Engineering Structures 32 (2010) 1199–1214

Contents lists available at ScienceDirect

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

An efficient mathematical programming method for the elastoplastic analysis

of frames
Konstantinos V. Spiliopoulos ∗ , Theodoros N. Patsios
Institute of Structural Analysis & Antiseismic Research, Department of Civil Engineering, National Technical University of Athens, Zografou Campus, Zografos 157-80, Athens, Greece

article info abstract

Article history: The elastoplastic nonholonomic analysis of frames is a nonlinear procedure in which the magnitude of the
Received 17 August 2009 structural loading is incrementally modified using a proportional load factor, in accordance with a certain
Received in revised form sequence of predefined loading patterns. It is an attempt by the structural engineering profession to
12 December 2009
estimate the strength as well as the deformations of framed structures under a given loading. In this work
Accepted 22 December 2009
Available online 26 March 2010
an analysis based on the force method and mathematical programming is presented. An elastic –perfectly
plastic material is assumed and conventional plastic hinges of zero length are used to model the plasticity
Keywords:
effects. The basis of the approach is the formulation of the incremental problem as a convex parametric
Numerical methods quadratic programming (PQP) problem between two successive plastic hinges. A novel numerical strategy
Quadratic programming is proposed that uses a fictitious load factor to convert the PQP problem to a QP one. The solution of
Force method the QP problem, by an effective standard algorithm, establishes a feasible direction on which the true
Graph theory solution lies. The real solution is then found, simply on the demand of the formation of a new plastic hinge
Plastic hinge that is closest to open. Possible plastic unstressing is automatically accounted for. The approach is first
Plane bending developed for pure bending behaviour and is then extended to cater for moment/axial force interaction.
Moment/axial force interaction Examples of application under monotonic, variable, and cyclic loading conditions are included. The whole
Limit analysis
procedure appears to be stable, robust, and computationally efficient as it requires much less time than
Pushover analysis
Cyclic loading the alternative displacement based direct stiffness method.
© 2009 Elsevier Ltd. All rights reserved.

1. Introduction in a subsequent paper [2]. The formulation of the problem as a

parametric linear complementarity problem (PLCP) has been given
Elastoplastic analysis is an important procedure to determine by Maier [3]. Linear programming (LP) is the main ingredient
the capacity of a structure beyond its elastic limits. In the course to solve such problems and a numerical solution based on the
of this analysis, the external loads are continuously applied with Simplex method has appeared in De Donato and Maier [4] together
more and more structural components yielding. A series of elastic with some examples and an extension to include non-proportional
analyses are therefore generated by modifying the mathematical loading.
model of the structure to account for reduced resistance of yielding Smith [5] and Maier et al. [6], almost at the same time,
components. The procedure consists of the superposition of these proposed numerical solutions of the PLCP problem based on the
analyses and stops when the structure cannot carry any further Simplex method and enforcing complementarity at each pivotal
load and becomes unstable, or until a predetermined load limit is step, thus, restricting the variables to enter the basis. With slight
reached. Thus a good estimate of the strength of the structure as modifications of the algorithms, they managed to extend the
well as of its ductility can be made. solutions to the case of nonholonomic plasticity. Extension to
In an early work, Maier [1] has shown that quadratic piecewise proportional loading is also reported in [6].
programming (QP) provides a unified theoretical framework for In an attempt to provide general purpose computer programs
the elastoplastic analysis of frames. In the context of holonomic Franchi and Cohn [7] produced a rather involved PLCP based
algorithm and applied it to a single storey plane frame. Kaneko [8]
plasticity various QPs for total quantities are written down taking
used the same formulation but worked directly on incremental
into account perfectly plastic, work-hardening or even softening
quantities and was able to assess nonholonomic plasticity without
behaviour. An incremental form of a QP in terms of kinematic
having to make any computational adjustments (as in [5,6]).
variables also appears in the same work. Duality QPs are presented
Wakefield and Tin-Loi [9] applied this method to grillages
and multi-storey frames. Formulations based on PLCP have
also recently been employed by Cocchetti and Maier [10] and
∗ Corresponding author. Tel.: +30 210 7721603; fax: +30 210 7721604. Tangaramvong and Tin-Loi [11,12] to account for local softening
E-mail address: [email protected] (K.V. Spiliopoulos). behaviour.
0141-0296/$ – see front matter © 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2009.12.045
1200 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214

