Sei sulla pagina 1di 9

2686

Decemb e r, 1969

E xp e rim e nt s ," Bul/ e tin 96, Iowa Engine e r i n g E xperiment S t a tion, February 19 , 1930 .

10. Mct ca lf , L. , a nd E ddy, H. P., Americ an S e we r age Pr acti c e , McGraw Hill C o mp a n y , New York

1

9 16, p p . 4 8 1 - 53 4 .

11 . R owe , R . R . , "Co n c re te

Pipe in Highw a y Co n st ru c tio n , "

T e c hnical Mem o ran d u m , A merican

C on c r e t e Pip e A s s o c i a tion , Januar y

1

2 . S p ang ler,

M . G . , " Th e Supp o rting

I , 19 62 .

Str e n g th o f R ig id Pip e C ulverts , "

Bul/et i n 1 1 2 . l o w a Engi.

n

ee rin g E xp e rim e n t S t a ti o n , Febru a ry 8, 1 933 .

13. Sp a n g l e r,

M . G ., " An a lysis

of Loads a nd S upp o rtin g

Strengths,

~

and Prin c ipies o f De s ign for , ~

High w a y C ulv e r t s, " Pr o c e edings, Highwa y R es e a r c h B o a rd , 1946, pp. 189 - 214 . , :

14

. Timrn e r s ,

J . H . , " L o a d

Study of Fl ex ible Pipe s Under High Fill , " Bul/e tin

Re s e a r c h B oa r d , 1 956 , pp . 1 - 11.

125 . Highw a y

"i':'

l

: ; ~

·1:'

. , ~ I

. ~ :

'i'

6943

Dec e mber, 1969

J ournal o f th e

STRUCTURAL

DIVISION

ST 12

Proceeding s o f the Americ a n So c iety of Civil Engin e er s

THEORY OF NONLINEARELASTIC STRUCTURES

By Graham H. Powell,l M. ASCE

INTRODUCTION

General theoretical formulations and computational techniques for the analysis of linear elastic structures have been very thoroughly studied in recent years. The theory can be elegantly and concisely expressed in standard matrix form. Efficient solution procedures for both the force and displace- m e nt meth o ds are well established. No such standard í zat í on exists, however, for the analysis of nonl í near elastic structures, as evidenced by recent papers (1,2,3,4,5 , 6).2 Indeed, it is still not clear what constitutes a consistent theo- retical formulation of the problem. In this paper , an attempt is made to separate, physically, the different types of nonlinearity which are associated with large displacements of elastic structures, to present a theoretical formulation in which these types of non- linearity can be easily identified, and to examine a number of problems as- sociated with solution techniques. The displac e ment method of analysis applied to discrete element systems is used throughout, and it í s assumed that stra í ns are small although displacements may be large. A member of a plane rigid frame í s selected to illustrate the theory. This member is selected because the results can be expressed compactly, and y et the features of more complex structural elem e nts are present. In particular, both flexural effects and axial straín effects are present, and therefore, the principIes developed can be extended to finite elements for plate bending, as well as for plane and three- dimensional stress. As the theory is developed, two alternative formulations are presented in

No te.- D i sc u ss l o n w ritt e n r e q u es t

o f th e c opy right ed

Society of C ivil E n gi n ee rs,

rnítted f o r r e vi e w f o r p o ssible publicati o n o n F e bru a r y 1 9, 1969.

o pen until m u s t be fil e d with th e Ex ec utiv e

J ou r n al

of the S tr uc t u r a l

M ay 1, 19 7 0 . To e x t e nd the closing da t e one m o nth, a

S ec re t a r y ,

ASC E. Thi s pa per is p ar t

Divis i o n, Pr oc eedings of the A me ri c a n

1969. Manus cr i p t w as s u b -

Vol. 95, N o . S T1 2 , De c em b e r,

1 A ssist .

2 N u m er a i s in p a ren thes es refer to co r respo n d in g i t em s in th e A ppen d ix.- R e f ere n c es .

P r o L C iv. E ngrg.,

Univ. of Ca li f . , Be rk e l ey ,

C a lif.

2687

December, 1969

'6, Iow a En gi neering Exp e riment Station , Febru a r y 19 , 1930 .

1. P . , Am e ri c an S ewe ra ge Pra c t i ce, McGra w HiII Co mpan y , N ew

ST 12

Y o r k ,

: P i pe in Hi g hw ay Con s tru c ti o n ,"

m , Jan ua r y I , 1 96 2.

Supp o rt i ng Strength on , Febru a r y 8, 1 93 3 ,

y si s o f L oad s a nd Supp o rtin g

T ec hni c al Mem o ra n du m , A m e ri ca n

of Ri g i d P i pe Cul v e rts , " Bull e t i n 11 2 , l o w a E n g i -

Str e n g ths ,

a nd Pr i n c ipi es o f D e s i g n f o r

' ee din gs , H ig h way Re s e a r c h Bo a rd, 1946 , pp . 1 8 9 - 2 14 .

