POINTWISE EQUILIBRATING POLYNOMIAL ELEMENT

FOR NONLINEAR ANALYSIS OF FRAMES

By Siu Lai Chan I and Zhi Hua Zhou z

ABSTRACT: A pointwise equilibrating polynomial (PEP) element is proposed for

nonlinear analysisof frames in which each member can be modeled by one element
in most cases. The formulation of the element is based on the imposition of com-
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

patibility conditions at end nodes as well as the satisfaction of equilibrium at mid-

span. The resulting expression for this new element is accurate in describing the
force-versus-displacementrelations at the element level, is simple, and it does not
lead to a significant numerical truncating error in the computer analysis. Its im-
plementation in a nonlinear analysis program is straightforward. The accuracy of
the analysis results by the element was found to be considerably higher than its
cubic counterpart in a number of well-known examples. The limitation or incon-
venience of the method of stability function such as the separation of the solution
for tensile, compressive, and zero load cases is eliminated.

INTRODUCTION

Second-order nonlinear analysis of framed structures has been p r o p o s e d

by the advanced design codes of practice such as the Load and Resistance
Factor Design ( L R F D ) (1986) and the Australian Standard 4100 (1990) for
accurate and robust analysis of steel structures. Generally speaking, this
type of analysis will automatically include various sources of nonlinearities
and thus eliminate the c u m b e r s o m e manual methods for considering the
nonlinear effects. Extensive efforts are being m a d e by researchers including
Ingraffea and Mink (1988), Bridge et al. (1991), W h i t e et al. (1991), Chan
(1991), and many others on the d e v e l o p m e n t of practical c o m p u t e r programs
for such a type of second-order nonlinear analysis.
One of the general requirements for the acceptance of the m e t h o d by the
engineer in a typical design office is the availability of an efficient c o m p u t e r
program which can be run on personal computers, complete an analysis
within a tolerable time, achieve numerical convergency and does not require
extensive data preparation and interpretation effort. To this end, the de-
velopment of an efficient element for second order analysis is essential to
achieve these aims. The element should be capable of modeling a m e m b e r
by an element in practical situation in o r d e r to reduce the complexity and
effort to an acceptable level. The objective of this p a p e r is to propose an
efficient element under this heading.
The simplest m e t h o d for developing a nonlinear element is to extend the
cubic Hermite element to the nonlinear case by including the geometric
stiffness with the linear stiffness matrix to form the tangent stiffness matrix.
This approach has been used by many researchers [see, for example, Bar-
soum and Gallagher (1970), M e e k and Tan (1984), and Chan and Kiti-
pornchai (1987)] and have been quite successful. H o w e v e r , the results of

ILect., Dept. of Civ. and Struct. Engrg., Hong Kong Polytechnic, Hung Hom,
Kowloon, Hong Kong.
2Lect., Dept. of Civ. Engrg., Southeast Univ., Nanjing, China.
Note. Discussion open until November 1, 1994. To extend the closing date one
month, a written request must be filed with the ASCE Manager of Journals. The
manuscript for this paper was submitted for review and possible publication on April
13, 1993. This paper is part of the Journal of Structural Engineering, Vol. 120, No.
6, June, 1994. 9 ISSN 0733-9445/94/0006-1703/$2.00 + $.25 per page. Paper
No. 5976.

1703

J. Struct. Eng., 1994, 120(6): 1703-1717

 using a single element for each member were noted to be somewhat different
from the more accurate equilibrium curves obtained by using more elements.
In fact, it has been shown by So and Chan (1991) that the element over-
estimates the buckling load of a simply supported strut by 21.6%. Recently,
Albermani and Kitipornchai (1990) proposed an improved analysis tech-
nique to allow least elements to be used and improvement was made by
addition of some terms for large displacement effects.
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