All the aforementioned procedures for the elastoplastic analysis a b

of frames are based on mathematical programming (MP). One
should only note here, that in the context of the evaluation of the
limit load of frames only, Marin-Artieda and Dargush [13] have
presented an approach that uses the Linear Matching method,
which is not formulated within MP but appears to have difficult
convergence properties. A more thorough presentation of the
approach together with improved convergence appeared quite
recently in Barrera et al. [14]. Fig. 1. Proportional loading: (a) limit load analysis, (b) prescribed loading analysis.
Although MP provides an ideal mathematical framework
for encoding the basic theory of elastic–plastic structures, the
solutions schemes referred to, as above, generally involve large
number of variables and constraints (Tin-Loi and Wong [15]). For
this reason the, displacement based, direct stiffness method has
been used almost exclusively and all commercial programs are
based on it (e.g. SAP2000 [16]). Following this method, an event-to-
event strategy is employed which takes into account the reduction
of the resistance whenever a plastic hinge occurs or the increase in
the resistance of a plastic hinge whenever local unloading occurs Fig. 2. Rotations and moments at member’s ends.
by re-formulating and re-decomposing the stiffness matrix.
The main issue in the elastoplastic analysis of frames is
the requirement of a statically admissible and safe distribution The procedure is formulated with respect to proportional or
of stresses throughout the whole structure. The most natural piecewise proportional loading. Since bending is the prevailing
formulation therefore appears to be within the framework of the mode of deformation for framed structures, in the Sections 2–
force method of analysis, since equilibrium may be expressed 4 of the present paper, the procedure is developed under the
accurately as a linear combination of the hyperstatic forces and assumption of a pure bending behaviour. The essential features
the applied loading. On the contrary, if one uses the displacement of the method are also discussed in these sections. In Section 5
method of analysis, a degree of approximation is needed (Pereira the procedure is extended to include also the effects of the axial
et al. [17]). Nevertheless, the displacement method has been force. Examples of application appear in Section 6 that indicate
almost exclusively in use, because it is easier to automate. the effectiveness of the method for either monotonic loading,
The main problem with the force method towards this loading scenarios or cyclic loading. The computational efficiency
automation is the way to pre-select the hyper-static forces, which of the procedure compared to the direct stiffness method is also
are the main unknowns. Approaches to deal with this problem are demonstrated.
to set up a displacement method based environment and transform
it to a force method based one, using algebraic methods (Damkilde 2. Governing equations for pure bending
and Høyer [18]). This transformation, of course, involves a degree
of approximation. Let us suppose that a frame, whose material behaves as
It has been realized, quite early (e.g. Spillers [19]), that graph elastic–perfectly plastic, is subjected to a proportionally changing
theory and the graph representation of a frame may provide a loading pattern of the form
direct way to automate the force method. It may be proved that, P = Pin + γ · rP (1)
for a ‘planar graph’ there is a unique number of closed loops
called cycle basis. Spiliopoulos [20] has proposed a relatively easily where, using bold letters to represent vectors and matrices
programmable algorithm which constructs such a basis using a throughout, Pin is an initial loading state, γ is a proportional
minimum path technique between two nodes of a graph. This loading parameter which controls the further application of
algorithm has been used together with LP for the optimum plastic loading, rP is the unit vector along the direction of the loading
design of frames under monotonic or variable loading (Spiliopoulos pattern (Fig. 1). Either for the case of a limit analysis or a prescribed
[20,21]). loading analysis, rP may be directly computed. In the prescribed
In this work, the numerical solution of the small displacement loading case, for a particular branch, with initial and final loading
elastoplastic analysis of frames, using the force method is states Pin and Pf respectively, rP = (1/kPL k) · PL , where PL =
presented. The material is considered elastic rigid-plastic with Pf − Pin (Fig. 1(b)).
nonholonomic behaviour, whereas the adopted model is a plastic In response to the external loading, every member of the
hinge model of zero length. The basis of the formulation consists in structure, which is defined as a line/curve lying between two
decomposing the increments of the stress resultants in two terms, nodes, develops two rotations at its ends, relatively to a local axis
one due to the indeterminacy of the structure and one due to the defined by its chord; these rotations may be decomposed into an
increments of the load. The problem is cast, not as the solution elastic and a plastic component (plastic hinge approach) (Fig. 2).
of a PLCP, but as a direct solution of the parametric quadratic One may write the elastic rotations in terms of the end moments
program (PQP) at the beginning of each incremental loading step, using the member’s flexibility matrix:
the parameter being the incremental load factor. A novel numerical
θ1 `
 el     
strategy is proposed which, by employing a fictitious load factor, 2 1 m1
= · · (2)
converts the solution of this PQP to the solution of a simple QP. θel2 6EI 1 2 m2
This QP may be solved using standard algorithms (e.g. Goldfarb and where ` and EI are the member’s length and bending stiffness.
Idnani [22]). Possible plastic unstressing, in a stepwise-holonomic By grouping all the individual flexibility matrices of the
fashion, is automatically detected by the solution of the fictitious structure in a block-diagonal matrix Fm , one may write for all the
problem. One may, thus, predetermine the direction on which the elastic rotations and the end moments of the frame:
real solution of the current incremental step lies. This solution is
then found merely on the demand that a new plastic hinge forms. θel = Fm · m. (3)
 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214 1201

Fig. 3. Rigid-plastic behaviour and plastic unstressing.

n +o
y∗
where y∗ = y−
collects the plastic potentials at every cross
∗
Due to the nonlinear (elastoplastic) behaviour, the solution section that are defined in Fig. 3.
will be acquired in incremental steps. At the end of such a step, If we denote by mk−1 the moments of the previously converged
an increment of the applied loading will create an increment of incremental step, the non-negativity of the plastic potentials
moments which may be determined, using the force (or mesh) makes possible to express the static admissibility condition of the
description: bending moments at the current incremental step k as:
1m = Bm · 1p + 1γ · Bo,m · rP . (4)
y∗ + NT · (1m + mk−1 ) = m∗ (10)
The first term is due to the indeterminacy of the structure with p
being a set of hyperstatic forces, called a statical basis. These forces n
m∗+o
may be introduced in the structure by making cuts at each closed where m∗ = m−
collects the plastic capacities of the cross
∗
loop of the indeterminate frame and converting it to a determinate sections in positive and negative bending, respectively (Fig. 3).
one. The second term is due to the equilibrium with the increment Eqs. (7)–(10), combined with (4), may serve as the Karush–
of the applied loads expressed through the increment 1γ of the Kuhn–Tucker relations that lead to the solution of the following
proportional load factor. quadratic program (QP) at every incremental step k:
The increments of the total rotations will then be given by:
1
Minimize z (1p) = · 1pT · (BTm · Fm · Bm ) · 1p
1ψ = 1θel + 1θpl . (5) 2
(11)
+ 1γk · (BTm · Fm · Bo,m · rP )T · 1p
From the principle of static kinematic duality (SKD), the
conjugate to the hyperstatic forces discontinuities at the cuts are Subject to: (NT · Bm ) · 1p ≤ (m∗ − NT · mk−1 ) − 1γk · (NT · Bo,m · rP )
related to the above increments of rotations through the matrix BTm .
with 1γk being the increment of the load factor at the current step.
Closing these cuts provides us with the conditions of compatibility,
It appears that a more accurate numerical solution may be
as was first proposed by Maxwell: achieved if each constraint is divided by its corresponding plastic
capacity so that the right-hand side of (10) is equal to one. So the
BTm · 1ψ = 0. (6) QP to be solved is:
Eq. (3) may be used to compute 1θ . By combining (4) and (5),
el
1
Minimize z (1p) = · 1pT · BTm · Fm · Bm · 1p


Eq. (6) becomes:
2
(12)
BTm · Fm · Bm · 1p + 1γ · B0,m · rP + BTm · 1θpl = 0. + 1γk · (BTm · Fm · Bo,m · rP )T · 1p
 
(7)
Subject to: (M∗ · Bm ) · 1p ≤ e − M∗ · mk−1 − 1γk · (M∗ · Bo,m · rP )
T T T
Depending on whether the bending moment that has reached
the plastic capacity m∗ of a cross section causes tension or where
compression at the bottom side of a cross-section (pre-selected
‘‘positive’’ fiber, Fig. 2) we may have positive or negative plastic
rotations with reference to the local coordinate system of the
member, respectively. So, in general, one may write:

