ARTICLE IN PRESS

Engineering Structures ( ) –

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Second-order elastic finite element analysis of steel structures using a single

element per member
C.K. Iu ∗ , M.A. Bradford
Centre for Infrastructure Engineering and Safety, School of Civil and Environmental Engineering, The University of New South Wales, UNSW Sydney, NSW 2052, Australia

article info abstract

Article history: Finite element frame analysis programs targeted for design office application necessitate algorithms
Received 19 October 2009 which can deliver reliable numerical convergence in a practical timeframe with comparable degrees of
Received in revised form accuracy, and a highly desirable attribute is the use of a single element per member to reduce computa-
23 April 2010
tional storage, as well as data preparation and the interpretation of the results. To this end, a higher-order
Accepted 26 April 2010
Available online xxxx
finite element method including geometric non-linearity is addressed in the paper for the analysis of elas-
tic frames for which a single element is used to model each member. The geometric non-linearity in the
Keywords:
structure is handled using an updated Lagrangian formulation, which takes the effects of the large trans-
Bowing lations and rotations that occur at the joints into consideration by accumulating their nodal coordinates.
Buckling Rigid body movements are eliminated from the local member load–displacement relationship for which
Elastic the total secant stiffness is formulated for evaluating the large member deformations of an element. The
Finite element influences of the axial force on the member stiffness and the changes in the member chord length are
Frame analysis taken into account using a modified bowing function which is formulated in the total secant stiffness re-
Geometric non-linearity lationship, for which the coupling of the axial strain and flexural bowing is included. The accuracy and
Snap-through efficiency of the technique is verified by comparisons with a number of plane and spatial structures, whose
structural response has been reported in independent studies.
Crown Copyright © 2010 Published by Elsevier Ltd. All rights reserved.

1. Introduction instability caused by axial compression, bowing of the deformed

Issues concerning geometric non-linearity and elastic buckling
in framed structures arise because of many factors. Economy and
optimisation of the material weight to unit area are important
in contemporary engineering structures, and minimisation of the
‘‘carbon footprint’’ of the structure is now crucial and essential
in its design. In steel structures, this inevitably results in consid-
erations in structural analysis in which geometric non-linearities
are produced by changes in geometry involving large deflection,
snap-through buckling, pre- and post-buckling as well as axial
shortening due to the effects of member bowing, since slender
members usually satisfy these economic and optimisation con-
straints. Because of this, and with the evolution of steel framed
structures with slender members and complicated frame topolo-
gies, much research has been devoted over the last half century to
the non-linear analysis of frames which are prone to the effects of
elastic instability.
In 1954, Masur [1] computed the critical loads of rigid-jointed
trusses by considering the equilibrium in which member bowing
deflections were formulated. By taking into account the effects of
∗ Corresponding author. Tel.: +61 2 9385 5029; fax: +61 2 9385 9747.
E-mail addresses: jerryiu@unsw.edu.au, iu.jerryu@gmail.com (C.K. Iu).
0141-0296/$ – see front matter Crown Copyright © 2010 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2010.04.033
members and finite deflections, Saafan [2] provided an approach
for the non-linear analysis of frames in 1963. Using the differ-
ential equation of equilibrium of a member in the frame, finite
deflection stiffness coefficients were presented in the form of
stability functions to account for buckling of a member, in the
formulation of Masur. Turner et al. [3] provided a pioneering pa-
per in 1960 in which a step-by-step stiffness procedure was de-
veloped, for which inherent difficulties encountered in non-linear
problems involving large deflections and initial stresses associ-
ated with thermal gradients were overcome. For the non-linear
stiffness analysis of prismatic elastic members reliant on stabil-
ity functions, Connor et al. [4] formulated a system of non-linear
load–displacement relationships in 1968, which were treated by
successive substitution and Newton–Raphson iteration. In 1972,
Chu and Rampetsreiter [5] studied the large deflection of space
frames using an analysis in which stiffness coefficients were de-
rived from Eulerian differential equation of a prismatic member in
the form of stability functions; the large deflection analysis needed
for the space frames was made with recourse to a constant-load
Newton–Raphson method. Oran [6,7] derived the tangent stiffness
for plane and space frames in deference to stability function for-
mulations, whose member bowing functions stemmed from the
work of Saafan [2], to tackle the geometric non-linearities pro-
duced by large deflections. In the later work by Oran and Kassi-
mali in 1976 [8], the earlier studies of Oran [6,7] were extended to

Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
 ARTICLE IN PRESS
2 C.K. Iu, M.A. Bradford / Engineering Structures ( ) –

