You are on page 1of 12

This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright

Author's personal copy

Engineering Structures 32 (2010) 32583268

Contents lists available at ScienceDirect

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

A simplified model for inelastic second order analysis of planar frames


Ahmed H. Zubydan
Civil Engineering Department, Faculty of Engineering, Port-Said University, Port-Said, Egypt

article

info

Article history:
Received 17 December 2009
Received in revised form
24 June 2010
Accepted 28 June 2010
Available online 5 August 2010
Keywords:
Residual stresses
Column
Beam
Nonlinear analysis
Inelastic

abstract
A simplified model for predicting second order inelastic behavior of steel frames is developed. New
empirical formulae are developed to describe the tangent stiffness of steel sections subjected to an axial
compression force and bending moment. The tangent stiffness formula is extended to evaluate the secant
stiffness that is used for the internal force recovery. The formulae are derived for steel sections considering
the residual stresses as recommended by the European Convention for Construction Steelwork (ECCS). The
tangent stiffness for steel sections is also evaluated for the case where residual stresses are neglected. A
finite element program is prepared to predict the inelastic second order behavior of plane steel frames
using the derived formulae for the steel cross sections. The updated Lagrange coordinates are used to
include the second order effect. The NewtonRaphson scheme combined with the minimum residual
displacement method is employed to satisfy the equilibrium between external and internal forces.
Comparisons with fiber model indicate good agreement with the present model. The analysis results
indicate that the new model is accurate and has a faster rate of convergence for problems involving
inelastic behavior.
2010 Elsevier Ltd. All rights reserved.

1. Introduction
During the past 20 years, researchers have developed various
methods to predict the second order inelastic analysis of steel
frames. Most of these studies may be categorized into two main
types: (1) the plastic zone method; and (2) plastic hinge method.
The plastic zone method is also called the distributed plasticity
method and it follows explicitly the gradual spread of yielding
throughout the members of the structure. Alvares and Birnstiel [1]
and Latona [2] first introduced the plastic zone analysis of steel
frames. The plastic zone method is based on the concept that each
member is subdivided along its length into several elements, called
fibers, so that the behavior of the element can be approximated
as a sum of the fiber contributions [36]. Because of the refined
discretization of the members and their cross sections, the plastic
zone analysis can predict accurately the inelastic behavior of the
structure and is generally considered an exact method of analysis [7,8]. However, the fiber model, which may be combined with
the plastic zone method, requires intensive computations and a
large disk storage due to the huge amount of data, thus incurring a
high cost.
A simpler way to represent the behavior of steel structures
is the plastic hinge model [917]. The plastic hinge approach is
the earliest and most widely used method in structural analysis. The principal feature of this method is that the outermost
fiber of a beamcolumn element reaches its yield strength when

E-mail address: Zubydan@gmail.com.


0141-0296/$ see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2010.06.015

the magnitude of the end-rotation of that element reaches its


yield moment capacity. When all fibers in the section yield, the
beam is said to have formed a flexural moment hinge. The concentrated plastic hinge method is easy to use for modeling material nonlinearity because of its simplicity and the potential use
of one element per member [18,19]. Models based on the elasticplastic hinge approach may overestimate the ultimate capacity
of steel frames because they do not include the effect of residual stresses or the plasticity distribution along the whole member
[20,21].
In the present paper a simple model for describing degradation
of member stiffness due to applied forces is proposed to simulate
the fiber model but without discretization of cross sections into
fibers. The formulae are derived by fitting the results obtained
using the fiber model. The proposed model is valid for steel frames
considering/neglecting the effect of residual stresses. The model
achieves the accuracy of the spread of plasticity method but in
an easier way. The present model saves a lot of calculations and
disk storage. The updated-Lagrangian formulation is applied in the
formulation of the incremental matrix equilibrium equations of
the proposed beam-element model [22,23]. The minimum residual
displacement combined with NewtonRaphson method is used to
satisfy the convergence when solving the nonlinear equilibrium
equations.
2. Numerical model
2.1. Basic assumptions
The following assumptions are made in the formulation of the
beamcolumn element:

Author's personal copy

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

3259

relationships are used to accumulate the internal forces under a


given deformation:
P =

n
X

as fs

(3a)

i =1

M =

n
X

as fs ys

(3b)

i =1

Fig. 1. Considered fiber model.

(1) A plane cross section remains plane after deformation.


(2) Local buckling and lateral torsional buckling are not considered. All members are assumed to be fully compact and adequately braced.
(3) Small strains but large displacements and rotations are considered.
(4) Only I and H-shaped sections subjected to flexure about the
strong axis are considered.
(5) Strain hardening is not considered.

where as and ys are the area and the coordinate of steel fibers,
respectively, and fs is the stress at each steel fiber. It is found that
the plastic strength surface is not affected by the presence residual
stresses. The proposed plastic surface differentiates between H and
I-shaped cross sections and it consists of two parts as follows: For
H-shaped sections
(4a)

p1r .5 + mr = 1 for pr < 0.2.