(8)

where 1θ+ ∗ and 1θ∗ are positive numbers and constitute the
−

elements of 1θ∗ . with Nc being the total number of critical sections. The product
Assuming rigid plastic behaviour, the relationship between a BTm · Fm · Bm is the flexibility matrix of the structure for the selected
bending moment and its corresponding plastic rotation, whether set of the hyperstatic forces, and eT = {1 1 · · · 1} is a column
we have further loading, or plastic unstressing (nonholonomic vector.
behaviour), may be seen in Fig. 3, and may be expressed by a single The way to solve this QP problem will be discussed in Section 4.
equation (complementarity relation) as follows: Here, we should note that, the product of the Lagrange multipliers
of the optimal solution with M∗ provides the increments of the
yT · 1θ∗ = 0 where y ≥ 0, 1θ∗ ≥ 0 (9) plastic rotations, 1θ∗ .
1202 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214

a b

Fig. 4. (a) Typical cycle basis and shortest path cantilevers, (b) Self equilibrating system of forces.

a b c

Fig. 5. Mesh base formation.

This algorithm is easy to implement because it is based on

initially setting the lengths of each member (not in the Euclidean
sense) equal to 1. The procedure then starts from the node that
has the maximum number of members incident to it, finds the
minimum path between the ends of each of the members, not by
going along the member but going around it. The minimum path
together with the generator member forms a cycle candidate to
enter the mesh basis. It will enter the mesh basis if the following
admissibility rule is satisfied:
‘‘The length of the path is less that 2*(nodes along the path-1)’’.
If this rule is satisfied the cycle enters the basis and all the
members of this cycle get the length of 2.
As an example of a cycle basis formation, the procedure that
was just described is used to establish a cycle basis in the sub graph
(Fig. 5(a)) that has been extracted from a main graph. Starting from
node k, the cycle klmk is selected (Fig. 5(b)) using km as a generator
member and all the lengths of the members of the cycle take the
value of 2. There is no way of re-entry in the basis of this cycle, since
the admissibility rule is not satisfied. By picking up a next member,
e.g. (mn), a next cycle may be selected to enter the basis (Fig. 5(c)).
Fig. 6. Fictitious (ρ ), and true incremental load factors (1γ ). There are cases of complicated graphs that this simple process may
break down, but there are remedies to overcome this problem [20].
By making a cut at each cycle, one may establish a pair of two
3. Selection of the hyperstatic forces unknown forces Xo , Yo along the x and y directions and an unknown
bending moment Mo at the point of the cut, with coordinates xo and
yo . These are the three indeterminate entities for the cycle at hand
The automation of the process depends on the selection of the (Fig. 4(b)). The bending moment at any cross section i along the
hyperstatic forces which is called statical basis. An algorithm has cycle with coordinates xi and yi , are given by:
been published (Spiliopoulos [20], where more details may be
Xo
( )
found), which finds such a basis by selecting a set of independent
mi = (±) (yo − yi ) (xi − xo )
 
−1 · Yo (13)
cycles which is equal to Betti’s number for planar graphs. Each
Mo
frame is such a graph (Fig. 4(a)) and Betti’s number is equal to
µ − ν + 1 where µ, ν are the number of members and nodes where the positive or the negative sign depends on whether the
that compose the graph. The ground is represented by an extra mesh orientation coincides with the member orientation, that
node and extra members are used to connect this node to each the cross section belongs to, or not. By filling in the appropriate
foundation node. positions the matrix Bm may be formed.
 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214 1203

Fig. 7. Various types of failure criteria considering bending moment & axial force interaction.

3.1. Equilibrium with external forces 1. Adopt a ‘‘fictitious’’ small initial value for 1γk = ρ .
2. Solve the QP problem (12) and obtain a ‘‘fictitious’’ set of incre-
Equilibrium with the applied loading is accomplished through ments of hyperstatic forces 1p̃ and, using the Lagrange multi-
the use of cantilevers, which mark the quickest way to the ground pliers of the optimal solution, a set of ‘‘fictitious’’ increments of
of the points of application of the loads (Fig. 4(a)). For a cross plastic rotations 1θ̃∗ . The QP algorithm [22] is used.
section i located along this way, the bending moment m is given 3. Normalize the increments of the ‘‘fictitious’’ set of hyperstatic
by: forces and plastic rotations:
 
Px 1 1
mi = (±) [(ya − yi ) (xa − xi )] · (14) 1p0 = · 1p̃ and 1θ0∗ = · 1θ̃∗ . (15)
Py ρ ρ
with xa and ya the coordinates of the point where the concentrated 4. Evaluate the corresponding incremental bending moment dia-
loads are applied. gram using (4), normalized with respect to ‘‘ρ ’’:
The positive sign in the parenthesis is valid if the orientations 1m0 = Bm · 1p0 + Bo,m · rP . (16)
of the member that this section belongs to, and the direction of the 5. Find the correct 1γk as the minimum 1γi,k that produces a new
cantilever, coincide. either positive or negative plasticization of a critical cross sec-
Using (14) for all the critical sections and all the loads, the tion i:
matrix Bo,m may be constructed.
mi,k−1 + (1γi,k ) · 1m0i = m+ or
Distributed loading of a member may be approximated by ∗,i
splitting it into a set of finite elements of equal length, and applying mi,k−1 + (1γi,k ) · 1m0i = −m−
∗,i (17)
statically equivalent point loads at their nodes. For a more precise
implementation of the distributed loading, one may include an for i = 1, 2, . . . , Nc , where Nc is the total number of critical
additional term in Eq. (3) that corresponds to the free elastic sections
6. Find the increments of the bending moments and the plastic ro-
rotations (see for example, [5]).
tations:
4. Proposed numerical strategy 1m = 1γk · 1m0 and 1θ∗ = 1γk · 1θ0∗ . (18)
7. Update the load factor and the various static and kinematic vari-
The QP problem (12) is a parametric one, since, although the ables:
basic unknowns are the hyperstatic forces 1p, the parameter γk = γk−1 + 1γk
1γk should also be supplied. This parameter may be estimated mk = mk−1 + 1m
requiring that each load increment ends with the formation of a θelk = Fm · mk (19)
new plastic hinge.
In this work, a novel numerical strategy to solve directly the QP
θpl pl
k = θk−1 + 1θ
pl

pl
problem (12) is suggested. Starting with an initial value of γ = 0 ψk = θk + θk
el

and k = 1, the following steps describe this strategy: where use of (8) is made.
1204 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214

a b

Fig. 8. (a) Search direction for plasticization from elastic state (b), (c) Further plasticization (1) or unloading (2).