An alternative method of including the effect of axial load is the method

of stability function [see, for example, Livesley and Chandler (1956), Oran
(1973), and Chert and Lui (1987)]. The method develops the element matrix
by solving the governing equilibrium equation of a beam-column under the
action of axial load. Different solutions will be obtained for the cases of
tensile and compressive loads, resulting in the loss of generality. In some
cases when the axial force is small, the function may lead to numerical
problem. Iterations are then performed at the element level in order to
determine accurately the stiffness relations between the moments and ro-
tations under a specified axial load. When the stiffness matrix is properly
derived, this method can predict accurately the buckling load by using only
one element per member. However, it relies on the solution of an equilib-
rium equation and cannot be classified as a finite element approach which
further implies that section properties cannot be varied along element length
unless the element matrix is rederived. Consequently, many finite-element
techniques such as numerical integration for inelastic analysis cannot be
directly used, at least in a straight theoretical sense.
Kondoh and Atluri (1986) developed a mixed element for the objective
of using one element to model each member. They abandoned the assumed
displaced shape for an element and formulated their element on the equi-
librium condition, which is then solved by the complementary energy method.
Similar to the method of stability function, the stiffness expression for the
tensile case is different from the compressive case. It appears that their
approach is similar in principle to the method of stability function which
also obtains the stiffness matrix by the solution of the equilibrium equation.
Fairly speaking, all these methods have their merits and limitations. For
example, the method of using the geometric stiffness to account for nonlinear
effects, although inaccurate when only a few elements are used, can be
easily used to include the instability effects due to moments (Chan and
Kitipornchai 1987; So and Chan 1991), which may pose complexity or dif-
ficulty to other currently available stiffness matrix methods. It can also be
used for numerical integration to account for spread of material yielding.
However, under the heading of nonlinear analysis of frames in which non-
linearity is dominated by axial force, the approach is not quite appropriate
for practical uses. Nevertheless, as pointed out by Bathe and Bolourchi
(1979), the cubic Hermite element exhibits stable behavior and is suitable
for general nonlinear analysis, although they have not commented on its
deficiency on modeling each member by an element.
Effort has been made by So and Chan (1991) to improve the displacement
based beam-column element for nonlinear analysis. They introduced a dummy
node at the midspan of an element and noted that the performance of the
element was significantly improved in general. However, depending on the
deflected shape of the structure, the element may not be able to model
accurately the deformation of the member in some cases and the accuracy
is thus reduced. This deficiency was reported and illustrated in an example
for the right angle frame (So and Chan 1991).
1704

J. Struct. Eng., 1994, 120(6): 1703-1717

 This paper is part of a continuing effort to develop a refined finite element
for nonlinear analysis of frames using a single element for each member in
which the second order effect is dominated by the axial load. As a prereq-
uisite for the element formulation, the displacement will not be solved
directly by the equilibrium equation, which is, however, imposed at the
midspan of the element. This measure is to transform the equilibrium con-
dition through the whole continuum via calculus to the much simpler al-
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

gebraic form of equilibrium condition and thus to maintain the generality

of the derived stiffness matrix in compressive and tensile cases. Further,
various finite numerical techniques can be applied directly to the analysis
without the need to check whether the displacement function obtained from
the equilibrium equation is still valid or violated.
In this paper, two additional constraints of equilibrium and its rate are
imposed at the midspan of an element. The obtained stiffness matrix for
the developed pointwise equilibrating polynomial (PEP) element is simple,
does not constitute odd numbers leading to serious truncating error and can
be easily incorporated into an analysis program. The need to separate the
compressive and the tensile load cases required in other rigorous methods
is eliminated and the matrix will be valid for positive, zero, and negative
axial load. Numerical performance of the new element is compared against
the method of stability function and the widely used Hermite cubic element
through a number of widely used examples for nonlinear analysis. This
explicit form for the element stiffness matrix with high accuracy and validity
irrespective of the sign of the axial load is derived in this paper. The element
represents a potential candidate for incorporation into a robust computer
program for nonlinear analysis of two and three dimensional frames per-
mitting the use of a single element for each member in most practical cases
and the application of various finite element numerical techniques for ele-
ment formation.

ASSUMPTIONS
The present theory is based on the following assumptions.

1. The element is prismatic and elastic.

2. The applied loads are conservative and nodal.
3. Warping and shear deformations are neglected.
4. Small strain but large deflection is assumed.

Assumption 4 requires that the angle between the tangent at a node and
the axis joining the two end nodes as shown in Fig. 1 is small. However,
the total angle of nodal rotation can be large. When this assumption 4 is
violated, additional elements are needed to model the member in order to
reduce the angle magnitude.

