Professional Documents
Culture Documents
1994
Copyright 0 1994 Elsevier Science Ltd
Pergamon 0045-7949(94)E025fM Printed in Great Britain. All rights reserved
0045-7949/94 $7.00 + 0.00
Abstract-The application of finite element analysis to lateral buckling problems, locating the critical
points and tracing the postbifurcation path, is treated on the basis of a geometrically nonlinear
formulation for a beam with small elastic strain but with possibly large rotations. The existing finite
element formulations for thin beams are examined in the aspect of application to bifurcation problems,
such as lateral buckling, and the choice of an appropriate rotation parameter for representing incremental
or variational rotations in finite element formulations is discussed in relation to locating bifurcation points.
This is illustrated through several numerical examples and followed by appropriate discussion,
I357
1358 Haengsoo Lee ef al.
dimensional deformations related to torsion as well form cantilever beam under transverse shear force. In
as bending, so that the effect of geometric tangent this section, we examine the lateral buckling of a
stiffness will be most significant. tapered cantilever beam under transverse shear force
We follow the line adopted in the finite element and obtain the buckling load by solving the linearized
formulation of Simo and Vu-Quoc [2] and Cardona eigenvalue problem by use of the shooting method;
and Geradin [3], but we assess these formulations the resulting solution is then employed in the sub-
together with lura and Atluri [4] from the viewpoint sequent chapter to confirm the finite element solution.
of applying them to the lateral buckling of conserva-
2.1. Equilibrium equations
tive beam structures in conjunction with the energy
criterion upon the stability limit. We rely upon the In the absence of forces or couples acting upon the
so-called ‘branch switching’ to calculate the accurate beam, except at the ends, we have the following forms
postbifurcation path without introducing any imper- of equilibrium equations [B. 91:
fection in geometry or in loading.
The content of the paper is outlined as follows. The dN,
dX-N2~,+N,~,=0
lateral buckling of a cantilever beam under a vertical 3
end load is treated in detail in Section 2. The govern-
ing equations of a tapered cantilever beam are de- dN,
z- N3~, + N,K,=O
rived and a shooting method is applied to find the 3
Substituting eqn (2.3) into eqn (2.2), we eliminate We thus have the six buckling equations in the six
M,, MI and M, and obtain the following: unknowns NZO,N,,, K,“, N,, K~ and K,. From force
equilibrium, it follows that
A,~+K,~-(A~-A~)K*K,=N~
3 3 N,, = P sin c,, N2”= P cos i,, (2.7)
dN,o
do - N3o’b = 0
3
dN3,
do + N2o’%o
=0
3
A,-A3dA3
‘ho dA,
4dX,=N,o-K~oc. (2.5) +A,A,z K1°K3
integration scheme combined with the shooting applicable once the strains increase beyond the pro-
method, which will yield almost exact values for the portional limit, or once the strains become sufficiently
buckling load P. The numerical results will be dis- large that the second order terms are not negligible
cussed together with the FEM solution in Section 5, anymore [3]. In the total Lagrangian finite element
wherein the solution obtained from this section will formulation of beams we follow the line taken by
be used to confirm the finite element solution. Simo and Vu-Quoc[2] and by Cardona and
Geradin [3] and, moreover, we critically assess these
3. FINITE ELEMENT FORMULATION FOR and other FEM formulations for large deformations
DEFORMATIONS OF A SLENDER BEAM of beams, such as those of Iura and Atluri [4], from
WITH SMALL ELASTIC STRAINS
the view point of applying them to a bifurcation
For versatile analysis of lateral buckling for general problem like lateral buckling.
beam structures, it is desirable to employ finite el-
3.1. Deformations of slender beams and representation
ement analysis. However, to treat lateral buckling
of rotations
problems, one needs a fully three dimensional finite
element model for a structure. Moreover, buckling in Let R denote the rotation operator mapping a
general involves finite rotations for a slender struc- material frame {E,(X,)},,,,2,,, fixed in space, into a
ture, even when the elastic strains involved remain body-attached moving frame {t,(.X’,)},= ,,2,3(Fig. 3):
small: if the prebuckling deformations are not small,
to evaluate the accurate stability limit as well as the t,(x,) = RW,)E,(X,) I= 1,293, (3.1)
postbuckling behavior, one has to consider finite
rotations or geometric nonlinearity in deformations where t, and E, are unit orthogonal base vectors. The
of a beam and calculate the accurate tangent stiff- deformed position X,E R3 of the centroid of the
ness, including the geometric stiffness and the initial cross-section is defined by the map 4,:
displacement stiffness in addition to the constant
linear stiffness (in the case of the total Lagrangian x, = 4,(X3). (3.2)
approach).