A physical explanation on the use of a ‘‘fictitious’’ starting load

factor is that it enables one to establish a feasible direction on
which the incremental step lies, whose true length is then found on
the grounds of the formation of a new plastic hinge. The procedure
may be pictured for two steps on a force–displacement diagram
(Fig. 6).

5. Extension to bending moment and axial force interaction

The method presented above may be extended to cases with

stress resultant interactions, the commonest of which is between
bending moments and axial forces.
For fully plasticized orthogonal steel sections, this interaction
is generally known to be represented by the symmetric, quadratic
and convex curve given by the following equation:
 2
|m| n
+ =1 (21)
m∗ n∗
Fig. 9. Frame’s geometry, mechanical properties, loading, and critical section
numbering.
where m∗ and n∗ represent the section’s bending moment and axial
force bearing capacities.
The above yield curve may be approximated by a finite set of
The displacements at the loaded points may be found using
‘‘ζ ’’ independent linear equations for each of the four quadrants:
SKD between rotations and displacements:
m n
uk = BTo,m · ψk . (20) f (m, n) = (±)s1 · + (±) s2 · −1=0 (22)
m∗ n∗
8. Return to step 1 and repeat the process for k = k + 1, until with the positive or negative signs in the parentheses depending
either on the particular quadrant.
(a) no solution of the QP may be found, meaning a collapse state We have ‘‘ζ ’’ distinct couples of (s1 , s2 ).
has been reached and γk is the limit load factor, or The simplest linearization (ζ = 1), consists of the four lines
(b) if we have a prescribed loading case and (a) has not oc- shown dashed in Fig. 7. For the AISC LRFD criterion [23], on the
curred, the process stops if |γk − kPL k| ≤ ρ , meaning we other hand, ζ = 2.
have reached the end of the current branch. Now, at every critical cross-section, one should supply, besides
the bending moment, the axial force also. Eqs. (13) and (14) can be
The algorithm automatically detects whether at the beginning
found to take the form:
of the incremental step we have further plasticization of an already
plasticized critical section (point A, Fig. 3, movement along the
direction (1) on the yield plane) or local unloading (direction (2)).
(23)
This is determined from the value of the corresponding Lagrange
multiplier, which is readily available after the solution of the
‘‘fictitious’’ QP problem: in the case where this value is positive
we have further loading, or if the value is zero we have plastic (24)
unstressing.
 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214 1205

Loading scenario

Vertical load (kN)

Horizontal load (kN)

Fig. 10. Loading scenario’s coordinates for each analysis step, plasticization/local unstressing sequence, quantitative bending moment diagrams (units: kN, m).

with
xf − xs where .
cos ϕ = p and
(xf − xs )2 + (yf − ys )2
The generalized plastic displacement now at a critical section
yf − ys
sin ϕ = p i consists of two components; a plastic rotation and an axial
(xf − xs )2 + (yf − ys )2 discontinuity, which may be computed with the help of (22):
where (xs , ys ) and (xf , yf ) are the coordinates of the two ends of
the member that the critical section i belongs to (Fig. 4(b)) and the
∂f
 pl 
1θi s1 /m∗i
 
pl
positive signs hold under the same assumptions as in Section 3. 1qi = = 1λi · = 1λi · . (28)
pl
1δi ∂Q i s2 /n∗i
Elastic axial elongations will now appear besides the elastic
rotations, so that (3) will now look like:
Collecting the plastic rotations at the top and the axial disconti-
qel = F̄ · Q (25)
nuities at
n theo bottom, of all the critical sections, one may form
where 1θpl
1qpl = 1δpl
.

With complementarity now holding between 1λi and the

section’s generalized plastic potential (distance from the yield
with Fn relating the elastic axial elongations at the two critical surfaces of Eq. (22)), the QP program (Eq. (12)) may be written as:
sections at the ends of the member with the corresponding axial
forces, through its axial flexibility, `/EA. 1
Collecting all the bending moments and axial forces of the Minimize z (1p) = · 1pT · (B̄T · F̄ · B̄) · 1p
2
critical sections in the matrices m and n we may write at the (29)
current incremental step k: + 1γk · (B̄T · F̄ · B̄o · rp )T · 1p
mk = 1m + mk−1 and nk = 1n + nk−1 (26) Subject to: (N̄T · B̄) · 1p ≤ e − N̄T · Qk−1 − 1γk · (N̄T · B̄o · rp )
where 1m and 1n may be computed with the aid of Eqs. (23) and
(24): where ‘‘N̄’’ is a matrix that stores in the appropriate positions
1m = Bm · 1p + (1γk ) · Bo,m · rp the various coefficients of the left-hand side of the constraints
 
1n = Bn · 1p + (1γk ) · Bo,n · rp (Eq. (22)). Depending on the number of the linear segments of
our yield criterion we may have ‘‘ζ ’’ distinct couples of (s1 , s2 ),
→ 1Q = B̄ · 1p + 1γk · B̄o · rp (27) i.e. (s1 , s2 ) = {(s11 , s21 ), (s12 , s22 ), . . . , (s1 ζ , s2 ζ )}. Therefore see
1206 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214

Load-Displacement Curves

Horizontal Load

Vertical Load

Load (kN)

Displacement (m)

Fig. 11. Load vs. corresponding displacement curves.

Box I.

Box I, where: multipliers 1λ̃i of the optimal solution, a set of ‘‘fictitious’’

increments of the generalized plastic displacements 1q̃pl . The
QP algorithm [22] is used.
3. Make a first correction to the fictitious set of hyperstatic forces
1p̃ and the various 1λ̃i . Also, use (28) to evaluate a set of
pl
fictitious 1q0 :
1
1p0 = · 1p̃
ρ
and (30)
s1 /m∗i
 
1 pl
1λ0 = · 1λ̃ → 1q0i = 1λ0i · .
ρ s 2 / n ∗i
4. Evaluate ‘‘fictitious’’ increments of bending and axial forces
using (27):