.2 .2 .,

P x

,- M1 L -I

FIG. 1. Basic Forces versus Displacements Relations in Element

1705

J. Struct. Eng., 1994, 120(6): 1703-1717

 FORMULATION OF DISPLACEMENT FUNCTION
Six conditions are i m p o s e d on the evaluation of the displacement function.
They are the compatibility condition at the two end nodes and the equilib-
rium conditions at midspan. Mathematically these conditions can be stip-
ulated as follows.
F o r compatibility
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

L
Atx = _--§
2' " v=0 (1)

L
At x = 2 ' 7) 01 (2)

L
At x = 2 ' ~) 02 (3)

For equilibrium

EIi) = Pv + MI + M2 ( L + x ) - M1; at x = (4)

Eli) = PO + M~ + M 2 . at x = 0 (5)
L '
in which v = lateral displacement; E I = flexural rigidity; M~ and M2 =
applied nodal m o m e n t s at nodes 1 and 2; L = u n d e f o r m e d m e m b e r length;
and a dot represents a differentiation with respect to the distance along
element length x.
A fifth-order polynomial will include the constraints from ( 1 ) - ( 5 ) and
therefore assumed as
v = ao + alx + a2X2 + a3X3 + a4X4 + a5x5 (6)
The coefficients are then eliminated from ( 1 ) - ( 5 ) and the final displace-
ment can be written as
v = (N, N2)(LOIL02) r (7)

in which N1 and N2 = shape function given by

A B
N1 = ~ + H-~ (8)

A B
N2 - (9)
H1 H2
and

A = - 2 0 Z + (80 - q) + 4q (10)

B = 6 - ~ ( 4 8 - q) - 2q (11)

1706

J. Struct. Eng., 1994, 120(6): 1703-1717

 H1 = 80 + q (12)

H 2 = 48 + q (13)
pL 2
q- E1 (14)

It can be seen that the displacement, v, in (7) will be converted to the

Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

cubic Hermite function when the axial load is zero (i.e. q is zero). The
discrepancy between the cubic and the present higher-order element in-
creases when the axial load is large. It is worth mentioning that equilibrium
conditions can be imposed on other locations along the element such as the
0ne-third point from the two ends and the resulting equation will be Slightly
different.

SECANT STIFFNESS MATRIX

The principle of stationary potential energy is employed to formulate the

element secant stiffness relations. The degrees of freedom will be the ro-
tations and lengtl!:ening and their corresponding moments and force are as
follows:

/"/1 = 01; /12 ~--- 02; U3 = e (15a,b,c)

F 1 = M1; F 2 = M2; F3 = P (16a,b,c)
The total potential energy functional can be expressed as the sum of the
strain energy and the external work done as
If L/2 l fL/2 P( L/2
II = ~ EA l~2 dx + E1 ~)2 dx @ 7)2 dx
, - L,2 -i , - L,2 , -

-- F l U 1 -- F2R 2 - F3u 3 (17)

By the principle of stationary total potential energy, the equilibrium con-
dition and thus the stiffness coefficients can be obtained by taking the first
variation of the functional as
OH OH Oq
8H = Ou--~+ 0q Ou; 0; i = 1, 2, 3 (18)

The final relations between the forces and displacements can therefore
be evaluated as

E1
M1 = ~ [cl(0i + 02) -~ c2(01 - - 02) ] (19)

E1
M2 : Z [C1(01 ~- 02) -- C2(01 -- 02)] (20)

P = EA [ L + bl(Oi + 02)2 + b2(O1 - 02)2] (21)

in which the coefficients ci, c2, b~, and b2 are given by

1707

J. Struct. Eng., 1994, 120(6): 1703-1717

 61 q2 23 q3
3(80) 2 + 10(80)q + -~- +
cl = H2 (22)

29 q2 11 q3
482 + 6(48)q + -ff + 4-~
c2 = Hzz (23)
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

46 23
2(80) 2 + _~ (80)q + ~-~ q2 + q3
b, = H3 (24)

66 11
2(48) 2 + (48)q + ~-~ q2 + 8 ~ q3
b2 = (25)

In the foregoing formulation, the higher-order terms are neglected.