Because of complexity in the postbuckling prob- An arbitrary deformed configuration of the beam
lem, more often than not, many authors prefer to use is now specified by the mapping of the central line
the so called imperfection approach, wherein a small 4,(X,) and the rotation of a cross-section R(X,):
perturbation of loading or geometry is introduced to
switch the bifurcation type problem into the limit 4(x3) = (4,@‘3), W’3)). (3.3)
point type problem [l, 2, 51. However, one needs the
correct behavior of a perfect structure, in order to Let !P denote the finite rotation angle of the
evaluate the imperfection sensitivity, and sometimes rotation tensor R and let e,, e, denote orthonormal
one may encounter a case wherein the imperfection vectors perpendicular to the axis of R. Then, the axial
approach yields less accurate results, depending upon vector e, of R is given as
the nature of a problem. In this section, we introduce
a total Lagrangian finite element formulation of e, = e, x e2 and Re, = e3, (3.4)
beam structures within small elastic strains with
possibly large rotations, which enables us to analyze where ‘x’ indicates the cross-product. The finite
the correct stability limit, as well as the postbuckling rotation vector representing R may be written as
behavior, regardless of the magnitude of prebuckling
deformations, as long as the strains remain small. The Y = ]P]e,. (3.5)
reason we limit ourselves to small strain deformations
is that the elastic constitutive relation (2.3) is not Moreover, the rotation tensor R may be written as
R=cosIPl(e,Oe,+e,Oe,)
I
A53 where ‘0’ denotes the dyadic product. Another way
of representing the rotation tensor R is to use the
exponential map,
Suppose a beam in the current configuration, The use of this conformal vector makes the full
whose orientation is represented in terms oft, in eqn linearization of the equilibrium equation feasible, and
(3.1), undergoes an incremental deformation so that the exact expression for the tangent stiffness is ob-
t, takes a new orientation t;. Let R’ denote a new total tained, but it still remains rather complex.
rotation from E, to t;, such that In contrast, the finite element formulation by use
of the representation (ii), used by Simo and Vu-
t; = R’E, . (3.9) Quoc [2], yields an unsymmetric tangent stiffness
during equilibrium correction but the tangent stiff-
Moreover, let A0 and Afi denote the incremental ness becomes symmetric for a conservative system
spatial rotation vector from t, to tj and the corre- once equilibrium is reached. Moreover, the symmet-
sponding skew-symmetric tensor, respectively, and ric tangent stiffness at an equilibrium configuration is
A@, A& the corresponding material rotation vector precisely related to the second variation of the total
and skew-symmetric tensor representing the same potential energy for a conservative system, so that the
incremental rotation. We then have the relation [3] energy criterion on the stability limit is applicable.
Moreover, the linearization leading to the tangent
A0 = RAQ stiffness is relatively simple. For this reason, we adopt
the representation (ii) for finite element formulation.