1Q0 = B̄ · 1p0 + B̄o · rp (31)

and establish the search direction 1Q0i , for each cross section,
to determine its next possible plasticization. This occurs at the
intersection with one of the yield planes (Fig. 8(a)).
For the simple criterion: ζ = 1, s11 = s21 = 1, whereas for the 5. Find the correct 1γk as the minimum 1γi,k among the non-
ASCI LRFD criterion [23], ζ = 2, s11 = 8/9, s21 = 1, s12 = 1, s22 = active constraints that produces a new plasticization at a critical
1/2. section (Fig. 8(a)):
The Lagrange multipliers of the optimum solution of (29) provide
the various 1λi . (αi · ni,k−1 + βi ) − mi,k−1
1γi,k = (32)
The numerical strategy, presented in Section 4, may now be 1m0i − αi · 1n0i
modified to take into account the moment/axial force interaction. for i = 1, 2, . . . , Nc , where Nc is the total number of critical
So, starting with γ = 0 and k = 1: sections. Note that parameters (αi , βi ) in the above relation may
1. Adopt a ‘‘fictitious’ small initial value for 1γk = ρ . be evaluated using Eq. (22):
2. Solve the QP problem (29), and obtain a ‘‘fictitious’’ set of s2 m∗,i m∗,i
increments of hyper-static forces 1p̃ and, using the Lagrange
αi = − · and βi = .
s1 n∗,i s1
 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214 1207

Fig. 12. Structure’s geometry; members’ & critical sections’ numbering; mechani-
cal properties; external load pattern.

Fig. 15. Frame’s geometry, loading, member & critical section numbering;
mechanical properties.

7. Update the load factor and the various static and kinematic
Fig. 13. Collapse mechanism.
variables:
γk = γk−1 + 1γk
Load-Displacement Curves
mk = mk−1 + 1m

→ Qk
nk = nk−1 + 1n

θelk = Fm · mk

(34)
→ qelk = F̄ · Qk
δelk = Fn · nk
pl pl
qk = qk−1 + 1qpl
pl
k + qk .
qk = qel
Load (kN)

The displacements at the loaded points may be found using SKD

uk = B̄T0 · qk . (35)
8. Return to step 1 and repeat the process for k = k + 1, until
either
(a) no solution of the QP may be found, meaning a collapse state
has been reached and γk is the limit load factor, or
(b) if we have a prescribed loading case and (a) has not
occurred, the process stops if |γk − kPL k| ≤ ρ , meaning we
have reached the end of the current branch.
Once again, the algorithm automatically detects any further
Floor Displacement (m)
plasticization of an already plasticized critical section (point A on
Left vertical Load Horizontal Load Right Vertical Load Fig. 8(b) and (c) — equivalent to the movement along the directions
(1) on the yield plane(s)) or local unloading phenomena (direction
Fig. 14. Load vs. corresponding displacement bearing capacity curves. (2)), based on whether the corresponding Lagrange multiplier of
the active constraint is positive or zero, respectively. We should
For each critical section, 1γi,k is the minimum positive among note here, that, only one constraint will be active when a cross
all numbers one would get using (32) for each of the four section is plasticized. Even when further plasticization continues
quadrants of the failure criteria. along a neighbouring constraint (Fig. 8(c)), the previously active
6. Find the increments of the bending moments, axial forces and constraint becomes inactive; the neighbouring constraint will
plastic displacements as: be now the only active constraint which gives rise to plastic
deformations’ increments. The only case where two neighbouring
1m = 1γk · 1m0 , pl constraints might be simultaneously activated is when the search
and 1qpl = 1γk · 1q0 . (33)
1n = 1γk · 1n0 direction (Fig. 8(a)) meets their point of intersection; two nonzero
1λi now appear for the same cross section. The plastic deformation
1208 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214

Table 1
Bending moment distribution for each analysis step (units: kN, m).
Member Section 1st 2nd 3rd 4th 5th 6th 7th 8th 9th

1 −29.53865 −31.26824 −36.14765 −39.64353 −42.10589 −43.49379 −45.00000 −45.00000 −45.00000

1
2 22.01398 23.40427 27.29530 30.10161 32.04830 33.37780 34.50878 35.11547 45.00000
3 22.01398 23.40427 27.29530 30.10161 32.04830 33.37780 34.50878 35.11547 45.00000
4 27.80468 28.73818 31.24522 32.82406 34.01334 33.74192 34.60057 35.00316 41.14113
2
5 27.80468 28.73818 31.24522 32.82406 34.01334 33.74192 34.60057 35.00316 41.14113
6 −60.18756 −62.96819 −69.00010 −72.64524 −74.66283 −78.00000 −78.00000 −78.00000 −78.00000
7 −14.63945 −15.43390 −16.60849 −17.50000 −17.50000 −17.41926 −16.05116 −15.58804 −10.00806
3
8 16.58555 17.50000 17.50000 17.50000 17.50000 17.50000 17.28631 17.50000 17.50000
9 −45.54811 −47.53430 −52.39161 −55.14524 −57.16283 −60.58074 −61.94884 −62.41196 −67.99194
10 64.13938 66.94652 72.86532 76.15108 78.00000 78.00000 78.00000 78.00000 78.00000
4
11 64.13938 66.94652 72.86532 76.15108 78.00000 78.00000 78.00000 78.00000 78.00000
12 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000
13 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000 −45.00000
5
14 41.03166 42.06610 45.00000 45.00000 45.00000 45.00000 45.00000 45.00000 45.00000

Table 2
Plasticization/local unstressing sequence & plastic rotations for each analysis step (units: rad).
Member Section 1st 2nd 3rd 4th 5th 6th 7th 8th 9th

1 – – – – – – • −0.00057 −0.00977
1
2 – – – – – – – – •
3 – – – – – – – – –
4 – – – – – – – – –
2
5 – – – – – – – – –
6 – – – – – • −0.00180 −0.00285 −0.01620
7 – – – • −0.00068 −0.00068 −0.00068 −0.00068 −0.00068
3
8 – • 0.00196 0.00338 0.00409 0.00439 0.00439 • 0.00858
9 – – – – – – – – –
10 – – – – • 0.00169 0.00533 0.00738 0.03310
4
11 – – – – – – – – –
12 – – – – – – – – –
13 • −0.00111 −0.00396 −0.00538 −0.00631 −0.00742 −0.00965 −0.01110 −0.03060
5
14 – – • 0.00100 0.00171 0.00205 0.00250 0.00292 0.00977