investigate the large deflections of framed structures under static

and dynamic loads, in which the coupling effect in the tangent stiff-
ness matrix between axial force and bending was considered in
detail.
The techniques used in the aforementioned studies gener-
ally handled large deflections and instabilities in elastic framed
structures based on stability functions in an elastic beam–column
approach. Although in this treatment of non-linearity the load– Fig. 1. Equilibrium condition and degrees of freedom of an element about z axis.
displacement relationship is somewhat complex, it requires fewer
elements than a standard non-linear finite element analysis, for 2. Finite element formulation
which each member of the frame may need to be discretised
into several elements to achieve sufficient accuracy. Gallagher and The vector of deformations along the element is taken as
Padlog [9] provided an alternative approach based on introduc- u = {u, v, w, ϕ}T , which comprises of the deformations u in the
ing non-linearity into the finite element method in lieu of using x-direction, v in the y-direction, w in the z-direction and the
stability functions. Employing a cubic interpolation polynomial twist φ about the x-axis. The member is idealised in a finite
for the transverse displacement, Mallet and Marcel [10] engaged element formulation from the standpoint of having only one el-
computational procedures for a finite element non-linear pre- and ement, for which the displacement functions are chosen as be-
post-buckling analysis, while Jennings [11] also relied on the con- ing required to satisfy the kinematic condition containing the P–δ
ventional finite element method to develop a non-linear analysis induced second-order moments as shown in Fig. 1. The nodal
of a structure experiencing changes of its geometry; second- displacement functions for the element are based on nodal dis-
order iteration or predictor–corrector methods were used to elim- placements in co-rotational coordinates, so that the dependent
variables for the transverse displacements v and w are replaced
inate the drift-off error due to the geometric non-linearities.
instead by the nodal rotations θz and θy about the z and y-axes
Powell [12] reported a thorough discrete finite element proce-
respectively. The element about z-axis is only shown in Fig. 1 for
dure, in which small strain but large deformation was assumed.
brevity. These rotations define the transverse displacements in the
Continuing work during the 1980’s included that of Wood and element stiffness formulation as follows.
Zienkiewicz [13], Meek and Tan [14], Chajes and Churchill [15], Linear functions are assumed for the axial deformation and
Oliveto and Cuomo [16], Chan and Kitipornchai [17], and later twist, and assumed no coupling effect between axial and torsional
Kitipornchai et al. [18] evolved their method into second-order in- effect, so that
elastic analysis. This type of research still extended in 1990’s. For
u = (1 − ξ )u1 + ξ u2 , (1)
example, Oda and Usami [19] developed the second-order elastic
analysis of plane frames. φ = (1 − ξ )θz1 + ξ θz2 , (2)
Finite element methods using one element per member for in which u = u1 at x = 0, u = u2 at x = L are the axial nodal
second-order frame analysis are very desirable for large-scale and deformations, φ = θx1 at x = 0, φ = θx2 at x = L are the twist
dense frame topologies because they both reduce the computa- nodal deflections, and where
tional effort and allow for rapid interpretation of the force resul- ξ = x/L. (3)
tants during the design process. Al-Bermani and Kitipornchai [20] This paper uses a fourth-order displacement function [27,28] to
firstly presented a method which is capable of accurately mod- simulate the member bowing behaviour; this necessitates the use
elling large deflections in structures such as transmission tow- of an additional equilibrium condition which constitutes a sec-
ers using only one element per member. Chan and Zhou [21,22] ondary or statical boundary condition, as illustrated in Fig. 1. The
developed a fifth-order finite element formulation to simulate primary boundary conditions are
second-order effects in a member using one element with initial v = 0 and ∂v/∂ x = θz1 at x = 0, (4)
imperfections, who implemented their non-linear analysis [23] to
practical structural design. Further, Gu and Chan [24] extended
v = 0 and ∂v/∂ x = θz2 at x = L (5)
to include flexural and torsional buckling. Izzuddin [25] subse- for the transverse displacement v in the y-direction, while the
quently formulated a fourth-order displacement-based finite el- equation of bending given by
ement method for structures under thermal loads. In 2000, Liew ∂ 2v
et al. [26] formulated the element stiffness matrix using stability EIz = P v − Mz1 (1 − ξ ) + Mz2 ξ (6)
functions (e.g. as in [2,7,8]) with initial imperfections, as in [22].
∂ x2
produces
In structural engineering design, many different trial frame
topologies are usually assessed in order to optimise the engineer- ∂ 2v Mz2 − Mz1 1
ing design, and so the numerical analysis must be as efficacious and EIz = Pv + at ξ = (7)
∂ x2 2 2
user-friendly as possible. This is best achieved using one element and which leads to the deformation
per member, which can represent different types of geometric non-
4 (24 + q) ξ 2 (48 + 5q) ξ 3 2qξ 4
 
linearities in the frame. As an evolution of the authors’ work in [27], v = ξ− + − Lθz1
the present paper proposes a robust, efficient and accurate finite el- 48 + q 48 + q 48 + q
ement formulation for second-order elastic analysis using one el- − (48 − q) ξ 2 3 (16 − q) ξ 3 2qξ 4
 
ement per member. It is able to include non-linearities associated + + + Lθz2 (8)
48 + q 48 + q 48 + q
with the large deflections, snap-through buckling, pre- and post-
buckling and the effect of member shortening due to bowing. The in which
approach is validated by being able to reproduce the structural re- q = PL2 /EIz (9)
sponse of plane and spatial structures which have been reported is an axial load or stability parameter and EI z the flexural rigidity
elsewhere, and these validations are used to provide a platform about the z-axis. The transverse displacement w in the z-direction
for discussion of the attributes of the formulation proposed herein. can be formulated in a similar fashion. It should be noted that when
Detailed descriptions of pertinent aspects of the formulation are the stability parameter q = 0, the transverse displacements reduce
discussed within each of these numerical examples. to the conventional cubic deformations.

Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
 ARTICLE IN PRESS
C.K. Iu, M.A. Bradford / Engineering Structures ( ) – 3

configuration as shown in Fig. 2. The approach is therefore an up-

dated Lagrangian formulation. It is worth mentioning that the ax-
ial force P in Eq. (12) is another form of the dependent variable q,
which leads to a complete and symmetric bowing function in the
following secant stiffness derivation. Since the strain energy func-
tional in Eq. (11) depends not only on the dependent variables uk
but also on the load parameter q, invoking Castigliano’s first theo-
rem of strain energy (as also given by Oran [6,7]) produces

∂U ∂U ∂q
Ks = + . (13)
∂ uk ∂ q ∂ uk
Using this, the secant stiffness formulation can be obtained from

∂U EIα
Mα 1 = = (C1 θα1 + C2 θα2 ) (α = y or z ) (14)
∂θα1 L
∂U EIα
Fig. 2. Member deformations in current configuration. Mα 2 = = (C1 θα2 + C2 θα1 ) (α = y or z ) (15)
∂θα2 L
3. Stiffness formulation for fourth-order beam–column ele- in which
ment
9216 + 3456q/5 + 68q2 /5 + q3 /105
C1 = (16)
The internal strain energy U caused by the axial strain εx and (48 + q)2
twist strain γx along the beam–column continuum can be accumu-
lated by the integration of δ UA = E εx δεx and of δ UT = Gγx δγx over 4608 + 576q/5 + 2q2 + q3 /42
C2 = ; (17)
the domain length x ∈ [0, L], in which E is the elastic modulus and (48 + q)2
G the shear modulus. Hence,
Z Z Z  and from
U = E εx d εx + Gγx dγx dV ∂U ∂U ∂q
εx γx
P = P1 − P2 = +
V
Z ∂e ∂q ∂e
1
( )
E εx2 + Gγx2 dV