(4b)

pr +
pr
2

mr = 1 for pr 0.2

+ mr = 1 for pr < 0.2

(1a)
(1b)

where pr is the ratio of the applied normal force P to the yield


value Py at the plastic strength envelope, and mr is the ratio of the
applied bending moment M to the plastic moment Mp at the plastic
strength envelope. This empirical formula is fitted as lower bound
capacity for general sections and so it provides an uncertain error
when used for plastic analysis.
Another interaction equation used by Duan and Chen [24] to
describe the limit plastic surface is given as follows:
p1r .3 + mr = 1.

mr = 1

9
11

(2)

This equation does not give the full correlation with the results obtained from the fiber model at all applied force ratios, as discussed
below in the present section.
In the present paper, other formulae are proposed based on the
results obtained from the analysis of many cross sections using
the fiber model in which the cross section is discretized as shown
in Fig. 1. The analyzed cross sections are selected to cover a wide
range of geometric varieties which may cover all popular universal
column and universal beam cross sections. Twelve universal
column sections (H-shaped section) are analyzed in which the
ratios B/T = 5.522.4, D/t = 1034.2 and D/B = 0.971.13,
where D, B are the cross section depth and the flange breadth,
respectively, and t and T are the thicknesses of the cross section
web and flanges, respectively. On the other hand, twenty universal
beam sections (I-shaped section) are selected in which the ratios
B/T = 5.715.8, D/t = 27.462.5 and D/B = 1.743.36.
For each cross section and at different axial compression forces,
the curvature is gradually increased until reaching the maximum
possible bending moment. For all fiber elements, the following

for pr 0.3

(4c)

for pr < 0.3.

(4d)

mr = 1

p1r .65 + mr = 1

For the present analysis, it is important to use an accurate function to control the yield of the cross section. Most previous analysis assumes simplified linear interaction equations to describe the
plastic surface. Based on AISC-LRFD the bilinear interaction equation for the plastic strength of the beamcolumn member is expressed as follows (AISC, 1993):
8

For I-shaped sections


pr +

2.2. Cross-section plastic strength

for pr 0.2

pr +

The graphical representations of both the proposed formulae and


Eq. (2) are compared with the fiber model results as illustrated in
Fig. 2. It is clearly observed that the conversion of the proposed
formulae with the fiber model results is better than Eq. (2).
2.3. Cross sectional stiffness degradation
The degradations of the tangential modulus (Et ) are determined for cross sections subjected to a pure axial force and also
for sections under combined axial compression force and bending
moments. The tangent modulus is determined for I and H-shaped sections with/without considering the effect of the residual
stresses. The residual stresses adopted in the present paper are
based on the recommendation by the European Convention for
Construction Steelwork (ECCS) [25]. The magnitude of residual
stress (r ) is assumed to be dependent on depth/breadth ratio, as
shown in Fig. 3. The tangent modulus is evaluated by applying the
concepts of the fiber model as shown in Fig. 1. The selected I and
H-shaped sections are analyzed using the fiber model in which the
forces are incrementally applied to the cross section and then the
tangent modulus is calculated at the average of each load increment.
2.3.1. Cross sections subject to axial force
It should be mentioned that many researchers [8,10] use the
empirical tangential modulus ratio proposed in AISC LRFD [26]
to describe the tangent modulus of columns under an axial force
taking the effect of residual stresses into consideration. Here the
tangential modulus ratio is given as follows:
Etr = 1 for Pr 0.5

(5a)

Etr = 4Pr (1 Pr )

(5b)

for Pr > 0.5

where Etr is ratio of inelastic tangent modulus (Et ) to the elastic


modulus (E ) for steel sections and Pr is the normalized axial force
(Pr = P /Py ). In the present paper, new formulae are proposed to
describe the tangent stiffness ratio for I- and H-shaped sections
under an axial force considering the effect of residual stresses
recommended by ECCS. The tangent stiffness ratio Etr for cross
sections subjected to axial forces are evaluated by analyzing the
cross sections under an incremental axial force using the fiber

Author's personal copy

3260

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

(a) H-shaped sections.

(b) I-shaped sections.


Fig. 2. Plastic strength for steel sections.

model. For each load increment, the tangent stiffness ratio Etr is
calculated using the following relationship:
Etr =

dP /d

(6)
EA
where dP /d is the change of axial compression force with respect
to the change of axial strain, E is the elastic modulus and A is the
cross sectional area. The tangent stiffness ratio Etr is plotted versus the average force ratio Pr for the analyzed cross sections. It was
found that the tangent stiffness ratio is equal to unity until a specific value of axial force ratio equal to Pr0 after which it gradually
decreases until it vanishes when Pr = 1. From the obtained results, the following formulae for describing the tangent modulus
ratio are proposed:
Etr = 1

for Pr Pr0

Etr = 1

Pr Pr0
1 Pr0

(7a)

 12

for Pr > Pr0

Fig. 3. ECCS residual stress for hot-rolled H and I sections.

(7b)

where Pr is the axial force ratio (Pr = P /Py ), Pr0 = 1 r /y , y


is the steel yield strength and r is the maximum residual stress
value as given in Fig. 3.
The formula proposed in Eq. (7b) is selected to obtain the best fit
of the fiber model results. A comparison of the proposed formulae
with both the fiber model results and the previous formulae in
Eq. (5) is shown in Fig. 4. It is found that the fiber model results are
the same for all sections that have the same residual stress ratio.
As shown in Fig. 4, the proposed formulae fit well the fiber model
results satisfying the ECCS recommendations.
2.3.2. Cross sections subject to bending moment and axial compression force
The empirical tangential modulus ratio proposed in AISC LRFD
(as shown in Eq. (5)) for sections subjected to an axial compression force is extended by Kim and Choi [8] to analyze members
subjected to a bending moment and axial force as follows:
for 0.5

(8a)

Etr = 4(1 ) for > 0.5

(8b)

Etr = 1

where is a force-state parameter that measures the magnitude of


the axial force and bending moment which may be calculated from
Eq. (2).