It is of interest to compare the coeffficients from (19)-(25) against those
by other methods and the exact coefficients obtained by solving the differ-
ential equilibrium equation. Their discrepancy due to a unit rotation at one
end and over the range for q varying from -2~r 2 to 27r2 is depicted in Fig.
2. This range will be of practical interest since too high buckling loads will
not be generally encountered.
As can be seen in Fig. 2, the error for the cubic function increases rapidly
as the axial load is increased. The error for the present element is small
over the whole range. Unlike the closed-form solutions by the stability
function method, the stiffness expression for the present element is the same
irrespectively of whether the force is zero, in tension or in compression.
The stiffness coefficients for the present method and the cubic Hermitian
element are also plotted in Fig. 3 against the exact stiffness coefficients. It
can be seen the error for the cubic element is large when the axial force is
high and that the error for the present element is small for a wide range of
axial force.

-- 0.25

8 0.20

a
o0.15

~ 0.10

7~ 0.05

0.00
-0.5 -0,3 -0.1 0 0.1 0.3 0.5
Distance along Element Length, xJL
FIG. 2. Deflection along Element Length under Compressive Axial Load

1708

J. Struct. Eng., 1994, 120(6): 1703-1717

 16

8 9 .

"O~ 4 I
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

-el- ' 3rd order

5th order
-12! Exact
-16 -2 0 2

Non-dimensional Axial Force, P/P.

FIG. 3. Variation of Bending Coefficients against Tensile and Compressive Axial

Force

The element matrix in three-dimensional space can be extended directly

from the foregoing equations by considering the stiffness in the y- and z-
axes as

m2. = [Eli
-~
n
[Cln(Oln + OZ.) -- C2.(01. -- 02.)]; for n = z, y (27)

GJ + Pr 2
M, - L O, (29)

in which GJ = torsional rigidity; M, = torsional moment; 0, = twist; and

r = polar radius of gyration.

TANGENT STIFFNESS MATRIX

The tangent stiffness matrix relates the incremental change in forces to
a corresponding change in displacements. This can be obtained by taking a
variation of the secant stiffness matrix with respect to the displacement
degrees of freedom and the axial force due to which the second-order effect
is considered. This operation can be stipulated as

8zII - OuiOuj02II~Ui~lA] = kil~UigUJ = LOuj[~ + OS~oqO~uj] BuiBuj (30)

To obtain the tangent stiffness matrix from (30), the following relations
are first established. For m = 1, 2; and n = y, z
1709
J. Struct. Eng., 1994, 120(6): 1703-1717
 C;n = 2 b l n (31)
'
C2n = 2b2. (32)
Oq Gmn
- - - (33)
OOmn H

0_.qq = 0 (34)
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

00t
Oq 1
- (35)
Oe LH

Gin = C[n(Oln -~- 02n) + C2n(01n -- 02n) (36)

G2n = C;n(Oln + 02n) - - C;n(OXn -- 02n) (37)
1 [b;n(Oln + 02n)2 -t- b;,(0ln - 02n)21
H = -X2
- _ ~ (38)
n=y,z ~n

GJ + Pr 2
"q - E1 (39)

-~2
h = (40)

in which a prime represents a differentiation with respect to qn; I and ~n =

respectively, reference second m o m e n t of area and ratio of the principal
moments of area a b o u t y- and z-axes to this reference value. When n is
equal to y and z, respectively, ~n is given by
lz
(z = 7 (41)
ly
~Y = 7

The basic stiffness matrix about the axis joining the two end nodes can
be determined by substituting (31)-(40) into (30) and can be written as
E/
[~,] = --
L
6L GlzGzz GlzGly GlzG2y 0 Glz
~z(Clz + C2z) + --~ ~z(Clz - C2z) + n n n
G~z G2zGly G2zG2y 0 G2x
~(C1~ + C2~) + - -
H n n Z-H
G~y G1yG2y 0 Gly
~y(Cly + C2y) + Y ~y(Cly - CZY) + H -~
G~y Gzy
~/Cty + c2y) + --if- o --ff
symmetrical "q 0
1
L2H

(42)