Ae= RAdRT (3.10) Deformation for rods or beams was discussed in
detail by Love [9]. He described Kirchhoff’s theory of
and have the following ways of representing the new thin rods as being largely kinematical and gave an
(total) rotation R’: alternative description of the nature of the strain in
the deformation of a thin rod, in order to relate the
(i) R’ = exp(l + Alp) (3.11) moments applied to a section of a rod to the curva-
ture and twist ([9], pp. 389-393). Some writers have
(ii) R’ = exp(A@R # exp(A8 + P) used finite elasticity theory to discuss deformations of
a rod more rigorously; for example, Parker [lo],
unless A& is coaxial with p (3.12a) Shield [1 l] and Shield and Im [12]. Others propose
useful technical theories; for example, Danielson and
R’ = R exp(A@) # exp(l + A&) Hodges [ 131 and references cited therein. For kin-
ematics of deformations, to clarify some points, we
unless Ad is coaxial with p (3.12b) begin with Shield and Im [12], wherein a consistent
formulation from the viewpoint of finite elasticity
where AP is the skew-symmetric tensor of the incre- theory is proposed for the analysis of elastic struc-
ment AY. Note that both of p and AY belong to the tures without a prior assumption about the magni-
same vector space, the tangent space to a nonlinear tude of the deformations, apart from the assumption
manifold, the so-called special orthogonal Lie group that strains are small. However, it is not the purpose
at the identity, so that the interpolation is allowed for of the present work to proceed to employ this kin-
‘P and A’P (see [3] for detail). ematics in the subsequent finite element formulation,
The representation (i), used by Cardona and but our attention is given to the point that this yields
Geradin [3] for finite element formulation, makes it a complete first order theory for deformations of a
possible to update R easily, but it is extremely difficult beam that accounts for cross-sectional warping, so
to obtain the full linearization of an internal force that it may be conceived of as an extension of the
when applying this representation to finite element kinematics with neglect of the warping displacement
formulation for deformations of a slender beam, in finite element formulation.
although the resulting tangent stiffness will always Let X indicate the undeformed position vector for
remain symmetric for a conservative system. Cardona a particle in a beam and X, the undeformed central
and Geradin [3] indeed omitted some term in their line. We suppose that the beam is initially aligned
geometric tangent stiffness because of its complexity, with E,, so that
but this does not give rise to any problems in tracing
a loadclisplacement curve. For locating a precise
X = X,, + X,E, (c( = 1,2, sum on do). (3.14)
bifurcation point, or for branch-switching for a con-
servative system, however, an exact expression for the
tangent stiffness, which is precisely connected with The small strain deformation of a beam can be
the second variation of the total potential energy, is thought of as effected by imposing the following three
needed so that the energy criterion on the stability of sequential deformations: first, particles are given
a conservative system may be applicable. Iura and small displacements II* which vanish on the line of
Atluri [4] introduced the conformal rotation vector, centroid; next, a section X, = const. is given a ro-
defined by tation R about the point (0,0, X,) which rotates line
elements to their orientation in the deformed state;
8*=4tanmY//lp[. (3.13)
4 finally, a translation of the section brings the centroid
TAS5316H
I362 Haengsoo Lee et al.
to its deformed location x,. Thus the deformation can 3.2. Variational principle and finite element formu-
be written as lation
Based upon the basic kinematics for deformations
x=x,,+R<, <=X-X,+u*=X,E,+u*, (3.15) of a beam in the foregoing section, we here discuss the
variational formulation for the finite element analysis
where R is a function of the X, coordinate alone and for a beam under conservative loading. The principle
u* is a displacement associated with local cross-sec- of virtual work states that for an arbitrary virtual
tional warping. The magnitude of u* is at most II O(c) displacement,
where a is the cross-sectional dimension and O(c)
represents the order of magnitude of the strain in a SW,-bW,.=O, (3.22)
beam. Because of complexity, ultimately the warping
displacement u* is not to be included in the finite where SW, and SW, are the internal and external
element formulation to follow, but we state this for virtual work, respectively.