Fig. 4. Inelastic stiffness reductions due to axial force.

In the present research, other formulae for the tangent modulus for cross sections subjected to a bending moment and axial
compression force are evaluated with/without the consideration

Author's personal copy

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

3261

forces exceeds the value Pr0 , the initial tangential stiffness ratio Etr
follows the path b0 c 0 d0 , as shown in Fig. 5. In this figure, the initial value of the tangent stiffness ratio is equal to Etr0 (at point b0 )
which is less than unity due to the plasticity caused by axial force
which exceeds Pr0 . The value Etr0 can be calculated from Eq. (7b)
according to the present value of axial force ratio Pr .
The proposed formulae are derived to give best fits of the results
obtained by analyzing the selected cross sections using the fiber
model. The values of Etr for sections subjected to a combined axial
force and bending moment considering the residual stress effect
can be mathematically expressed as follows:
For Pr Pr0 :
Etr = 1 for Mr Mr0

"
Etr = (1 Etr1 ) 1

(10a)

r1 # r12

Mr Mr0
Mr1 Mr0

+ Etr1

for Mr0 < Mr Mr1


Fig. 5. Inelastic stiffness reduction due to bending moment and axial compression
force.

of residual stresses. As discussed before in the case of sections subjected to a pure axial force, the elastic modulus ratios here are determined for the sections with the help of fiber model. Using the
fiber model, the moment curvature relationships for the analyzed
cross sections are evaluated under different values of axial compression force. For each value of axial force, the tangential modulus
ratio Etr is evaluated at each load increment as follows:
Etr =

dM /d
EI

(9)

Etr = Etr1 1

(10b)

 r1

Mr Mr1

mr Mr1

for Mr1 < Mr mr .

(10c)

For Pr > Pr0 :

"
Etr = (Etr0 Etr1 ) 1

Mr

r1 # r12
+ Etr1

Mr1

for Mr0 < Mr Mr1

where EI is the cross sectional elastic rigidity. The relationships


between the tangent modulus ratio Etr and the moment ratio Mr
are plotted for each value of axial compression force. Two possible
paths may be obtained as shown in Fig. 5. When the value of axial
compression force ratio Pr is less than Pr0 , the relationship follows
the path abcd. For this case the value of Etr is equal to unity until a
specific value of moment ratio Mr0 (at point b), after that it sharply
decreases to reach a value equal to Etr1 at moment ratio Mr1 (at
point c) and finally the stiffness ratio gradually decreases to vanish at point d when the moment ratio reaches its maximum value.
On the other hand, when the value of applied axial compression

(a) UC sections.

Etr = Etr1 1

Mr Mr1
mr Mr1

(10d)

 r1

for Mr1 < Mr mr

(10e)

where Pr is the axial force ratio (Pr = P /Py ), Pr0 = 1 r y , Mr


is moment ratio (Mr = M /Mp ), mr is the moment ratio at the
plastic strength surface, which is calculated from Eq. (4), Mr0 =
(Pr0 P )Zx /Sx , Zx and Sx are the elastic and plastic modulus of
cross section, respectively. The values of constants r1 , r2 , and r3
are dependent on the cross section type and the applied axial force
ratio Pr . The values of r1 , r2 , r3 , Mr1 and Etr1 are represented in the
Appendix. When the effect of residual stress is neglected, the value
of Etr is only calculated from Eqs. (10a)(10c).
The proposed tangent modulus ratios given in Eq. (10) are
compared to those obtained from the fiber model, as shown in

(b) UB sections.

Fig. 6. Inelastic stiffness ratio for cross sections under bending moment and axial compression force considering the effect of residual stress (- - - - Fiber model;
Proposed model).

Author's personal copy

3262

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

(a) UC sections.

(b) UB sections.

Fig. 7. Inelastic stiffness ratio for cross sections under bending moment and axial compression force neglecting the effect of residual stress (- - - - Fiber model;
model).

Fig. 6 for UC sections (H-shaped) with residual stress ratio r /y =


0.5 and for UB sections (I-shaped) with r /y = 0.3. It is clearly
observed that the proposed formulae give good correlations with
the fiber model results for various values of Pr . On the other hand,
when the effect of residual stress is neglected, the comparisons
between the proposed model and the fiber model are represented
in Fig. 7(a) for UC sections (H-shaped) and in Fig. 7(b) for UB
sections (I-shaped). The conversion between the two models is
clearly observed in the figures.
It should be mentioned that, in the presence of residual stresses,
the derived tangent plastic modulus for cross sections subjected
to compound actions (bending moment and axial force) is only
valid when the axial force is compressive, while the cross sectional
behavior under compound bending moment and axial tensile
forces is quite different in the presence of residual stresses. On the
other hand, the derived model is valid for cross sections subjected
to bending moments and axial tensile or compressive forces in the
absence of residual stresses.
2.4. Cross section incremental secant modulus
As the forces are incrementally applied to the structure as
shown in Fig. 8, the tangent stiffness given in Eq. (10) can be
approximately used to evaluate the internal bending moment and
axial force at the end of increment j + 1 as follows:
Mj0+1 = Mj + d Etrj EI
Pj0+1

