Professional Documents
Culture Documents
1. Introduction
The effect of the axial force within a member on the flexural stiffness that is dis-
cussed by Gere1 is commonly referred to as the beam-column effect or secondary
effect in structural engineering. This effect leads to the beam-column equations
that are used to generate the stiffness matrix for plane frames. Being a function of
the axial load this stiffness matrix may become singular for a certain compressive
load that will lead to buckling. In cases of tensile loads there will be an increase in
loading capacity known as stress hardening.
The question of what to do when it comes to space frames has been discussed
by Levy and Spillers2 who simply suggest that “if beam-columns are used in plane
frames then 3D beam-columns should be used in space frames”. This paper takes the
formulation for 3D prismatic beams a step further to 3D beam-column equations
for beams of varying cross section.
Engineering practice will often dictate that the effect of axial load on the stiff-
ness is by itself sufficient even if a 3D analysis is performed. While this might
∗ Associate Professor.
† Ph.D. Candidate.
395
September 10, 2002 9:4 WSPC/165-IJSSD 00064
be true for some cases it is not the general case and is definitely not consistent
with incremental analysis where a complete initial state of stress and not only
compression is involved. Consistency should dictate the use of the appropriate 3D
beam-column effects. Collapse analysis, for example, will require such use if mean-
ingful results of the capacity of the structure are sought. Since 3D beams have
bending about two axes and torsion in addition to axial deformations a rather com-
plex beam-column like set of interactions may occur and to disregard them might
prove quite unsafe. Predicting in advance how the coupled state of affairs will affect
the stiffness matrix, especially for cases of varying cross section, is quite a task.
It is not intended here to characterize individual interactions but to include them
all “at no cost” to the usual analysis routines by providing the appropriate linear
elastic stiffness matrix that contains their contribution. This is done by deriving
the implicit governing equations and using a finite difference scheme to solve them
for a set of discontinuities that are applied consecutively.
Geometrically full nonlinear analysis is usually performed iteratively for each
load increment with subsequent updating, of coordinates and internal stresses. Such
a type of analysis handles buckling since the determinant of the total stiffness ma-
trix can be monitored as the loading progresses. Typically for a given fixed joint load
matrix, P, the unbalanced load P0 is computed as P0 = P−NT F where N is a given
incidence matrix and F and P are vectors of internal stresses and joint loads respec-
tively. It is assumed that NT F = P describes deformed equilibrium. Incremental
displacements δ, are then computed from the system of equations (KE +KG (F))δ =
P0 . Here KE and KG (F) are the linear and geometric stiffness matrices in the global
coordinate system respectively. Coordinates are then updated and new member
stresses computed. Computation stops when the unbalanced load goes to zero.
If beam-columns are to be included then this paper assumes that their effects
simply correct the basic linear stiffness matrix, KE , and that geometric effects
are comprised of effects due to large rotations only. It is true that beam-column
effects are sometimes called geometric effects in the literature and in these cases are
interlocked with large rotations effects through large strains. This paper, therefore,
only replaces the linear elastic stiffness matrix of a tapered member to accommodate
the beam-column effects for a given set of initial forces.
Going back to 3D beam problems, the best known interaction is probably the
phenomenon of torsional (lateral) buckling that is discussed by Timoshenko.3 In
the simplest case, the presence of bending moment about the strong axis of a
beam weakens its effective stiffness about its weak axis (stress softening) leading
eventually to lateral buckling. The presence of axial compression, of course, only
makes matters worse. Biot4 in his classic book discusses another case of interaction
in which axial tension increases the torsional stiffness in a prismatic bar. More
generally, not knowing what to expect from the interaction of the entire initial
state of stress with the member stiffness matrix, only strengthens the need for a
complete 3D beam-column formulation.