= (10) e X
b1 (θα 1 + θα 2 ) + b2 (θα 1 − θα 2 )
2 2
 
2 V
= EA +
L α=y,z
which using an appropriate expansion of the Green–Lagrange !
strain tensor becomes e X
EA
Z
P
Z
P
Z
EIz = EA + Cb (18)
U = u02 dx + v 02 dx + w 02 dx + L α=y,z
2 L 2 L 2 L 2
Z Z Z in which
EIy GJ
× v 00 dx + w002 dx + φ 02 dx, (11)
L 2 L 2 L
6 × 482 /5 + 18 × 48q/5 + 128q2 /35 + q3 /40
b1 = (19)
in which EA is the axial rigidity, EI y and EI z the flexural rigidities (48 + q)3
and GJ the torsional rigidity.
2 × 482 + 14 × 48q/5 + 66q2 /35 + 11q3 /840
The elastic stiffness relationship for a general fourth-order b2 = . (20)
element is derived from the total potential energy Π in terms of (48 + q)3
the displacements; the total potential for non-linear analysis being In Eq. (18), e = 1u = u1 − u2 is the axial shortening of the
the sum of the internal strain energy U in Eq. (11) and of the work original chord length to the deformed chord length, as illustrated
done, which can be written as in Fig. 2 with respect to the axial member load P as a dependent
Z Z Z
EA P P EIz variable, b1 and b2 are bowing functions representing the effect
Π = u02 dx + v 02 dx + w02 dx + of bowing on coupling between the axial load and the rotational
2 L 2 L 2 L 2
Z Z Z deformations and the bowing term Cb is the length correction fac-
EIy GJ
× v 002 dx + w002 dx + φ 02 dx − uTk fk (12) tor due to the effect of member bowing. The bowing function b2
L 2 L 2 L of Eq. (20) is equivalent to the corresponding function of Chan and
where uk and fk are column vectors of the displacements and Zhou [21]; the bowing functions b1 and b2 are functions of q and can
internal applied forces with respect to the corresponding freedom; be regarded as being complete and symmetric containing a con-
uk = {1u, θz1 , θz2 , 1θx , θy1 , θy2 }T in which 1u = u1 − u2 stant term, whereas the formulation in [27] contains only a simple
and 1θx = φ1 − φ2 . The higher-order element proposed is quadratic function of q and is not complete and symmetric.
formulated in the co-rotational coordinates shown in Fig. 2, to The higher-order stiffness formulation is necessary but not suf-
which the deformed chord length of the element is referred for ficient for the analysis using an element capturing the second-
the derivation of the second-order stiffness formulation of this order effect of a member. For a formulation with one element per
paper; the incremental transverse displacements 1v and 1w are member, large deformations of the member necessitate the total
eliminated in the subsequent derivation, being separated from the load–displacement relationship for formulating the stiffness ma-
natural member deformations in the element stiffness formulation trices. Eqs. (14)–(20) thus rely on the total deformations (excluding
and accumulated as rigid body movements in the coordinates of rigid movements) to which the initial configuration is referred, as
the structure. shown in Fig. 2. It is worth nothing that, in accordance with Iu and
The principle of virtual displacements is exploited in the pre- Bradford [27], the large deformation behaviour of a member, which
sent paper to derive the secant stiffness matrix Ks and tangent stiff- is discretised into several elements, is handled within the analy-
ness matrix Kt , which are formulated with reference to the current sis of the complete structure, so that natural nodal deformations

Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
 ARTICLE IN PRESS
4 C.K. Iu, M.A. Bradford / Engineering Structures ( ) –

along the member are considered in the secant stiffness matrix Ks

with reference to the total load–displacement relationship, whilst
the large deformations of the member in its nodal response are
treated as rigid body motion updates and accumulations in the co-
ordinate system in the analysis of the complete structure. In short,
the present analysis adopts the higher-order element with its total
deformations in order to capture large deformations at the element
level, and it is the total secant stiffness formulation which is used to
evaluate the member resistance in the member coordinate frame.
Because of the large deformations and the participation of the
total axial load parameter q in the element formulation, using
a standard solution formulation would make convergence in the
structural analysis difficult and slow. In addition, the member axial
force parameter q results from the bowing functions b1 and b2 ,
which in turn are functions of q. And because of these, Eq. (18) can
be written as
 
I e
Q (q) = q − Fig. 3. Non-linear solution procedure for the arc-length method.
AL2 L
X and N is a stability matrix to allow for the work done by rigid
b1 (θα 1 + θα 2 )2 + b2 (θα 1 − θα 2 )2 ,

− (21) body motions or changes in the geometry of the structure. The
α=y,z
geometry of the structure is then updated by accumulating both
in which q is the only unknown when equilibrium is being imple- the natural member deformations and the rigid body movements
mented iteratively in the element formulation; this has also been in an updated Lagrangian approach. Because of the non-linear
noted by Chan and Zhou [22] and Kassimali [29]. nature of governing equation of Eq. (12), an incremental-iterative
If qi is an approximate solution of Eq. (21), then using a Taylor procedure is needed to trace the non-linear equilibrium path.
expansion of this equation to first order produces
Q (qi + 1qi ) = Q (qi ) + (∂ Q /∂ q) · 1qi (22) 4. Non-linear solution procedure
while from Eq. (21)
An incremental-iterative solution procedure is developed in
∂Q I X
this section, based on various load techniques, in order to trace the
= 2− (∂ b1 /∂ q) (θα1 + θα2 )2
∂q AL α=y,z
non-linear load–deformation equilibrium path in each load incre-
ment and to handle loading decrements after the limit point. Such
+ (∂ b2 /∂ q) (θα1 − θα2 )2 = H