= Pj + d Etrj EA

(11a)
(11b)

where j + 1 is referred to the present step and j is referred to


the previous step. d and d are the change in the curvature and
axial strain of the cross section from step j to step j + 1. E , A and
I are the elastic modulus, the area and the moment of inertia of
the cross section, respectively. Etrj is the tangent modulus ratio at
step j which can be calculated from Eq. (10) using the force values
at that step (Pj and Mj ). The use of tangent stiffness in calculating
the internal forces causes a slightly overestimation of the member
strength, as shown in Fig. 8. As the increments d and d are
smaller, the accuracy of the calculated internal forces increases.
The results accuracy essentially depends on the accuracy of
calculation of the internal forces. The accuracy of internal forces
may be highly increased by additional mathematical calculations

Proposed

Fig. 8. Load-deformation curve for cross section.

in which the incremental secant stiffness is derived. Since an


incremental procedure is adopted, the change of Etr from step j to
j + 1 may be considered linear. Assuming that, at step j, the actual
values of bending moment, axial force and the tangent stiffness
ratio are Mj , Pj and Etrj , respectively, and the values Mj0+1 , Pj0+1 and
0
Etrj
+1 are the corresponding approximate values at step j + 1 that
can be calculated using Eqs. (11a), (11b) and (10), respectively.
For the incremental procedure, it is believed that the approximate
values at step j + 1 are very near the exact values. Then, a linear
equation of Etr versus the Mr , at a constant value of axial force, from
step j to j + 1 is given as follows:
Etr = c1 Mr + c2
where c1 =

(12)

0
Etrj
+1 Etrj
0
Mrj
+1 Mrj

and c2 = Etrj c1 Mrj .

Now, the accumulation of curvature at step j + 1 can be written


as follows:

j+1 = j +

Mj+1

dM
Etr EI

Mj

(13)

Substituting dM = dMr Mp and Etr = c1 Mr + c2 , Eq. (13) becomes

j+1 = j +

Mp
EI

Mrj+1
Mrj

dMr
c1 Mr + c2

(14)

Author's personal copy

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

3263

in which = x/L. Substituting Eq. (19) into Eq. (20) and using the
displacement shape functions, the material and geometric stiffness
matrices of the element in its local coordinates are derived and
given as follows:

[km ] = Etr
u1

Fig. 9. Plane frame element.

Putting y = Mp /EI and integrating Eq. (14) leads to

j+1 = j +

y 

ln c1 Mrj+1 + c2 ln c1 Mrj + c2

c1



EA/L

Mj+1 =

where k =

Mp e c2

c1 j+1 j
y

(16)

+ ln(c1 Mrj + c2 ).

1 Mj+1 Mj

EI j+1 j
where EI is the elastic flexural rigidity of the cross section.

(17)

2.5. Finite element model


A stiffness method approach for the analysis of plane frames is
developed considering both geometric and material nonlinearities.
The equation of equilibrium in terms of geometry of the deformed
system is given as follows:

[Km + Kg ]{1D} = {1F}


(18)
where {1F} and {1D} are the force and displacement vectors,
respectively. [Km ] is the stiffness matrix of the structure considering material nonlinearity, and [Kg ] is the geometric stiffness matrix which represents the change in the stiffness that results from
deformation effects. The geometric stiffness matrix may be developed in a number of ways.
Consider a prismatic element of a symmetric cross section in the
plane of the frame. This element is subjected to an axial force, and
bending moments and flexes in xy plane as shown in Fig. 9. It is
assumed that the deformations occur only in the plane of the frame
and that the shear deformation effect is negligible. The total finite
strain efin of a segment of length dx is given by McGuire et al. [27]
as follows:
efin =

du

dx

d2

dx2

1
2

"

du
dx

2


+

2 #

dx

(19)

where u and are the longitudinal and transverse displacements,


respectively. Applying the principle of virtual displacements to the
reference configuration, the change in the internal virtual work
Wint due to the virtual displacement is

Wint =

x efin d(vol)

u2

0
0
EA/L

0
12EI /L3
6EI /L2
0
12EI /L3

0
6EI /L2
2EI /L
0
6EI /L2
4EI /L

EA/L

(20)

vol

where x is the normal stress. Now, appropriate flexural and axial


shape functions are introduced into Eqs. (19) and (20). For this
purpose the following shape functions are adopted:
u = (1 ) u1 + u2

(21a)



= 1 3 2 + 2 3 1 + 1 2 + 2 x1


+ 3 2 2 3 2 2 x2

(21b)

(22a)

(15)

The incremental secant modulus ratio (Esr ), which passes


through points j and j + 1 in Fig. 8, is then calculated as follows:
Esr =

0
6EI /L2
4EI /L

u1

c1

0
12EI /L3
Sym.

Then the internal moment at increment j + 1 can be easily written


as follows:
1

[kg ] =

P2
L

u2

0
1.2

0
L/10
2L2 /5

0
6/5
L/10
0
1.2

0
L/10
L2 /30
0
L/10
2L2 /15

Sym.