September 10, 2002 9:4 WSPC/165-IJSSD 00064
2. 3D Beam-Column Equations
The 3D beam-column equations are obtained by subjecting to a small load pertur-
bation a given straight beam that is in equilibrium under given forces. The starting
point is the equilibrium of a 3D beam segment ds that is described in Fig. 2 by
Reissner’s6 force and moment equilibrium as:
ylocal
11 (5) 5 (3)
8
2
7 10 1 (1) 4 (2)
xlocal
9 3
6 (4)
12 (6)
zlocal
Fig. 1. Degrees of freedom for the expanded stiffness matrix (numbers in brackets indicate basic
D.O.F.s).
M(s+ds) t
P(s+ds)
M(s) P(s) p
t m
ds
P0 + p = 0
(1)
M0 + m + t × P = 0 .
Here P and M are the usual force and moment stress-resultants with p and m
the applied forces and moments and the prime symbol refers to the differentiation
with respect to arc length. The vector t is of course the unit tangent vector that will
also be referred to as for a straight beam. When going from initial to a perturbed
configuration the zero-order equations are assumed to satisfy equilibrium in the i0 ,
j0 , k0 directions and in the initial configuration. These are three force equilibrium
equations [Eqs. (2)–(4) below] and three moment equilibrium equations [Eqs. (5)–
(7) below] written as:
These equations [Eqs. (8)–(13)] are obtained by writing the equilibrium equations
in the perturbed configuration in the i0 , j0 , k0 directions and keeping terms which
are linear in a small parameter ε. The perturbed quantities are described as
P1 = P0 + εP̄
M1 = M0 + εM̄ (14)
i1 = i0 + εī
and the bar is used to highlight the perturbation term. For convenience this small
parameter can then be set to one in which case the terms indicated by bars represent
the full perturbation. Moreover the assumptions of small displacement theory are
September 10, 2002 9:4 WSPC/165-IJSSD 00064
is now inserted into the equilibrium equations [Eq. (1)] and terms collected in each
of the initial coordinate directions i0 , j0 , k0 . Before doing so it is important to
introduce the rotation vector, w, associated with any beam element in order to
determine the perturbed base vectors from
i1 = i0 + εī = i0 + w × i0
j1 = j0 + εj̄ = j0 + w × j0 (16)
k1 = k0 + εk̄ = k0 + w × k0 .
Here use has been made of the fact from rigid body mechanics7 that the change,
(di0 or εi) in a vector i0 undergoing a small rotation w is w × i0 .
If in the initial configuration the beam is straight and lies along the x-axis as
in Fig. 3, for the case of small rotations, w can be written as
k1 j1
i1
perturbed
0 configuration
k ,z
j0, y i0, x
initial configuration
Furthermore at this point all the member loads can be eliminated as not of interest
so that,
m0x = m0y = m0z = m̄x = m̄y = m̄z = 0
(18)
p0x = p0y = p0z = p̄x = p̄y = p̄z = 0 .
This implies that the initial axial thrust and torque, Px0 , Py0 , Pz0 and Mx0 , must be
constant and that the initial bending moments, My0 and Mz0 , can at most be linear
functions of x. This can be seen from Eqs. (2)–(7) that reduce to:
Px00 = Py00 = Pz00 = Mx00 = 0
Mz00 = −Py0
and Eqs. (8)–(13) then reduce to the following four equations:
P̄x0 − Py0 δy,xx − Pz0 δz,xx = 0
• The last three equations are coupled and must be solved simultaneously. Then
the first equation can be integrated to complete the solution.
• The last two equations are fourth order in the beam displacements (like the linear
elastic beam equations). The other two equations are second order.
• The equations themselves are linear in x since the initial moment diagrams My0 ,
Mz0 may be linear in x.
• A unit axial discontinuity (δx (L) = 1.0) while the others are held fixed, i = 1.
• A unit torsional discontinuity (θx (L) = 1.0) while the others are held fixed, i = 2.
• Four flexural discontinuities (δz,x (L) = −1.0; δy,x(L) = 1.0; δz,x(0) = −1.0;
δy,x (0) = 1.0; ) while the other held fixed, i = 3, 4, 5, 6 respectively.