J,I Stud - II. y o f Fl exible Pip es U nder High Fill " "

B

ull e tin

12 5

H ' h

'

Ig w a y

6943

Dec e mber, 1969

ST 12

J o urnal of th e STR U CT U RAL DIVISIO Proceedings of th e Am e rican So c iet y of Civil Engineers

THEORY OF NO N LINEAR ELASTIC STRUCTURES

B y Graham H. Powe ll ;' M. ASCE

INTRODUCTION

General theoretical formulations

and computational

techniques for the

analysis of linear elastic structures have been very thoroughly studied ín recent years. The theory can be elegantly and concisely expressed in standard matrix formo Efficient solution pr o cedures for both th e force and displace- ment methods are well established. No su c h standardization ex ists, how e ver,

for the analys í s of nonlinear elastic structures, as evidenced b y recent papers

( 1 , 2 , 3 , 4,5,6 ) .2

retical formulation of the problem . In this pap e r , an attempt is made to separate, ph y sicall y , th e different

types of nonlinearity which are associated with large displ ac ements

structures, to present a theoretical formul a tion in which these types of non-

linearity can be easily identified, and to examine a number of problems as-

sociated wíth solution techniques. The displacement

to discrete

are small although displacements may be larg e . A member of a plane rigid

fram e ís seI e cted to illustrat e th e theor y . This member is sel ected be c ause the results can be ex pressed c o mpactly, and y et the featur e s of m o re complex

Indeed,

í t is still n o t clear what constitut e s

a c on s istent theo-

of elastic

method of analysis applied

element systems ís used throughout, and it is assumed that straíns

structural eIements

str ain effects are pr e sent, and therefor e , the p rincipI e s d e v e Ioped c an be ext e nded to finite e lements for plat e bending, a s well as for plane and three- dimensional stress.

formulations are present e d in

are present. In particular, both flexur a l eff e cts and a xi al

As the theor y is developed, two alternative

I \ ' o t e .-Dis c u ss i on

op en un t il

M ay 1 , 19 7 0. T o ext e nd t he cl osi n g da t e o ne m onth , a

w r itt e n request of th e copyrighte d J o urnal

Soc i e t y of Civil E ngin ee rs, Vol. 95 , No . ST12, Decembe r, 1969. M a n usc ript wa s sub - m itte d f o r revi ew f or pos sible pub li c a t i o n o n F ebr u a r y 1 9 , 196 9 ,

A SC E. Th is paper is p ar t

m u s t be filed with th e E xecuti ve Secreta ry,

of th e S t r uctural

Di v i sio n, Proceedin gs of the Am e r í ca n

1 A ss ist.

2 N ume rals in p a r e nth eses re f e r to c orres p o ndin g it ems in th e App e nd i x .- R e f ere n c es .

Pr of . C i v. En grg . ,

U ni v . o f Ca lif. ,

B e r k e i ey , Ca lif .

2687

r

2688

parallel,

one of a " traditional"

December, 1969

engineering type and one in general matrix

form. The development proceeds as follows:

1. Basic definitions and equations are presented.

2. Geometric

compatibility

relationships

and joínt displacements

are examined.

between member deformations

3. Equilibrium equations for members and joints are developed.

4. The essential equations of the problem are identified.

5. The role

of the tangent stiffness

in analysis is described,

and expres ,

síons for tangent stiff ne ss are deriv e d .

In this developm e nt,

in order

member deformations

a theory

and joint displacements

which is applicable when rigid body

in which at the end

has commonly been used, but it ís as the less dir ec t formulation.

are

separated, rotat í ons

member deformations

of th e paper.

ar g ued herein that ít is not as satisfactor y

to present

of the members

are large.

A more direct formulation,

is considered

are not expl ' í cttly examined,

This type of formulation

DEFINITIONS AND BASIC RELATIONSHIPS

S tr uctu re . - The structure

consists of a number of rigid body [oínts linked

study the members are

together by deformabl e structural

members.

In thís

SI(

F I G . 1.- ; \ 1EM BER A C TI O N S A N D DE FORM A TI O N S

initially straight, and their elastí c properties are assumed to be concentrated

along theír axes. External

loads are applied only at the joints.

Di f fe re ntial El eme nt . -Each

member

The deformations

differential elements.

is made up of an infinite number of

of all differential

elements

are as-

sumed to be are aSSumed of material

ment of a plane frame member

of beam , with stress and strain resultants

small, and the relationships between actions and deformations

to be linear. In general, differential

elements will be elements

wíth stress

and straín

related by Hooke's law. However, an ele-

is more simply taken as an ele mental length

related by

and

CJ 1 =EIE1

CJ 2 = E 2 E 2

•••• • ••••••• • •••••••••••••••••••

.

.

.

.

.

.

.

.

.

.

(1)

(2)

r:

'::~;;

ST 12

NONLINEAR ELASTIC STRUCTURES

2689

in which CJ 1 = bending moment; CJ 2 = axial force; E 1 = E I; E = Young's modu- lus; I = second moment of area of cross section; E 2 = E A; A = area of cross

section; E 1 = curvature of

Memb e r Actio n s and Deformatio ns .-ln order to develop a theory which is applicabl e when member rigid body rotations are lar g e, the deformations and

rigid body displacements

member supported so that its chord undergoes no ri g id body displacement.

The member g eneralized deformations ,

sponding actions, 51, S2 and S3' are selected for the d e rivation

mem b e r axis; and E 2

= axial strain.

of the members are separat e d.

Fig. 1 shows a frame

v 1 , v 2 and v 3 , as shown, and the corre-

of the theory.

The m e mber deformations

a re a ssumed

to be small.