(23) techniques are the arc-length method [30,31] and the minimum
so that the expression for H can also be used in the stiffness coef- residual displacement method [32]. Fig. 3 illustrates schematically
ficients in the tangent stiffness formulation (given in Appendix). It the arc-length method; the verification studies in this paper make
is interesting to note that the bowing function b1 is stationary with use of both the arc-length method and the residual displacement
respect to q. An updated value for q is thus obtained using method.
The equation of equilibrium at a solution point is expressed as
Q (qi )
qi+1 = qi + 1qi = qi − . (24) f − KS u = KT 1u, (28)
H
The tangent stiffness matrix can be obtained from the second in which KS is the secant stiffness in global coordinates, KT is the
derivative of the total potential function in Eq. (12) with respect tangent stiffness matrix given in Eq. (27) and f is the applied load
to the dependent variables uk and the load parameter q, which vector. Eq. (28) can then be rewritten in incremental form as
contains the second-order P-δ effect. When the external work done 1f = KT 1u, (29)
is linear, the second derivative of U with respect to uk produces the
tangent stiffness matrix in the form where 1f is the prescriptive load increment or the unbalanced
force and 1u the incremental displacement due to a load incre-
∂ 2Π ∂ ∂U ∂ ∂U ∂q ment (i = 1) or an unbalanced force (i 6= 1).
   
Kt = = + . (25)
∂ uj ∂ uk ∂ uj ∂ uk ∂ uk ∂ q ∂ uj For a variable load method, a constraint equation should be im-
posed in the incremental equilibrium equation (29) as
The tangent stiffness matrix for the fourth-order element, which
relates the incremental deformation to the corresponding external 1λ1f̄ = 1λKT 1ū (30)
loads imposed on the element in the member coordinate system, is
then given in Box I in which I is the second moment of area about and this is adjusted according to different constraint conditions
the relevant axis for which second-order effects are considered, germane to the relevant iterative-control scheme. The force vector
ζy = Iy /I , ζz = Iz /I, and the coefficients Gαi and H are given in 1f̄ in Eq. (30) is of arbitrary length and parallel to the total applied
Appendix. When the axial force parameter vanishes, the terms Gα i load vector, and 1ū is the displacement vector conjugate to 1f̄
also vanish and the matrix in Eq. (26) (see Box I) reduces to that for based on the current tangent stiffness formulation. The load cor-
a conventional cubic element. rection factor 1λ in Eq. (30) engages two functions so as to adjust
For the global coordinate system, the tangent stiffness matrix both the load increment and the iteration scheme and it changes
KT is represented as at different load increments and iterations.
X X When Eqs. (29) and (30) are superimposed, the resulting
LKe LT = L TT Kt T + N LT ,


KT = (27) incremental equilibrium equation is
elements elements
1f + 1λ1f̄ = KT (1u + 1λ1ū) .


(31)
where T is a transformation matrix which relates the member
forces to the element forces in the local coordinate system, L is At the first iteration (i = 1), the load corrector factor 1λi controls
a transformation of the local coordinates to global coordinates, the load increment. The load increment at the nth load cycle is then

Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
 ARTICLE IN PRESS
C.K. Iu, M.A. Bradford / Engineering Structures ( ) – 5

 A 1

Gy1 Gz1 Gy2 Gz2 
+ 0
 I L2 H LH ! LH LH LH 
G2y1
   
Gy1 Gz1 Gy1 Gy2 Gy1 Gz2
ζy C1 + ζ y C2 +
 
 0 

 H H H H 
 
G2z1
  
Gz1 Gy2 Gz1 Gz2
 
EI  ζz C1 + 0 ζz C2 +
 
Kt = (26)

 H H H 
L 
η 0 0

 ! 
G2y2
 
Gy2 Gz2
symm. ζy C1 +
 
 

 H H 
 2
 
G
 
ζz C1 + z2
H

Box I.

given by multiplying the total load by the load correction factor at derived from minimisation of the displacement error 1uni+1 +
the first iteration as 1λi+1 · 1ūi+1 . This constraint condition can therefore be written as
1fni = 1λi f. (32) ∂ 
T

1uni+1 + 1λi+1 1ūi+1 1uni+1 + 1λi+1 1ūi+1 = 0,
For the arc-length method, the load correction factor at the first ∂ 1λi+1
iteration (i = 1) for the load increment is defined by the arc-length (40)
S, which can be computed from the geometric scalar mean of the and hence the load correction factor is
incremental displacements at the first iteration according to
1uTi+1 1ūi+1
1λi+1 = − .
q
(41)
S 1uT1 1u1 1ūTi+1 1ūi+1
1λi = √ = √ (i = 1). (33)
1ūT 1ū 1ūT 1ū The incremental displacements in the global coordinate system ob-
The incremental displacements can be determined from the tained from Eq. (36) can be transformed to the member deforma-
tangent stiffness equation tions 1unei using

1uni = K− 1unei = LT 1uni (42)

T 1fi
1 n
(34)
once the load increment is known. The determinant of the tangent and the incremental member force 1Rnei using
stiffness matrix is negative on the unloading path, and when 1Rnei = Ks 1unei (43)
this occurs the load correction factor 1λ is negative, and so the
incremental applied force in Eq. (32) is negative and the load is in which Ks is the incremental stiffness matrix evaluated at 1unei .
reduced on the unloading path. After the load increment at the After determining the member forces, the incremental member
first iteration is obtained, the load correction factor is then used force is accumulated to produce the total member force in the
to determine the total load from the equation global coordinate system as
Rni = TL1Rnei . (44)
fni = fn−1 + 1λi 1f̄ (35)
The unbalanced force 1fni+1 at the second iteration (i = 1) is ob-
and the total displacement from the equation
tained as
uni = un−1 + 1uni + 1λi 1ū. (36) 1fni+1 = fni − Rni (45)
In this case, the load correction factor 1λi+1 (i ≥ 1) in Eqs. (35) where fni and Rni are the total applied load and total member force
and (36) governs the iteration scheme, which differs for different at the nth loading cycle as given in Eqs. (36) and (44) respectively.
constraint equations. For the arc-length method, the arc-length S This incremental procedure is repeated until an equilibrium solu-
changes for the second and subsequent iterations, so that after the tion is achieved, or until divergence is detected according to the
first iteration the constraint equation is written as tolerance criteria
1uTk 1uk 1fTk 1fk
T
S 2 = uni + 1uni+1 + 1λi+1 1ūi+1

≤ εtol and ≤ εtol (46)

uTk uk fTk fk
uni 1uni+1 + 1λi+1 1ūi+1 .


