Вы находитесь на странице: 1из 39

Computer Methods in Applied Mechanics and Engineering 96 (1992) 133-171

North-Holland
On a stress resul t ant geomet ri cal l y exact shel l
mo de l . Part V. Nonl i ne ar plasticity: f ormul at i on
and i nt egrat i on al gori thms*
J. C. Si mo and J. G. Kennedy
Division of Applied Mechanics, Department of Mechanical Engineering, Stanford University,
Stanford, California, USA
Received 28 March 1989
Revised manuscript received 20 January 1991
The continuum basis and numerical implementation of a finite deformation plasticity model
formulated within the framework of the geometrically exact shell model presented in Parts I and |II of
this work, is discussed in detail. The model is formulated entirely in stress resultants, and hence the
expensive integration through the thickness associated with the traditional degenerated solid approach
is entirely by-passed. In particular, the classical llyushin-Shapiro plasticity model for shells is extended
to accommodate kinematic and isotropic hardening, and consistently formulated to accommodate finite
deformation. The corresponding closest-point-projection return mapping algorithm is shown to reduce
to the solution of a system of two nonlinear scalar equations, and proved to be amenable to exact
linearization leading to a closed form expression of the consistent elastoplastic tangent moduli.
Numerical simulations are presented and comparisons with exact and approximate solutions are made
which demonstrate the excellent performance of the proposed methodology.
1. I nt r oduc t i on
In Part s I, II and I l l of this wor k, we have pr esent ed t he f or mul at i on, numeri cal analysis
and i mpl ement at i on of a nonl i near shell t heory f or mul at ed ent i rel y in t erms of stress
resul t ant s. Thi s t heor y t hus falls wi t hi n t he real m of ' classical' shell model s, whi ch are typically
f or mul at ed in stress resul t ant s (and stress couples). It is by now well known t hat t he
mo me n t u m equat i ons of shell t heor y ( f or mul at ed in stress resul t ant s) t ake a canoni cal (or
generi c) f or m whi ch is, in fact, i ndependent of t he met hod of deri vat i on. The general st ruct ure
of t he const i t ut i ve equat i ons in shell t heory is also known in t he specific case of elastic
response. A mai n t hr ust of our previ ous work has been to demonst r at e t hat these canoni cal
equat i ons can be r ef or mul at ed in a f or m which ci rcumvent s t he appar ent compl exi t i es f ound in
classical exposi t i ons of t he subj ect , and is directly amenabl e to numer i cal i mpl ement at i on.
Fr om an engi neer i ng perspect i ve, however, if stress resul t ant shell t heor i es are to become a
st andar d engi neer i ng tool t hat repl aces t he widely used cont i nuum degenerat ed sol i d appr oach,
t he crucial issue t hat r emai ns t o be add~'essed concerns t he f or mul at i on and i mpl ement at i on of
i nel ast i c const i t ut i ve model s. Two di fferent appr oaches can be adopt ed: (i ) use of t hr ee-
di mensi onal pl ast i ci t y model s and numeri cal comput at i on of the stress resul t ant s and stress
coupl es by i nt egr at i on t hr ough t he t hi ckness of t he shell, or (ii) use of const i t ut i ve model s
f or mul at ed di rect l y in stress resul t ant s.
* Research supported by AFOSR under contract numbers 2-DJA-544 and 2-DJA-771 with Stanford University.
Elsevier Science Publishers B.V.
134 J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model
The through-the thickness pre-integrated approach has been advocated by Stanley [1],
among others, and represents a step forward in the simplification of the traditional solid
formulation. Note, however, that the balance laws should remain formulated in resultants, in
contrast with typical formulations adopting this approach, including that of Stanley [1]. The
advantages of this method are as follows:
( i a) Conceptual simplicity is inherited from the three-dimensional theory.
( i b) Direct applicability of existing three-dimensional constitutive models is maintained. For
plasticity, for instance, this leads to the use of the standard return mapping algorithms,
see e.g. [2] for a review.
There are, however, two major disadvantages associated with this widely used approach:
(i c) Computational cost: even if the balance laws are formulated in terms of stress resultants,
a numerical integration through the thickness is required to compute the stress resultants.
( i d) Prevalence of rate formulations of the 'stress-strain' equation in terms of objective stress
rates. In this type of shell theory, rate formulations are used as a means of enforcing
constraints present in shell theory on the constitutive model; in particular, the plane-
stress assumption. As a result, objective integration algorithms are required to define the
elastic predictor (trial stress); see [2, Chapter 7]. This leads to added cost in the
formulation, and precludes the desirable use of the so-called algorithmic tangent modul i
I31.
In the second approach, which has not been widely used in the literature until recently,
constitutive models are formulated directly in stress-resultants. An example of this approach is
the work of Crisfield [4, 5]. The primary advantage of this methodology lies in the following
feature:
(iia) Computational cost: in this approach, the integration through the thickness associated
with the degenerated solid formulation is entirely eliminated from the computational
procedure.
This reduction in cost, however, is accomplished at the expense of introducing considerably
more complex functional forms in the constitutive response functions. For elastoplasticity, for
instance, even the simplest yield criterion; e.g., the Von Mises condition, leads to a rather
complex functional form when expressed in stress resultants, as is demonstrated in the classical
work of llyushin [6], Shapiro [7] and Ivanov [8]. These yield criteria often exhibit a lack of
regularity which, from an algorithmic point of view, requires a careful treatment. To
summarize, the disadvantages of this second approach are as follows:
(iib) Implementation of three-dimensional material models may prove to be a difficult task. It
is not a priori clear how to perform a closed-form, analytical reduction of complex
three-dimensional constitutive models to resultant form.
(tic) The complexity of the algorithmic treatment is typically increased. For elastoplasticity,
for instance, the formulation and implementation of proper return mapping algorithms
for models with general yield surfaces intersecting in a possibly nonsmooth fashion is not
a trivial task.
Despite the aforementioned shortcomings, we believe that the low cost advantage more
than offsets the difficulties associated with this latter approach.
The objective of this paper is to present a rather general treatment of plasticity, in the
context of shell theory, for constitutive models formulated in stress resultants. Our contribu-
tions, we believe, lie in the following features.
(1) The formulation discussed herein is completely general, at least within the context the
classical kinematic assumption of shell theory that straight fibers off the mid-surface remain
straight, and is not restricted to the case of infinitesimal kinematics. That is, all kinematic
quantities such as the displacements, rotations and strains may be large. Of course, the utility
J.C. Sima, J.G. Kennedy, On a stress resultant geometrically exact shell model 135
of the classical kinematic assumption degrades as the transverse shear and membrane strains
become large, but this speaks of the underlying shell theory itself and not the current
formulation of that theory.
We illustrate the theory by means of a properly invariant extension of the classical
Ilyushin-Shapiro yield condition for J2-flow theory.
(2) Our extension of the llyushin-Shapiro plasticity model to the nonlinear theory includes
both ki nemat i c and isotropic hardening. The flow rule is associative leading, therefore, to
symmetric tangent moduli. Accordingly, the model constitutes the counterpart in nonl i near
shell theory of the classical J2-flow theory with kinematic/isotropic hardening.
(3) Within the framework of our previous work on multi-surface plasticity [9], we construct
an unconditionally stable return mapping algorithm which, at the stress-point level, involves
only the solution of two nonl i near scalar equations. This algorithm is amenable to exact
linearization leading to a closed-form expression for the consistent elastoplastic tangent
moduli.
(4) In the proposed formulation, the elastic response emanates from a hyperelastic form of
the free energy function, an example of which is discussed. The elastic predictor is therefore
exact, and is computed without resorting to incrementally objective algorithms.
An outline of the paper is as follows. In Section 2 we give an account of the basic
kinematics, strain measures and rate of deformation tensors employed in the formulation of
the general theory. The structure of the general elastoplastic model is outlined in Section 3,
and the concrete application to a model of the Iiyushin-Shapiro type is undertaken in Section
4. In Section 5 we give an explicit construction of the return mapping algorithm. Numerical
simulations which illustrate the performance of the proposed formulation, including com-
parisons with available solutions and large-scale simulations exhibiting very large deforma-
tions, are given in Section 6. Details pertaining to the linearization of the algorithm and the
thermodynamic structure of the theory are given in two appendices.
We close this introduction with a remark on our finite element implementation. In our
previous work of Parts II and III, we have employed a mixed (assumed-stress) finite element
method for the membrane and bending fields based on a Hellinger-Reissner variational
formulation. Conceptually, the extension of this computational framework to accommodate
elastoplastic response follows identical lines to those considered in detail by Simo et al. [10].
For simplicity, however, we have chosen to present the methodology developed herein within
the much simpler and classical context of a strain-driven method. In particular, our numerical
simulations employ a nonopt i mal displacement formulation for the membrane and bending
field. Recently, however, we have constructed an assumed stress method which inherits similar
accuracy properties for coarse meshes as our previous assumed stress formulation without the
need to modify the return mapping algorithm at the stress-point level. We defer the discussion
of this method to a subsequent publication. The assumed strain framework presented in Part
III for the transverse shear strains, on the other hand, will be used here (i.e. the shear strains
are linearly interpolated between mid-side nodes).
2. Ki nemati c relations: summary of field equations
In this section we examine in detail the geometric structure and alternative definitions of the
strain measures, rates-of-deformation tensors, and stress resultants associated with the
geometrically exact shell model considered in Parts I and III of this work. These notions,
which where not discussed in depth in our previous work, play a central role in the
136 J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model
formulation of inelastic constitutive equations. For further details on this and related topics we
refer to Simo and Fox [11], and Simo et al. [12, 13].
2.1. Conf i gurat i ons: ki nemat i c description of the shell
Following Parts I and III of this work, we recall that from a geometric standpoint, a
shell-like body is described in terms of the following two objects (see Fig. 1):
(i) The mi d-surf ace of the shell, viewed as a parametrized surface; that is, as a map
~," M- - - ~ 3, where M C [~2 is the domain of the parameterization; a compact set with
smoot h boundary OM and points denoted by ~: : = ( ~' , ~52) ~ ..~ : = M U OM. In the context
of the finite element met hod, this parameterization is defined by a collection of charts
constructed via the iso-parametric mapping. We denote by
be : = I # --- (2.1)
the actual mid-surface; i.e., the graph of q~.
(ii) The director f i el d of the shell, viewed as a vector field i : be---, S 2 which assigns to each
point ~ E be of the mid-surface a vector i (. ~)~ S 2. Here, S 2 denotes the unit sphere
defined as
s - ' : - { t ' I Iltll = 1 } . (2.2)
As in Part I, we parametrize the director field by the map t : = i o q~ : s----~ $2. '
With these two objects in hand, the kinematic assumption underlying the shell model under
consideration is that any placement of the shell in Euclidean space, denoted by ~ C R 3, is
given as
~ : = ( x ~ 3 l x = ~ + ~ : t wh e r e ( ~ , t ) ~ a n d E[ h - , h + ] } ,
(2.3)
where [h-, h*] C R with h + > h- is interpreted as the thickness of the shell, and c~ is the
manifold of admissible parameterizations of the mid-surface and the director field; i. e. ,
a-' s' - = - - + I t ' +., x ~.2 > 0 and Ilq~., x +.2ll # o } . (2.4)
Note that the conditions appended to (2.4) preclude the physically unreasonable situation of
Fig. 1. Illustration of the geometry which defines the kinematics of the shell model.
' As shown in Part IV of this work, the inextensibility constraint requiring that t ~ S" can be easily removed. We
introduce this assumption here only for simplicity.