That is, v and v a re

1

2

in com-

smalI ín compartson

with un í ty , and vi, v~, and v s / L are negligible

parison with unity. Memb e rs which may undergo lar g e flexural deformations must be subdivided into shorter members. Lar g e axial deformations are not

permitted. In order for the behavior

of th e complete

structur e

to be defined uniquely

e le-

ín terms of its joint displacements,

the deform a tions

of alI differ e ntial

FIG. 2 .-DI S PLACEMENTS ELE ME NT

OF DI FF ERENTIAL

-

RI,rl

YLx

F IG. 3 .-JOINT L O AD S AND DI S PLACEMENT S

ments in a member must be defin e d uniquely in terms of the member g ener-

alized deformations.

to vary cubically with the distance along the member chord, and lon g itudinal displacements, u , to vary linearly (Fig. 2). The followin g linear transformation is then consistent wíth the small deformations assumption:

Transvers e

displacements,

w , ar e therefore constrained

tl = [LU? - :P' .p)

L(P3 _ p2)

o

:J l::\

(3)

in which the deformed length, L, is not significantly different from the ori g inal length, Lo. The differ e ntial element deformations are given by

2

d

w E 1 = dx2

1

= L

(6p - 4) v 1

1

+ L (6p -

2) v 2

••

•• ••.•

(4)

" i ~

"t1

2690

December, 1969

ST 12

which is a linear function of the displacements, and

du

E 2 = d x +"2

1 (dW)2

dx

E3 + .! {(3p2 _ L

2

4p + l)v, + (3p' - 2p) v ,)' (

(5)

in which the quadratic term introduces the effects of large displacements.

More

generally, if {v} = < v 1 v 2 v s > T, then

Ei = < ai(p»{ v }

+ {{ v } T

<bi(P»T

< bi(P»

{v}

(6a)

in which the first and second terms on the right-hand side are, respectively,

the linear and quadratic

components of deformation, and in which for E 1

<

a 1 > =

<

1

L (6p - 4)

1

L (6p - 2)

O>

(6b)

and < b 1 > = and for E 2

<

a 2 >

<O

<O

O O>

O

1

L>

(6c)

(6d)

and < b 2 > = « 3p2

- 4p + 1) (3p2 - 2p) O>

(6 e )

. -The displacements of any joint consist of

translations, r 1 and r2' along the fixed coordinate directions, X and Y, and a rotation rs, as in Fig. 3. The joint loads are the corresponding forces Rl1 R and R3.

2

a vanishingly small change of geometry of

the structure about any deformed or undeformed configuration, a linear d í s- placement transformation can be written as

Joint Loads a n d Displac e ments

Geometric

Compatibility.-For

{dv} = [c] {dr}

in which {dv} = matrix of deformation increments

(7)

for a single member or any

g roup of members; {dr} = matrix of displacement increments at one or more joints; and [c] = displacement transformation matrix which is instantaneously

correct in the

particular configuration of the structure. For the single mem ber

in Fig. 4, the transformation is well known, as follows:

dv [_l

{dvl-i

dv

s

-c

c

L

1

c

L O

-5

O

s

L

5

L

C

-c

L

-c

T

5

;j

âr;

âr;

drs

dr

4

drs

âr,

!

• ••

(8)

in which 5 = sín ê ; and c = cos«. For large displacements, the transformation is not, of course, linear. For purposes of computation, the nonlinear transformation for a frame member

2691

ST 12

displacing from an initial state, O, to a new state, A, can be formulated as follows (Fig. 5):

NONLINEAR ELASTIC STRUCTURES

XA

A

Y

= Xo + r 4

= Yo + rs

-

r 1

- r 2

•••••••••••••••• • •••• • •••••••

(9 a )

(9b)

v

3

=

(x3t + y3t)1/2

- Lo

.

(9c)

e o = tan- 1 (~)

(9d)

More

OA = tan- 1 (~)

v

1

v

2

= 0 0 + r

3

= e o

+ r

6

-

-

 

(ge)

eA

•••

•.•

•.•

(9f)

e A

.•

••••••

••

•.••

(9 g )

g enerall y, as a structure

deforms from state O to state A, it moves

y

~/-

\.:

~rl

Cx

's t '{6

-r 4

FIG. 4.-ME MB ER ARBITRARIL Y ORIENTED IN SPACE

r 5

Yo

rl •. l.

~n

XA

~r4

_I

FIG. 5 .-INITIAL AND F INAL STAT E S OF MEMBER

through an infinite number of intermediate states, and the transformation can be written as

{ V A} = J

o

{ r A}

[c

r ]

{dr}

(10)

in which {VA} = matr í x of member deformations in state A; {r A} = matrix of joint displacements in state A; and [c r ] = instantaneously correct displace- ment transformation matrix at any intermediate state. Eqs. 9a-9 g are s í mple to use in computation for a frame member, and it should generally be possible to establish similar direct relationships for other

structural elements. However, it may be convenient to evaluate member de- íormations by numerical integration of Eq. 10 . For example, the form

 

N

{

V A} = ~

~

[c n ]

{rA }

n=l

.•

(11)

can be used, in which {r A} is divided into N equal parts and [cn ] is the instan-

 

.,

~

"

.

.• ;

.