Table 3 included to implement the algorithm of [22]. Overall, a value of

Loading cycle (units: kN). ρ = 10−3 to 10−4 has proved sufficient for all examples presented
Loading cycle point P1 P2 P3 P4 herein.
1 86.0 43.0 258.0 129.0 Five examples of application are used to demonstrate the
2 0.0 0.0 0.0 0.0 versatility of the proposed procedure. The first four examples
3 86.0 43.0 0.0 0.0 assume a pure bending behaviour. All kinds of inelastic analyses
4 0.0 0.0 0.0 0.0
like monotonic loading, piecewise proportional loading, cyclic
loading and pushover analysis are tested. In the fifth example
increments for this cross section will now be evaluated as the moment/axial force interaction is considered. Both the simple and
composition of two vectors [1], one for each yield plane, that are the AISC LFRD criteria have been applied.
determined using (28). Nonholonomic behaviour is assumed throughout as it was
The procedures described are stable, robust and computation- shown above that it may easily be accounted for and it represents
ally efficient. The flexibility matrix is established at the first step a more realistic behaviour, despite the fact that most commercial
and is decomposed only once. There is no need of any further refor- packages do not have or have improper facilities to accommodate
mulation or re-decomposition whenever a new plastic hinge or any it (e.g. SAP2000 [16]). The computational efficiency of the
local unloading might take place. The QP program is solved only approach is demonstrated on a problem of a relatively large
once at each incremental step and the step length that marks the scale.
next plasticization is automatically determined without having to
perform unnecessary intermediate elastic steps of fixed length as 6.1. Simple frame under prescribed loading
would be the case in any of the existing computer packages that
are based on the direct stiffness method. The frame of Fig. 9 is analyzed as a first example. The bending
The parameter ‘‘ρ ’’ is a pure number, and does not depend on stiffness, as well as the bending plastic capacity of the various
the adopted units. Although the procedure is stable for any ‘‘ρ ’’, members is shown in the same figure.
for good accuracy reasons, its’ value is chosen so as to be able to This example was solved by hand in Smith and Munro [25]. It is a
capture all possible plasticization events, no matter how close they prescribed loading case, with the horizontal load being applied first
are to each other. up to the level of 50 kN, then the vertical load is applied up to the
level of 160 kN with the horizontal load being kept constant, and
6. Numerical examples finally the two loads are proportionally varied, the horizontal being
decreased to 20 kN, whereas the vertical is increased to 180 kN. The
A computer program that follows the above described proce- loading scenario may be seen in Fig. 10. All the important events
dure was written in FORTRAN. The IMSL routine DQPROG [24] was are shown in the same figure. It may be seen that a plastic hinge
 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214 1209

Load- Displacement Curve

Base Shear (kN)

Roof Horizontal Displacement (m)

Fig. 16. Base shear vs. roof horizontal displacement curve.

Fig. 17. (a) Structure’s geometry, seismic loads, member numbering. (b) Uniform dead load on beams.
1210 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214

Fig. 18. (a) Bending moment diagram on collapse (units: kN m). (b) Collapse mechanism.

forms at section 2 first. Then, it unloads at the third step resulting This example has been solved in the work of Franchi and Cohn [7],
(2)
to a remaining θpl = +3.33E − 02 rad, while another hinge forms using a rather complex algorithm.
at section 7. At the fifth step, this section unloads resulting to a The evolution of the bending moments may be seen in Table 1.
(7)
remaining θpl = −1.33E − 01 rad, while another hinge forms at The sequence of plasticization/local unloading events, as well as
(4) the plastic rotations’ values, may be seen in Table 2, where a bullet
section 4, leaving a θpl = +8.33E − 03 rad, at the end of the sixth
(•) symbolizes the activation of the corresponding plastic hinge
step (see Fig. 10).
at the current step. Local unstressing of a previously activated
The evolution of the bending moments is shown in Fig. 10. hinge is denoted with italics. As it can be seen from Table 2, the
Results coincide with the ones obtained in [25]. cross section 7 starts unloading at the sixth step and continues
The load–deflection curves for the two loads and the corre- unstressing till the end. On the other hand, the cross section 8
sponding displacements are shown in Fig. 11. unloads at the seventh step, but reloads at the next step and
remains plasticized through the end. The collapse mechanism
6.2. Limit analysis of a two-bay frame appears in Fig. 13 and the bending moment distribution at the state
The second example of application is the limit analysis of a two of collapse in the last column of Table 1.
bay frame shown in Fig. 12, with λ being the load factor. Despite Finally the various load vs. corresponding deflection curves for
the fact that the loading is monotonic, plastic unstressing occurs. the three loads appear in Fig. 14.
 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214 1211

Load–Displacement Curves
Base Shear (kN)

(SAP2000)

(Proposed Method)

Roof Displacement (m)

Fig. 19. Static pushover analyses curves.

6.3. Two-storey frame under cyclic loading

Cyclic loading, which is a special case of prescribed loading, is

chosen as the next application. The example consists of a two-
storey frame that is subjected to horizontal and vertical loads
(Fig. 15). The mechanical properties of the members can be seen
in Fig. 15. Results are presented for L = 1 m.
The loading cycle consists of applying both horizontal and
vertical loads at first, then unloading to zero, then applying
the horizontal loads only, and unloading to zero again. The
maximum values that the loads reach inside the cycle (Table 3)
have been selected so that they are above their corresponding
shakedown values Psh = {P1 , P2 , P3 , P4 }sh = {82.296, 41.148,
246.888, 123.444} (units: kN) that have been estimated by
Nguyen and Morelle [26]. The bending moment diagram that
develops inside a cycle may be seen in Table 4. From these results Fig. 20. Frame geometry, member numbering and initial loading coefficients.
(units: kN).
one realizes the development of plastic hinges at sections 4, 6, 8,
11 and 14 (see Fig. 15) in the first part of the loading cycle which
all eventually unload in the second part, whereas a new plastic a
hinge opens at section 1 at the third part of the cycle. These hinges
b
form an incremental collapse mechanism. Also, in the same table,
one may see the values of the plastic rotations at the end of the
first cycle, as well as their increments over the second cycle. These
increments become constant over the third and subsequent cycles
leading to incremental collapse. The base shear (loads P1 + P2 ) vs.
roof horizontal displacement curve for the first four loading cycles
may be seen in Fig. 16. Despite the fact of the formation of plastic
hinges very close to each other, the procedure with ρ = 0.0001
is capable of capturing all of them, thus proving the method’s
effectiveness once more.