× + (37)
Eq. (37) may be expanded to produce the load correction factor where the preselected tolerance εtol = 10−3 is used in the present
1λi+1 for the arc-length method from the quadratic equation [30] paper; smaller tolerances are more demanding of computational
effort.
α1 1λ2i+1 + α2 1λi+1 + α3 = 0 (38)
5. Numerical verifications
in which
α1 = 1ūT 1ū; The proposed analysis in this paper is validated firstly by
T analysing a cantilever undergoing large tip displacements. A two-
α2 = 2 uni + 1uni+1 1ū;

(39) bar toggle frame and a two-member right-angled frame are also
T
investigated, in which most types of elastic instability occur.
α3 = uni + 1uni+1 uni + 1uni+1 − S 2 .


Further, a hexagonal spatial frame is studied to investigate its
In order to avoid doubling back on the original load–deflection snap-through buckling, pre- and post-buckling as well as its large
path, the positive root of Eq. (38) is chosen. deformation behaviour. Finally, a three-dimensional reticulated
Another efficient iterative scheme is the minimum residual shell structure containing 168 members is investigated to demon-
displacement method [32], for which the constraint equation is strate the capacity of the approach proposed. In the modelling

Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
 ARTICLE IN PRESS
6 C.K. Iu, M.A. Bradford / Engineering Structures ( ) –

Fig. 4. Large displacement analysis of tip-loaded cantilever.

0.60

0.50

0.40

0.30

0.20

0.10

0.00
0.00 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 18.0 20.0

Fig. 5. Load–deflection curve and snap-through buckling of toggle frame.

of this paper, a single element was used for each member in the 0.9L whilst the transverse deflection v increases without bound.
frame, truss or shell structure. The analysis herein to a load factor of 70 took 5 s.

5.1. Large displacement of a cantilever 5.2. Snap-through buckling of a toggle frame

A benchmark solution of an elastic cantilever beam which in-
volves large elastic deformations was first presented by Bisshopp
and Drucker [33], in which an exact theoretical solution was devel-
oped. Their analysis allows for expressions of the exact curvature
and the effect of shortening due to the deflection of the tip load on
the cantilever. This problem was also investigated by Saleeb and
Chen [34], and others.
Fig. 4 shows the normalised horizontal and vertical deflections
of the free end of the cantilever, plotted against the dimensionless
applied load PL2 /EI. It can be seen that the axial deformations u
and transverse deflections v from the present analysis are in good
agreement with the theoretical solution [33] throughout the entire
range, especially the axial deformation. The stiffness formulation of
a single element proposed herein, and in particular its total secant
stiffness matrix, therefore allows for large displacement behaviour
due to flexure and axial shortening. It is interesting to note that
while the axial load P in the present second-order elastic analysis
reached a load factor of 70, the axial deformation u approaches
A shallow two-bar toggle frame was proposed by Williams [35]
for demonstrating the snap-through buckling behaviour of a struc-
ture, and he presented experimental and analytical solutions for a
toggle frame with fixed end supports under a vertical load P at its
apex. The elastic modulus of the bars is 71.02 kN/mm2 and they
are of rectangular cross-section with a width of 19.126 mm and a
depth of 6.1722 mm. This is a useful benchmark solution for vali-
dating second-order elastic analyses using one element per mem-
ber because of the simplicity of its geometry, and the incorporation
of snap-through and pre- and post-buckling responses.
Fig. 5 shows Williams’ toggle frame with a rise of 100 mm and
a horizontal span of 657.5 mm between the supports, for which
the vertical deflection at the apex increases to the critical load
(limit point) for the frame; the structure then softens slightly with
increasing deformations as the load decreases, and then finally
hardens under tension. It can be seen from Fig. 5 that the experi-
mental results are disparate from the analytical ones when flexural

Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
 ARTICLE IN PRESS
C.K. Iu, M.A. Bradford / Engineering Structures (
Fig. 6. Horizontal reaction and snap-through buckling of toggle frame.
–20 –15 –10
Fig. 7. Large displacements of the right-angle frame.
shortening is excluded from the analysis, but they agree very well
with the results of the present paper and the second-order pro-
gram NIDA [36] when one element per member is used, for which
flexural shortening reliant on both bowing functions b1 and b2 in
Eqs. (19) and (20) is included in the element formulation. These
results also agree with those of Williams [35] when flexural short-
ening is included. The solution of Iu and Bradford [27] requires at
least four elements to achieve reasonable accuracy, as can be seen
in Fig. 5. This is partly attributable to the incomplete and asym-
metric bowing function formulated in [27], as the component with
(θ1 + θ2 ) in the bowing function, which was neglected in [27], con-
tributes considerably to the flexural shortening for the case of a
frame member, which usually bends in double curvature.
Wood and Zienkiewicz [13] analysed the toggle frame using
the finite element method with a modified Newton–Raphson
technique, and presented results for the horizontal reaction RH
shown in Fig. 6. In their modelling, five finite elements were used
for each member, and their result for the load–deflection curve
in [13] was consistent with that of Williams [35]. The results of the
present formulation are in good agreement with those of Wood
and Zienkiewicz [13] as shown in Fig. 6, in which the softening
and hardening behaviour of the frame are reproduced accurately.
The snap-through buckling is caused by a change of geometry of
the frame, whose flexural buckling load if it were a straight fixed-
ended column is 2.4 kN. It can therefore be seen that the present
Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
) – 7
1.4
1.4
0.8
0.6
0.4
0.2
0
–5 0 5 10 15 20
–0.2
stiffness formulation for one element per member can replicate
member bowing and snap-through buckling involving a change of
geometry. The present analysis took two seconds for the solution
of the toggle frame.
5.3. Load–deflection response of right-angled frame
Koiter [37] provided the first analytical solution of a right-
angled frame, presenting a formulation for studying its buckling
and post-buckling response. Chan and Zhou [18] have also stud-
ied this frame in order to verify their higher-order element for-
mulation. It consists of a right-angled frame with pin supports and
equal column and beam lengths, subjected to a point load P with an
eccentricity e to the beam-to-column joint. The section, geometry
and material properties are shown in Fig. 7, which also shows the
joint rotation plotted against the dimensionless load P /PE , where
PE is the Euler load. The proposed non-linear modelling using one
element per member produces results consistent with those of
Chan and Zhou [21], and moreover predicts the post-buckling re-
sponse of a perfect frame reported by Koiter [37]. The solution of
the right-angled frame using the present analysis was completed
within five seconds.
The development of instability in this frame is germane to Eu-
ler buckling in a column; as when the axial load P approaches PE
the joint experiences large rotations as shown in Fig. 7. When this
 ARTICLE IN PRESS
8 C.K. Iu, M.A. Bradford / Engineering Structures ( ) –