,

" : ,

, 'tt t

~

2692

Decemb e r, 1969

ST 12

ST 12

NONL1NEAR ELAST1C STRUCTURES

2693

taneously correct displacement each parto

transformation

at the be g inning or middle of

+ L: t {dv } T

o

< bi>T <bi> { V A } e, <ae { V A} . Ldp

EQUILIBRIUM

+ L: "2

1 J1

o

{dV}T

<bi>T < b i > {V A} e, { v AF

< bi>T < b i> { V A} . Ldp

(19)

Joint Equil i brium.-All joínts and m e mbers must b e in equilibrium in the

deformed configuration of the structure.

veniently established by application of the Virtual Displacements

The equilibrium

equations are con-

PrincipIe.

For an infinitesimal imaginary displacement, {dr}, about the deformed

state, A, the member deformations,

{dV}, are given by

The imaginarydeformation,

{dv}, can be cancelled, and the integraIs of Eqs.

18 or 19 can be evaluated analytically for a frame member

to give the result:

2

1

O

{d v}

= [ C A]

{dr}

(12)

l::(. 'fI

1

O

2

O

O

A

21

in which [c A]

matrix

is the instantaneously

correct

displacement

transformation

1::\

in state A. The virtual work equation is then

{dj:} T {RA}

= {d v } T {SA}

(13)

in which {RA } isthe

matrix

of [oínt loads in state A; and { SA} isthe

matrix of

member

actions

in state

A.

Therefore,

Irom

Eq. 12, and because

{ dr }

is

arbitrary

+ EA I

{RA}

= [cAl T {SA }

 

(14)

1~0 ( 4 v i - v 1 V Z + V Z ) 2

280 -L

( 2

2)

V i + v 2

280 -L

( z

2)

v 1 + v 2

1~0 (vi - V 1 V 2 + 4 v ~)

4 V i

-

30

2

v

I

I v 1

4 V 23~ V 1 I \

V 2 ) (20)

The force transformation matrtx, [c A] T, is dependent on the displacements of the structure. Memb er Equilibrium . -The chain of equations linking joint displacements { r A} and loads {R A} in state A is completed byestablishing a re lationship be- tween member actions {SA} and deformations { V A}' For a single frame member, an infinitesimal imaginary displacement {dv} about its deformed position leads, by differentiation of Eqs. 4 and 5, to imag- inary differential element deformations dE 1 and dEz g iven by

- dE 1 = L (6p - 4)d v 1 + L (6p - 2)d v z

-

-

1

1

•.

(15)

4

V 1~

60

4v 1 - v 2

60

O

I

I v g

This equation can be written symbolically as

 

{sAl

= [[k 1 ]

+ [k z ( V A)]J {V A} = [ks]

{ V A }

(21)

Matrix

[k 1 ] is the ordinary linear stiffness

of the member.

Matrix [k 5] is the

secant stiffness of the memb e r, and is obviously deformation dependent.

If pseudostrains

Yi are defined by

and dEz = T

-

d v

+

{

(3p2 - 4p + l) v 1

+ (3pZ - 2p)v z

}

x {(3pZ - 4p + 1)d v 1 + (3pZ - 2P)dv 2 }

More generally, the following equation is obtained by differentiation

dEi = < ai>

{ãv} + { v }T

<lu» T < b i > {d v }

The virtual work equation ís

{dvF

{sAl

= L: t a , ai

o

lilp

= L: t a, s,

o

or, from substitution

of Eqs. 17 and 6 into Eq. 18

{dv}T

{SA} = L: t {dvF

o

< ai>T z,

< ai> {VA} . Ldp

E i Ldp

(16)

of Eq. 6:

(17)

(18)

Yi = < b i > {liA}

then Eq. 19 can be rearranged

,

to give

{SA} = L: t <ai>T z, < ai> Ldp {VA }

+ "2 L:

+ L: J

o

+ '2 L:

1

1

J1

o

1

o

<ai>T s, Yi < bi > Ldp { V A}

<bi>T s, Yi < ae lilp {VA}

f1

o

<bi>TYi

EiYi « b ,» lilP { V A}

 

. (22)

(23)

+ L: ~ t {dvF

o

<ai>T Ei {vAF

<be T

< b t > {VA} . Ldp

These integraIs can be evaluated exactly or approximately by numerical inte-

gration. The procedure

over surfaces and volumes as wellas in one dimension. A substantial simpli-

í s general because the integration

can be carried

out

. , . ' i

2694

December, 1969

ST 12

fication, possibly w í th no g reat loss in accuracy, can be achieved by assigning constant values to Yi'

ESSENTIAL EQUATIONS

The essential equations which must be satisfied can be summarized as follows:

{r A }

{VA} = J

o

{SAl

[a n ]{dr}

{VA}

= [ks]

{RA} = [aAF {SAl

.

.

(10)

(21)

(14)

Nonlinearity is present because the displacement transformation is nonlinear (Eq . 10), the members behave nonlinearly (Eq. 21), and the equilibrium equa- t í ons must be formulated in the d e formed confi g uration (Eq. 14). In computational algorithms, one or more of these nonlinearites may be ignored , and approxírnate results obta í ned. For example, for finite element analyses in which the elements are smal l , a common assumption mi g ht be that the member nonlinearity, represented by Eq. 21, can be i g nored. The influence of this particular assumption on the solution technique is noted subsequently.