completeness of formulation regarding the kin- Let S,, denote the components of the second
ematics of beam deformations. Piola-Kirchhoff stress tensor S. Then, the expression
Let e denote the extensional strain of the central for the internal virtual work 6 W, is written as
line. Then the tangent vector t along the central line
may be written as
(3.23)
I dx,
t=---. (3.16)
1 +edX, Employing the expressions (3.20) for the Green
strain components for deformations of a beam, we
The tangent vector t is, in general, not in agreement can show that, up to O(t),
with the unit vector t, attached to the cross-section
because of the shear strain and the warping. The
curvature resulting from bending is related to the rate 6wi= $+gx(x-x,,)
7 3
of rotation R, with respect to distance along the
undeformed central line, and the curvature tensor C
is defined as -68 X 2 I dA,dX,
3
dR
ss
ri’ =-RR’. (3.17) L
dX,
+ &K~ekdAdJ’3, (3.24)
0 AlI
The curvature vector K, which is the axial vector of
li, is then given as where X, = 0 and X3 = L denote both ends of the
dt beam and T, is the traction vector per unit unde-
K =ft, XL. (3.18) formed area on the cross-section, which had the
dX,
outward normal m = E, in the undeformed configur-
Note that K is related to K’as ation, i.e. T, = E3SFr where F is a deformation
gradient. Note that the first term in the expression
above agrees with the result obtained in the absence
Rv=K XV VVER’. (3.19)
of the cross-sectional warping [2, 31 and the second
term represents the work resulting from the virtual
Now the Green strain components for the beam
cross-sectional warping displacement 6u*.
deformation (3.15) can be obtained up to O(c) as
The consideration of the cross-sectional warping in
the formulation of the large deflection of a beam has
K,, = es (&B=l,2) been treated in a different context by Simo and
Vu-Quoc [14]. Their formulation is rather compli-
E,,=e,*,+i(l +a,,)[~ x R(X-X,)lt,+6,,e cated, but the numerical results do not yield a
significant difference from the results obtained with
(no sum on Z, I = 1, 2, 3), (3.20) neglect of the warping.
The inclusion of the warping term in eqn (3.24)
where e& is the strain related to the warping displace- would yield an extremely complex finite element
ment u* and is given by formulation because the expressions for u* in Appen-
dix A are not simple. We therefore do not try this, but
simply state that the contribution of virtual work
from the cross-sectional warping, which is given in
the form of the second term in eqn (3.24), is neglected
The detail of u* may be found in Shield and Im [12] when only the first term is retained in eqn (3.24).
(or see Appendix A). Retaining only the first term with neglect of the
Analysis of lateral buckling 1363
warping displacement u* and introducing the stress where b, and /I2are shear correction factors, GA is the
resultant n and the stress couple m: shear stiffness and EA is the axial stiffness of the rod.
In the absence of body forces and distributed
tractions on the lateral surface of a beam, the external
n=n(X,)= T,(X,) d&, virtual work by a prescribed traction T, may be
s A”
written as
m=m(X,)= (x - x,) x T,(X,) dA, (3.25)
s A” cSW<=[~~o’P3~hxdAo]+-
A”{6x,x3
and the deformation measure y, the corotational rate
of which is conjugate to n, = T,
dxo
U
=L
1
Y =dX;-t3,
+sR(X -X0)} dA, (3.34)
X, =cl
we have
In terms of 60, we may write this as
SW,= L(n,cfy +m.&c)dX3, (3.27)
s II SW,= T, .6x, dA, + {(x - x,,)
where the corotational variation gy and &, which is [S A” j A”
the variation taken by an observer moving along the Xx=L
deformed central line, are given as [3] XT,}.60 dA,
1 X, =II
=[ii~6x,+iii~Se]~I,L, (3.35)
(3.28) where ii and rir indicate the resultant force and stress
couple vectors resulting from the prescribed traction
Setting T,, i.e. a prescribed force and moment. We can easily
show that the principle of virtual work, eqn (3.22)
with the aid of eqns (3.24) and (3.35), yields the weak
form as given by
(3.32)
because the shear strains involved in deformation of
a beam are O(E) at most. We remark that the strong
In addition to the stress couplecurvature relation form, eqns (2.1) and (2.2), the six equilibrium
(2.3), we introduce the constitutive relation for the equations of resultant forces and moments are noth-
resultant forces on the cross-section, ing but equations obtained from integration of the
local equilibrium equation (3.36) multiplied by some
N,=BrGArl, N2=/&GAr2, N3 = EAT,, (3.33) weight factors [lo].
I364 Haengsoo Lee et al.