Fig. 8. Load–deflection curve at the vertex of the hexagonal spatial frame.

Load factor P (kN)

Dimensionless axial force components

Fig. 9. Components of axial load resistance of the spatial hexagonal frame.

occurs, the coupling of the axial load and the rotational deforma-
tion contributes to the large displacements associated with the
buckling of a column for which the bowing functions b1 and b2 in
Eqs. (19) and (20) are very significant in the element stiffness for-
mulation. In particular, the constant terms in the numerators in
these Eqs. (19) and (20) mean that this coupling still exists for small
values of the axial load parameter q in the tangent stiffness formu-
lation, and they result in large rotations through the coefficients Gα i
in the tangent stiffness matrix in Eq. (26) as shown in Box I. Fig. 7
shows that the load–deflection curve from the theory of [27] us-
ing one element does not predict these large rotations accurately,
and two elements are needed, although the frame buckling load
from [27] converges to that of the theoretical result in [37].
5.4. Twelve-member hexagonal space frame
A three-dimensional shallow space frame of hexagonal shape,
which is composed of 12 identical members subjected to a vertical
load P at its apex, is shown in Fig. 8 (which also lists the material
and section properties). This hexagonal frame was tested experi-
mentally by Griggs [38], in which genuinely large rotations were
recorded. The pre-buckling behaviour of the frame was predicted
numerically by Chu and Rampetsreiter [5], while Papadrakakis [39]
traced the solution through the post-buckling regime, and more
recently Meek and Tan [14] and Chan and Zhou [21] validated
their analyses of geometric non-linear frames with this benchmark
structure.
The boundary nodes of this structure are restrained against ver-
tical settlement only, and the horizontal deformations of these
nodes is influenced by the stiffness of the six horizontal members
around its edge. Rigid body movement of the frame was eliminated
by restraining the horizontal movement of the apex node in the x
and y directions. In the pre-buckling loading regime, the inclined
members are subjected to compression, and to tension in the post-
buckling loading regime; the non-linear structural response is re-
liant on the vertical load P, the stiffness of the inclined members
and on the deformation of the six horizontal members around its
edge.
Fig. 8 plots the load–deflection curves at the apex obtained
from the present analysis, where they are compared with various
solutions [14,21,39]. It can be seen that the results from the present
analysis agree completely with those of Papadrakakis [39] and
Chan and Zhou [21], but the results of Meek and Tan [14], who used
a cubic finite element, are discrepant with the other results in the
snap-through and post-buckling range. The maximum load before
buckling is close to the Euler load of each member of 0.6 kN, and so
frame instability as a result of member bowing and snap-through
buckling due to a change of geometry is of significance.
Fig. 9 shows the dimensionless axial loads in the leaning me-
mbers, which are all the same because of symmetry. The axial

Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
 ARTICLE IN PRESS
C.K. Iu, M.A. Bradford / Engineering Structures (
Fig. 10. Geometry and dimension of a 168-member reticulated shell structure.
250
200
150
Load factor P/EAx104
100
50
0
0 100
–50
–100
Fig. 11. Vertical deflection at node 1 of reticulated shell truss.
resistance in Eq. (18) embodies two major components, viz. the ax-
ial strain e/L and the bowing function Cb . It can be seen that the
hexagonal frame has the same snap-through buckling characteris-
tics as the toggle frame in Fig. 6. In the course of the snap-through
buckling in Fig. 8, the dimensionless load decreases from P /(EA) =
0.254 to 0.0324 and the frame then stiffens with an increase of
loading. In the pre-buckling regime in Fig. 9, the axial strain com-
ponent e/L dominates, with the bowing function Cb making a neg-
ligible contribution. However, in the post-buckling regime, the
parameter e/L increases with a decreasing rate, whereas the bow-
ing function Cb increases in order to counteract the axial strain
completely, and so the dimensionless axial force P /(EA) plummets
to form a large loop. In short, the bowing function contributed to
be the large rotations dominates the axial strain effect e/L for snap-
through buckling. In the post-buckling range, the axial strain e/L
decreases sharply because the deformed chord length Lc is close
to the original member length L, while the bowing function con-
tributed to by the large rotations still dominates the post-buckling
response of the structure. In summary, the genuinely large rota-
tions experienced by this structure ensure that its behaviour moves
from axial compression dominance to member bowing, which
leads to difficulties in reliable convergence as the total deforma-
tions are formulated in the element stiffness formulation.
Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
) – 9
vertical deflection (mm)
200 300 400 500 600
5.5. 168-member reticulated shell structure
Fig. 10 shows a reticulated shell structure containing 168 mem-
bers, whose post-buckling analysis was undertaken by Paradiso
and Reale [40]. Papadrakakis [39] verified his vector iteration
methods using this shell structure, for which the material was as-
sumed to be linear elastic, and the members are assumed to be
identical and pinned so that bending and torsion effects vanish at
the joints. This reticulated shell with many components, with a
central concentrated load P at its apex, experiences snap-through,
pre- and post-buckling characteristics, as well as large displace-
ments and rotations at its joints, and so is an ideal benchmark
structure for demonstrating the capacity of large displacement
second-order numerical methods.
Paradiso and Reale [40] reported the critical load Pcr in the pre-
buckling regime as being 84.756 kN; this is extremely close to
the value of Pcr = 84.657 kN (0.1%) obtained from the present
analysis as shown in Fig. 11. The vertical displacement of the
apex joint (joint 1) is also shown in Fig. 11 as a function of the
dimensionless load P /(EA) (×104 ), in which it can be seen that
the results are very close to those of [39] throughout the loading
and unloading portions. Figs. 12 and 13 plot the vertical and
horizontal displacements of node 2 (Fig. 10) respectively, which
 ARTICLE IN PRESS
10 C.K. Iu, M.A. Bradford / Engineering Structures ( ) –