TANGENT STIFFNESS

Ge ne ral.-In "exact" computational algorithms for large nonlinear struc- tures, the non-linear problem is most commonly solved by a Newton type of

{R}

TANGENT

AT O

O

FIRST DISPLACEMENT

ESTIMATE

SECOND DISPLACEMENT

ESTIMATE

EXACT SOLUTION

B

SECOND UNBALANCED

FORCE

FIRST UNBALANCED

FORCE

{r}

FIG. 6.- D I A G RA M M A TI C

IL LUS TRATI O N O F NEWTO N PROCEDURE

method in a series of linear steps. The procedure is well known andphysically reasonable, being one in which a solutíon which satisfies compatibility is sue- cessively corrected until it also satisfies equilibrium . The basie method is illustrated diagramatically in Fig. 6. For this method to be applied, tangent stiffnesses for the structure are required. In this sectíon , expressions for the tangent stiffness matrix of a single member are developed . The tangent stiffness of the complete structure is then considered in the following section.

investigations, the tangent

It should be emphasized that, except for stability

ST 12

NONLINEAR ELA$TIC STRUCTURES

2695

5tiffness í s largely just a tool which is a part of some solut í on techniques. The fundamental equations are Eqs. 10, 21 and 14, and if a solution can be Iound which satisfies these equations, by any technique whatsoever, this is a valid solution.

any state , A, an

equation of the íorrn

Member Tangent Stiffness in Memb e r Coo r dinates . -In

{dS} = [tA] {dv}

(24)

ís sought, in whieh [IA] is the mem ber tangent stiffness in state A. This ma- trix is obtained most directly for a frame member by differentiating Eq. 20

with respect to {v}. The result is

1 O j r'l

2 o

dV

2

as, \

dS 2 ) = 22

2

1

dS s '

L O

o

~

dvs

+ EA

[( L(12 , : - 3 v , u + vj)

140

.

+

~)

15

( - L(3 ' : - ''', "

280

+ 3,:)

_ "- )

30

( L(v~ - 3v)v2 + 12vV 5 ) symme t rrc . 140 + 15
( L(v~ - 3v)v2 + 12vV
5 )
symme t rrc .
140
+
15
( 4v)3~ v 2 )
l
dVi (
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
dV
(25)
•••••• •
2
( 4V23~ Vi )
dv
s
o
whieh ean be written as
{zs } = [[ti]
+ [12(VA)]] {dv} = [tA] {dv}
(26)

More generally, the tangent stiffness is obtained by differentiating Eq . 19. This differentiation is conveniently, and instruetively, carried out as follows. Two states are eonsidered, namely {VA}, {SA} and {VA + tov}, {SA + toS}. ln each of these states a virtual displacement is imposed and an equilibrium equa- tion in the form of Eq. 19 is developed. The first equation is then subtracted

from the seeond to give

an equation for {toS}. Both {toS} and {to v} ar e then made

to tend to zero, 50 that high order terms in {tov} can be i g nored. The result

ean then be arranged as

{dS} = I r < ai>T e, < ai> Ldp . {dv}

o

+ I f <ai>T e, Yi <b i > Ldp . {dv}

o

"

.

. "; ; ' 1~ · 4

,-.

;;:' , ' ,

.

f r ;

; ' ~I

I

::

i'"

2696

+ L t <bi>TEiYi<ai>Ldp·

o

December, 1969

{dv}

+ L f

1

<bi>T r, z, r, <b i > Ldp . {dv}

+ L t <bi>T ai <b i > Ldp . {dv}

o

The first term on the ríght-hand side of this equation is the ordinary linear stiffness of the member. The remaining terms depend on the deformations or the member, and may be collectively termed the " inítíal straín= stiffness

of the member. The t a ngent stiffness numerical evaluation of the integraIs

matrix will always be symmetrical, and is possible. 25, the flexural deformations, v l and v

,

2

in Eq.

are made equal to zero, the result is

Initial Stress Stiffness.-If,

[t 2 (vA»)

or [12(VA»)

~

15

EA I _!a.

2

15

-1

SsL I 30

O

30

O

-1

30

2

TI

O

-~

30

~

15

O

ol

:J

:1

.

.

.

 

.

(28)

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

(29)

This matrix now corresponds to the last term on the right-hand side of Eq . 27. It is appropriately termed the "initial stress" stiffness matrix of the member, following the terminology used by Turner, Martin and others (6,5). A further contribution to the structure tangent stiffness, more appropriately identified as a "geometric" stiffness, fol lowing the terminology of Argyris (1), is considcred in the next sectíon, It can be seen that the true member tangent stiffness is not í n general ob- tained by adding the ordinary linear stiffness to the "ín í tíal stress= stiffness. Nevertheless, because convergence can be obtaíned by a Newton type of method without the use of an exact tangent, it may be satisfactory ín practice to use thís approximation to the true tangent stiffness. For stability investigations of such structures as axially loaded, geometrically perfect columns, the as- sumption that the flexural deformations are initially zero is, however, correct. It may also be noted that the derivations of Eqs. 23 and 27 are identical in principIe to the use of the strain energy expressions

l

Si = .E!.-

ôVi

.

.

.

.

.

.

.

.

.

.

"

(30)

ST 12

1l 2 U

and tij = -",-",-

uV i uV j

ín which U = strain

NONLINEARELASTIC STRUCTURES

2697

,

,

(31)

energy; and t ij = term of tangent stiffness matrix.