Note that the four flexural discontinuities are those which are used in moment
distribution: a unit rotation about an axis of flexure is applied at one end of a
beam while the other end is held fixed. Two beam ends and 2 axes of flexure then
imply four flexural discontinuities.
The 6 × 6 stiffness matrix is first obtained from the displacements that are
computed for each discontinuity, i, from the following relations and then corrected
for deformed geometry:
Transformed components in the perturbed coordinate system P̄x1 (L), M̄x1 (L),
M̄y1 (L), M̄z1 (L), M̄y1 (0), M̄z1 (0), are computed from the components of Eq. (22)
by considering the change in the unit vectors due to a small rotation vector w.
Thus
0 (L) 0 (L)
P̄x1 (L) P̄x2 (L) P̄x3 (L) + Pz P̄x4 (L) − Py P̄x5 (L) P̄x6 (L)
M̄ 0 (L) 0 (L)
M̄x4 (L) − My
x1 (L) M̄x2 (L) M̄x3 (L) + Mz M̄x5 (L) M̄x6 (L)
0 (L) 0 (L)
M̄y1 (L) M̄y2 (L) − Mz M̄y3 (L) M̄y4 (L) + Mx M̄y5 (L) M̄y6 (L)
k = .
0 (L) 0 (L)
M̄z1 (L) M̄z2 (L) + My M̄z3 (L) − Mx M̄z4 (L) M̄z5 (L) M̄z6 (L)
M̄ 0 (0)
y1 (0) M̄y2 (0) M̄y3 (0) M̄y4 (0) M̄y5 (0) M̄y6 (0) + Mx
M̄z1 (0) M̄z2 (0) M̄z3 (0) M̄z4 (0) 0 (0)
M̄z5 (0) − Mx M̄z6 (0)
(26)
It is appropriate to state here that the 3D beam-column stiffness matrix is
generally nonsymmetrical. This is physically acceptable since equilibrium is required
in the perturbed configuration making the usual energy theorems inapplicable.
1
(P̄x )n− 12 = (kx )n− 12 (−(δx )n−1 + (δx )n ) (31)
h
1
(P̄x )n+ 12 = (kx )n+ 12 (−(δx )n + (δx )n+1 ) (32)
h
1
(M̄x )n− 12 = (kT )n− 12 (−θn−1 + θn ) (33)
h
1
(M̄x )n+ 12 = (kT )n+ 12 (−θn + θn+1 ) (34)
h
(kz )n
(M̄z )n = ((δy )n−1 − 2(δy )n + (δy )n+1 ) (35)
h2
(kz )n−1
(M̄z )n−1 = ((δy )n−2 − 2(δy )n−1 + (δy )n ) (36)
h2
(kz )n+1
(M̄z )n+1 = ((δy )n − 2(δy )n+1 + (δy )n+2 ) (37)
h2
(ky )n
(M̄y )n = ((δz )n−1 − 2(δz )n + (δz )n+1 ) (38)
h2
(ky )n−1
(M̄y )n−1 = ((δz )n−2 − 2(δz )n−1 + (δz )n ) (39)
h2
(ky )n+1
(M̄y )n+1 = ((δz )n − 2(δz )n+1 + (δz )n+2 ) (40)
h2
will finally yield the following four finite difference equations corresponding to
Eq. (20).
! ! !
(kx )n+ 12 (kx )n− 12 (kx )n+ 12 (kx )n− 12
Xn+1 − + Xn + Xn−1
h2 h2 h2 h2
! ! ! !
Py0 Py0 Py0 Pz0
− Yn+1 + 2 Yn − Yn−1 − Zn+1
h2 h2 h2 h2
! !
Pz0 Pz0
+2 Zn − Zn−1 = 0 (42)
h2 h2
! ! !
(kT )n+ 12 (kT )n− 12 (kT )n+ 12 (kT )n− 12
θn+1 − + θn + θn−1
h2 h2 h2 h2
! ! ! !