0
0
1

(22b)

where 1 , and 2 are the rotations at nodes 1 and 2, respectively,


and L is the member length. EA and EI are the axial and flexural
rigidities, respectively. It is noticed that the geometric stiffness
matrix is a function of the total axial force P2 acting at node 2 of
the element in the reference configuration (note that P2 = P1 ).
The components of the nodal point displacement are incremental
values referred to this configuration. Since the matrices [Km ]
and [Kg ] are displacement dependent, Eq. (18) cannot be directly
solved. Various procedures can be used to solve the equilibrium
equations. Generally, members must be subdivided into several
elements to produce satisfactory results. The modulus ratios Etr or
Esr are evaluated at each member end and then the average value
is used in the stiffness matrix [km ].
2.6. Internal force recovery
The employing of the equilibrium equation in conjunction with
the incremental analysis requires that the structural geometry
includes all accumulated deformations. For the current analysis,
the node coordinates are updated after each iteration. That is,
the coordinates of each node are modified or updated to include
the translational displacement components that occur during
iterations. In updating the coordinates of nodes or element ends,
the deformed geometry of the structure is achieved by changing
the position and hence the orientation of each element with
respect to the global coordinates system. For all the elements of
a structure, the element stiffness equation, as given in Eq. (22),
can be assembled to yield the stiffness equation of the structure
for an incremental step. For an incremental-iterative nonlinear
analysis, the element incremental displacement vector {1d} is
used to calculate the incremental strains (d and d ) and so the
secant modulus at element ends can be evaluated as illustrated
in Section 2.4. The material stiffness matrix [km ] for each element
is reformulated using the average secant plastic modulus (Esr )
instead of the tangent modulus. On the other hand, the geometric
stiffness matrix [kg ] is also formulated again using the new axial
force P2 , which is calculated by adding the incremental axial force
to the total previous value as follows:
P2 = (P2 )previous + d Esr EA.

(23)

The increments of internal forces for an element can be calculated


as



{1fint } = km + kg {1d} .

(24)

Author's personal copy

3264

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

By summing the element forces at the structural nodes and


comparing them with the applied loads, the unbalanced forces for
the structure can be obtained. Finally, by treating the unbalanced
forces as applied loads, other iterations can be repeated.
2.7. Nonlinear iterative solution
For the purpose of performing iterations at each incremental
step, the incremental structural equation as given in Eq. (18) can
be rewritten for the jth iteration of the ith incremental step as

  i  i  i

1Dj = Fext j Fint j1
(25)
 i 
 i
where Kj1 is the tangent stiffness matrix, Fext j is the applied
 i

load vector, Fint
j1 is the internal resistant forces of the structure
 i
and 1Dj denotes the incremental displacements generated at
Kji1

the jth iteration. To avoid numerical instabilities, the iterations


should not be performed under constant applied load. For that
purpose, an incremental load factor is introduced as follows:

i
i
i

Fext
j = Fext j1 + j 1Fext

(26)

where the ij is the load increment factor associated with the jth

iteration of the ith incremental step, and 1Fext is the reference


load vector.
Let {Rij1 } denote the unbalanced forces arising from the

i
difference between the applied loads Fext
j

and internal forces

i
Fint
j1 in the previous (j 1)th step:

 i
 i

{Rij1 } = Fext
j1 Fint j1 .

(27)

By the use of Eqs. (26) and (27), Eq. (25) reduces to


Kji1

n o 
  i

1Dj = ij Fext + Rij1

(28)

which can be decomposed into two parts as

o n o
1D ij = Fext
 i   i  i
j = Rj1 .
Kj1 1D


Kji1

n

(29)
(30)

Consequently, the displacement increment vector can be accumulated as


(31)

For first iteration (j = 1) the displacement increments is calculated


as

n
o
 i
1D1 = i1 1D i1

(32)

and the total displacement vector Dij of the structure at the end
of the jth iteration can be computed as
Dj = Dij1 + 1Dij .

(33)

The load increment factor in Eqs. (26), (28) and (32) remains
undetermined. This factor is important to trace the postpeak path.
For that purpose, the minimum residual displacement method developed by Chan [28] is used. The method guarantees a minimum
value for the equilibrium error in each iteration. By generalized
displacement control method, the load increment factor ij is determined from a constraint condition. For j = 1, which means the
beginning of the ith incremental step, the load increment factor i1
is given as
i
j

i 1/2

= |GSP |
i
1

1
1

(34)

where is the load increment parameter for the first incremental


1
1

step and GSP is the generalized stiffness parameter which can be


defined as

n
o
 i  i
1Dj = 1D j + ij 1D ij .

 i

Fig. 10. Flow chart for the present nonlinear analysis.

oT n

o
1D 11
GSPi = n
oT n
o.
1D 1i1
1D i1
1D 11

(35)

The sign in Eq. (34) determines the path of the loaddeflection


curve. This sign is taken positive for positive values of GSP, while
the loading direction is reversed (negative sign is considered) if
GSP < 0.
For j 2, the load increment factor ij is

oT 

1D ij1
1D ij
ij = n
oT n
o
1D ij1
1D ij
n

(36)

ij1 is the displacement increments generated by the


where 1D
reference loads

Fext

at the previous iteration, and

o
1D ij and


1D ij are the displacement increments generated by the reference

loads and unbalanced forces, respectively, at the jth iteration of the


ith incremental step, as defined in Eqs. (29) and (30).
A flow chart showing the sequence of the incremental-iterative
analysis for a load increment is illustrated in Fig. 10.

Author's personal copy

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

3265

3. Numerical analysis and results


A computer program is developed to predict the second order
behavior of plane frames using the derived model. The analysis results obtained from the proposed model are essentially compared
to those obtained from the fiber model.
3.1. Inelastic analysis of cantilever column

Fig. 11. Geometry and load configuration of the analyzed cantilever column.