STRUCTURE TANGENT STIFFNESS

Member Tangent Stiffn e ss form

{dR} = [TA)

{dr}

in Structure

Coordinates .-An equation of the

(32)

ís sought, in which [TA 1 is the tangent stiffness of a singIe member or a group of members in the fixed coordinates, X, Y . The following equations can be written:

{dS}

= [tA 1 {dv}

(26)

{dv} = [cAl {dr}

(33)

and {dR} = [CA]T{dS} + [dcF{SA}

(34)

Eq. 34 is obtained by differentiation of Eq. 14, in which the force transforma-

tion matrix, [CA]T, is a functí on

33 can be substituted into Eq. 34 to g ive

of the joint dispIacements, {r}. Eqs. 26 and

[tA) [cAl {dr} + [dcl T {SAI t

(

{dR} = [cAF

= [Tl) {dr} + [dcF

{SA}

(35)

Matrix [Tl] is the memb e r tangent stiffness transformed into structure co- ordinates, and accounts for the chan g es in ma g nitude of the memb e r end forces as th e structur e deforms. Term [dc F {SA} accounts , essentially, for the chang e s in direction of th e member e nd forces as the structure deforms. This se c ond term must be rearranged as

[dc]T {SAI = [T 2 )

{dr}

(36)

in which matrix [T 2 ) can be termed the "geometric" stiffness. Two methods of makin g this rearrangement are considered herein. The first method, which is essentially that used by Argyris (1), involves differen- tíation of the force transformation matrix, [ c A) T, and í s "exact" but does not appear to guarantee that a symmetric stiffness matrix will resulto The second method follows more cIosely the derivations of the prevíous sectíons, but re- quires that some terms be ignored a prior i. Method 1, Differentialion of Force Transformation.-The matrtx, (cAF, for a frame member í s a function of the in í tial geometry of the structure and the joint displacements, r l through re (Fig. 4). Therefore

(dc)

- /.J

T _

\ '

t = 1

and [dcJT {SA} =

[ SC JT

Ilri

,'[SCJT J

. ú

z.= 1

Ilri

dri'

lSA} drj

 

, (37)

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

(38)

,

o''::

"v-

'

. '1-/'.':"

2698

December, 1969

ST 1.

Eq. 38 c a n now be arran ge d

in the form of Eq. 36, in which column i of [T

l

i

[8 e/ er il T {S A}' If the multiplication

be expressed in the mor e convenient form:

sequence is reordered,

matrix

[T 2

2

l C

[T 2 l = SI [T 2 ] + S2 [T 22 ] + S3[T 2 3 l

 

.

in which column i of [ T 2 j 1 is column j of [ôe/ õr i J T.

 

For a planeframe 8, usin g

memb e r, matrices [é l e / 8ri]T

[

~JT

8

ri

= [ 8eJT

88

~ 8 r i

+ [~JT

si .

u:

can be evaluated

from Eq.

.

The result

is

 

[T21l = [T 22 ] = i2 I

S2

- se

e

2

(e 2 _ S2)

2se

symmetri c

O

O

_S2

se

O

8 r i

2se

( S 2 - c2)

O

-2se

( S 2 _ e 2 )

-2 s e

O

( e 2 _ S2)

O

O

O

O

O

O

(41)

 
 

2

se

O

 

O

se

O

_ e 2

O

and [T 2 3 ] = -b I

O

symmetric

O

S2

O

O

- se

O

I

(42)

 
 

c 2

O

 

O

in which s = s í ns, and e = cos e . For more complex structural

elements, it should generally

be possible

to

carry out the corresponding

manipulations.

The member

tangent sttifness

in

structure

coordinates

is then given by

[TA)

= [TIl

+ [T2 l

(43)

 

However, by this method there appears ways be symmetric.

to be no guarantee

that [ T 2 1 will al-

If the manipulations

required

to develop Eqs. 41 and 42 are examined,

it is

found that for [T21 ]

and (T 2 2 ] there are contributions

of similar

orders

of

magnitude from the differentiations

with respect

to both

e and L. That

is, the

influence of the change in mangitude of the end shears,

(SI + S2)/ L,

as the

member changes length is of the

same order of magnitude

as the influence

of

the changes in direction

has already b e en assumed

therefore, for consistency

of those shears

as the member

rotates.

However, it

and,

On the other

that change of member length is negligible,

[ T 21 ) and [ T 221 should both be ignored.

hand, the terms

in [T23 ] result solely from differentiation

with respect

to e,

and are not negligible.