J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model 137
i nf i ni t e t ransverse s he ar d e ) o r ma t i o n , and require regularity of the paramet eri zat i on. In t erms
of t he unit vect or field 5" be---> S 2 normal to the mid-surface, with paramet eri zat i on v : = i; o q~
defi ned as
a ! x a 2 where a,, : = q~.~ , (2.5)
" " I l a l X a 211 '
t hese two requi rement s are equi val ent to the conditions that
v # O and t . v > O . (2.6)
Consequent l y, at each point E be we have the well-defined fl ame { a~, a 2, t } called t he
nat ur al f r a me . We denot e by a ~ the component s of t he induced Ri emanni an metric (i . e. , the
first fundament al form) and set 2
a ~ t 3 " =a ~ . a ~ and y , ' = a ~ . t . (2.7)
The r e c i pr oc al or c ov ar i ant nat ural frame, denot ed by { a ~, a 2, a3}, is t hen defined by the
st andard expressions
a : 3. a, , =6, ~, , a ~ . t =0 , a 3 . t =0 and a 3 . a , , =0 . (2.8)
It is clear from the preceding discussion that, by virtue of the kinematic assumption, a given
map @ : = (~p, t) in ~ uni que l y defines a pl acement of t he shell. Accordingly, one refers to q~ as
t he abst ract c onf i gur at i on space. Boundar y conditions of place are appended to definition (2.3)
by requiring t hat
~o=q~ onO~s and t =i " ono~, . ~, (2.9)
for any @ "= (~p__:_, t) E ~. Here ~ 1 and 0,M are disjoint parts of the boundar y as such t hat
O~M U 0,M = 0M. We recall that c is a differentiable manifold with t angent space at @ E c
denot ed by T, t, ~ and defined by
T. < = : - # T,S 2 II - 0 a n d ~ t l ~ , ~ - 0 } . (2.10)
As in Part I we adopt the following convention. We denot e by ~ 0 the r ef er ence p l a c e me n t of
0 t o} the reference
t he shell, we let 5e be its r ef er ence mi d- s ur f ac e , and denot e by { a , a 2,
nat ural frame. In general, unless otherwise explicitly stated, a superscript ' 0' will refer to an
obj ect associated with the reference placement ~ o.
Finally, we denot e by { e t, e 2, Ca} the f i x e d i nert i al f r a me in Eucl i dean space, and choose
e~ = 8~Et, where { E~} is the s t a n d a r d basi s in R 3. We then have the component expressions
i i
~p = tp e i , ~p.~ = q~ ,~e i , ~tp = 5~i ei ,
t = t i e~, ~t - - ~ t i e i , A = A~lei E I .
(2.11)
2Throughout this presentation, unless otherwise stated, Greek indices range only over { 1, 2}; a, /3, & %
o . . . . E{ 1, 2}.
138 J . c . Si mo, J. G. Kennedy, On a stress resultant geometrically exact shell model
Here, A : s~---> SO(3) is the orthogonal matrix that maps a pre-selected vect or of the st andard
basis into the director field; for instance, t = AE 3.
2.2. The strain f i el ds on the mi d-surf aces S and S
The formul at i on of constitutive models in stress resultants requires a careful definition of
the strain measures associated with the current and reference pl acement s of the shell. To
make mat t ers precise we define the following two linear spaces (see Fig. 2):
(i) Tangent pl ane to S at a poi nt ~ E ~. This is a two-dimensional subspace, denot ed by Tz 5e,
and given by
L A e : = { h(.~) I h q,., x .2 =0 } . (2. 12)
An analogous definition holds for T~,,~".
(ii) Tangent space at a poi nt ~ E S. This is a three-dimensional space, isomorphic to [~3 and
denot ed by ~ , which is given by
~ : = { v = a h + ~ t l h ~ TiO and a, fl E[R} .
(2. 13)
An identical definition holds for ~, , .
With these definitions in hand, we consider the following surface tensors.
(iii) (Surf ace) def ormat i on gradient. A two-point t ensor F : ~o--~ ~x (for each x ~ 5e ) given
by
/ ~ = a . a ''~ + ta 3 . ( 2 . 1 4 )
(iv) (Surf ace) unit tensor on ~. A bilinear form 1:7/' A x ~x--*R with associated covari ant
rank-two tensor given by
l " = a.i j a-" a tJ + ~ / . ( a" a 3 + a3~a " ) + a3a 3
(2. 15)
(v) Di rect or curvature tensor on ~. A bilinear form : ~., x ~, - , R with associated covari ant
rank-t wo tensor given by
t~ 1
K = K,,~a a ; K,, b : = ~(t,~ . a o + t,~. a~) . ( 2 . 1 6 )
Identical definitions hold for the corresponding t ensor fields defined on 5e and denot ed
}'~o nO T~oS
No 1 o, ~;o
S ~
Fig. 2. Tangent planes and tangent spaces on the current and reference configurations.
J . C. Si mo, J . G. Kennedy, On a stress resultant geomet ri cal l y exact shel l mo d e l 139
by 1 and K , respect i vel y. Not e t hat r ank [/~] = 3 by vi rt ue by assumpt i on (2. 6); i . e. ,
J ' = det I / r ] >0 . (2. 17)
Ther ef or e, t he map /~" W,,,,---, ~x is i n v e r t i b l e at each .~" = q~"(~:~, ~2) ESe and ~7= q~(~ l,
s~2)(E 5e, and we t r ansf or m surface t ensors from Ux. to 7/'~ ( p u s h - f o r w a r d ) or from W~ t o o//.
( p u l l - b a c k ) . Use of t hese oper at i ons l eads to t he strain t ensors r ecor ded in Tabl e 1.
We not e t hat t he component s of t he strain t ensors defi ned on ~0 and rel at i ve t o t he basis
{ a ~, a 2, a 3 } are i dent i cal to t he component s of t he cor r espondi ng strain t ensor s defi ned on
rel at i ve to t he basis { a' , a 2, a3}. That is, we have
o - o ~at~ (a~,t3 - a ~ ) =" e,t 3
a~ E , at3 - - a~ =
(2. 18a)
t o - o
E ~ a ~ - t . ~ a ~ = l ( ~ / ~ . ~/ o) =: e3, =: ~, ,
and similarly
0 - 0 0 _ _ .
a,~ Et at 3 =-- a, , ~t a~ = K,~fj - K ~:j p ~ ,
(2. 18b)
t " / ~ a - - t . $,a~ - 0
t Ot "
Thi s concl usi on follows from t he fact t hat , according t o definition (2. 14),
- " Ft" ( 2 . 1 9 )
a , = F a ~ and t = ;
i . e. , t he frame { a~, a , t } is ' c o n v e c t e d ' t o the frame { a, , a 2, t} by t he surface def or mat i on
e,
gr adi ent fi " ~/~,:,, ~ 7/~.
2 . 3 . T h e r a t e o f d e f o r m a t i o n t e n s o r s o n t h e m i d - s m f a c e s b ~') a n d b ~
For conveni ence, we use t he not at i on g = ( ~ ', ~ 2) E M, so t hat ~? = ~o(~ ). A mot i on of t he
shell is defi ned by a t i me dependent curve of confi gurat i ons in ~, t hat is, by t he map
t ~ [ 0 , T],--> q~(. , t ) - 0 p ( " , t), t ( . , t ) ) ~ ~ . (2. 20)
The m a t e r i a l v e l o c i t y field t hen consists of t he vel oci t y of t he mi d-surface and t he vel oci t y of
t he di r ect or field; i . e. ,
@( ~, t ) " = (~/,( ~, t), i( ~, t)), ( ~, t) ~ ~ / [0, T] . (2. 21)
As in our di scussi on of strain measur es, spatial and mat eri al rate of def or mat i on t ensors are
Table 1
Strain tensors on the reference and current surface b " and 90
On 5e" On 5e
Surface strain tensors /~,o "= ~[ # '1/~ - 1"] d~ "= ~[1 - F- ' I "/ ~- ' ]
Director strain tensors /~, := [P' KP - K"] i, "= ~[K - ,~-'K",~-'I
Natural basis { a , a~, t") { a I , a 2 , t}
140 J.C. Simo, J. G. Kennedy, On a stress resultant geometrically exact shell model
defined as follows:
(i) Mat eri al r at e- of - def or mat i on tensors on 5e , are the time derivative of the correspondi ng
strain tensors; i. e. ,
/ ) ~, ' =/ ~, and /), "=/ ~, . (2. 22a)
In component form, in view of relations (2. 14), definition (2. 18) yields
L)~ = ~i,~t~a"" a ' ' + ~ ~/ ' ( a' ' a"3 + a ''3 a " ~) ,
Dt " (w i)/3
= K~a a .
(2. 22b)
(ii) Spat i al r at e- of - def or mat i on tensors on b ~, are defined as the Lie derivatives of t he spatial
strai~ tensors on b . Accordingly, we have
_ , [ o ( p , ~ p ) ] p _ ,
J ~ : = L o ~ = V
d , : =L o g , = p - ' ~ ( ~ 6 t e , p) p - ' .
(2. 23)
Fr om (2.18) and (2. 19), and noting that
a" = f f -ta" and a 3 = f f - t a 3 ,
(2. 24)
in component s relative to the basis { a~, a' , a3} , we have
dt
= p - , l ~ ] p - ,
- p - , [ . ,,,, a, , ~ ]
- K,,#a p- t
= % (P -' a' " ' ) (P -' a" " ),
(2. 25)
and a similar expression for d,~. Consequent l y, in view of (2. 24), we concl ude t hat
d r = ti, ~a a @a 0 + ~,~(a ~ t ~ a 3 + a 3 @a ~)
= e ~ a " ( ~ a 0 + t ~( a ' ( ~ a 3 + a 3 ~ a ~ ) ,
C ~
d, = K ~a a s = p~a a ~
(2. 26)
Agai n, we observe that the component s of the spatial rates of deformat i on t ensors d r and
d, on 5e relative to { a~, a 2, t_} are identical to the component s of the mat eri al rat e-of-
o to} .
deformat i on tensors D r and D, on b relative to the basis { a{ ~, a2,
2. 4. St ress resultants and stress coupl es
With the kinematic quantities defined above in hand, we next i nt roduce the stress resul t ant s
and stress couples defined in Part I.
(i) St ress resultants on ~ . The ef f ect i ve stress resultants and stress coupl es on t he current
J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model 141
surface 6e are denot ed by 3
= n % a a + q ( % t + t % ) ,
"- " a/3
m = m a~ a~ . (2.27)
We recall t hat an i mport ant consequence of balance of angular moment um is the
symmet ry of the membrane effective resultants r~ ' ' . This propert y is also assumed to hold
for n~ "a. Consequent l y, one has
r~ " ~=r 7 ' " and n3 " a =n~ a " , a =l , 2 . (2.28)
(ii) St ress resul t ant s on ~0. As in Part III, the effective stress resultants and stress couples on
the reference surface Ae" are defined via pull-back operation wi t h/ ~" ~x, ---> o//. as
AI "= ] . S- ' i ' i ~' - ' and /14 : = ] / ~- ' n~/ ~- ' . (2.29)
- I}
Agai n by exploiting the fact t hat a, = F a , , from (2.27) and (2.29) we conclude that, up to
a factor of J, the component s of g and ~ relative to { al, a2, t} are equal to the
component s of / V and M relative to { a , a , t}; for example,
N~ : J[n- x~#_o% a~ + 4. ( aO t o + to aO)],
/14 : jnq"# a a~.
(2.30)
It can easily be shown that t he stress power expression takes the following form:
v=
-
+
f~ ~,, ~-a. - d/xO
= [ fi~ t~ ti,a + q 7 . + m r . # ] J ,
where d/z "= ]" d~ :l d~ :2.
(2.31)
2. 5. Mat r i x f or mul at i on: s u mma r y o f not at i on
It should be clear from the precedi ng discussion that the only rel evant objects in t he
definition of t he state of stress and deformat i on of the shell are the t ensor component s {r7 ~a,
q~, m } and { a, a, y, , r , a }: These objects can be i nt erpret ed as component s of tensors ei t her
on the reference surface 6e, or on the current surface ~, according to the definitions given
above. It t hen proves conveni ent to introduce matrix notation and set
{ o}
a l l - - a]~
0
e(@) = a22 - a2z ,
2 ( a , 2 - a'~2 )
a( # ) =
o}
~1 - - ~ l
0
7 2 - - 7 2
I 0]
Kl l - - KI01
0
(2. 32)
3 This is at variance with Parts I and III in which t~", a E { 1, 2}, are not included in fi, but rather are treated
separately.