My0 My0 My0 Mz0
− Yn+1 + 2 Yn − Yn−1 − Zn+1
h2 h2 h2 h2
! !
Mz0 Mz0
+2 Zn − Zn−1 = 0 (43)
h2 h2
September 10, 2002 9:4 WSPC/165-IJSSD 00064
! ! ! !
Py0 Mz0 Mz0 Mz0 Py0 Mx0
− 2 θn+1 + 2 θn − − θn−1 + Yn+2
h h h2 h2 h 2h3
! ! !
Mx0 Mx0 Mx0
−2 Yn+1 + 2 Yn−1 − Yn−2
2h3 2h3 2h3
! !
(ky )n+1 2(ky )n 2(ky )n+1 Px0
− Zn+2 + + + 2 Zn+1
h4 h4 h4 h
!
(ky )n−1 2(ky )n (ky )n+1 2Px0
+ − + − − 2 Zn
h4 h4 h4 h
! !
2(ky )n 2(ky )n−1 Px0 (ky )n−1
+ + + 2 Zn−1 − Zn−2 = 0 (44)
h4 h4 h h4
! ! ! !
Pz0 My0 My0 My0 Pz0 Mx0
+ 2 θn+1 − 2 θn + − θn−1 + Zn+2
h h h2 h2 h 2h3
! ! !
Mx0 Mx0 Mx0
−2 Zn+1 + 2 Zn−1 − Zn−2
2h3 2h3 2h3
! !
(kz )n+1 2(kz )n 2(kz )n+1 Px0
+ Yn+2 + − − − 2 Yn+1
h4 h4 h4 h
!
(kz )n−1 2(kz )n (kz )n+1 2Px0
+ − + + 2 Yn
h4 h4 h4 h
! !
2(kz )n 2(kz )n−1 Px0 (kz )n−1
+ − − − 2 Yn−1 + Yn−2 = 0 . (45)
h4 h4 h h4
Here
n =
segment number (node count starts at zero)
h =
space length
(kx )n = EAn =
axial stiffness at node n
(kT )n = GIxn =
torsional stiffness at node n
(ky )n = EIyn =
bending stiffness at node n
(kz )n = EIzn =
bending stiffness at node n
An , Ixn , Iyn , Izn =
area of cross section, polar moment of inertia and moments
of inertia about y and z respectively
E = Young’s modulus
G = Shear modulus
X, Y , Z = δx , δy , δz .
September 10, 2002 9:4 WSPC/165-IJSSD 00064
5. Numerical Example
This section addresses two points for the purpose of establishing some level of
confidence to the finite difference technique that has been presented. The accuracy
of a computed stiffness matrix is discussed and the accuracy of obtaining buckling
loads compared to Timoshenko’s3 closed form results.
The stiffness matrix of a three dimensional beam element with no initial loads
is first computed. Then it is compared to that obtained via the flexibility method
where unit loads are applied to a statically determinate basic element. A pro-
gram coded in FORTRAN accommodates the finite difference equations [Eqs. (38)–
(41)], boundary conditions and fictitious points solving for the displacements at
each nodal point. With these displacements at hand first and second derivatives
are approximated using central differences and stiffnesses computed according to
Eq. (23). The flexibility coefficients and the inverse of the flexibility matrix were
computed using the symbolic algebra program, MAPLE.
For a beam with base dimensions of 0.22 m × 0.6 m, a length of 10.0 m, top
2 2
dimensions of 0.2 m × 0.4 m, E = 2 × 107 t/m , and G = 8 × 106 t/m the stiffness
matrix is given in Fig. 4 and each element contains three values. The top value is
obtained using finite differences, the middle value (the exact value) comes from the
flexibility matrix and the third value is the percentage difference.