Method 2, Use of Nonlin e ar

Deformati o n-DisPlaeement

Relationship . -As-

l

2699

ST 12

surne that for moderately small changes of [oint displacernent {~r} about some

d e for rned postt í on of the structure , nonlinear relationships between change of rnember d e forrnation and change of joint displacement can be written, and that these relationships can be approximated by the linear and quadratic terms

onl y as

NONLINEAR ELASTIC STRUCTURES

t.Vi

= < di> {t.r}

+ ~ {~ r F

[eil

{t.r }

(44)

For an infinitesimal

increas e ,

a pplying the virtual displacements

b y differentiating Eq. 44:

d~ Vi

= < di > {d~r}

+ {~rF

{d~r} ,

in {~r},

such as will be ímposed in

principIe, the following equation is obtained

[ ei l

{dt. r }

(45)

Consider, now, the equilibrium between the [oint loads and the member actions

in two adjac e nt

r e mains constant. In the first state, the virtual work equation is

states

{r}, {R} , {S} and {r + t. r} , {R + t.R } ,

{S}, in which{S }

{d~'tF

{R} = ~ dt. Vi

Sil,6 r= o = {dt.'t } T

L: S i < d i >T

 

(46)

In the second state, the virtual

work equation is

 

{dt.rF

{R

+ t.R }

= ~ dt.Vi

Si l,6r

= M

~

 

(47)

 

= { d t.rF

L: S i [ < d i >T

+ [ ei V

{~r } l)

Therefore ,

if Eq . 46 is subtracted

from Eq.

47 the result

í s

{t.R }

= [~

Si [e i 1T ]

{~r}

, (48)

mem ber, it is found that

there

is no simple and direct way to

For a frarne determine

[eil for the flexural

effects,

if ttís

as-

sumed, a priori,

that the "geometric"

i = 1 and i = 2. However, stiffness effects associated

with the

flexural actions

SI and S2 ar e

zero, then this problem

can be ignored.

The

quadratic term inthe equation

for ~ v 3 ís then L(~e)2, and

therefore

matrix [ e 3 1

is readi1y written as

[e 3 1 = <f 3 >T <f3>

(49)

in which

<f 3> = L -1/2

< sin O

Therefore,

matrix

(T 2 1 becomes

-cose

O

-sine

cos e

O>

(50)

S3[ e 3 ], in which [ e 3 1 is found to be identical

to matrix [T 23 1 in Eq. 42. "Correct" Tangent Sliffn e s s fo r Us e in Large tnepia ce ments Analys i s.-

For any rnathernatically

vergence can be expected

As noted previously,

which is used is exact

idealized structure,

the most reliable

and rapid con-

by a Newton type of method if the tangent stiffness

for the idealized structure.

a

common assumptíon in the analysís of finite element systems is that the mem-

ber actíon deformation relationship of Eqs. 21 and 23 can be linearized as

,
.

. ' \

" : ' \ ~. : \! ' . :

"

-

11

2700

{SA}

\' J1

[k1l {VA } = i.J

o

D e c e mber, 1969

< a i > T

s , < a i > JJlp . { V A }

ST 12

.

(51)

If this as s um p tion

id e a li za tion .

Eq s . 26 and 27 is

is m a d e , it b e c o m e s

an int e g r a l

It follows imm e diat e l y

that th e m em ber

part of th e math e m a tical

tan ge nt s tiffn e s s ,

from

{dS} = [t 1 1 {di'}

~t' < ai> T

t t , < a i> Ld p . {d d

(

52)

T h a t is, th e "i nitial

tion a l contr i bution to th e s tructur e

s t i f f n e ss. If a more e l a b o r a te f o rm of the t a n ge n t stiff n e ss i s us e d , it w iU

a c tu a lly

s t rain "

s tiffn ess

i s impli c itl y ta n g e nt stiff n e s s

i g nor e d ,

a nd the onl y a dd í -

must b e t he " g e om e tric"

b e l e s s e xact , an d m ay b e e x p e cted

to l e a d to less rapid conv e r g e nce.

DIRECT FORMULATION

In the p r e c e ding

th e o r y ,

mem be r d e f ormations

a r e

in o rd e r

w e re d e l i b e rately s e p a rated

to dev e l o p a g e ne r al

f r o m m e m b e r ri g id bod y di s pla c em e nt s , p e r m itti n g l a r g e rigid b o d y rotations.

th e memb e r ri g id bod y r o t a t í ons

line a r r e l a ti o nships

di s placement s c an b e writt e n as

th e ory

n

If the [ oint displa ce m e n ts known no t t o be lar g e,

are su ch that th e n dir ec t

and joint

o n-

b e t w e en th e dif fe rential

e l e ment d e formations

Ei = < f[ i > {r} + ~ {r} T <h i > T « l t {r}

(

53)

for w hich appropriate

for the s ec ant and tan g ent stiffness e s

be d e riv e d , b y the sam e m e thods a nd with simil a r form as Eqs . 23 and 27 .

m a tr ic es

<f[ i? a nd « h c an b e det e rmin e d .

E x pr es

sions c an now

dir e ctly in structur e coordinat e s

If the jo i nt displa ce m e nt s

are n o t m oderat e ly

s mall, how e v e r,

s hip repr es e nt e d

b y Eq . 5 3 ma y n o t be suffi c i e ntl y

ac c ur a t e .

th e r e l a tion-

In th í s c ase,

e ith e r hi g h e r order t e r ms must b e i nc lud e d in E q . 5 3 or th e th e or y o f the

previous

dir e ct formul a tion do e s not requir e s ubstantiall y

the

s e ctions

must b e applied.

Ev e n if Eq. 5 3 is suffi c i e ntl y

accur a t e,

l e ss co mputational

o r c om-

puter codin g e ffort to appl y than th e indirect formulation.

formulation

th at th e "initial

s train"

Fu rth e r , th e indi r ect an d th e " ge om e tric"

nonlinear

ha s the advanta g e e ff ec ts ar e distinctl y

s e p a rat e d,

so th a t the "i niti a l

strain "

e ff e ct

ca n b e ignor e d if the initial a ssum p tions

th e indir e ct formulation should be pref e rabl e

requir e

it. It appe a r s ,

therefor e,

that

in a ll cases .