142 J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model
along with
f , ~ " ]
I . , ~ ' U
q =] ~ , m=] . (2.33)
12
Furt hermore, for convenience we also set
[ : m }
e . - -
[ " e ( q b ) }
, ~ ( ' ) = '~ , ~ ( ' ) ,
L p ( m )
, ( ~ ) = i ~ ( ~ ) ~ ,
L#(~)J
(2.34)
where e., with Y scaling the Cauchy stress resultants, is interpreted as components of Kirchhoff
stress resultants. The rate-of-deformation ~( ~) is related to the velocity field through the
matrix expression
/:l Im
~'(~)=B(~) , B(~)= B, , , , B~.b , (2.35)
B;,,,, B;,bJ
whe we have used the notation summarized in Box 1 of Part III. The weak form of the
equations of equilibrium then takes the form (see Parts I and III)
G ( , r , ~ . ~ ) : = B ( ~ ) ~ r . , r - ~ x , ( a a , ) = 0
' ,/
(2.36)
for all admissible variations ~ =( ~ q ~ , ~it)E T,~,~, where _T~,~ is the tangent space of
variations defined in Part I. Recall that ~it= A ~T where A is a 3 x 2 submatrix of the
orthogonal matrix A mapping a vector E 3 ~ S 2 into the current director t, as explained in
detail in Part III. As usual, ~g~,, ( ~ ) is the weak form of the external loading given by (2.32)
of Part III.
Finally, for later reference, we record a form of the linearized discrete weak form which is
valid for elastic or elastoplastic constitutive equations. Recall that linearization of the discrete
weak form, following the notation of Section 5 in Part III, is split into geometric and material
parts as
DG(@,,+~; ~@). A@,,+~ = DGo(@,,+,; ~i@). A@,,+~ + DGM(@,,+,; ~@). A@,,+,.
(2.37)
In the case of elastoplasticity, the geometric part DGG(~,,+~; ~@). A@,,+~ takes exactly the
same form as in the purely elastic case discussed in Section 5 of Part III, except the stress
resultants {n, q, m} are now evaluated simply using the discrete elastoplastic constitutive
discussed below. The material part, DGM(P,, +~; 5@) . A@,,+,, on the other hand, takes a
somewhat different form, which for now will be represented in the form
f~{ ~i ~} t [de. "+ ] { A~P,,+'}d/z
OMa( C' , , + l ; ~ C' ) ' a ' , , + , = ~ r a' ' a aT, , +
d E n + 1 1
(2.38)
Once de',,+ 1 / de,, + ~is specified in (2.38), DG(~k; ~i ~). A ~k becomes completely defined. We
refer to de.,+l/dE,,+l as the consistent algorithmic tangent moduli.
J . C . S i m o , J . G . Ke n n e d y , O n a st ress resul t ant ge ome t r i c al l y exact s he l l m o d e l 143
3 . St ruct ure o f t he el ast opl ast i c cons t i t ut i ve model
In this section, we outline the structure of the elastoplastic constitutive model considered in
this paper. First, we consider the formulation of the model in 'intrinsic' form in terms of the
kinematic quantities defined in the preceding section. Subsequently, we revert to matrix
notation and show that for implementational purposes, the structure of the model is identical
to that of the linearized theory. For further information on elastoplasticity at finite strains, we
refer to the monograph of Simo and Hughes [2].
3. 1. Pl ast i c st rai n a n d pl as t i c rate o f d e f o r ma t i o n t ens or
- p
To characterize plastic flow, we introduce spatial strain tensors defined on ~, denoted by %
and (P, with components (cf. (2.18a))
- p p a at3 p ot 8~(a a 3 a 3
% = e~,tja + + a " ) ,
, , a
P ,~t3a .
(3.1)
Note that by pull-back with F" ~.~, ---> ~.~ we obtain plastic strain tensors on ~x,,, denoted by ff, P
and EP, respectively, and given by
/ ~ p = p (W ! R P [ nO a a Os a o3 a o' ~)
e~t3a a~ + ~,, ~x,, +
= p, , t~a .
(3.2)
Rates of deformation tensors are defined exactly as in the preceding section; i.e.,
aP~ = 1.,o~, p and dP=Lo~~,
6 p = / ~ and /),P =/~,P.
(3.3)
Again we observe that the components of { d p, d, p} and { /)P,/~,P} relative to the basis { a~, t}
and { a~, t}, respectively, coincide. For instance, we have (cf. (2.26))
d p p a a a/ 3 p a 3 a 3 a a
e~ a + 8 ~ ( a ~ + ) ,
d P = p . a a a a .
(3.4)
This conclusion is the result of using convected coordinates.
3. 2. Hy pe r e l as t i c st ress r es pons e
We characterize the stress response by means of a free energy function of the form
- . - p - . ) .
, e , , % , e P , P . l ( 3. 5)
Frame indifference (or covar i ance) requires that W depend on the strains and/ ~ through the
pull-backs of the strains with F; i.e.,
W - W ( E, p , i f , t ; E, p , - p i f , r , 10 ) .
(3.6)
144 J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model
In component s we simply have
I~/'( P 1
w = e. , ao, p ) . (3.7)
As in expressions (3.5) and (3.6), the dependence of if' in (3.7) on the reference metric 1 is
understood.
Standard arguments then lead to the constitutive equations
aft,'
jrT~ _ O f f " y~,~_ yr~,,~ _ ( 3 . 8 )
ae ,,~ ' aa,, ' a p ~ "
In our numerical implementation, for reasons of simplicity, we restrict our attention to the
following simple constitutive equation:
]ff,,tJ = E h
2 H'~"7(e~,,7 P e y ~ ) ,
1- - V
J q " = GhKa""~(~t3 - aP~), (3.9)
E h 3 p ) ,
Jn5"~ = 12(1 - 2 ) H~" ( Pv o - P~,
where h : = h - h is the thickness of the shell, E > 0 and G > 0 are interpreted as elastic
moduli, and u ~ [0, ] as Poisson' s ratio, and H "a~a is given by
H ''~va "= va"~a ~ + (1 -- v)[a"~a ' , ~ + a " " ~ a ' ' " l .
(3.10)
In (3.9), the rate of deformation tensors d r and t~ are assumed to a d d i t i v e l y d e c o m p o s e into
-- rl
elastic and plastic parts, denoted { a~, d,~} and { d~, d~,'}, respectively; i.e.,
- e - p
d~ = d~ + d~ ,
d, - - + a, .
(3.11)
Consequently, from (2.26) and (3.4) the strain measures, using the matrix notation e I~ 8 of
(2.34), additively decompose into elastic and plastic parts, denot ed by ee E I~ s and P E ~8,
respectively; i.e., e = e e + e p.
3. 3. Y i e l d c o n d i t i o n , f l o w r ul e a n d h a r d e n i n g l aw
Conceptually, one characterizes the plastic response by means of a yield function formu-
lated in stress space of the form
~b = t~(j~ "~, J ~' , .I~"~; pS, 10),
(3.12)
where p", s = 1 , . . . , m, are the components of m internal variables characterizing the
hardening response of the material. We assume an associative flow rule of the form
J. C. Si mo, J. G. Kennedy, On a stress resultant geometrically exact shell model 145
1 ~
. p
e ~ t ~ = 4/ y O ff,,t~ ,
1 0 ~ (3.13)
~P = ,~ y ~4o'
1 0 ~
-p
0 . 8 = ? y o. 5o~ , ,
and a general hardening law given in terms of generalized hardening moduli h S ( J ~ ~ , J ~ ,
jth,,a; pS) as
p " = 4 / h " ( J ~ t 3 , ] ~ t J , ] ~ , , t 3 ; p. , 1 " ) , s = 1, 2 , . . . , m . (3.14)
Notice that the yield function (3.12) and the flow rule (3.13) are formulated in terms of the
Kirchhoff stress resultants, consistent with (3.8). In these evolution equations, 4/I> 0 is the
plastic consistency parameter; a function satisfying the Kuhn-Tucker complementary condi-
tions
4/ >~ 0 , Jp <~ O , 4 / ~ = 4/ ~ = 0 , (3.15)
which complete the formulation of the plasticity model.
R E M A R K S 3 . 1 . (1) In view of relations (2.27) and (2.30), the yield condition (3.12) may be
interpreted either as a function depending on the spatial quantities if, hi, the deformation
gradient F (and the reference metric tensor 1 ) as
= 8( Jn, Jn]; p", F, 1"),
(3.16)
or as a function of N, A/according to the pull-back relations
= ~ ( j p - , ~ p - , , j p - , , ~ p - , , p~, ~,,)
= ~ ( / V, A I, p", 1") .
(3.17)
However, when use is made of convected coordinates, the coordinate expression of & reduces,
in (3.16) and (3.17), to (3.12).
(2) Similarly, the flow equations (3.13) can be interpreted in the spatial description, as
defining evolution equations of the form
1 0 ~ and alP=4/ 1 0~ (3.18a)
'~P = ? y 0,i } O,a;
or, equivalently, in the material description as defining evolution equations of the form
O~ 0~ (3.18b)
/~P=4/ O/V and EP = 4 / 0 / ~ ;
the equivalence of both descriptions is again the result of using convective coordinates.
146 J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model
(3) In the formulation of (3.12) and (3.18) we have assumed for simplicity that the function
4, is smooth. Such an assumption is not realistic for models formul at ed in terms of stress
resultants, where 4, is typically non-smooth. This situation is considered in detail in Section 4
following the general development presented by Simo et al. [9].
4 . Co n t i n u u m el as t opl as t i c mo de l i n st ress res ul t ant s
In this section we summarize a general framework which encompasses a broad class of
resultant based elastoplastic constitutive models suitable for shells. Many common elastoplas-
tic models for resultant based shells fit into this framework (see [4-8] and the review article by
Robinson [14]). In addition, we focus on a specific model based on a generalization of the
two-surface yield criterion proposed by Shapiro [7]. Throughout this development, the stress
measures { n, q, m} and strain measures { e, 6, p} are assumed to be expressed in the local
Cartesian f rame, as defined in Section 3.1. of Part II.
4. I. General multisurface elastoplastic model: basic equations
To aid in constructing the elastoplastic model, the matrix notation introduced in Section 2.5
for the stress resultants or E R ~, strain measures e ~ I~ ~ and plastic strains ePE R 8, is used,
where the stress resultants { n, q, m} and strain measures { e, 6, p} are hereafter assumed to
be resolved in the local Cartesian frame.
Following Simo et al. [9, 10], the set of internal variables p", s = 1 . . . . , m, introduced in
Section 3, is supplemented by a conjugate set of internal variables a~, s = 1 , . . . , m, t hrough
the transformation
p = - V ~ ( a ) - - De ,
(4.1)
where the hardening potential ~ ( a ) ' = e' De, with D E R'" x R'" constant, is assumed to be
strictly quadratic, for simplicity. The hyperelastic stress response (3. 8), in matrix notation, then
becomes
o" = VW( e - eP) , (4.2)
where, again, or are the Kirchhoff stress resultant components (cf. (3.8)). The dissipation
f unct i on @ P, representing the energy dissipated in plastic processes, then takes the form
~P[o. , p; [P, t ~]: = o. ' [ p +p' & ~>0.
(4.3)
Relations (4. 1)-(4. 3) have a sound thermodynamic basis which is discussed briefly in
Appendi x B.