For the present beam model to be applicable with A0 of eqn (3.12a) to represent the incremental
no end effects, i.e. no boundary layers near the ends rotation and update the rotation vector according
of a beam, the traction condition on the end cross- to this equation and similarly use 60 to represent the
section has to be satisfied pointwise. The normal variation of rotation 6R.
traction component from bending stress is normal to The weak form for the equilibrium equation is now
the end cross-section and the corresponding traction obtained by making the first variation of the total
has to be applied normal to the cross-section, so that potential energy vanish. From eqn (3.40) we can
the applied traction, within the beam model, is obtain the weak form as
configuration dependent under large rotations, and in
general not conservative in the local sense. In the
global sense, however, it may be conservative under
special circumstances [ 151. When the traction con-
dition is satisfied pointwise, i.e. T, = E,T = Tj,e,, the
applied moment rir is configuration dependent. More-
over, the axial vector of dRR’, which we may write
as 68, is not an exact differential. Hence the moment
term rir. 60 in eqn (3.35) has no potential, i.e. = L(N4- +M.?!K)dX,
s0
lil 68 # s[a 01. (3.37)
-[ii'6x,p;=o, (3.41)
Argyris et al. [I] and Simo and Vu-Quoc [2] also
pointed out that an applied moment about an axis where 6f and 6K are given by eqn (3.29) and thus
fixed in space is not a conservative force in general. directly related to &#I = (Sx,), 6R) or, equivalently,
As discussed by Simo and Vu-Quoc [2], the applied s4 =(6x0,68).
moment may be conservative only when the following Assuming that an equilibrium configuration is
condition is met: reached at a time step t’“), the solution for the
subsequent time step t In+‘) is calculated through
m (~30x A0) = 0. (3.38) linearization:
The symmetry of the material stiffn_essis now appar- Taken together, the tangent stiffness is now given
ent from the symmetry of c and C. as the summation of the material stiffness, eqn (3.44)
To calculate the geometric stiffness term in eqn and the geometric stiffness, eqn (3.48). If we are
(3.43), we first linearize 6r and 6K as concerned only with calculating equilibrium paths in
the load-deflection space, we may neglect relatively
d6x, unimportant terms in the expressions for the tangent
DsrA.9 =(A0 xR)~ =-68 x2 stiffness, or take an approximation to the tangent
3 3>
stiffness during equilibrium iterations in nonlinear
finite element analysis; we can still enforce the sol-
-Rr(&? ~2) ution to converge to equilibrium configuration by
updating the internal force correctly. With a view to
applying the energy criterion for locating the critical
DbKA4 = (A0 x R)‘$ (3.45) points, however, we have to calculate the tangent
3
stiffness accurately to assess the vanishing of the
second variation of the total potential energy strictly.
Substituting these in the geometric stiffness term and
This point will be discussed again in relation to the
employing n = RN and m = RM, we have
choice of the variable for representing incremental
rotation AR.
[N (DW) + M. (D6K)]A4
Let N,(X,) denote a shape function that takes 1 at
the node I. Then the vectors 4 = (x,, R), representing
= -s.(n x At’) a configuration, may be interpolated as
3
A8 = 1 N,(X,)A8,. (3.50)
I
+g. (m x A0). (3.46)
3
As pointed by Cardona and Geradin [3], the interp-
olation (3.50) is not allowed in a strict sense because
To write this in matrix form, we introduce the different points in an element will take in general
following matrix D, as in Simo and Vu-Quoc [2], different rotations and therefore each of the corre-
sponding incremental rotations A0 belongs to a
different vector space, so that the interpolation is not
allowed. However, the difference of rotations in one
0 0
D= (3.47) element is small in practice and there are no
e 0 difficulties in the interpolation.