Load factor P/EAx104

250

200

150

100

50

vertical deflection (mm)

0
–30 –20 –10 0 10 20 30 40

–50

–100

Fig. 12. Vertical deflection at node 2 of reticulated shell truss.

Load factor P/EAx104

250

200

150

100

50

Horizontal displacement (mm)

0
–20 –15 –10 –5 0 5 10 15 20

–50

–100

Fig. 13. Horizontal displacement at node 2 of the reticulated shell truss.

also agree very well with those of [39]. It is interesting to note Lagrangian approach for the global analysis, and by a sophisticated
that there are two zero-loaded configurations for this shell; at the element stiffness formulation which is able to capture the geomet-
first the vertical deformations of nodes 1 and 2 are approximately ric non-linear response of a member. This higher-order element
196 mm and −24.5 mm respectively and at the second they are is able to replicate geometric non-linearity within the member
441.5 and 0 mm. Because node 1 is located 220.75 mm above and for a framed structure with many members which involves
the horizontal plane containing the internal ring nodes before large displacements, as well as bowing of the members and the
loading and deflects 220.75 mm below this plane in the second associated shortening, snap-through buckling and pre- and post-
zero-load configuration in the course of loading, the latter state buckling, leading to a versatile and powerful approach for struc-
is symmetrical to the initial unloaded state with respect to this tural analysis. The approach is very useful for analyses involving
plane and the horizontal deflections are zero, as can be seen from optimisation of the structural form, for which a great many such
Figs. 12 and 13, at which the entire structure is unstrained. This analyses may be needed with varying frame topologies which must
is a common characteristic of general snap-through buckling in be assessed and compared.
trusses of this kind. The present method was able to analyse this When using one element per member, the element stiffness for-
reticulated shell with 168 members in about 90 s, and so this mulation relies on the total deformations of the member to trigger
numerical example demonstrates the capability as well as the its geometric non-linearity. In the presence of genuinely large de-
efficacy of the proposed formulation. formations, the large unbalanced forces may therefore develop in
some instances, which provoke inefficient rates of convergence. If
6. Concluding remarks several elements are used for a member in a particular case of gen-
uinely large deformation, this behaviour can then be modelled as
This paper has described the formulation of an efficacious non- the rigid body movement through the system analysis. In view of
linear finite element technique for elastic second-order frame anal- this, not only the total deformations, but also the unbalanced forces
ysis, which requires in most cases only one element per member. for a member are reduced by using several elements, thereby lead-
The accurate modelling is affected both by an efficient updated ing to rapid convergence.

Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033
 ARTICLE IN PRESS
C.K. Iu, M.A. Bradford / Engineering Structures ( ) – 11

The approach proposed herein can be considered as an augmen- References

tation of the finite element method with a stability function ap-
[1] Masur EF. Post-buckling of redundant trusses. Trans ASME 1954;119:699–712.
proach, and it removes the limitations of the finite element method [2] Saafan SA. Nonlinear behavior of structural plane frames. J Struct Div, ASCE
in that one element per member is adequate for the geometrically 1963;89(ST4):557–79.
non-linear modelling. Using one element per member, the tech- [3] Turner MJ, Dill EH, Martin JC, Melosh NJ. Large deflections of structures
subjected to heating and external loads. J Aerosp Sci 1960;27(2):97–106.
nique was compared with several analytical, numerical and exper- [4] Connor JJ, Logcher RD, Chan SC. Nonlinear analysis of elastic framed structures.
imental results reported independently, and was articulated to be J Struct Div, ASCE 1968;94(ST6):1525–47.
in very good agreement, particularly with a multiplicity of mem- [5] Chu KH, Rampetsreiter RH. Large deflection buckling of space frames. J Struct
Div, ASCE 1972;98(ST2):2701–22.
bers and highly non-linear behaviour. [6] Oran C. Tangent stiffness in plane frames. J Struct Div, ASCE 1973;99(ST6):
973–85.
[7] Oran C. Tangent stiffness in space frames. J Struct Div, ASCE 1973;99(ST6):
Acknowledgement 987–1001.
[8] Oran C, Kassimali A. Large deflections of framed structures under static and
dynamic loads. Comput Struct 1976;6:539–47.
The work reported in this paper was supported by the Aus- [9] Gallagher RH, Padlog J. Discrete element approach to structural instability
tralian Research Council through a Federation Fellowship awarded analysis. AIAA J 1963;1:1437–9.
[10] Mallet RH, Marcel PV. Finite element analysis of nonlinear structures. J Struct
to the second author. Div, ASCE 1968;94(ST9):2081–105.
[11] Jennings A. Frame analysis including change of geometry. J Struct Div, ASCE
1968;94(ST3):627–44.
Appendix. Stiffness matrix terms [12] Powell GH. Theory of nonlinear elastic structures. J Struct Div, ASCE 1969;
95(ST12):2687–701.
The terms Gα i (α = y or z , i = 1 or 2) in Eq. (25) are: [13] Wood RD, Zienkiewicz OC. Geometrically nonlinear finite element analysis
of beams, frames, arches and axisymmetric shells. Comput Struct 1977;4:
∂ Mα1 725–35.
[14] Meek JL, Tan HS. Geometrically nonlinear analysis of space frames by an
∂q incremental iterative technique. Comput Methods Appl Mech Engrg 1984;47:
( 261–81.
12 (48)2 /5 + 36 (48) q/5 + 252q2 /35 + q3 /20