Motivated by the fact that yield criteria for resultant based shell models typically entail
multiple yield functions, we consider the case in which the elastic domai n, denot ed by
[E,, C R 8 x R p, and its boundary, denoted by OE~,, are defined as
IE,. := { (o', P) ~ ~ RP I 4,.(~r, p) < O, for all/x ~ [1, 2 , . . . , ml } ,
(4.4)
0[E,,' = { (o., p ) E~ x RP[ 4,.,(~r, p ) = O, f or s ome/ z ~ [ 1 , 2 , . . . , m] } ,
where 4,~,(o', p) are m I> 1 smoot h functions which are assumed to define i ndependent
J. C. Simo, J. G. Kennedy, On a stress resultant geometrically exact shell model 147
constraints at any (~r, p ) ~ 19~r 4 and may intersect in a non-smooth fashion. The closure
IF,, U 01E,, is assumed to be a closed convex set.
The evolution of the plastic strains e p and the internal variables a t ake the form
5 , " a , . , / , . ( o - , p) ,
. =1
nl
a = Z ,~,z gpf~.(O', p ) .
g =l
(4.5a)
where, for simplicity, attention is restricted to the associative case. Her e, b represent s
ordi nary time differentiation of to. Notice that so-called ' objective rates' are avoided in t he
component expressions (4.5a), which, in terms of the notation i nt roduced in Section 3, t ake
t he equivalent forms
1 O , ~ . ,
t n
Lva= E
. = !
dP = ~ ~/" l o,~qb,
=l J '
(4.5b)
where { &,/i} are resolved in { a ~, a 2, a 3 } a n d { a~, a 2, t } , respectively, with component s { a, p} .
Fr ame invariance is guarant eed in (4.5a) by the use of the Lie derivatives (4.5b) in convect ed
coordi nat es. In (4. 5), ~," are m i> 1 plastic consistency parameters, which satisfy the following
Kuhn- Tucker complementarity conditions for # = 1, 2 , . . . , m"
, ""~ b (o- , p ) = 0 , ( 4 . 6 )
~," I>0 ~b,(~r,p)~<0 and 3' ,
along with the consistency requi rement
~ , ' ~ . ( e r , p ) - O .5 (4.7)
Condi t i ons (4.5a) I and (4.6) are essentially the multisurface plasticity count erpart of those
in Koi t er [6, eq. (2.19)], and have been employed by several authors, not abl y Maier [16] and
Mai er and Gri erson [17]. The associative expressions (4.5a) evolve from the classical principle
o f maximum plasticfdissipation (see [101).
The above framework may be immediately ext ended to viscoplasticity by means of a
generalized Duvaut - Li ons model as discussed in [9]. Not e that viscoplastic models of the
Perzyna type (e. g. , [18]) are not meaningful in the presence of multiple yield ~,arfaces
intersecting in a nonsmoot h fashion.
4.2. Generalized llyushin-Shapiro elastoplastic model
We now specialize the general framework above to a particular model based on a proposed
generaiization of the yield criterion proposed by Shapiro [7] to include hardeni ng. This model
is consi dered here not only as a good approximation of the Von Mises criterion for shells, but
also as a represent at i ve model for a general class of multi-surface resultant based elastoplastic
formulations for shells. For simplicity, the stored energy function W and the hardeni ng
4 For example, if p = 0 and dim IF., = 8, then at mast eight independent surfaces can intersect at one point.
.s The summation convention on repeated dices is not enforced in (4.6) or (4.7).
148 J. C. Si mo, J. G. Kennedy, On a stress resultant geometrically exact shell model
potential ~ are assumed to be strictly quadratic; i.e.,
W(E- E p) : - I ( E - E p) t c( E - E P) ,
Y[(a)' = arDor.
(4.8)
Here, C and D are constant and are taken to be of the form
C :'-- V2 W( E- e p) --
I 2 ^
O~W 0
2 ^
0 O~W
0 0 o~ w o Cm
0 ~ 0 '
(4.9)
where lk'(e - e p, y - TP, K - K p) ' = W(e - eP), ~( a , &) "= : ~( a) , and
K(}
D: =
Kf
b' -~H' l~.
(4.10)
The constants { K0, K', H' } are yield parameters (defined below). The hardening variables
a E ~ and & E ~8 are those associated with so-called isotropic and kinematic hardening of the
yield surface, respectively. The quadratic forms (4.8) imply linear elastic and linear hardening
behavior. For isotropic elastic response, we have
C,, := hC,
C,i := GhK~I,.,
h 3
c,,, := i ~ ( : ,
C: =
E
1- - /,2
0]
1 0
0 1 - v
2
(4.11)
where E is Young' s modulus, v is Poisson' s ratio, GK is the effective (transverse) shear
modulus, h is the shell thickness, C is the standard plane stress elasticity matrix, and 1,,
denotes the rank-n identity matrix. The stress resultants tr E [~8 and hardening variables
p E R" (conjugate to e - e p and a, respectively), from (4.1) and (4.2) then become
~ r : = VW( e - eo) = C( e - e o ) ,
[oo]
o ' - = - V~ ( ~ ) = - Da = - f ~ .
(4.12)
The generalization of the yield criterion introduced by Shapiro [7] proposed here in terms of
J. C. Siren. J. G. Kennedy. On a stress resultant geometrically exact shell model 149
the Kirchhoff stress resultants ~r takes the form
~ . ( o r , g ) =-- ~ ) ( e r + , ~ , p ) " = f ~ ( o r + , 8 )
K2(p)
2
Ko
~<0, / x( E{ 1, 2} ,
(4.13)
where
A# "'-"
x ( p ) ' = K~t + K' p,
f . (o" +/ / ) ' =( or +/ i ) ' A. (~r +/ ~) ,
1_..~ p 0
no
0
1
--~ 12
%
m
sign(/x)
2V'~n.m P
0
sign(/z) 1
_"" v o P 0 mo2 P
+1, if/x = 1,
sign( /.~) : = _1, i f / z = 2 ,
P : = - ~ 1 ,
O- 0
( 4 . 1 4 )
and n,,, q. and m()are the yield parameters associated with membrane, transverse shear and
bending responses, respectively. These yield parameters are typically related to the uniaxial
yield paramet er x()through the relations n. =[ h] K. , m. =[h2/ 4]x, and q.=[h/V'3]Xo. The
paramet ers x' and H' are the hardening moduli associated with (linear) isotropic and
kinematic hardening, respectively. Note t hat / i is the negative of the so-called hack ~tress. If
K' = 0 and H' = 0 (or, equivalently p --- 0 and , 6 - 0), then the llyushin-Shapiro yield criterion
[7] is recovered. Nonlinear isotropic and kinematic hardening can also be accommodat ed
without difficulty for suitable choices of the functions K(p) and H' ( p) .
The evolution equations (4.5) for the plastic strains e p and hardening variables a take the
form
~ P = ~ '~ ~ , ' 2 A u ( e r + / ~ ) ,
tzE{ I, 2}
+ t i ) J
(4.15)
which implies that & = e P Finally, with/~ "= - D&, it follows that
~ P
, { (4.ti6a)
where ~P is defined in (4.3). That is, (4.16a)~ is a generalization of the notion of equivalent
plastic work W p ' = o' " gP as an internal variable (e.g., see [19,20] for a discussion of WP),
whereas (4.16a)2 is an extension to the current model of the Pr ager - Zi egl er kinematic
hardening law for J2-flow theory. In tensor form, (4.16a)2 is equivalent to
150 J . c . Simo, J. G. Kennedy, On a stress resultant geometrically exact shell model
L O P = - 5 LdPj '
(4.16b)
where/ ~ is resolved in the basis { a~, a 2, t} with component s/ i .
To gain insight into the nature of the isotropic hardening mechanism defined by (4.14)! and
(4.16a)1, we consider the case of simple uniaxial tension. In a uniaxial tension state, this
isotropic hardening mechanism evolves linearly with the total strain magnitude, as in classical
Jz-fl6w models with linear isotropic hardening (for classical Jz-flow models, see, e. g. , [3]).
This is shown explicitly in Fig. 3 in which an isotropically hardening specimen is l oaded in
uniaxial tension using both the shell model above and a classical (plane stress) Jz-flow model.
The isotropic hardening mechanism in the latter case is defined by
4 , ( , 7 , ~ o ) : = Il sl l - ( ~ , , + ~ , ~ 0 ) ,
e ~ = ~ I 1 ~ ' 1 1 , s ' - d e v o ' ,
(4.17)
where o" and e p have their classical definitions in Jz-flow theory. Both responses depicted in
Fig. 3 correspond to E = i0, K0 = 0.2, and K' = 9.0. This comparison is particularly useful
since, for plane stress conditions, (4.13) reduces precisely to the classical plane stress Von
Mises yield criterion (ignoring hardening). From Fig. 3, it is evident that the classical isotropic
hardening variable 6P leads to considerably more severe hardening than ~ p that associated
with the shell model here. One can, however, adjust the slope corresponding to the shell
model in Fig. 3 via K' in (4.14)~ to replicate any degree of linear hardening present in the
classical model. Similarly, the kinematic hardening mechanism (4.16a)z also evolves linearly.
0
0
S S
S f
S S
S S
s S
s f
s J
s S
s S
s S
s S
s S
s j
s S
s S
s S
Sj SS
Pl ~l e stress hardening model . . . .
' o'., ' o:2 ' o:3 ' , : , ' 0'.5 ' o:6 ' o.7
De f l e c t i o n
Fig. 3. Illustration of the linear isotropic hardening response for the proposed shell constitutive model. Corn-
parison with the linear isotropic hardening response of classical J,-flow theory.
J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model 151
5. E l a s t o p l a s t i c r e t u r n ma p p i n g a l g o r i t h m: c l o s e s t p o i n t p r o j e c t i o n
In this section we outline a general algorithm to integrate the continuum elastoplastic
problem 6iscussed in Section 4 on t El 0, T]. The algorithm has the standard geometri~
interpretation of a closest-point-projection, in the energy norm, of a trial state onto the elastic
domain. The crucial difference between the present multi-surface algorithm and standard
single surface algorithms (see e.g. [2] for a review) conceros the determination of the active
surfaces in the return mapping procedure.
5.1. Discrete algorithmic problem
Let ~ c denot e a given point on the current mid-surface of the shell. Consider a time
discretization of the interval [0, T] C R of interest. The algorithmic probl em addressed in this
section is local in the sense that E 6e is assumed fixed, but arbitrary. At .17 E 6e and time
P a,,} are known. Let (A~,,+,, At,,+~) be a t,, E [0, T] we assume the state variables { e,,, e,,,
given increment in the kinematic measures on the interval t E It,,, t,,+ ~]. The basic probl em,
P a, , } at t, , ~ [ 0 , T] t o { e, , +~ , t~ , o, ~ , t~ , , +~ } at t, , +~ ~ [ t , , , T 1 in then, is to update the fields { e,,, e, , +
a manner consistent with the continuum elastoplastic constitutive equations developed in
Section 4. To this end, for the general model of Section 4.1, application of an implicit,
backward Eul er difference scheme to the evolution equations (4.5) leads to the following
nonlinear coupled system for the state variabies { e,,+~, eP+~, a,,+~} at time t,+~"
l~n + I
= +(Atp.+~,At,,+t) (gi ven),
ttl
n +l t t + l + '
g = l
i n
, +, , +, = ,+,, + N ( , , - , p) , , +, , ( 5 . 1 )
g = l
,' )
= v w ( e , , + , - e , , + , ) = c ( e , , + , - e, , + , ,
P,,+l = -'9'Oe(a,,+ l) = - Da . +t ,
where we have set Y,,+tt' "= At3,,,+z,'" and t~(A~0,,+z, At,,+~) is the identical geometric update
considered for the elastic case in Part III, regarded here simply as a given quantity in the
return mapping algorithm. As not ed previously, the use of convected coordinates precludes
the need for ' objective rates' in the component forms of the evolution equations (4.5a).