1710

J. Struct. Eng., 1994, 120(6): 1703-1717

 That corresponding displacement vector is given by the following:
[U] = [OlzOezOlyO2y~be] T (43)
Note that, apart from the exact values for the coefficients in the tangent
stiffness in (42), the general expression for the matrix is slightly different
from the one by Oran (1973). The terms corresponding to the coupling of
variables about the two principal axes, k13, kx4, k23, and k24 in (43) are
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

nonzero, and that these coefficients in (36) of the paper by Oran (1973) are
zero, which is believed to be incorrect. These nonzero coefficients are due
to the fact that they are both functions of the axial load functional, q.
Further, the variables G13 and G12 in the coefficients/22 and/44 in Oran's
(1973) paper should be read G23 and G22.
The tangent stiffness matrix can finally be determined as
Nele Nele
[Kr] = Z [LI[kTI[L]T= ~ [Ll([Tlr[ke][ T] + [NI)[L] r (44)
in which [T] = transformation matrix relating the member basic forces to
the element force in the local coordinate; [L] = local to global transfor-
mation matrix; and [N] = a matrix to account for the work done by the
initial force and the translational displacements. For completeness and con-
venience of the readers, these matrices are detailed in Appendix I.

NUMERICAL PROCEDURE

The success of a nonlinear analysis depends very much on the performance

of the element as well as on the nonlinear numerical methods used. The
conventional constant load method is incapable of tracing the postbuckling
equilibrium path and a modified numerical technique such as the displace-
ment control (Batoz and Dhatt 1979), the arc-length (Crisfield 1981), the
constant work (Yang and McGuire 1985),. and the minimum residual dis-
placement (Chan 1988) methods must therefore be used in such cases. It
should be noted that the recently introduced constant work method is in
fact similar to the displacement control method with the steering displace-
ment degrees of freedom coincided with the locations of the applied loads
(Chan and Ho 1989). Consequently, if the structure snaps back at this loaded
degree of freedom, the numerical method fails to converge, as in the case
of the displacement control method, which diverges at snapback problems.
Nevertheless, when the applied loads are at several locations, the possibility
of all the displacement degrees of freedom under loads to snap back is
lowered than the single displacement control method.
In the present studies, the numerical method used is a combination of
the arc-length and the minimum residual displacement methods. The load
size is chosen to fulfill the requirement of a constant arc-distance and the
iterative scheme is to minimize the equilibrium error measured in terms of
residual displacements. Numerical studies by Chan and Ho (1989) have
shown that this mixed approach is one of the most reliable schemes currently
available and simpler to apply. Further, unlike the arc-length method, it
avoids the solution of a quadratic equation in which the root may be imag-
inary. The details of the method has been given by Chan (1988) and Chan
and Ho (1989) and will not be recapitulated. However, the procedure used
in conjunction with the tangent and secant stiffness equations derived in
this paper is outlined for clarity of the readers. The tangent stiffness rela-
tions, [KT] in (42) is used to determine the incremental displacements as
1711

J. Struct. Eng., 1994, 120(6): 1703-1717

 [Au + AkAt/] = [Kr]-I[AF + AkAIT] (45)

in which AF = applied incremental load vector for the first iteration or the
unbalanced force vector in second iteration onward; Au = corresponding
displacement increment due to this force; Af" = a reference force vector
parallel to the applied load vector; Aa = conjugate displacement solved;
AK = a load parameter determined according to the imposed constrained
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

conditions as follows.
For the first iteration in each load cycle, Ak is obtained as

arc distance
&k - for the first iteration (46)
X/(aaF(aa)

For the second iteration onward, A~ is calculated as

O(equilibrium error) O[(Au + 8kaa)~(Au + aXAa)]

= = 0 (47)
OAk OAk

or except at the first iteration in each load cycle, AF is determined as the

difference

ax - [[ aa aa V for second iteration onward (48)

] ' [[ aa a. ]]

between the applied [F] and the resistance force [R], which is computed
through (26)-(29), as
[AF] = [F] - [R] (49)

NUMERICAL EXAMPLES

The computer program G M N A F (1993) is modified to include the element

described in addition to the conventional cubic Hermite element that was
used for past analyses (Chan 1990). To demonstrate the difference between
the new and the old elements, several common and widely studied problems
are selected for reanalysis and the improvement is illustrated by using the
new element. In most cases, the new PEP element generates results close
to the exact stiffness equation using the stability function which, unfortu-
nately, involves other numerical deficiencies mentioned previously.