Fig. 12. Comparison between the tangent and the secant modulus for internal
forces recovery.

(a) With residual stress (r /y = 0.5).

The cantilever column shown in Fig. 11 is assumed to be made


of steel with yield strength y = 250 MPa and elastic modulus
E = 200,000 MPa. and the column is of 4 m height. The column
is subjected to an incremental lateral load H at constant values
of vertical load P. The cantilever is first analyzed to compare the
recovering the internal forces using the tangent stiffness in Eq. (10)
and also by using the secant modulus that is given in Eq. (17). In
that case the column cross section is assumed to be UC254 254
107 with residual stress ratio equal 0.5. The cantilever is analyzed
under different values of vertical load ratios Pr = 0.3, 0.5 and 0.7.
The comparisons of result analysis obtained from the two methods
are shown in Fig. 12. As expected, the use of material tangent
stiffness for calculating the internal forces slightly overestimates
the stiffness and the capacity of the column. In spite of the clear
conversion between the two methods of analysis, the secant value
in Eq. (17) is used in the present research.
The column is reanalyzed to compare the results obtained by
using the present model to those obtained by the fiber model
with/without consideration of the residual stresses. Also, the
present model is compared to the model proposed by Kim and
Choi [8] using the tangent stiffness ratio given in Eq. (8). First,
the cross section of the column is assumed to be HEB320 (the
slenderness ratio L/rx = 28.9) and it is analyzed under vertical
load ratios Pr = 0.25, 0.5 and 0.75. Fig. 13 shows the comparisons
between the different models with/without the effect of residual
stresses. It is clearly observed that the proposed model gives a good
correlation with the fiber model. On the other hand, the formula
proposed by Kim and Choi [8] may overestimate or underestimate
the capacity of the column compared with the fiber model.
The column is reanalyzed again using IPE240 as a cross section
for the column (L/rx = 40). In this case the residual stress ratio
is assumed to equal 0.3, as recommended by ECCS. The column
is analyzed under vertical load ratios of Pr = 0.1, 0.3 and 0.5.

(b) Without residual stresses.

Fig. 13. Lateral load versus maximum drift for the cantilever column with cross section HEB320.

Author's personal copy

3266

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

(a) With residual stress (r /y = 0.3).

(b) Without residual stresses.

Fig. 14. Lateral load versus maximum drift for the cantilever column with cross section IPE240.

Fig. 15. Geometry and load configuration of the analyzed simple beam.

The analysis results are shown in Fig. 14(a) for the case of residual
stress consideration and in Fig. 14(b) when the residual stresses
are neglected. It is also observed that the proposed model gives
a good conversion with the fiber model. On the other hand the
formula given in Eq. (8) always underestimates the capacity of the
column with respect to the fiber model in the case of residual stress
consideration.
3.2. Interaction equation for simply supported beam
The simple beam shown in Fig. 15 is analyzed to evaluate the
effect of residual stresses on the capacity of the beam under different values of axial compression forces and bending moments. The
beam section is considered as UC152 152 30 and the steel yield
strength y = 250 MPa with elastic modulus E = 200,000 MPa.
The beam is analyzed under different values of slenderness ratios
L/rx = 20, 60 and 100. The beam is discretized into 10 equal elements and the bending moments are incrementally applied at
different values of axial compression forces until reaching the maximum capacity. The beam is analyzed for the case of residual stress
consideration (r /y = 0.5) and also for the case of neglecting
the residual stresses using both the proposed and the fiber models. The interaction curves obtained from the analysis of the beam
are shown in Fig. 16. From the figure, it is observed that the proposed simplified model correlates well with the results of the fiber
model for all slenderness ratios L/rx . For the case of low slenderness ratio (L/rx = 20), the ultimate strength of the beam is almost
the same for the cases of when considering and neglecting residual stresses. As the value of slenderness ratio increases (L/rx = 60
and 100), the neglect of residual stresses overestimates the beam
ultimate strength, especially for the cases of higher values of axial forces, compared with the case where the residual stresses are
considered.

Fig. 16. Interaction curves for simply supported beam.

Fig. 17. Geometry and load configuration of El-Zanatys frame.

3.3. EI-Zanatys portal frame under fixed gravity loads


The portal frame shown in Fig. 17 was originally analyzed by ElZanaty [3]. The frame cross section is W8 31 (L/rx = 40) and the

Author's personal copy

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

3267

vertical load P is considered as H /P = 0.02, 0.05, 0.1 and 0.2.


In that case, the cross section of the frame members considered is
W6 20 (L/rx = 51.9) and the steel yield strength y = 250 MPa
and the elastic modulus E = 200,000 MPa. The loads are proportionally applied to the frame until failure occurs. The frame
is analyzed using both the proposed simplified and fiber models.
Fig. 19(a) shows the analysis results when the residual stresses are
considered (r /y = 0.5) while the results of nonresidual stressed
sections are shown in Fig. 19(b). It is clearly observed that the simplified model correlates well with the fiber model results.
From the previous analyzed examples, it was found that the
average run time of the frame using the fiber model ranges from 7
to 9 times the run time required using the simplified model. These
run times include the formulation and the solving of nonlinear
equilibrium equations and they also include the time consumed
to equilibrate the fibers on the cross sectional level which is saved
when the proposed simplified model is used.
4. Conclusions
Fig. 18. Loaddeflection relationship for El-Zanatys frame with fixed gravity loads.