Consequently, so-called ' incrementally objective' algorithms of the type discussed by Hughes
and Winget [21] to integrate such objective rates are not needed here. Incremental frame
invariance of the algorithm is guarant eed by the use of the Lie derivative in convected
coordinates. The discrete count erpart of the Kuhn- Tucker loading/unloading conditions (4.6)
take the form
" I>0 ~bg(tr, p) , , +~< 0 and " =--0 (5.2) p) , , +,
" Y n + 1 '
f or / z = 1, 2, but no sum on/ z. By specializing the update algorithm (5.1) to the case of the
current model in Section 4.2., (5.1)2.3 are replaced by
152 J. C. Simo, J. G. Kemwdy, On a stress resultant geometrically exact shell model
e, P, +t E . + 1 2 A~ i '
t . ~{ I . 2 }
5;
= = ~ , , +
d]~n + I ~ n +l ~tE{ 1,2}
- - 2K' K( p) }
9
T, " , + , K, ; ,
2 a . ( ( r +/ i ) ,,+,
(5.3)
whereas (5.1)~.4.s remain unchanged. It follows that &,,+~ = e,,+~. The elastic strain measures
e ~ ar/d the stress resultants o" are regarded as dependent variables and are obtained t hrough the
hyperelastic stress-strain equations e",,+m := e,,+~ -- e~ +~ and o',, +~ = VW(e~+~).
The nonlinear system (5.1) (or (5.3)) subject to the unilateral constraints (5.2) defines the
discrete algorithmic problem. Convexity of the admissible region IE,, U 01:,, (which is guaran-
teed by a positive definite A, , IX = 1, 2, in (5.3)), renders this probl em a convex mathematical
programming problem with a unique solution. In particular, it is a convex minimization
probl em equivalent geometrically to the closest-point-pro]ec',ion of the trial state { O',,+triall ' Pn+trial| ~
onto the boundary of the elastic region on:,, in the energy norm. The trial state is defined by
' freezing the plastic flow' in the interval t E [t , , t,, + ~]. Accordingly, setting y,+~' ~ = 0, t* = 1, 2,
in (5.1) we obtain
ptrial
En +1 "-- EP '
t ri al
t2 Irial ,
En +I = En +l - - Ep '
t r i a l . . _ ~ TW( t2trial
l[lrn t I E. +I ) '
Iri al t ri al
P , , ~ I ' = - Da, , + I ,
(/ } t ri al . ( l[]r t ri,| l
p. , . I- I "- ~1 . n+ I ~
- t r i a l
P , : t l ) '
(5.4)
TIle trial state arises naturally in the context of an elastic-plastic operator split.
5.2. Numerical solution strategy
The iterative solution algorithm for the general elastoplastic model in (5.1) is considered in
[9]. Specialization of this iterative algorithm to the current model in (5.3), with combi ned
isotropic/kinematic hardening, follows without fundamental difficulty. However, the particu-
lar form of ~, , _IX ~ { 1, 2}, in (4.13) allows for a significantly more efficient solution strategy.
In particular, 4,,, IX ~ { 1, 2}, in (4.13) may be expressed exclusively in t erms of the
consistency parameters T,",+~ ~ R, Ix ~ { 1,2}. Consequently, the return mappi ng reduces to
the solution of the following nonlinear, rank-2 system:
6 . ( T' . T" ) . +, ' =q ~ . ( ( r +/ ~ . P). +, = 0 . I X E{ 1. 2} . ( 5. 5)
Al t hough the explicit definition of the functional forms of ~,, IX E { 1, 2}, is a cumber some
and notationally intensive task, the development is conceptually straightforward, as is shown
below.
5.2.1. Reduced rank-2 systems
The reduction (5.5) follows directly by expressing { o" + ,6, P},,+t exclusively in { 3 ,~, T2},,+~.
J. C. Si mo, J. G. Kennedy, On a stress resultant geometrically exact shell model 153
t ri al
Use of t he elastic stress-strain equat i ons o-,, ~ = C(e,, + ~ - el~ I ) = o-,, ~ - C Ae, p, + ~ al ong wi t h
(5.1).s and (5.3)_, t ot p,,+, yields
Y,,+l H' A# -t- 2CA, ~!,,+, = r/,,+l ,
~ 1 1 , 2 1
wher e
I
~, , ~ : = ~q "= q+/ J q
11, . , , + | ~ +/ : i , n . + |
[6,}
g , + , = ' / p _ , ,
I, P,,, ,,!
(5. 7)
Fur t her , by not i ng t hat
-~2jC,,P 0 si gn(/ z)
2V~nom, , C, , P-
1
,..,-,a-'A# = 0 -512 0
qo
sign(/x) Cm p 0 1 C,,,P
2v' ~n.m. m, 2,
( 5 . 8 )
we see t hat t he l i near syst em (5. 6) is uncoupl ed in q,,+l. In fact, from (5. 6), we concl ude t hat
q, , +, = ~ ' ( y ', y' - ) , , + , ql , ~ ' l ,
~' (y ', y'~),, +.
" ~ 2 )
1 + GhK - - ( y ~ + y
q. ,, + i
(5. 9)
wher e (4. 11), has been used for C,~. For isotropic el ast i c response, t he syst em (5. 6) t akes a
r emar kabl y si mpl e form. In part i cul ar C,,, C,,, and C defi ned in (4. 11), al ong with P have t he
same charact eri st i c subspaces. That is,
= - 1 1 , ( 5 . 1 0 )
P = QApQ t , C QAc_Q' , O: = ' ~ 0 0 "v~
wher e Q-~ = Q' is or t hogonal and t he diagonal mat ri ces Ap and Ae are gi ven by
0
A p ' = 1_ 3 j 0 3 , A~ ' = 0 1 + v E
0 0 0 0 2(1 + v)
0
(5.11)
It follows t hat P and C: commut e; i . e. , PC =CP. El i mi nat i ng q,,+t from (5. 6) via (5. 9), t he
154 J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model
remaining system (5.6), after premultiplication by [Q', Qt], takes the form
/ ~ trial '
~' ~3 ~'~'4 n + 1
L = m n+l
(5.12)
where
~:,, "- Q tn,,, ~m := Q t'i~m (5.13)
and
..-(3 l, 2) , , +,
4 t
. = 13 + ( I + 2 ) n + l _ _ [~H Ap + hAcAp],
no
., 1 4
~%( , y 1, ' Y' ) n + l " - =( ~ / t l - ~ 2 ) n + l X/-~nllm, I [~H'Ap + hac ap] ,
- %( ~, ' , ~ ) , , + , : = ( ~, ' - ~,
[ h3 ]
1 4
2),,+ 1 X/-~nom ~H'Ap +- ~ Af:Ap
[ h3 ]
2 4
, ~%(, . ylj "Y' ) , ' , +I - - 13 -1" (, . yl + ,,} t2)n+, mE ~H'Ap + -1~ AcAp .
(5.14)
Inversion of (5.12) leads to
r, 1 =E. -lE rll_ _2
L g , , , J , , + , - ~ ~ ~ : ' " " ' '
J--r3 ~ 4 ~ m J n + I
B
where "~k, k E { 1 , . . . , 4}, are diagonal and are given by
(5.15)
- % ( ~ , ' , ~ , ' ) , , +, "= [ .% - Z , ~ . ; ' ~ ' ~ l - ' .
- % ( ~ , ' , ~ , ;) , , +, ' = [ Z ~ - ~ , ~ " ~ . ; ] - ' ,
- - ~ 4 ~.w3 ,
(5.16)
With these results at hand, f~,.,,+ t, ~ E { 1, 2}, may be expressed as
f . . , , +, = a . .
~ 2 , , , , , + , L ~ 2 , , , J , , + ,
where
(5.17)
- 1 sign(~)
-~ Ap 0 2X/3nomo
0
1
0 q212 0
sign(/x) 1
-"" mnom v ~ Ap 0 2 Ap
mo
I
Ap
(5.18)
J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model 155
Substituting (5.9) and (5.15) into (5.17) leads to
r. e, . . , '
/ E ,ri.!
I _ m m .
n+ !
where
~ ' ( ~ ' . ~ ) , , + , a . ~ ( ~ ' . 2) , , + ,
io o o ] ~ ( ' y 1, y2) , , + 1 : = ~ ' n + , 1 2
L~.,3 0 , , ~ 4 J n +
I
~,,?a,]
.:a' /
n + !
(5.19)
(5.20)
In (5.19), only 5(y~, "y2)n+l i s dependent on 3'.+1, /z 1. 2. Next, it follows from (5.1)s,
(5.3)2 and (4.13) that, for plastic loading,
where
~( , y !, , y2)n+ !
/~(1'1, 1,2),,+,' = p,, + ~'~ y.+,~' 2
pE{ !,2} KII
p,, + ~] y~' 2~//~(y ', 3/2
= ,,+I ),,+l , (5.21)
/~E{ i,2}
/ ~( , yl ~/ 2) n+! _,= K,,~]~.(3~I, y2) . +, . (5.22)
The yield functions, ~b~.,,+ ~, may then be expressed exclusively in { y i, , y 2 } n + I as
kz(yl, ~,2),,+, =0 /~ 1 2 (5.23)
$ , , ( ~ " , ~ ' ~ ) , , + , = L ( ~ " , ~ ' ~ ) , , + , - , - , = ,
KO
Consequently, (5.23) replaces (5.5) as a reduced nonlinear, rank-2 system for ~/~', ~ = 1, 2
subject to the unilateral constraints (5.2). It may be shown that 4..~,,, + ! , / z = 1, 2 monotonically
decrease with ~' y,,+~,/z = 1, 2 and that
lim ~,.,,+, = O, /.~ = 1, 2 . ( 5 . 2 4 )
{~, k~,21---,0
Thus, for monotonically increasing hardening laws, (5.23) has a unique solution "/),+1
2
3,,,+ i ~ > 0. An iterative solution procedure to solve (5.23) is discussed below.
~>0,
5.2.2. Iterative solution scheme
The reduced system (5.23) is ideally suited for a local iterative solution scheme using
Newton' s method. The crucial difference between the solution algorithm for the multi-surface
system (5.23) and standard solution algorithms for analogous single surface systems concerns
the determination of the active surfaces during the return mapping. For convenience, the
iterative solution algorithm for the solution of (5.23) subject to the unilateral constraints (5.2)
is summarized in Boxes 1-4, in which,
a, , L 2 ~q ( . ~. +) a ~( a v , 5 . + ) nq'"'
L .[~=ia' .+l L~2'ria'J.+,
156 J . C. Si mos J . G. Ke nne dy s On a stress resul t ant geomet r i cal l y exact s hel l mo d e l
~ox 1. Elastic predi ct or
I. Comput e elastic predictor
,r,,,, = V W( t , , d, , ) , . a,
- P , , . t = P, , O ' n ~ I + I s
z t r i a l t r i a l x {
, t , ' r i a l :=~b t o - , I p,,+~) f o r / ~ E 1 , 2 }
I . , t . t ! I , s
2. Check for plastic process
IF .,-'htrial~,.. + t "~-< 0 for al l / z E { 1,2} THEN:
Set ( . ) . , i = (" trial
) . + t and EX I T
ELSE:
k = O
d ] ( " ) a, ''ial >0}
,,~, = { t t e 1 1 , 2 } 1 . ~ , , .... ,
GO TO BOX 2.
ENDI F.
Box 2. Ne wt o n i t e r a t i o n f o r T i, y . ~
3. Initialize ret urn mappi ng
/ z ( ' ~ ) . / , t ( x )
Y,,.I =0 ~ ay, , ~ I = 0 s
4. Check convergence
I1~,11-':=((~,')-" + ( -'--'-,*, Y ) ),,+l
, , , 2 x 2 x ( k )
IIA ~ 'II-':= ( ( A ~ " ) " + t " ~ , j ),,+,
i F: IIArll -" < TOLIl Yl l ~ THEN:
EX I T
ENDI F
5. Co mp u t e
( ~ ( k , j ~ ( I s 2 x ( k ,
, . , , t = ")' Y ) . ~ l -
b,,.~ = a . ~( * )
~ , l ~ J l ~ . n + I
GO TO BOX 3.