Two-Bar Toggle Frame

The two-bar toggle frame with rise 2 cm, length 100 cm and cross-sectional
dimensions of 1 cm by 1 cm as shown in Fig. 4 is analyzed using the cubic,
the stability function and the present refined PEP element. Each member
is modeled by a single element in all cases and the corresponding equilibrium
paths for the three methods are traced in Fig. 4.
It can be seen that the solutions using the stability function and the PEP
element are in close agreement but differ considerably from the cubic ele-
ment. This is due to the presence of a high axial force in the member which
has properly been accounted for in the method of stability function and the
PEP element, but not in the cubic element. Nevertheless, when more cubic
elements are used, the solution path obtained is basically the same as the
solution using the single PEP element.
1712
J. Struct. Eng., 1994, 120(6): 1703-1717
 2.2

0..
1.8
! 1
"0
,'--,, 1= L -I /
3 /~\\cross section = l cm x lc 4
1,4 / Y\ 1 etement / member ///
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

0.6 / cuo,eX /
o
PEP Element ~,///
Stability Function "" "
0.2
0 11 3I 5
Displacement v(cm)
FIG. 4. Postbuckling Analysis of Toggle Frame

1.5

(11
0.
/// elL = 0.01
e/L = -0,01 \ \

),
0.5 . L' 1
A = 600 inF
L = 1,000 in.
----11 ;EPiEl;~e2:nt
I = 60,000 in:
2 PEP Elements E = 10,000 psi
f

920 -1'0 0 1'0 20

Rotation | (in degree)

FIG. 5. Large-Deflection Analysis of Right-Angle Frame

Right-AngleFrame
The right-angle frame of length 1,000 in., cross sectional area 600 sq in.,
Young's modulus of elasticity of 10,000 psi arid second moment of area of
60,000 in. 4 was studied analytically by Koiter (1962) and subsequently an-
alyzed by a number of researchers using the stiffness matrix method of
analysis. The geometry of the frame is depicted in Fig. 5.
Kondoh and Atluri (1986) used two mixed elements per member to trace
the equilibrium path of the structure and accurate solutions were obtained.
Using one and two PEP elements to model each member and assumed load
eccentricities of _+0.01 of member length l, the equilibrium paths are plotted.
The solution paths by using two elements per member (or four elements
for the whole structure) coincides with the result by Kondoh and Atluri
1713
J. Struct. Eng., 1994, 120(6): 1703-1717
 (!986). The buckling loads for using one and two elements are similar and
the error for using one element increases with the magnitude of the rotation,
due to its assumption of small angle relative to the chord joining the two
nodes of the element. In most realistic cases, such a large rotation should
not exist in practice (Kondon and Atluri 1986) and therefore the present
element should be of sufficient accuracy to model one member by one
element for practical purpose. The results by stability function was found
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

to be close to the present analysis when the same number of elements are
used and were not plotted for clarity.
Also plotted in Fig. 5 is the solution path by using a single cubic element
for each member. It is obvious that the error for using the conventional
cubic element is unacceptably large, even in the determination of buckling
load.

Hexagonal Frame
The hexagonal flame of dimensions shown in Fig. 6 has been studied
extensively by many researchers. Chu and Rampetsreiter (1972) traced the
prebuckling path of the flame and Meek and Tan (1984) completed the pre-
and postbuckling path tracing via the use of the cubic element and the arc-
length method. As pointed out by Chan (1988), the use of a single element
for each member is insufficient; he models each member by two cubic
elements, obtaining a more accurate equilibrium path.
Using a single PEP element for each member, the equilibrium path is
traced and shown in Fig. 6. It was found that the curve is close to the one
by Chan (1988), who used two cubic elements per member. When a single
cubic element is used, an equilibrium path similar to the one traced by Meek
and Tan (1984) was plotted, which is noted to differ considerably from the
solution by the PEP element. This example further illustrates clearly the
remarkable performance of the proposed PEP element over the conven-
tional element, although both are displacement based elements.