We may employ the incremental finite rotation
vector AY for representing rotations, as in Cardona
when ii and I are the axisymmetric tensors whose and Geradin [3]. For this, we rely upon the relation
axial vectors are n and m, respectively. Then, we can between AY and AB
recast the geometric stiffness into
A0 =RTAY or 68 =RTS’P, (3.51)
[N.(D~~)+M.(D~K)].AI$ = $,$%I
3 3 where T = T(Y), derived by Cardona and Geradin
[3], is given in Appendix B. We first express 6fI in
dAx, dA8 terms of 6Y’, not in terms of 60, and then linearize the
D. - ~ A8 . (3.48)
dX3 ’ dX,’ resulting weak form 6lI = 0 to reach the incremental
relation, such as eqn (3.42). During this process,
Since the matrix D is not symmetric, the geometric however, the second derivative of T(Y), which would
stiffness is generally nonsymmetric. As shown by be extremely complicated to calculate, is involved in
Simo and Vu-Quoc [2], however, the geometric stiff- the tangent stiffness. In [3], this second derivative was
ness for conservative loadings recovers the symmetry indeed omitted in calculating the tangent stiffness,
for most of the typical end conditions when the owing to its enormous complexity. As discussed
equilibrium configuration is reached. earlier, there are no difficulties in calculating an
I366 Haengsoo Lee er al.
equilibrium path in the configuration space, although For a proportional loading, the equilibrium
some minor terms in the tangent stiffness are neg- equation or the weak form, like the type of eqn (3.41),
lected or some approximation to the exact tangent may be recast into
stiffness is taken for equilibrium iterations. However,
accurate values of tangent stiffness are required for F(q, 6q, L) = 0 for an arbitrary 6q
locating the bifurcation points on an equilibrium
path in the present work. This is why we choose A0 *f(q, I) = iP, -g(q) = 0,
for representing an incremental rotation, rather than
A’P, and this point will be manifested via the numeri- where i is a scalar loading parameter and P,> is a
cal example in Section 5.1. reference load vector.
Linearizing this equilibrium equation about a state
(4, I), we obtain the incremental equation of type
4. THE ENERGY CRITERION ON THE STABILITY AND A (3.42) as
POST BIFURCATION EQUILIBRIUM PATH
For a conservative system, the change of the total f(iYj+ Aq, 2 + An), = f(ij, 2)
potential energy An due to an admissible variation
+DfAl +DfAq+....
SI$ may be written as
= f(Q, 2) + Ai P,>
An(4,64) = ol-I&#~ + D(Dl-I6~)@ +. ..
-KAq+....=O
The first variation Dn@ = 6l-I vanishes for an If (&I) is a state on an equilibrium path, this reduces
equilibrium configuration and the sign of AII(+, 64) to
just depends upon the next dominant term KAq = Ai P,, (4.6)
D(DID+)&$, unless D(DIIc?#)&#J happens to be
zero. The energy criterion on the stability limit of a For the existence of solution for eqn (4.6) at a critical
conservative system is now stated as: the stability point, characterized by eqn (4.3), we have the follow-
limit (or a critical point) on an equilibrium path is ing consistency relation [ 161,
reached at the instant that the second variation of the
total potential energy D(DII&$)&$ ceases to be AE.v P, = 0. (4.7)
positive definite.
Let q = (q,, q2, ) denote the nodal degrees of To meet this condition, we have
freedom corresponding to the vector 4 = (x,, R) =
(x,, Ip), which represents the configuration. In terms A>. = 0 or v. P,, = 0. (4.8)
of the tangent stiffness, the condition for stability
The first condition corresponds to a limit point and
may then be written as
the second condition to a bifurcation point. The
critical points, whether they are limit points or bifur-
D(DI-I&$)@ = K,,Sq,Sq, > 0 cation points, are detected by eqn (4.4).
For implementing eqn (4.4) in finite element analy-
sis, we employ Newton type iterations in conjunction
for an admissible 6q, (4.2)
with the arc-length method [17]. Let s”” denote an arc
length when convergence is reached for time step
where K,, is the tangent stiffness. It is recalled that K,, (loading step) n in a load-displacement space and
become symmetric at an equilibrium configuration suppose that a sequence of pre-critical points of a
under conservative loading [2]. Hence the stability primary path has been computed. Then the updated
limit is reached at the point where the symmetric new guess for the arc length, s”‘+ ‘I corresponding to
tangent stiffness K,, ceases to be positive definite, i.e. a critical point is given as
at the point that one or more eigenvalues of K,,
become zero first. That is,
For the lateral buckling, the critical point is of a arc-length method, until convergence to a point on
bifurcation type, unless an imperfection in geometry the secondary path is reached.