/ z (E ,11 ( * '
- - al zt
/ ~ 2 ( , . ~ I s _ 2 x ( k )
Y ) , l
"s
K ( I
from eq. (5.25)
Box 3. ( Cont i nued) Newton iteration for y J, y"
6. Solve linearized system
(a) Case !: dll,~ = { I, 2}, ( n, m ) p # 0; general loading
( 4 .
= ;) . i l k) = b~' r ~ ,,el l .' .} / , , , i g v . / j c v l j ~ v . n I ~ + E
Ay "d,,, I gl l ga: - gl:g21 - g 2 , g' . 2 ~aa, , , I
(b) Case 2: .n'*) _,,~., = { 1, 2} , ( n , m) , , - - O; pure membrane or pure bending
( ' )(*)
-~' *' = b 8K~ ( t , + )
g t l = Ov ' ~ l .... I II yl bl l ,,+l
K o
(c) Case 3: d]l~ ~ = { 1} or { 2}; single active surface
~" = { fl (E { 1, 2} I dllk~ = { /3}}
4KK' \ ( * )
=~ 3.(* , = b~ 2 ( ] + y' b~. ) ) , , , ,
g ~ . ~ . t ' v ~ u , ) t n I
. K o
. o.(*, for/ 3 = ~" (no sum)
A yn+ I = , ~ l '
for ~ # ~"
GO TO BOX 4.
( n o s u m )
J. C. Si mo, J . G. Kennedy, On a stress resultant geometrically exact shell model
Box 4. (Continued) Newton iteration for "),', .),2
7. Updat e y~'
_ / t ( k I ) ~ ( k ) . /. t, k )
Y,,+, : =Y, , +, +AT, , , , [ t r i a l y ~']
IF: y,,+, < 0 , / x E THEN:
- - a c t ~
_ v . l k ~ I1
Reset S l,~, +. ' = {V. e S ~ : l y , , + , >0}
S e t k = k + l and GO TO3
ELSE:
ELSE:
,~ t . *, l k + I ) , ~ v . , l k + I )
n + l = / n + l
S e t k = k + l and GO TO4
ENDI F.
157
where
n 4 t
2 [~H Ap + hA tAP]
n~
m
sign(/3) [ .~ H' Ap + hA t Ap]
V~nom.
0 0~v~',l+ t 12 0
+ -~ A t A P 0 --v ~H
m?~ 'At" + -i2 Ac Ap
sign(/3) [
_X /~n,,m,, ~ H'Ap
- Ghr 2
- " 7
q~
0,o~,,+ i = (5.26)
1 + Ghr 2] (Y),+I + ]t~+l)
Not e that special precaution is required when ( n, m) p - 0; ( n, m) p := ntpm, since the matrix
g in Box 3 becomes singular in this case. Once y~' ,/x = 1, 2 are det ermi ned from the solution
of (5.23), the remai ni ng state variables are obtained from (5.21), (5.15), (5.14)1, (5.1) and
(5. 2) as follows:
_ f ~ t r i al )
{ " 1 o o_ ,] l
! t r i a l /
1 ~ q = ~, , + ! | 2 0 l ~q
n ! J
~' 12A~ '
P = P + ~n+ + ,
n + i ~ n 1
(5.27)
p { p, , + Y~ ~" 2~( y~' y2) "+l }
, , +l
--" ~ txE{ 1,2} K0
P n +! n +i
- - 2
The stress resultants are then obt ai ned from ~r,,+ ~ = ~,, ~ -fro 1.
158 J. C. Si mo, J. G. Kem~edy, On a stress resultant geometrically exact shell model
5.3. Algorithmic elastoplastic tangent moduli
The linearization of the weak f orm (2.37) is completed by specifying the form of
do',,+~/de,,+~; i.e. the consistent elastoplastic tangent moduli. An important advantage of the
algorithm summarized in Boxes 1-4 is that it can be exactly linearized in closed f orm to obtain
these consistent tangent moduli. Use of convected coordinates allows the tangent moduli to
take a remarkably simple form. In fact, in matrix notation, the form is identical to that in the
kinematically linear theory. Restricting our attention to isotropic hardening, the linearization
of the return mapping algorithm leads, in matrix notation, to the following algorithmic
elastoplastic tangent moduli :
- g , , + I N / L , , + i N ~,,,, + I '
d e I n + l = E , . , , + t , s s . 0 , , . o, s J . ~ ,
(5.28)
where
m ] - 1
2; " E . , " = O - ' + " / , ' ~ + , O;p6,~ ,
t a+l o~=1
]-'
[ 2
E ~ r , , + l - ' - C , ~ + l I " + " ' ) / , i + I O q o r O ' ~ O t , , + l '
o t = l
. t , 9p4 , , + , ] - I
g, , +, = --[(0o. 0/ 3. , , +1 ) E,,.,,+,(0o.6o,.,,+l ) + Op6~.,,+~Ep,,+, .,,
(5.29)
N,,.,,+ I := _ E,,,,+t 0,,~b,~.,,+l
We remark that the structure of (5.28) is entirely analogous to the expression for the
continuum elastoplastic tangent moduli. To obtain the continuum tangent moduli all that is
needed is to replace the algorithmic moduli E,,,,,, in the expression for the algorithmic
elastoplastic moduli by the elastic moduli C,,+t. A derivation of (5.28) as well as a similar
result for the case of combined isotropic/kinematic hardening are provided in Appendi x A.
With do.,,+t/de,,+~ specified here, the linearization of the discrete weak form (2.37) is fully
defined. This consistent linearization of the weak form leads not only to quadratic rates of
asymptotic convergence in a global Newton iteration scheme, but also to robust continuation
methods for post-buckling analysis such as those discussed in Part III.
6. Numerical examples
Four numerical examples are considered to illustrate both the physical behavior of the shell
under the proposed generalization to the Ilyushin-Shapiro yield criterion as well as the
performance of the corresponding return mapping algorithm. The objective of these simula-
tions is to demonstrate the reliable performance of the new plasticity model and its
implementation in practical calculations. All calculations are performed on a Convex C1
comput er by implementing the algorithm in Boxes 1-4 in an enhanced version of the
nonlinear finite element computer program FEAP, developed by R. L. Taylor, and described
in Chapt er 24 of [22]. A global Newton solution procedure enhanced with a line search
algorithm is used throughout. The finite element spatial discretization consists of 4-node
isoparametric quadrilateral elements with bilinear displacement interpolation and 2 x 2 Gaus-
sian quadrat ure. Attention is directed toward the excellent convergence characteristics of the
J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model 159
global Newton procedure due to use of the consistent tangent operator developed in Section
5.3.
6.1. Built-in, perfectly plastic beam
A perfectly plastic beam built in on each end with a concentrated load at the 3 point along
its span is considered first. The geometry, material properties and finite element mesh,
consisting of 60 elements with added refinement neighboring the locations of the onset of
plastic hinges, are shown in Fig. 4. The objective here is to compare the global response
predicted by the shell model with the Bernoul l i -Eul er (bending dominated) elementary
analytical solution (see e.g., [23]) in predicting the onset of plastic hinges. Consequently, only
the shell model with linear kinematics is considered (see Part II). Furthermore, the rigidities
associated with membrane (Eh), bending (Eh3/12) and transverse shear (GhK) response are
taken as independent parameters in order to simulate the elementary bending dominated
solution. Similarly, the yield parameters n~D, m0 and qo are also treated independently.
The transverse load P applied to the beam is plotted against the transverse displacement
under the load in Fig. 5. The break points in the plot associated with the elementary solution
correspond to the formation of plastic hinges, first at the boundary closest to the load
application, and next at the point of the load application. A third hinge forming at the
boundary furthest from the load corresponds to the collapse load of the beam. The effect of
the shell model, as is evident in Fig. 5, is to smooth or blunt these break points upon
formation of plastic hinges. The mechanism for this smoothing process is depicted in Fig. 6.
Elastic loading occurs along the m l~ axis until the first yielding occurs at m~ = m 0. Next, in
contrast to the elementary solution which relies on a one-dimensional yield criterion and stress
state, further increase of the load P allows additional increases in m~ (pseudo-hardening) as
well as nonzero values of m22 due to the orthogonal projection of the trial state onto the yield
surface. Consequently, m~ and m22 are allowed to increase progressively through points B
and D. Hence, locally in the beam, the multi-dimensional nature of the shell yield criterion
acts as a pseudo-hardening mechanism from point A to D in Fig. 6. An analogous process
occurs at subsequent hinge points. Furthermore, the shell yield criterion serves to distribute
the plastic zone neighboring each hinge through a finite longitudinal region, in contrast to the
longitudinally point-wise hinges predicted in the elementary solution. In spite of these
differences, there is good overall qualitative agreement between the two solutions.
h = 1.0
v = 0 . 0
E h = 0.25 x I0 v
G h ~ = 0.II x I011
E h s
= 0.13 x l0 T
n o = 0.41 x l0 s
q~ = 0.24 x 10 s
m o = 0.253 x 102
~' = o.o p
H ' = 0.0 ,
Built-in 2 2 . ~ /
E n d
i IllllllLIllllltlllll[il1111111111111111111 II IIIL IIIIIIll
Built-in
f 10 ~ Y" End
Fig. 4. Geometry, finite element mesh and material properties for a built-in, perfectly plastic beam.
160 J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model
35 , , , , , , '
j , , J
25 j "
j *
j . ~ "
2O /
?
15
10
5 Elementary I-D Analytical Solution ,
Shell Model Finite Element Solution . . . . . .
i , i , i , t , , ' '
E-5 Deflection
Fi g. 5. L o a d - d i s p l a c e me n t c ur ve s f or a bui l ~-i n b e a m: e l e me n t a r y mo me n t d o mi n a t e d a na l yt i c a l s o l u t i o n , a n d
b e n d i n g d o mi n a t e d shel l mo d e l s o l u t i o n .
l
. . . . . .
ml t
6.2. Point-loaded, simply supported, perfectly plastic plate
A perfectly plastic, simply supported plate with a point load at its center is considered next.
Due to symmetry, only ~ of the plate will be considered. The -~ domain geometry, material
properties and finite element mesh, consisting of 1200 elements with added refinement
neighboring the locations of the onset of diagonal plastic hinges, are shown in Fig. 7. The
objective here is to compare the shell response to the predictions of the elementary bending
dominated limit load analysis (see e.g. [23]) within the context of the Kirchhoff-Love
kinematic hypothesis. Consequently, as with the beam in Section 6.1, only linear kinematics is
considered and the rigidities and yield parameters associated with membrane, bending and
transverse shear response are treated as independent parameters. The transverse load P
applied to the plate is plotted against the transverse displacement under the load in Fig. 8.
J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model 161
Si mp l y - Su p p o r t e d
- I
h = 0 . 5
v = 0 . 2
Eh = 3. 5 106
Gh,c = 5. 0 x 10 T
Eh 3
- ~ = 1.2 102
no = 1.22 x 104
qo = 7.0 x 1 0 a
mo = 1.519 x 10 I
,c' = 0. 0
H' = 0. 0
Fig. 7. Geometry, finite element mesh and material properties for a point-loaded, simply supported perfectly
plastic plate.
No t e that the l oad val ues shown in this figure are .~ of the load applied to the correspondi ng
full plate due to the enf orcement of the symmetry condi ti ons. A s shown in the figure, the
elastic sol uti on matches the el ement ary Ki rchhof f - L ove plate theory sol uti on identically ( see
[ 24, p. 143] ). The plastic sol uti on al so shows good qualitative agreement with the upper b ound
on the limit l oad ob tai ned using classical el ementary met hods ( see e. g. [ 23] ) . 6 A gai n, as in the
b eam exampl e ab ove, the physical influence of the shell yield criterion is to provi de a s moot h
transition b et ween the elasticity domi nat ed and plasticity domi nated sol uti ons.