90
E = 439.8 ksi
G = 159.0ksi / \ / \
A = 0.494in.' / V \
70 I,= I==0.02in.4 ~ A /
J = 0.0331in.' 24 i n . ~ / ~/
~- 60 t
• , t
. 50
"113

q 40

30

20 [ " StabilityFunction~k\ ~"

--ProposedElement ~ . J /
10
.. .. .. .. . . 3rd LOrder
. . . . .Element
.. ~ ,,, ,,,/J
0
I I I
0.5 1.5 2,5 3.5
Displacement v(in)

FIG. 6. Postbuckling Analysis of Hexagonal Frame

1714
J. Struct. Eng., 1994, 120(6): 1703-1717
 CONCLUSIONS

The Stiffness matrix of a pointwise equilibrating polynomial (PEP) ele-

ment for nonlinear analysis of two- and three-dimensional frames is for-
mulated and presented. Because it is based on as assumed displacement
shape, it is valid for compressive, tensile, and zero-axial load cases, and it
permits the use of various techniques for finite-element analysis. The for-
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

mulated element matrix is accurate that it permits the use of a single element
for each member in most practical cases. Its accuracy is believed to be higher
than other displacement-based one-dimensional elements, while simulta-
neously it does not suffer from the deficiencies of other exact elements
solved directly from the equilibrium equation (which, strictly speaking, should
not be classified as finite element, because it does not assume a displacement
or force function in their element formulations).

APPENDIX I. MATRICES

The transformation matrices [L] and [T] and the stability matrix due to
rigid-body motion [N], in (44) are given by the following:

ILl = L0
|0
L'
00
0
L0'
!
0t (50)
0 0 '
where

- DXD Y - DZ
DX
" ~ / D X 2 -+- D Z 2 "~v/DX 2 -4- D Z 2

[L'] = DY " M / D X 2 q- D Z 2 0 (51)

- D YDZ DX
DZ
~v/DX 2 + DZ z ~v/DX 2 + DZ 2

and D X , D Y , and DZ = direct cosine about the x-, y-, and z-axes, re-
spectively, as

DX = x2 - xl
(52a)
~v/(x2 - Xl) 2 + (Y2 - Ya)2 + (z2 - - Zl) 2

DY = Y2 - Ya (52b)
k/(x2 - xl) 2 + (Y2 - Y02 + (z2 - zl) 2

DZ = zz zl -

(52c)
~v/(X2 - Xl) 2 + (Y2 - Yl) 2 q- (z2 - - Zl) 2
If DY is equal to unity, the [L'} should be replaced by the following:

[L,] =
El li] o
o
(53)

1715

J. Struct. Eng., 1994, 120(6): 1703-1717

 m

1 1
0 i o 0 0 1 0 -~ 0 0 0 0
1
0 i o 0 0 0 0 -1 0 0 0 1
L
0 o -• 0 1 0 0 0 --1 0 0 0
L L
[T] = (54)
0 1 1
0 0 0 0 0 0 -- 0 i 0
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

L L
0 0 0 -1 0 0 0 0 0 1 0 0

0 0 0 0 0 1 0 0 000

[NI =

Mlz + Mzz Mly + M2y Mlz + M2z Mly + M2y

0 0 o 0 0 000
L2 L2 L2 L2
P Mu + M2z P
- 0 0 0 0 0 000
L L2 L
P Mly + Mly P
0 0 0 o -Z 00o
L
0 0 0 0 0 o 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 o 00
MI~ + M2~ Mly + M~,
000
L2 L2
P
symmetric -s o ooo
P
-- 000
L
000
0 0
o_

(55)

APPENDIXII. REFERENCES
AI-Bermani, and Kitipronchai, S. (1990). "Nonlinear analysis of thin-walled struc-
tures using least e l e m e n t / m e m b e r . " J. Struct. E n g r g . , ASCE, 116(1), 215-234.
Australian Standards AS4100-1990, steel structures. (1990). Standards Association of
Australia, Sydney, Australia.
Barsoum, R. S., and Gallagher, R. H. (1970). "Finite element analysis of torsional-
flexural stability problems." Int. J. N u m e r i c a l M e t h o d s in E n g r g . , 2, 335-352.
Bathe, K. J., and Bolourehi, S. (1979). "Large displacement analysis of three-di-
mensional beam structures." Int. J. N u m e r i c a l M e t h o d s in E n g r g . , 14, 961-986.
Batoz, J. L., and Dhatt, G. (1979). "Incremental displacement algorithms for non-
linear analysis." Int. J. N u m e r i c a l M e t h o d s in E n g r g . , 14, 1262-1267.
Bridge, R. Q., Clarke, M. J., Hancock, G. J., and Trahair, N. S. (1991). "Trends
in the analysis and design of steel building frames." Civ. Engrg. Trans., Sydney,
Australia, CE33(2), 87-94.
Chan, S. L. (1988). " G e o m e t r i c and material nonlinear analysis of beam-columns
and frames using the minimum residual displacement m e t h o d . " Int. J. N u m e r i c a l
Methods in Engrg., 26, 2657-2669.
Chan, S. L. (1991). " A generalised numerical procedure for nonlinear analysis of
frames exhibiting a limit or a bifurcation point." Int. J. Sp. Struct., 6(2), 99-114.

