage the current study has its stiffness decreased at a faster rate and
deflects more at failure. Figure 8 shows the stress distribution in com-
parison with that found by Rhodes and Harvey. Note that the load levels
are not exactly identical. However, the distribution patterns are quite
similar. The yielded zone at the ultimate load is shown in Fig. 9.
4 CONCLUSIONS
problems are used to illustrate the capabilities of the program. Results are
c o m p a r e d with previous theoretical work and found to be in good agree-
ment. Even though the demonstrated problems are based on elastic-
perfectly plastic material, the program can also handle elastic-linear
hardening and nonlinear hardening materials. In such cases the isotropic
hardening rule is used to define subsequent yielding surfaces. Elastic
unloading is also permitted in the program. With this program many
aspects of detailed'studies of thin-walled sections such as distortion of the
cross-section, the effect of lateral restraints, the determination of
deflection in the post-buckling range and the growth of the yielding area,
which cannot be analysed satisfactorily by other methods, can be handled
easily.
REFERENCES
APPENDIX
2b
Z,W Z, Oz
V y,v y,0y
Translation
X,U V ~
Rotation
X,0 x
ui
.0x
w 0.:
I
• . ~o~+~
i - m
• . r ~ . ~
_~ _-- + E
+~ ~"~ +
• @
-~ ,
- "gc
~. + ~ ~ + ~-
+ ~*
i rl.~ rl~
~y ~r ~ r ~ ~
t
374 H. P. Lee, P. J. Harris, Cheng-Tzu Thomas Hsu
I
I I[ I
+ + /
i
~1~ ~1~ /
~' /
1
~1~ ~~l ~~ /
+ + #
~ ~, /
.0 -i,-,
L
× I ×
N
I
I ....
b~
1 V
/
/ ~l <~
+ + + +
~1~<
<~ ~
~
~4
o,
<~1 ~
U ~' 2 ..4'
7
J
II
Nonlinear finite element program for thin-walled members 375
I
I ! /
/
4-
/
/
• + + •
~1~ ~1~ / +
/
/
/
~1~ ~4~ /
/ ~4
• • + + •
~1~ ~n~ /
~4 u4 /
/ + 4-
~1~ ~1~ / +
• • + + •
+ 4-
+ + + + I;~ I~
~ ~ ~ /
/
+ + + +
,~ /
+ +
• • • fO
f
• • t0 j
f
• /0 ~
II rl II
f Z
Of
II
4~
376 H. P. Lee, P. J. Harris, Cheng-Tzu Thomas Hsu
Txy O'y
where o-~, O-y and rxy are, respectively, the normal stress in the x and y
directions, and the shear stress on the x-y plane.