35 , , , , '
3 0
l
I
I
I
I
!
!
10
Elastic Anal yt i cal Solution . . . . .
El ement ar y Analytical Li mi t Load ( Upper Bound) . . . . .
Moment Domi nat ed Shell FEM Solution
5
o ' 10 ' 2'0 ' ' 20 ' 5'0
De f l e c t i o n
|
~0
Fig. 8. Load-displacement curves for a simply supported plate: elementary Kirchhoff-Love elastic solution
appended to elementary upperbound plastic collapse solution, and bending dominated shell model solution.
"I t is useful to note that better bounds (upper and lower) are available for the uniformly loaded plate problem in
[25]. Thi s was kindly brought to our attention by one of the reviewers of this manuscript.
162 J. C. Simo, ]. G. Kennedy, On a stress resultant geometrically exact shell model
t x
( a ) ( b ) ( c )
D ELASTIC
PLASTIC BOUNDARY
PLASTIC
( d ) ( e )
Fig. 9. Evol ut i on of plastic zone in a si mpl y-support ed pl at e: (a) P = 11.0; (b) P = 18.5; (c) P = 25.6; (d) P = 26. 3;
(e) P = 27.0.
The evolution of the plastic zone in the plate is depicted in Fig. 9(a-e) for increasing loads.
Note that, although the elementary limit load analysis assumes that the plastic hinges occur
along infinitely thin lines, the shell model predicts finite width plastic zones neighboring each
hinge, a direct analog to the finite zones in the beam. These plastic zones have considerable
width in comparison to the plate thickness, even for a highly refined in-plane mesh, as is
shown in Fig. 9(d-e). Nevertheless, there again is good qualitative agreement between the two
solutions.
6.3. Pinched cylinder with isotropic hardening
As a third example, a short cylinder bounded by two rigid diaphragms at its ends, loaded
with two radial pinching displacements at the middle section, and characterized by an isotropic
hardening plastic response is considered. Due to symmetry, only one octant of the cylinder is
modeled. The geometry, material properties and finite element mesh of the octant, consisting
of 1024 equally spaced elements, are shown in Fig. I0. Full finite deformation kinematics are
considered here.
The pinching loading P is plotted against the radial displacement under the load in Fig. 11.
The step-like regions are due to snap-through-like mechanisms which arise as a result of the
relatively coarse mesh in comparison to the width of the indentation ridge forming about the
J. C. Si mo, J. G. Kennedy, On a stress resultant geometrically exact shell model 163
Yl h o 3 0 I
J 300
Fig. 10 Geometry, finite element mesh and material properties for pinched cylinder.
E3
, , , , ,
~ , i , i , , , I i
0 0 5 0 100 150 21111 2511
Deflection
Fig. 11. Load-displacement curve for a pinched cylinder.
point load upon loading (see Fig. 12(c-e)). That is, such regions are due to the nature in which
the ridge, which has a width equal to or below the element width, passes through the elements
as the ridge moves outward. Note that the pinching displacement in the final configuration is
nearly the cylinder radius. Progressive states of the deformed configuration as well as the
evolution of the plastic zone in the cylinder are shown in Fig. 12.
6.4. Pinched hemisphere with isotropic hardening
A hemisphere bounded by a free edge, loaded by two inward and two outward forces 90
apart, and characterized by an isotropic hardening plastic response is considered last. Due to
symmetry, only one quadrant of the hemisphere is modeled. The geometry, material
properties and finite element mesh of the quadrant, consisting of 768 elements, are shown in
164 J.C. Simo, J.G. Kennedy, Opt a stress resultant geometrically exact shell model
( a )
, r
( b ) ( c )
ELASTIC
PLASTIC BOUNDARY
I PLASTIC
( d ) ( e )
F i g . 12. E v o l u t i o n o f p l a s t i c z o n e i n a p i n c h e d c y l i n d e r : (a) t~ = 1 5 0 ; ( b ) ~ = 2 0 7 ; (c) 6 = 2 4 7 ; ( d ) 6 - 2 6 8 ; ( e )
t~ = 280,
y
/ ~ s , j m m e t ~
Z Ra d i u s = I 0
h =O5
t, ~ ( } 2
E : 1. 0 ~ HI t
Fi t - (I 5 ~ I O I
G/, , ~ : 3 1 ~ II1"
Eh I
~, , ~ 2 0 x I o - I
~' = 9 0
H' = 0 . 0
F= 1,0
Fig. 13, Geometry, finite element mesh and material properties for a pinched sphere.
J. C. Si mo, J. G. Kennedy, On a stress resultant geometrically exact shell model 165
3 5 , , , , , , , , ,
E- 3
2 0
I
i
I
i
i
i
i
i
i
i
/
S
/
.,:/
I i I , I , I
l 2 3 4 5 6
I nwar d l oad
Out wa r d l oad . . . . . . .
, I i I , I ,
0 0 7 8 9 l O
De f l e c t i o n
Fig. 14. Load- di spl acement curves for pinched sphere: inward load and outward load.
( a ) ( b ) ( c )
ELASTIC
PLASTIC BOUNDARY
PLASTIC
( a ) ( e )
Fi g. 15. E v o l u t i o n o f pl ast i c zone f o r a pi nc hed sphere: (a) P = 0. 010; ( b) P = 0. 016: (c) P = 0. 019; ( d) P = 0. 023;
(e) P = 0.029.
166 J.C. Simo, J. G. Kennedy, On a stress resul tant geometri cal l y exact shel l model
Fig. 13. A s with the cylinder in Secti on 6 . 3 , full finite deformati on ki nemati cs are represent ed
here.
The appl i ed loads P are pl ot t ed against the radial di spl acement under each respect i ve l oad
in Fig. 14. Not e that the inward di spl acement in the final configuration is nearly equal to the
hemi sphere radius. Progressive states of the def ormed configuratio~ l as well as the evol ut i on of
the plastic zone in the hemi sphere are shown in Fig. 15.
6.5. " Convergence rates for global N ewton i terati ons
Us e of the exact, cl osed- form consi stent tangent operator de ve l ope d in Secti on 5. 3 l eads to
quadratic rates of asymptoti c convergence in a glob al Ne wt on iterative procedure. Thes e
excel l ent convergence properti es are present even for reasonab l y large prob l ems such as t hose
consi dered in Secti ons 6 . 2, 6. 3 and 6. 4. A s an illustration, val ues of the E ucl i dean norm of the
gl eb al residual are reported in Tab les 2 and 3 within typical plastic l oad steps for t he exampl es
in Sect i ons 6. 3 and 6. 4. These results clearly exhib it the quadratic rates of asymptoti c
convergence.
Tab l e 2
Pi nched cylinder: residual norms for glob al Ne wt on iteration
L oad step
Iteration P = 2(10" P = 690 P = 2491 P = 8753
I 0. 254E + {16 0. 516E + 04 0 . 3 7 5 E + 04
2 I). 105E + (15 11.919E + 02 ( I . 211E + 03
3 0. 127E + 04 0. 202E + 00 0 . 2 5 0 E + 01
4 0. 445E + (13 0. 526E - 04 0 . 2 6 3 E + (10
5 0. 962E + 112 0. 191E - 07 (1. 121E - (14
6 0. 143E + (12 - 0 . 1 9 0 E - 117
7 0. 169E + (}(I - -
8 0 . 6 0 7 E - ( 1 4 - m
9 0 . 2 1 4 E - - 0 7 -- --
0 . 3 5 0 E + 04
0 . 9 6 4 E + 02
0 . 7 9 2 E + (10
0 . 1 7 2 E - 04
0 . 1 8 2 E - (17
m
" Elastic step (larger ti me st ep) .
Tab l e 3
Pi nched hemi sphere: residual norms t br gl oba l Ne wt on iteration
L oad step
Iteration P = 0. 006" P = 0. 015 P = 0. 021 P = 0. 029
1 0. 212E - 02 0. 212E - 02 0 . 2 1 2 E - 02 0 . 2 1 2 E - 02
2 0. 419E - 01 0. 314E - 01 0 . 3 2 4 E - 01 0 . 3 1 4 E - 01
3 0. 365E - 03 0. 179E - 02 0 . 4 3 7 E - 02 0 . 1 0 9 E - 01
4 0. 115E - 04 0. 248E - 04 0 . 3 5 4 E - 03 0 . 1 0 2 E - 02
5 0. 213E - 08 0. 595E - 09 0 . 2 8 9 E - 04 0 . 4 4 2 E - 04
6
- 0 . 1 2 4 E - 08 0 . 2 1 4 E - 05
7
- - 0 . 2 4 0 E- l 0
~' Elastic step (larger ti me st ep) .
J. C. Si mo, J. G. Kenned),, On a stress resultant geometrically exact shell model 167
7 . Concluding remarks
Within, the context of the geometrically exact shell model discussed in Part III of this work,
the formulation and numerical implementation of a general constitutive theory for elastoplas-
ticity formulated entirely in terms of stress resultants and stress couples has been presented.
As an application, an extension of the classical llyushin-Shapiro yield criterion to include both
kinematic and isotropic hardening has been considered. The return mapping algorithm
circumvents the use of objective integrators to define the trial stress, is unconditionally stable,
involves only the solution of two nonlinear scalar equations at the Stress-point level, and is
amenable to exact linearization.
Future work will extend the present formulation in two directions. From a physical
modeling perspective, the theory and finite element method is generalized to include transient
temporal response. From a numerical analysis perspective, an assumed stress me t hod is
developed for the elastoplastic problem which possesses identical accuracy for coarse meshes
in elastic problems as the assumed stress method considered in Part III for the membrane and
bending fields. This new assumed stress method has the advantage of requiring no modi f i ca-
tion of the return mapping algorithm (at the stress-point level) discussed within the context of
a displacement formulation in Section 5.
A cknowledgment
We arc indebted to M.S. Rifai for his involvement in the numerical implementation of the
formulation described in this paper. Support for this research was provided by AFOSR Grants
AFOSR-86-0292, AFOSR-28169-A and LLNL Grant LLNL-2254903 with Stanford Univer-
sity. J.G. Kennedy was supported by a Fellowship from the Shell Development Company.
This support is gratefully acknowledged.
A ppendix A . L inearization: algorithmic tangent moduli
The exact linearization of the return mapping algorithm summarized in Boxes 1-4 is
sketched below. For simplicity, the development will consider only isotropic hardening,
although the end result for combined isotropic/kinematic hardening will also be reported.