1716

J. Struct. Eng., 1994, 120(6): 1703-1717

 Chan, S. L., and Ho, G. W. M. (1989). "A comparative study on the nonlinear
algorithms." 3rd Int. Conf. Numerical Methods in Engrg., Theory and Applications,
Swansea, U.K., 552-565.
Chan, S. L., and Kitipornchai, S. (1987). "Geometric nonlinear analysis of asym-
metric thin-walled beam-columns." Engrg. Struct., 9,243-254.
Chan, S. L., and So, A. K. W. (1992). "Buckling behaviour and design of unre-
strained steel beams." Trans., Hong Kong Inst. ofEngrs., Hong Kong, 4, 23-26.
Chen, W. F., and Lui, E. M. (1987). "Effects of joint flexibility on the behavior of
Downloaded from ascelibrary.org by Orlando Mireles on 07/22/21. Copyright ASCE. For personal use only; all rights reserved.

steel frames." Computers and Struct., 26(5), 719-732.

Chu, K. H., and Rampetsreiter, R. H. (1972). "Large deflection buckling of space
frames." J. Struct. Div., ASCE, 98(12), 2701-2722.
Crisfield, M. A. (1981). "A faster incremental-iterative solution procedure that han-
dims snap-through." Computers and Struct., 13, 55-62.
GM-NAF: geometric and material nonlinear analysis of frames; user's reference man-
ual, 9303-03. (1993).
Ho, W. M. G., and Chan, S. L. (1991). "Semi-bifurcation and bifurcation analysis
of flexibly connected steel frames." J. Struct. Engrg., ASCE, 117(8), 2299-2319.
Ingraffea, T., and Mink, K. (1988). "Project SOCRATE: fostering a new collegi-
ality." Academic Computing, 20-63.
Koiter, W. T. (1962). "Post-buckling analysis of a simple two-bar frame." Recent
progress in applied mechanics; the Folke Odquist volume, B. Broberg, J. Hult, and
F. Niordson, Admquist and Wiksell, Stockholm, Sweden, 337-354.
Kondoh, K., and Atluri, S. N. (1986). "A simplified finite element method for large
deformation, post-buckling analysis of large frame structures, using explicitly de-
rived tangent stiffness matrices." Int. J. Numerical Methods in Engrg., 23, 69-90.
Livesley, R. K., and Chandler, D. B. (1956). Stability functions for structural frame-
works. Manchester University Press, Manchester, U.K.
Load and resistance factor design (LRFD) specification for structural steel buildings.
(1986). American Institute of Steel Construction (AISC), Chicago, I11.
Meek, J. L., and Tan, H. S. (1984). "Geometrically nonlinear analysis of space
frames by an incremental iterative technique." Computer Methods in Appl. Mech.
and Engrg., 47, 261-282.
Oran, C. (1973). "Tangent stiffness in plane frames." J. Struct. Div., ASCE, 99(6),
973-985.
So, A. K. W., and Chan, S. L. (1991). "Buckling analysis of frames using t element/
member." J. Constructional Steel Res., 20, 271-289.
White, D. W., Liew, J. Y. R., and Chen, W. F. (1991). "Second-order inelastic
analysis for frame design: a report to SSRC Task Group 29 on recent research
and the perceived state-of-the-art." Rep., CE-STR-91-12, Purdue University, West
Lafayette, Ind.
Yang, Y. B., and McGuire, W. (1985). "A work control method for geometrically
nonlinear analysis." Proc., Int. Conf. Adv. in Numerical Methods in Engrg. , Swan-
sea, U.K., 913-921.

1717

J. Struct. Eng., 1994, 120(6): 1703-1717