CONC LUS10NS

In this p a p e r the thr ee

major sourc e s of nonlin e arity hav e b e en d i sti n guish e d ,

undergoin g l a r g e displa ce ments

in e lastic structures a nd the esse ntial

e quations of the probl e m have b ee n stated. It has been shown that th e non-

linearity of th e s tructure

r e lating m e mber deform a tions to joint displ ace ments, from th e ne ce ssity of

results from the nonlin e arity of th e transform a tions

.

f!!,

~ 1.'

{ .

~.

~

, .

?r

:

'

\

formulatin g

th e joint e quilibrium

e quations in the displac e d confi g uration

of

the structure,

and from th e nonlin e ar action-d e formation

r e l a tionship s

of the

structural m e mbers . Th e nonlineariti es

orginatin g within th e structural m e rn-

2701

re latin g diff e rential

ha s

ST 12

bers r e s ult

e l e m e nt d e formati o n s to m e m b e r defor ma tions, and f r o m th e n e c e s s ity o f

f ormul a tin g th e me mb e r e quilibrium e qu a ti o n s in th e d e f o r m ed co n f i g ur a tion s

o f th e me mbers. Th e nonlin e arit y

b ee n t e rm e d the m em b e rs

t h e " geom e tri c "

NONLINEAR ELAST'IC STRUCTURE S

th e n o nlin e ar

t r ans form a ti o n s

in turn f r om

whi c h o ri g inates wi thi n the mem b e rs

th e " ini t i a l

s train "

e f f ec t h e r e in, an d th a t w h ic h ori g i nates

e ff ec t .

outsid e

ha s b ee n t e rm e d

E x pr e ss ions

f or th e secant a nd tan ge nt s t iff ne ss es o f m e mb e r s

a nd stru c -

tnr e s h a v e b ee n pr e s e nt e d

e lem e nt sys tems o f a v a riel y

t an ge nt s tiffn ess e s

in a corn pa c t I o r m , whi c h c a n be ex t e nd e d to finit e

o f ty p e s. Th e i m porta n ce

se lectin g co nsist e nt

b y m e t ho d s

o f N e wt on t y pe ha s b ee n e mph a -

f o r s oluti o n s

siz e d, an d it h a s b ee n shown t h a t th e "i ni ti a l

str a in"

er f c ct

ma y h a v e to b e

a s oluti o n . Consid e r-

abl e w ork r e mai ns t o b e carri e d o ut to s e l e c t e f f i cien t comp utation a l a l g o rithm s a nd to d e t e rmin e th e r e lati v e imp o rtan c e of th e diff e r e nt t y pes o f non linearit y .

i g nor Of e d co . ur se , a sta t e men t

o í th e o r y do e s n o t co n s titut e

AP PEND N.-R E FERENCE S

I. Ar gy ri s , J . H., K e l s ey , S . , an d K a me l , H

"

M a t r ix

Me th o d s i n S tru c tural

A n a l y s i s . " Agard o -

g

r a ph 72 , P e rga m o n Pr es s , New Y o rk, 1 96 4, pp . 1 - 1 64 .

2. Co nn o r , J . J ., L og c h e r . R . D . , a nd C h a n. S . c., "No nlinea r

A n a l y si s o f El astic Fr a m e d Stru c -

urr es. " .I o urnal o/ l h e S tructura l

pp . 1 52 5 - 1 5 48.

õi v i s i o n , ASCE. Vo l . 9 4, No. ST6, Pr oc. P a p er 60 11 , June 1 968 ,

No nlin e ar T wo- Di m en sio n a l

3 . F e lipp a, C . A ., "R e fin e d

Fin i t e E l e m e nl A n a l ys i s o f Lin ea r a nd

Stru r. tur es, " SES M R e p o r/66 -22. U ni ve rsity o f C a l i f o rnia, B e r keley, 1966.

4 _ M a lle u , R . H . , a n d M a r ca l, P . V ., " Fi n i te El em e n t Ana l y sis o f N o nl i n ea r Stru c tur e s . " .Iourn a l 01 ih e S tr u c t u ra ! Di vis i o n , A SCE , V o l. 94 , No. ST 9, Pr o c . P a p e r 6 11 5, S e pL. 1 968 , pp . 2 0 8 1- - 210 5 .

5 . M a rtin , H _ c. , "O n lhe Derivat io n o f S tiffnes s M a tri c e s for th e A n a l ys i s of L a r g e D e flection a n d

S t a bility Probl e rn s , " Pr ucee d i n gs . Co nf e ren ce

Wright P a tters o n A ir Fo rce Ba se, O h i o , Oct ober , 1 9 6 5, pp . 6 97 - 7 1 6 .

o

n M a trix

M et h o d s

in St r u c tur a l Mech a ni cs ,

6. T urn e r , M . J ., M a rt in , 1-1 .c. , a n d W ei k e l, R . C

"Further

D e v e l o pm e nt an d A ppl i c a tion o f th e

S t i ffn e s s Meth od. " Aga rd o gr a ph 7 2 . P c r ga m o n Pr e ss. New Y o r k , 1 96 4 , pp. 203 - 26 6.

~

. ~~