steel yield strength y = 248.2 MPa and modulus of elasticity E =


200,000 MPa. The frame is subjected to constant values of gravity
load ratios (Pr = 0.2, 0.4 and 0.6). Each of girders and columns
are discretized into 10 equal elements. The horizontal load H is
incrementally applied until failure occurs. The frame is analyzed
using both the proposed model and the fiber model considering
the residual stress pattern as recommended by ECCS with residual
stress ratio = 0.5. The fiber model is also used to reanalyze the
frame using another residual stress distribution as considered by
White [29]. The analysis results shown in Fig. 18 indicates that
both the proposed simplified model and the fiber model using
residual stress distribution recommended by ECCS almost coincide
while the residual stresses considered by White underestimate the
ultimate strength especially at Pr = 0.2 and 0.4.
3.4. EI-Zanatys portal frame under proportional gravity and lateral
loads
The portal frame shown in Fig. 17 is reanalyzed under proportional lateral and vertical loads. The ratio of lateral load H to

(a) With residual stress (r /y = 0.5).

A simplified model was developed to predict the second order inelastic behavior of steel frames. The model predicts the
behavior of steel frames with/without the consideration of residual stress effect. The residual stress distributions were considered
as recommended by the European Convention for Construction
Steelwork (ECCS). New simplified formulae for tangent stiffness
of the steel cross sections are proposed, moreover the section incremental secant stiffness was derived. The derived formulae were
implemented into a finite element program based on stiffness analysis to predict the full behavior of steel plane frames. The NewtonRaphson method was used with the minimum residual stress
method to evaluate the second order inelastic behavior of the
frames. The simplified formulae were used to analyze the frames
without any discretization of the cross section. The analysis results
of the simplified model are compared to those predicted by using the fiber model. The comparisons indicate that the simplified
model correlates well with the results of the fiber model without
the need of cross section discretization. It was found that neglecting the residual stress may overestimate the ultimate capacity of
planar steel frames especially at higher levels of axial loads and
slenderness ratios. The main advantage of the present model is its

(b) Without residual stresses.

Fig. 19. Loaddeflection relationship for El-Zanatys frame with proportional loads.

Author's personal copy

3268

A.H. Zubydan / Engineering Structures 32 (2010) 32583268

simplicity of use and the saving of computational time and data


storage.

Values of Mr1 , Etr1 .


For H-sections (with no residual stress)
Mr1 = 0.96Pr + 0.97

(A.1a)

Etr1 = 3.12Pr + 0.06

(A.1b)

for Pr 0.05

Etr1 = 0.22 for 0.05 < Pr 0.67

(A.1c)

Etr1 = 0.62Pr + 0.64

(A.1d)

for Pr > 0.67.

For H-sections (with residual stress)


Mr1 = 4.37Pr2 + 0.14Pr + 0.98 for Pr 0.15

(A.2a)

Mr1 = 1.06Pr + 1.07

(A.2b)

for Pr > 0.15

Etr1 = 0.04 for Pr 0.4

(A.2c)

Etr1 = 0.04 (1 (Pr 0.4)/0.6)

for Pr > 0.4.

(A.2d)

For I-sections (with no residual stress)


Mr1 = 0.91Pr + 0.93

(A.3a)

Etr1 = 5.04Pr + 0.13

(A.3b)

for Pr 0.05

Etr1 = 0.14Pr + 0.39

for 0.05 < Pr 0.67

(A.3c)

Etr1 = 0.89Pr + 0.91

for Pr > 0.67.

(A.3d)

For I-sections (with residual stress)


Mr1 = 2.9Pr2 + 0.36Pr + 0.95

for Pr 0.15

Mr1 = 1.1Pr + 1.11 for Pr > 0.15


Etr1 = 0.08 for Pr 0.4

(A.4a)
(A.4b)
(A.4c)

Etr1 = 0.08 (1 (Pr 0.4)/0.6)

for Pr > 0.4.

(A.4d)

Values of r1 , r2 .
For H and I-sections (with no residual stress)
r1 = r2 = 1.

(A.5)

For H-sections (with residual stress)


r1 = 1.5 0.3 (Pr /Pr0 )3

for Pr Pr0

r1 = 1.2 + 0.8 ((Pr Pr0 )/(0.75 Pr0 ))


for Pr0 < Pr 0.75

r2 = 3

for Pr > 0.75

(A.6a)
(A.6b)
(A.6c)

for Pr Pr0

for Pr > Pr0 .

(A.6d)
(A.6e)

For I-sections (with residual stress)


r1 = 1.5 + 0.1 (Pr /0.15)

for Pr 0.15

(A.7a)

r1 = 1.6 0.3 ((Pr 0.15)/(Pr0 0.15))


for 0.15 < Pr Pr0

(A.7b)

r1 = 1.2 + 0.8 ((Pr Pr0 )/(0.75 Pr0 ))


for Pr0 < Pr 0.75

(A.7c)

r1 = 2

for Pr > 0.75

r2 = 1.5 + 1.5 (Pr /Pr0 )


r2 = 3

for Pr > Pr0 .

Values of r3 .

(A.8a)

r3 = 1 0.3(Pr 0.15)

The values of Mr1 , Etr1 , r1 , r2 and r3 imbedded in Eq. (10) are


given as follows:

r2 = 1.5 + 1.5Pr /Pr0

r3 = 1 for Pr 0.15
0.1

Appendix

r1 = 2

For H-sections

(A.7d)
2

for Pr Pr0

(A.7e)
(A.7f)

for 0.15 < Pr 0.25

r3 = 0.7 for Pr > 0.25.