6.4. Pushover analysis of a multi-storey frame

This example is a static pushover analysis on a five-storey plane

frame, and is presented here to demonstrate the algorithm’s effi- Fig. 21. Collapse mechanism and plasticization/local unstressing sequence for each
criterion: (a) (m/m∗ ) + (n/n∗ ) = 1. (b) AISC LFRD.
ciency on relatively large-scale problems of common engineering
practice.
In Fig. 17(a) the structure’s geometry and member numbering (with EI = 65751.0 kN m2 , Mp = 555.28 kN m) were assigned
may be seen. The assumed material is S220. In order to achieve a to the columns, and to all beams (including the inclined members)
strong column-weak beam collapse mechanism, sections HEM260 sections HEB160 (with EI = 5233.2 kN m2 , Mp = 77.88 kN m). The
1212 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214

Load–Displacement Curves
a

Base Shear (kN)

Roof Horizontal Displacement (m)
b Fig. 23. Horizontal loads vs. corresponding horizontal displacement curves.

Pushover analysis with the proposed method is performed in

the form of a prescribed loading. At first, only the vertical (dead)
loads are applied; then the earthquake loads are added to the
existing load/stress state and are monotonically increased until
collapse occurs.
The best accuracy was acquired by using ρ = 0.0001. Plasticiza-
tion started from the small horizontal beams that are connected to
the exterior column on the right. In the course of the analysis, the
characteristic of a numerical nonholonomic analysis, i.e. the for-
mation of hinges that were loaded, unloaded and reloaded, or per-
manently unloaded, often appeared.
Results show that the plastic hinges develop close to or exactly
at the members’ ends. The values of the earthquake loads that
brought the structure to collapse are shown in Fig. 17(a). The
bending moments’ diagram at the state of collapse, for the whole
structure, appears in Fig. 18(a). The hinges in the final collapse
mechanism may be seen in Fig. 18(b).
The results of the proposed method were compared to those
of a widely used commercial package (SAP2000 [16]), that uses
the direct stiffness method. In order to match the total number of
the critical sections, 256 finite elements were used to model the
Fig. 22. (a) Bending moments distribution on collapse. (units: kN m). (b) Axial problem examined. As one may see from Fig. 19, the base shear
forces distribution on collapse. (units: kN). forces versus the roof displacement curves for both the analyses
are almost identical. The collapse mechanism was also identical.
earthquake loads were computed according to the EAK 2000 [27] The time required by SAP2000 to solve the problem, with the
having linear distribution along the height and are applied at the default analyses options, may be seen in Table 5. Note that, only
level of each floor (Fig. 17(a)), while the dead load for the various for inverting the stiffness matrix, 13.31 s were needed, while
types of beams of the structure may be seen in Fig. 17(b). the total time for calculations was 22.55 s. On the other hand,
For the sake of checking computational efficiency in terms using the proposed method, solution was acquired within only
of the number of constraints, and in order to approximate the 3.78 s. Computations were performed on an Intel Core2 Duo T8100
exact locations of plastic hinges along the beams more accurately, microprocessor (2.1 GHz), using only one of the CPU’s cores.
distributed loads are modeled as a finite set of quite close and
equally positioned, statically equivalent point loads. Two types of 6.5. Limit analysis of a three-storey frame with {m, n} interaction
beam vertical loading were appointed; beam members 26 to 40
& 46 to 54 were subjected to point loads equivalent to a uniform The fifth example shows the application of the proposed
loading of magnitude q = 9 kN/m, while members 41 to 45 & 55 algorithm to a 3-storey frame when considering (m, n) interaction.
to 59 to point loads equivalent to 10 kN/m (see Fig. 17(b)). Due to This example was first presented as a limit analysis problem by
this simulation, the structure ends up having a total of 502 critical Cohn & Rafay [28].
sections. Since for each critical section we have two inequality The structure’s geometry, loading, member numbering and
constraints, this yields a total of 1004 constraints. section types for each member, may be seen in Fig. 20; the three
 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214 1213

Table 4
Bending moments values inside a cycle; changes of plastic rotations over loading cycles (units: kN, m).
Member Section Bending moments for every loading cycle point θpl at the end of the 1st 1θpl at the end of the 2nd 1θpl at the end of each
cycle cycle subsequent cycle
1st 2nd 3rd 4th

1 −156.000 −67.108 −162.000 −42.106 – −3.300E−03 −3.300E−03

1
2 −17.357 −28.961 44.665 −28.942 – – –
3 −30.000 −28.397 77.704 −26.442 – – –
4 162.000 9.749 7.152 7.152 2.800E−02 6.600E−03 6.600E−03
2
5 162.000 9.749 7.152 7.152 – – –
6 −162.000 47.895 −63.401 40.745 −3.090E−02 −6.600E−03 −6.650E−03
7 −86.357 49.252 −34.102 39.504 – – –
3
8 162.000 11.105 146.233 26.339 4.280E−03 3.300E−03 3.305E−03
9 12.643 −0.564 −33.039 −2.500 – – –
4
10 −15.000 1.330 36.000 2.039 – – –
11 −81.000 3.252 −30.662 3.298 −2.780E−02 −6.600E−03 −6.650E−03
5
12 75.643 1.357 29.299 −1.241 – – –
13 −15.000 1.330 36.000 2.039 – – –
14 81.000 2.291 2.669 2.669 2.450E−02 6.600E−03 6.600E−03
6
15 81.000 2.291 2.669 2.669 – – –
16 −81.000 3.252 −30.662 3.298 – – –