EI [15] Chajes A, Churchill JE. Nonlinear frame analysis by finite element methods.
= J Struct Eng, ASCE 1986;113(6):1221–35.
L (48 + q)3 [16] Oliveto G, Cuomo M. Incremental analysis of plane frames with geometric and
material nonlinearities. Eng Struct 1988;10:2–12.
× (θα1 + θα2 ) [17] Chan SL, Kitipornchai S. Geometric nonlinear analysis of asymmetric thin-
walled beam–columns. Eng Struct 1987;9:243–54.
4 (48)2 + 28 (48) q/5 + 132q2 /35 + 11q3 /420
 
[18] Kitipornchai S, Ai-Bermani FGA, Chan SL. Geometric and material nonlinear
+ analysis of structures comprising rectangular hollow sections. Eng Struct
(48 + q)3 1988;10:13–23.
[19] Oda H, Usami T. Stability design of steel plane frames by second-order elastic
)
analysis. Eng Struct 1996;19:617–27.
× (θα1 − θα2 ) [20] Al-Bermani FGA, Kitipornchai S. Nonlinear analysis of thin-walled structures
using least element/member. J Struct Eng, ASCE 1990;116(1):215–34.
  [21] Chan SL, Zhou ZH. Pointwise equilibrating polynomial element for nonlinear
EI EI analysis of frames. J Struct Eng, ASCE 1994;120(6):1703–17.
= {2b1 (θα1 + θα2 ) + 2b2 (θα1 − θα2 )} = Gα 1 (47) [22] Chan SL, Zhou ZH. Second-order elastic analysis of frames using single
L L imperfect element per member. J Struct Eng, ASCE 1995;121(6):939–45.
∂ Mα2 [23] Chan SL, Zhou ZH. Nonlinear integrated design and analysis of skeletal
structures by 1 element per member. Eng Struct 2000;22:617–27.
∂q [24] Gu JX, Chan SL. A refined finite element formulation for flexural and torsional
( buckling of beam–columns with finite rotations. Eng Struct 2005;27:749–59.
12 (48)2 /5 + 36 (48) q/5 + 252q2 /35 + q3 /20

EI [25] Izzuddin BA. Quartic formulation for elastic beam–columns subject to thermal
= effects. J Eng Mech, ASCE 1996;122(9):861–71.
L (48 + q)3 [26] Liew JYR, Chen H, Shanmugam NE, Chen WF. Improved nonlinear plastic
analysis of space frame structures. Eng Struct 2000;22:1324–38.
× (θα1 + θα2 ) [27] Iu CK, Bradford MA. Higher-order non-linear analysis of steel structures. I:
elastic second-order formulation [submitted for publication].
4 (48)2 + 28 (48) q/5 + 132q2 /35 + 11q3 /420
 
[28] Iu CK, Bradford MA. Higher-order non-linear analysis of steel structures. II:
− refined plastic hinge formulation [submitted for publication].
(48 + q)3 [29] Kassimali A. Large deformation analysis of elastic–plastic frames. J Struct Eng,
ASCE 1983;109(8):1869–86.
)
[30] Crisfield MA. An arc-length method including line searches and accelerations.
× (θα1 − θα2 ) Internat J Numer Methods Engrg 1983;19:1269–89.
[31] Ramm E. Strategies for tracing the nonlinear response near limit points. In:
  Wunderlich et al. (Eds.), Nonlinear finite element analysis of structures, Berlin;
EI EI 1981, p. 63–89.
= {2b1 (θα1 + θα2 ) − 2b2 (θα1 − θα2 )} = Gα 2 (48) [32] Chan SL. Geometric and material nonlinear analysis of beam–columns and
L L frames using the minimum residual displacement method. Internat J Numer
∂q 1/L Methods Engrg 1988;26:2657–69.
= [33] Bisshopp KE, Drucker DC. Large deflection of cantilever beams. Quart J Appl
∂e I
−
P h −16(48)2 /5−64(48)q/35 i
(θα1 − θα2 )2 Math 1945;3(3):272–5.
AL2 (48+q)4 [34] Saleeb AF, Chen WF. Elastic–plastic large displacement analysis of pipes.
α=y,z
J Struct Div, ASCE 1981;107(4):605–26.
1/L 1 [35] Williams FW. An approach to the non-linear behaviour of members of a rigid
= = (49) jointed plane framework with finite deflections. Quart J Mech Appl Math 1964;
I /AL2 − b02 (θα 1 − θα 2 )2
P
LH 17(4):451–69.
α=y,z [36] NIDA. Non-linear integrated design and analysis. NAF-NIDA. Version 7. 2007.
[37] Koiter WT. Post-buckling analysis of a simple two-bar frame. In: Broberg et al.
∂q 2b1 (θα 1 + θα 2 ) + 2b2 (θα 1 − θα 2 ) Gα 1 (Eds.), Recent progress in applied mechanics. New York; 1967. p. 337–354.
= = (50) [38] Griggs HP. Experimental study of instability in elements of shallow space
∂θα1 I /AL2 − b2 (θα 1 − θα 2 )2
P 0
H frames. Research Report. Cambridge (Massachusetts): Department of Civil
α=y,z Engineering. MIT; 1966.
[39] Papadrakakis M. Post-buckling analysis of spatial structures by vector iteration
∂q 2b1 (θα 1 + θα 2 ) − 2b2 (θα 1 − θα 2 ) Gα 2 methods. Comput Struct 1981;14:393–402.
= = . (51) [40] Paradiso M, Reale E, Tempesta G. Nonlinear post-buckling analysis of
∂θα2 I /AL − b2 (θα 1 − θα 2 ) 2
2
P 0
H recticulated dome structures. In: Proceedings of IASS world congress on shell
α=y,z and spatial structures. 1979, p. 1.67–1.80.

Please cite this article in press as: Iu CK, Bradford MA. Second-order elastic finite element analysis of steel structures using a single element per member. Engineering
Structures (2010), doi:10.1016/j.engstruct.2010.04.033