(A.8b)
(A.8c)

For I-sections
r3 = 1 for Pr 0.2

(A.9a)

r3 = 1 3(Pr 0.2) for 0.2 < Pr 0.3

(A.9b)

r3 = 0.7 for Pr > 0.3.

(A.9c)

References
[1] Alvares RJ, Birnstiel C. Inelastic analysis of multistory multibay frames. J Struct
Eng Div, ASCE 1969;95(11):2477503.
[2] Latona RW. Nonlinear analysis of building frames for earthquake loading. Ph.D.
thesis. Cambridge (Mass): Massachusetts Institute of Technology; 1970.
[3] El-Zanaty M, Murrary D, Bjorhovde R. Inelastic behavior of multistory steel
frames. Structural engineering report no. 83. Alberta (Canada): University of
Alberta; 1980.
[4] White DW. Material and geometric nonlinear analysis of local planar behavior
in steel frames using iterative computer graphics. M.S. thesis. Ithaca (NY):
Cornell University; 1985. p. 281.
[5] Vogel U. Calibrating frames. Stahlbau 1985;10:17.
[6] Clarke MJ, Bridge RQ, Hancock GJ, Trahair NS. Design using advanced analysis.
SSRC Annual Tech. Session Proc. Bethlehem: Lehigh Univ.; 1991. p. 2740.
[7] Jiang XM, Chen H, Liew JYR. Spread-of-plasticity analysis of three-dimensional
steel frames. J Constr Steel Res 2002;58:193212.
[8] Kim SE, Choi SH. Practical second order inelastic analysis for three dimensional
steel frames subjected to distributed load. Thin-Walled Struct 2005;43:
13560.
[9] Ziemian RD. Advanced method of inelastic analysis in the limit states of steel
structures. Ph.D. dissertation. Ithaca: Cornell University; 1990.
[10] Liew JY, White DW, Chen WF. Second order refined plastic hinge analysis of
frames design: part I. J Struct Eng, ASCE 1993;119(11):3196216.
[11] King WS. A modified stiffness method for plastic analysis of steel frames. Eng
Struct 1994;16:16270.
[12] Kim SE, Chen WF. Practical advanced analysis for braced steel frame design.
J Struct Eng, ASCE 1996;122(11):126674.
[13] Kim SE, Chen WF. Practical advanced analysis for unbraced steel design. J Struct
Eng, ASCE 1996;122(11):125965.
[14] Chen WF, Kim SE. LRFD steel design using advanced analysis. Boca Raton (FL):
CRC Press; 1997.
[15] Chan SL, Chui PT. A generalized design-based elastoplastic analysis of steel
frames by section assemblage concept. Eng Struct 1997;19(8):62836.
[16] Liew JYR, Chen H, Shanmugam NE, Chen WF. Improved nonlinear plastic hinge
analysis of space frame structures. Eng Struct 2000;22:132438.
[17] Kim SE, Lee DH. Second order distributed plasticity analysis of space steel
frames. Eng Struct 2002;24:73544.
[18] Ziemian RD. Advanced method of inelastic analysis in the limit states of steel
structures. Ph.D. dissertation. Ithaca: Cornell University; 1990.
[19] White DW. Advanced analysis/design of typical moment frame. In: Proc., 10th
struct. cingr. compact papers, ASCE. 1998. p. 3303.
[20] Challa MV. Nonlinear seismic behavior of steel planar moment resistance
frames. Rep. No. EERL 82-01, Earthquake Engrg. Res. Lab., Pasadena (Calif.):
California Institute of Technology; 1992.
[21] Liew JYR, Chen WF. Chapter 1: trends toward advanced analysis of steel
frames. In: Chen WF, Toma , editors. Advanced analysis of steel frames: theory,
software, and applications. Boca Raton (Fla): CRC Press; 1994. p. 143.
[22] Nanakorn P, Vu LN. A 2D field-consistent beam element for large displacement
using the total Lagrangian formulation. Finite Elem Anal Des 2006;42:
12407.
[23] Yang YB, Lin SP, Leu LJ. Solution strategy and rigid element for nonlinear
analysis of elastically structures based on updated Lagrangian formulation.
Eng Struct 2007;29:1189200.
[24] Duan L, Chen WF. Design interaction equation for steel beamcolumns. J Struct
Eng ASCE 1989;115(5):122543.
[25] European convention for constructional steelwork, Ultimate limit state
calculation of sway frames with rigid joints, ECCS. Technical working Group
8.2, Systems, Publication No. 33. 1983.
[26] Load and resistance factor design specification for structural steel buildings.
1st Ed. Chicago: American Institute of Steel Construction (AISC); 1986. p. I11.
[27] McGuire W, Gallagher R, Ziemian R. Matrix structural analysis. New York:
Wiley; 2000.
[28] Chan SL. Geometric and material non-linear analysis of beamcolumns and
frames using the minimum residual displacement method. Internat J Numer
Methods Engrg 1988;26:265769.
[29] White DW. Material and geometric nonlinear analysis of local planar behavior
in steel frames using interactive computer graphics. Report No. 86-4. Ithaca
(NY): Department of Structural Engineering. Cornell University; 1986.

You might also like