or in loading is introduced. In the present paper, we
are concerned with accurate calculation of the buck- 5. NUMERICAL EXAMPLES
ling points for beam structures with no imperfections In this section, we take several numerical examples
in geometry or in loading. For bifurcation type to determine the validity of the foregoing finite
buckling, a finite element solution for postbifurcation element formulation in application to the lateral
path in general is supposed to trace the symmetry buckling of beam structures. We first consider the
preserving (or the primary) path, which is unstable on lateral buckling of a cantilever beam under shear
the postbifurcation region, unless an appropriate force, which has been discussed in Section 2, and the
branch switching scheme is implemented. For switch- critical points from the finite element solution are
ing the finite element solution path to the symmetry compared with those obtained from the shooting
breaking (or the secondary) path, which structures method in Section 2. Moreover, the postbifurcation
will actually follow for loadings beyond the buckling path will be obtained for this problem. Next we
point, we impose a constraint such that the incremen- consider the lateral buckling of a cantilever right-
tal nodal vector be orthogonal to the primary angled frame under end loads and the lateral buckling
path [16, 181. Then this will restrict the solution path of a hinged right-angle frame. The postbifurcation
traced in finite element analysis and will prevent it solutions obtained through the branch switching are
from returning to the primary path. compared with those reported earlier [I, 2, 51, which
Let Aqp denote the incremental nodal vector at the were obtained by imposing small imperfection upon
bifurcation point along the primary path, and v the loading.
eigenvector associated with the zero eigenvalue of the In all of the three numerical examples, we employ
tangent stiffness matrix at the critical point, i.e. the the quadratic beam element and uniformly reduced
eigenvector defined by eqn (4.3). We then let the trial two point Gaussian integrations to avoid the shear
increment of nodal vector Aqs along the secondary locking.
path be given as [ 161
5.1. Lateral buckling of a cantiletier frame under
vertical end load
Aqs = a(AqP + bv). (4.10)
This problem has been described in detail in
As stated earlier, Aqs is constructed such that it is Section 2 and it is depicted in Fig. 1 and 4. Young’s
orthogonal to the primary path: modulus and Poisson’s ratio used in the numerical
example are 200GPa and 0.3, respectively, and 20
Aqs Aqp = 0. (4.11)
quadratic beam elements are employed, regardless of
the beam geometry.
Because the cantilever beam under a vertical end
From this and eqn (4.10), it follows that
load is a conservative system, there are no difficulties
in applying the buckling criterion, eqn (4.4). The
(4.12) buckling loads for uniform cantilever beams are first
calculated and compared with those reported by
Hodges and Peters [8], who employed the shooting
The constant a is now determined from the prescribed
method as in Section 2. The results are plotted in Fig.
arc length introduced in the arc-length method.
5. The agreement between the two solutions is excel-
lent even for a large value of AZ/A,, which represents
AqS Aqs = Aqp. Aqp = I’, (4.13) the aspect ratio of a beam cross-section. Note that
prebuckling deformations become large as the value
where 1 is a prescribed arc length. of Al/A, increases. The buckling points for uniform
When the eigenvector v is orthogonal to the pri-
mary path at the bifurcation point, eqn (4.12) fails
because b goes to infinity. In this case, we can simply
take Aqs as
Aqs = av. (4.14)
ho= 40126
&=PLZ
as
7
‘*
0 Present Solution
C>\ --- Shooting Method
\ \
k = b/he
‘,j,
Y
k = 15 (tapered)
?$,A
\
\ ‘>
b‘
k = 1 0 (urnform)
0 1 7
0 50 0
1 0
UZ
Fig. 5. The bifurcation load vs the aspect ratio for uniform Al/A, increase, the prebuckling deformations will
and tapered cantilever beams. become large and then the transformation tensor
T(Y), which reduces to identify tensor as Y goes to
cantilever beams are calculated also by employing zero, i.e. as the prebuckling deformations come to
AY (or 6Y) for representing the incremental (or disappear, will be away from the identity tensor and
variational) rotations. As discussed in Section 3.2, its second derivative will not be negligible any more.