Differentiation of the elastic stress-strain relations (4.12)~ and the discrete flow rule (5.1)2
yields (noting that d,,p~b~(tr, p ) = 0 for strictly isotropic hardening)
d ~ , , + , = C,,+l(de,,+ I - d e ~ + l ) ,
. 2 d~r,, + dT,~+, # .(~r. p), , +, ].
de,P,+, = [~/,,+l 0~b.(~r, p),,+, +,
a =l
(A.1)
By combining these two equations one obtains the relation
[ ] de , , +, - dT,:+, O,,~b.(~r, p),,+, , (A.2)
do',,+l = E~,,+~ . : l
where E~,,+, are algorithmic moduli now given by the expression
[ ] ,
E, , 2 ( A . 3 ) = + .
t i n + I
+i a =l
168 J. C. Si mo, J. G. Kennedy, On a stress resultant geometrically exact shell model
Similarly, differentiation of the discrete hardening law (5.1)3 yields
ttl
dp, , +, =- Ep, t + ' Z dy,~+, O p d p , , . , , + , ,
a = l
r
m -] - 1
l - E
" = D 1 .4_ ' Yn + ! O p p ( ~ o t
J E p n + , v t = l
L
(A. 4)
Next, the coefficients dy,'~+ ~ are determined from the algorithmic version of the consistency
condition obtained by differentiation 4,~(o', p ) , , + ~ =0; i.e.,
(0, , 6, , ) t dtr,,+l + ( 9 p 6 , ~ dp,,+~ = 0 , a ~ 3 a c t
(A. 5)
Substitution of (A.2) and (A. 4) into (A.5) then yields
_ , d e , , + l ]
d , / ~+, - Z [g.~.:.ll(o.4~o...+.)E~..+. ,
a E, ~a c t
(A. 6)
where g , ~ "= [g~.,,+1] -t, and g~.,,+l is defined by
0 p, b , , , , , + , ] - I
a ~ ' 0 . . 6 . ~ . , , + I + O p ~ b ~ . , , + l E p , , + , g , , + , = [ 0 ~ , ~ . , , + , ) E ~ , , + ,
(A. 7)
Finally, substitution of (A. 6) into (A.2) gives the desired expression for the algorithmic
elastoplastic tangent moduli
I - X X de ,,+1 = E,,,,+~ tJes,,~, aes,,~.,
_ . E c r n N o t J l + ' + , O 0" ~[~ tlt , l l q" l
g,,+l t~.,,+ . I ,
(A. 8)
A similar calculation for the case of combined isotropic/kinematic hardening leads to the
following expression for the algorithmic elastoplastic tangent moduli:
dcr I _ ~ ~ ,tJ,, ,
= o . + I N~ . , + I N. , + l
de ,,+l E,,,,, ~ ~ e . D , , ~ , ,,E~,,~,
, , + , = [ ( 0 , . 6 . ) ' . ( a , . 6 . ) ] E , , + , a , . 6 ~ . + o p 6 z E , ' , , + , O p 6 . , , + , . ( A . 9 )
.,, " - + ~ :, ~ , , +, ] 0 . 6 ~ . , , + , . N, , +, - [ [ : , , , , + ,
is defined in (A. 4), E,, t and E,,,,+, are given by In (A. 9), E l , , , + , +
Et l + l " ~
??1 i t ! " 1 _
[
" 2 o ' - - y "
~ / , , + t , . , . e ~ . , , + t I ~ - ~ + 3 ' , , + t a ; . , ~ b . . , , +
c t = [ o r = !
- E , , ~ , , + l E ~ , , + ! '
(A. 10)
and E,,,, ~ and E,,p,,+. + are defined by
J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model 169
En-+l' = " E . , ~ . + , E~ "+~ ' " ( A . 1 1 )
In (A. 11), E,,+~ may be inverted in closed form, analogously to the inversion in (5.15) and
(5.16).
A p p e n d i x B. S u m m a r y o f t h e r m o d y n a m i c r e s u l t s i n e l a s t o p l a s t i c i t y
Here we provi de a brief t hermodynami c motivation for results cited in Section 4 regarding
t he elastoplastic constitutive equat i ons. Here, plastic processes are assumed to be fully
charact eri zed in t erms of the history of three state variables: the strain measure component s
e E [R 8, the plastic strain measure component s e p E R, and a statable set of mternal variables
a E R p oft en referred to as hardening variables. Accordingly, plastic flow at each point x ~
up to the current time t E R + is charact eri zed in t erms of the histories ~- E ( - ~ , t] ~ { e(x, r),
eP(x, 1"), a( x, ~')}. The stress measure component s ~r E R 8 are then dependent functions of t he
variables { e, e p} t hrough elastic stress-strain relations, as discussed below.
As discussed in Section 3, the strain measure component s e E R 8 are assumed to be
additively decomposed into an elastic and a plastic part , denot ed by e e ~ IR ~ and ePE ~8,
respectively; i. e. ,
e = e c + e p . ( B . 1 )
The Helmholtz free energy ~" RSx R~x ~P""> [1~ is assumed to take the form
I~/ (E, E p GI~): -- W( E - - e p ) -~- ~(~11~) , (B. 2)
where W" R 8---> R is the elastic stored energy function, and ~ " P--> is the hardening energy
function. The dissipation function @ p is defined, in component form, according to
(00 '
= ( v w ( e - -
( B . 3 )
The hyperelastic constitutive equations and the restriction placed on the dissipation function
are obt ai ned by exploiting the second law of t hermodynami cs in the form of the Cl ausi us-
Duhem i nequal i t y and, following st andard arguments (see [26, 27]), t ake the form
o ' =VW( E- eP) and @P >10, (B. 4)
where ~r are t he Kirchhoff stress resultant component s (cf. (3.8)). Non-negat i vi t y of t he
dissipation fimction ensures that plastic flow is a dissipative process. Not i ce t hat the compo-
nent expression (B. 3)2, using t he not at i on of Section 3, may be expressed as
= j a . a, o + L o a , ( B . 5 )
where & is resol ved in the basis { a 1, a 2, a 3} with component s a.
Thus far t he constitutive equat i ons are present ed strain space; i.e. t he response functions
17{1 J. C. Si mo, J . G. Kennedy, On a stress resultant geometrically exact shell model
are expressed in terms of the state variables { e, e p, t~}. In classical plasticity t he response
functions, e.g. the yield condition and the flow rule, are formul at ed in stress space in t erms of
the variables ~ ' ~ ( - ~ , t]~-->{ o'(x, z), p(x, ~')}, where o' Ell~ s is a function of { e, e p} and
p E R p denot e a compl ement ary set of internal variables which arc functions of a E I~ p t hrough
the Legendre transformation
= - f a -
(B. 6)
where O' ~ P " ~ [ ~ is the compl ement ary potential associated with ~. One refers to { ~r, p} ,
which are constrained to lie in the closure of the elastic range IF,, defined in Section 4, as the
fluxes conj ugat e to the variables (affinities) { e - e p, a }. Typical examples for t he functions
~ ( a ) and O(p) which fit many classical plasticity models; e.g. J2-flow t heory, t ake t he form
~ ( a ) = at Da and O(p)= ~ptD-~p,
(B. 7)
where D is assumed constant. 7 By differentiation of the Legendre t ransformat i on (B. 6) we
obt ai n the relations
p = - V~ ( a ) and a =- VO( p ) . (B. 8)
Maki ng use of (B. 8), the dissipation function given in (B. 3) is expressed in stress space as
~ P[o', p; ~P, i~] "-- o' t ~ p "~" pttk. (B. 9)
It is shown in Section 5 that use of the Legendre t ransformat i on (B. 6) as a means of
expressing the flow rule and hardening law in stress space has i mport ant algorithmic
implications, i.e., it preserves symmetry of the algorithmic consistent elastoplastic t angent
moduli.
Ref erences
[1] G. Stanley, Continuum-based shell elements, PhD dissertation, Applied Mechanics Division, Stanford
University, 1985.
[2] J.C. Simo and T.J.R. Hughes, Elastoplasticity and Viscoplasticity: Computational Aspects (Springer, Berlin,
1989).
[3] J.C. Simo and R.L. Taylor, Consistent tangent operators for rate-independent elastoplasticity, Comput.
Methods Appl. Mech. Engrg. 48 (1985) 101-118.
[4] M.A. Crisfield, On an approximate yield criterion for thin steel shells, Department of the Environment,
TRRL Report 658, Crowthorne, Berkshire (Transport and Road Research Laboratory), 1974.
[5] M.A. Crisfield, Ivanov's yield criterion for thin plates and shells using finite elements, Department of the
Environment, TRRL Report 919, Crowthorne, Berkshire (Transport and Road Research Laboratory), 1979.
[6] A. A. llyushin, Plasticity (in Russian) (Gostekhizdat, Moscow, 1948).
[7] G.S. Shapiro, On yield surfaces for ideally plastic shells, in: Problems of Continuum Mechanics (SIAM,
Philadelphia, PA, 1961) 414.
[8] G.V. lvanov, lnzh. Zh. Mekh. Tverdogo Tela 74 (6) (1967).
[9] J.C. Simo, J.G. Kennedy and S. Govindjee, Non-smooth multisurface plasticity and viscoplasticity. Loading/
unloading conditions and numerical algorithms, Internat. J. Numer. Methods Engrg. 26 (1988).
7 For example, in classical J2-flow theory with linear isotropic hardening, ~( ot ) = ~aK' a and O(p) = p ( 1 / K' ) p ,
where a E It~ is the equivalent plastic strain and p E is the associated conjugate variable.
J.C. Simo, J.G. Kennedy, On a stress resultant geometrically exact shell model 171
[lo]
[ l l l
[12]
[131
[141
[151
[16]
I171
I181
[191
12Ol
[211
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31l
J.C. Simo, J.G. Kennedy and R.L. Taylor, Complementary mixed finite element formulations for elastoplas-
ticity, Comput. Methods Appl. Mech. Engrg. 74 (1989) 177-206.
J.C. Simo and D.D. Fox, On a stress resultant geometricaliy exact shell model. Part I: Formulation and
optimal parametrization, Comput. Methods Appl. Mech. Engrg. 72 (loS9) 267-304.
J.C. Simo, D.D. Fox and M.S. Rifai, On a stress resultant geometricail/exact shell model. Part II: The linear
theory; Computational aspects, Comput. Methods Appl. Mech. Engrg. 73 (1989) 53-92.
J.C. Simo, D.D. Fox and M.S. Rifai, On a stress resultant geometrically exact shell model. Part Ill:
Computational aspects of the nonlinear theory, Comput. Methods Appl. Mech. Engrg. 79 (1990) 21-70.
M. Robinson, A comparison of yield surfaces for thin shells, Internat. J. Mech. Sci. 13 (1971) 345-354.
W.T. Koiter, Progr. Solid Mech. 6 (1960).
G. Maier, A matrix structural theory of piecewise linear elastoplasticity with interacting yield planes,
Meccanica (1970) 54-66.
G. Maier and D. Grierson, Engineering Plasticity by Mathematical Programming (Pergamon, New York,
1979).
P. Perzyna, Thermodynamic theory of viscoplasticity, in: Advances in Applied Mechanics, Vol. 11 (Academic
Press, New York, 1971).
L.M. Kachanov, Fundamentals of the Theory of Plasticity (Mir, Moscow, 1974).
L. Malvern, An Introduction to the Mechanics of a Continuous Media (Prentice Hall, Englewood Cliffs, N J,
1969).
T.J.R. Hughes and J. Winget, Finite rotatioa effects in numerical integration of rate constitutive equations
arising in large-deformation analysis, Internat. J. Numer. Methods Engrg. 15 (1980) 1413-1418.
O.C. Zienkiewicz, The Finite Element Method, 3rd Edition (McGraw-Hill, London, 1977).
W. Johnson and P.B. Mellor, Engineering Plasticity (Wiley, New York, 1983).
S. Timoshenko and W. Woinowsky-Kreiger, Theory of Plates and Shells, 2nd Edition (McGraw-Hill, New
York, 1959).
P.G. Hodge and T. Belytschko, Numerical methods for the limit analysis of plates, J. Appl. Mech. 15 (1968)
796-801.
B.D. Coleman and W. Noll, The thermodynamics of elastic materials with heat condition and viscosity, Arch.
Rational Mech. Anal. 13 (1963) 167-178.
B.D. Coleman and M.E. Gurtin, Thermodynamics with internal variables, J. Chem. Phys. 47 (1967) 597-613.
J.E. Dennis and R.B. Schnabel, Numerical Methods for Unconstrained Optimization (Prentice Hail,
Englewood Cliffs, N J, 1983).
D.G. Luenberger, Linear and Nonlinear Programming (Addison-Wesley, Menlo Park, CA, 1984).
P.M. Naghdi, Stress-strain relations in plasticity and thermoplasticity, in: Proc. 2nd Syrup. on Naval Structural
Mechanics (Pergamon, London, 1960).
J.C. Simo and R.L. Taylor, A return mapping algorithm for plane stress elastoplasticity, Internat. J. Numer.
Methods Engrg. 22 (1986) 649-670.

Вам также может понравиться