Table 5 0.062 s. Computations were performed on an Intel Core2 Duo

SAP2000 v.14.0.0 computational time table. T8100 microprocessor (2.1 GHz), using only one of the CPU’s cores.
Time for initializing analysis = 0.10
Time for controlling analysis = 3.59
Time for forming stiffness matrix = 0.35 7. Concluding remarks
Time for solving stiffness matrix = 13.31
Time for calculating displacements = 4.11 A force based numerical procedure that deals with the
Time for determining events = 0.15 nonholonomic elastoplastic analysis of frames has been presented.
Time for updating state = 0.94
Total time for this analysis = 22.55 A relatively simple algorithm is used to select the hyperstatic forces
which are the basic unknowns.
The method is developed within the framework of mathemati-
cal programming. It is formulated as an incremental PQP problem.
decimal point accuracy is due to conversion from Imperial to S.I.
The PQP problem is converted to a QP problem through the use of
units.
a fictitious loading step. In this way, a good QP algorithm may be
Two types of analysis were performed; one based on the simple
used for the solution. No extra numerical care needs to be taken,
criterion, as in [28], and one based on the AISC LFRD criterion [23].
since plastic unstressing, in a stepwise holonomic way, is naturally
The value ρ = 0.001 was used for the fictitious load increment in
accommodated in the procedure.
both cases.
The method was first formulated for frame structures of pure
For the simple criterion case, the limit load factor was found bending behaviour. It was then extended to cater for axial force
λc = 1.94177, almost identical with the one given in [28] (λc = effects also. Examples of application for both types of behaviour
1.952), whereas for the AISC LFRD criterion case, the limit load have been presented.
factor was λc = 2.06649. The procedure turned out to be accurate, stable and computa-
The collapse mechanism and the plasticization sequence for the tionally superior as compared to the direct stiffness method that
two criteria are shown in Fig. 21(a) and (b), where the numbers is almost exclusively used. It is therefore believed that this could
correspond to the analysis step in which the event takes place. One enhance the use of mathematical programming methods towards
may notice the difference in the plasticization sequence between the numerical solution of elastoplastic problems.
the two cases. In the AISC LFRD case (see Fig. 21(b)), the critical The proposed method may be extended to 3D structures.
section at the right end of member 7 that was plasticized in step Material hardening may also be included.
3 is unstressed in step 8, while – in the same step – a new plastic
hinge forms at the upper end of member 3.
References
The bending moments’ and axial forces’ diagrams on the state
of collapse for each criterion case are shown in Fig. 22(a) and (b), [1] Maier G. A quadratic programming approach for certain classes of non linear
with the values inside the parentheses corresponding to the AISC structural problems. Meccanica 1968;3:121–30.
[2] Maier G. Quadratic programming and theory of elastic–perfectly plastic
LFRD criterion analysis results. structures. Meccanica 1968;3:265–73.
In Fig. 23, one may see the base shear vs. roof horizontal [3] Maier G. A matrix structural theory of piecewise linear elastoplasticity with
displacement curves, for both bending moment and axial force interacting yield planes. Meccanica 1970;5:54–66.
[4] De Donato O, Maier G. Historical deformation analysis of elastoplastic
interaction criteria. structures as a parametric linear complementarity problem. Meccanica 1976;
The same example was solved using SAP2000 [16], yielding 11:166–71.
identical results (Fig. 23). For the simple criterion, 2.70 s were [5] Smith DL. The Wolfe–Markowitz algorithm for nonholonomic elastoplastic
analysis. Eng Struct 1978;1:8–16.
required in total, while 1.86 s were required for solving only the [6] Maier G, Giacomini S, Paterlini F. Combined elastoplastic and limit analysis via
stiffness matrix. On the other hand, using the proposed method, restricted basis linear programming. Comput Methods Appl Mech Engrg 1979;
the total computational time was only 0.047 s. For the AISC 19:21–48.
[7] Franchi A, Cohn MZ. Computer analysis of elastic–plastic structures. Comput
LFRD, 3.17 s were required in total, while 2.19 s were required Methods Appl Mech Engrg 1980;21:271–94.
for solving only the stiffness matrix. On the other hand, using [8] Kaneko I. Complete solutions for a class of elastic–plastic structures. Comp
the proposed method, the total computational time was only Meth Appl Mech Engrg 1980;21:193–209.
1214 K.V. Spiliopoulos, T.N. Patsios / Engineering Structures 32 (2010) 1199–1214

[9] Wakefield RR, Tin-Loi F. Large scale nonholonomic elastoplastic analysis using [19] Spillers WR. Application of topology in structural analysis. J Struct Div ASCE
a linear complementarity formulation. Comp Meth Appl Mech Engrg 1990;84: 1963;89:301–13.
229–42. [20] Spiliopoulos KV. On the automation of the force method in the optimal plastic
[10] Coccheti G, Maier G. Elastic–plastic and limit-state analyses of frames with design of frames. Comput Methods Appl Mech Engrg 1997;141:141–56.
softening plastic-hinge models by mathematical programming. Int J Solids
[21] Spiliopoulos KV. A fully automatic force method for the optimal shakedown
Struct 2003;40:7219–44.
design of frames. J Comput Mech 1999;23:299–307.
[11] Tangaramvong S, Tin-Loi F. Limit analysis of strain softening steel frames under
pure bending. J Const Steel Res 2007;63:1151–9. [22] Goldfarb D, Idnani A. A numerically stable dual method for solving strictly
[12] Tangaramvong S, Tin-Loi F. A complementarity approach for elastoplastic convex quadratic programs. Math Program 1983;27:1–33.
analysis of strain softening frames under combined bending and axial force. [23] Load and resistance factor design specification for structural steel buildings.
Eng Struct 2007;29:742–53. 2nd ed. Chicago: American Institute of Steel Construction; 1993.
[13] Marin-Artieda CC, Dargush GF. Approximate limit load evaluation of structural [24] IMSL Fortran Library V 6.0. 2007.
frames using linear elastic analysis. Eng Struct 2007;29:296–304. [25] Smith DL, Munro J. On uniqueness in elastoplastic analysis of frames. J Struct
[14] Barrera O, Cocks ACF, Ponter ARS. Evaluation of the convergent properties Mech 1978;6:85–106.
of the Linear Matching Method for computing the collapse of structural
[26] Nguyen DH, Morelle P. Optimal plastic design and the development of
components. Eur J Mech A Solids 2009;28:655–67.
practical software. In: Smith DL, editor. Mathematical programming methods
[15] Tin-Loi F, Wong MB. Nonholonomic computer analysis of elastoplastic frames.
Comput Methods Appl Mech Engrg 1989;72:351–64. in structural plasticity. CISM courses and lectures, No. 299. 1990. p. 207–29
[16] SAP2000, v14.0.0. User’s manual. CSI Inc. 2009. [Chapter 12].
[17] Pereira NZ, Borges LA, Hecke MB. A force method for elastic–plastic analysis of [27] EAK 2000. Greek Code for earthquake resistant structures. 2000.
frames by quadratic optimization. Int J Solids Struct 1988;24:211–30. [28] Cohn MZ, Rafay T. Collapse load analysis of frames considering axial forces. J
[18] Damkilde L, Høyer O. An efficient implementation of limit state calculations Eng Mech ASCE 1974;100(4):773–94; J Eng Mech ASCE 1975;101(4):493–4
based on lower-bound solutions. Comput Struct 1994;6:953–62. [errata].