The terms in the stiffness matrix were calculated using 50 segments. This choice
was made by studying the convergence characteristics of the stiffness elements as
the number of segments was increased. A convergence plot for K44 is given in Fig. 5
below. It is seen that results with less than 20 segments may be inaccurate. With
20 segments the accuracy is about 0.3% and with 50 segments about 0.06%.
Timoshenko3 provides a closed form solution of the buckling compressive load
for a column of varying moment of inertia. An example having top dimensions of
2063.45
(2063.48) 0 0 0 0 0
(−0.001%)
1201.06
0 (1201.01) 0 0 0 0
(+0.004%)
22924 . 4
8224. 4
0 0 (22921.1) 0 (8240.7 ) 0
(+0.01%) (−0.19%)
K =
3595.62 1508.8
0 0 0 ( 3597 . 88 ) 0 (1510. 8 )
(−0.06%) (−0.13%)
8238.3 11878.7
0 0 (8240.7 ) 0 (11893.8) 0
(−0.03%) (−0.13%)
1508.8 2544.89
0 0 0 (1510.8 ) 0 (2546.79 )
− (+0.07%)
( 0. 13 %)
3.8
3.6
3.4
stiffness ∗ 10e-3
3.2
3
2.8
2.6
2.4
2.2
2
0 10 20 30 40 50
No. of segments
3.5
3
determinant ∗ 10e-4
2.5
2
1.5
0.5
0
0 0.25 0.5 0.75 1 1.25
compressive load ∗10e-2
T I1 /I2
2π 1.0
2.076π 0.906
2.24π 0.823
2.376π 0.683
2.629π 0.572
2.794π 0.48
2.96π 0.41
3.245π 0.301
3.527π 0.197
4.0π 0.1066
4.5
T=0 (classical 2D beam-column) M0xL
4 T=
Normalized Bending Stiffness, k
T=3 E I yIz
3.5
T=4
k=KyL/(EIz)=KzL/(EIy)
3
T=5
2.5
2
T=6
1.5
1
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Normalized Compression Load, P*
P* = Px0L2 /(EI y )
Finally a set of interactions, for a uniform member, showing the effect of twist
on the classical 2D beam-column curves (effect of compressive loads upon bending
moments) are presented in Fig. 7.
6. Conclusions
General nonlinear analysis programs for space frame analysis that possess tangential
stiffness matrices that separate large rotations from 3D beam-column effects may
now incorporate, without difficulty, cases of variable stiffness elements by inserting
the coded subroutine of this paper in place of the usual linear stiffness matrix.
These nonlinear analysis programs will thus possess the ability of determining
buckling loads of rather intrinsic interactions that may result from the 3D states
of stress. The effect of a state of stress (at any increment after convergence) on
the elastic stiffness matrix that is to be used in the next increment may now be
accounted for since this paper can accommodate this general state of initial stress
in its beam-column formulation.
The finite difference formulation for the solution of the coupled 3D beam-column
equations proved rather accurate in determining both the stiffness matrix and buck-
ling loads.
This paper handled examples of beams with linearly varying cross sectional
dimensions but the formulation presented in Eqs. (38)–(41) is quite general for
handling varying axial, torsional and bending (in both directions) stiffnesses.
References
1. J. M. Gere, Moment Distribution, D. Van Nostrand Co., Princeton, New Jersey, 1963.
2. R. Levy and W. R. Spillers, Analysis of Geometrically Nonlinear Structures, Chapman
and Hall, New York, 1994.
3. S. P. Timoshenko, Theory of Elastic Stability, McGraw Hill, New York, 1936.
4. M. A. Biot, Mechanics of Incremental Deformation, John Wiley and Sons, New York,
1965.
5. M. Eisenberger, “An exact element method,” Int. J. Num. Meth. Eng. 30, pp. 363–370,
1990.
6. E. Reissner, “Variational considerations for elastic beams and shells,” J. Eng. Mech.
Div. Proc. ASCE 88(EM1), pp. 23–57, 1962.
7. H. Goldstein, Classical Mechanics, Addison-Wesley Publishing Company, Reading,
MA, 1950.