this involves the second derivative of T(P) in eqn This example clearly illustrates that the formulation
(3.51), but this is neglected, as in Cardona and of employing AY (or 6Y) is not appropriate for
Geradin [3], because of its complexity. The results are application to locating the critical points.
shown in Fig. 6 and compared with the aforemen- For tapered beams with k = I .5, which is the ratio
tioned results. As seen in this figure, the predicted of the heights at the two end cross-sections, the finite
buckling points for small values of AZ/A,, for which element solution is again compared with the results
the prebuckling deformations will be small, is in good obtained from the shooting method in Section 2 in
agreement with the results in Fig. 5. However, the Fig. 5. We notice that the two solutions are in good
results obtained by employing A’P (or glp) deviate agreement, except for minor differences for very small
away from the shooting solution as the value of AJA, values of A,/A, or short beams, for which discretiza-
increases, in contrast with the results obtained from tion for tapered beam sectional properties involves
the present A0 (or 80) formulation. As the values of relatively large errors compared with slender beams.
For the tapered beams, the average values of A,, A2
and A, at both ends are employed for the plot in
Fig. 5.
1
A F. E M
(T(+) neglected in
08
Cardona & Geradln[3]:
‘I Primary path /
0 present Solution
k/6 h /
-\ -~ Shooting Method ,,’
‘\
06 \ ,/“
‘\ /
/’
Fig. 6. The bifurcation load vs the aspect ratio for a Fig. 8. Load vs vertical tip displacement for a uniform
uniform cantilever beam. slender cantilever beam.
Analysis of lateral buckling 1369
Section A-A
2 400
“E
200
2 i
Fig. 10. Load vs lateral tip displacement of the free end for Fig. 12. End moment vs abscissa of left hinged end for a
a right-angle frame. hinged right-angle frame.
1370 Haengsoo Lee et al.
12. R. T. Shield and S. Im, Small strain deformations 0(X,, I’,) is the warping function of the St Venant solutions
of elastic beams and rods including large deflections. (v is Poisson’s ratio). The functions are harmonic in the
J. Appl. Math. Phys. (ZAMP) 37, 491-513 (1986). cross-section of the rod and satisfy the conditions
13. D. A. Danielson and D. H. Hodges, A beam theory for
large global rotation, moderate local rotation and small
rotation. J. Appl. Mech. 55, 179-184 (1988). gn, +$n,= -{X:+fv(Xy-$)jn,
14. J. C. Simo and L. Vu-Quoc, Rod model incorporating
shear and torsion-warping. In!. J. Solids Struct. 27,
-(2 + v)X, X,n,
371-393 (1991).
15. K. Schweizerhof and E. Ramm, Displacement depen-
dent pressure loads in nonlinear finite element analysis. gn,+$n,= -{X:++v(X:-Xi)}n,
Compw. Strucr. 18, 1099-l 114 (1984). 2 2
16. E. Riks, An incremental approach to the solution of
snapping and buckling problems. Inf. J. Solids Strucf. -(2 + v)X,X,n,
15, 5299551 (1979).
17. M. A. Crisfield, A fast incremental/iterative solution do do
-n,+-nn,=X2n,-X,n,
procedure that handles ‘snap-through’. Comput. Struct. ax, ax,
13, 55-62 (1981).
18. R. Kouhia and M. Mikkola, Tracing the equilibrium on the boundary of the cross-section; n, is the unit outward
path beyond simple critical points. Int. J. Numer. Meth. normal to the surface.
Engng 28, 2923-2941 (1989).
APPENDIX B
APPENDIX A
The linear transformation T(P) is defined by
Guided by the St. Venant solution for flexture of a beam,
we now assume that
u,= -veX,+fK2V(X:--X:)--,vX,X2
where
u2= -veX,+fK,v(X:-_X:)+h.2vX,X2
‘p
~,=~.;(X,+X,X:)-K;(X~+X,X:)~K,~.
(R.1)
64.1) e=Wi
Here a prime denotes differentiation with respect to X,, From this equation, it is clearly seen that T(P) reduces to
x,(X,, X,) and xI(X,, X,) are the flexure functions, and the identity tensor as P goes to zero.