Вы находитесь на странице: 1из 228

ROMANIAN JOURNAL OF TECHNICAL

SCIENCES APPLIED MECHANICS

formerly REVUE ROUMAINE DES SCIENCES


TECHNIQUES SRIE DE MCANIQUE APPLIQUE

Volume 59, Nos 12, January August 2014

SPECIAL ISSUE STABILITY AND NONLINEAR ANALYSIS


OF STEEL STRUCTURES RESEARCH ADVANCES

Guest Editors: Dan DUBINA (Politehnica University of Timisoara , Romania)


H.H. SNIJDER (University of Eindhoven, The Netherlands )
L. Simes da SILVA (University of Coimbra, Portugal)

CONTENTS

DAN DUBINA, H.H. SNIJDER , L. SIMES DA SILVA, Editorial of the Special Issue on
Stability and Nonlinear Analysis of Steel Structures Research Advances .................. 3

PART I: Theoretical background , numerical and experimental advanced


studies
DAN DUBINA, VIOREL UNGUREANU, Erosion of interactive buckling load of thin-
walled steel bar members: contribution of Timisoara School ..................................... 9
GIOVANNI GARCEA, ANTONIO BILOTTA, ANTONIO MADEO, GIUSEPPE ZAGARI,
RAFFAELE CASCIARO, A numerical asymptotic formulation for the post-buckling
analysis of structures in case of coupled instability......................................................... 38
RODRIGO GONALVES, DINAR CAMOTIM, Effect of distortion on the structural
behavior of thin-walled steel regular polygonal tubes ..................................................... 56
MARIA KOTEKO, ARTUR MODAWA, MARCIN JANKOWSKI, Axial impact of open-
section twcf columns experimental study ..................................................................... 72
JOO PEDRO MARTINS, L. SIMES DA SILVA, LILIANA MARQUES, MARTIN
PIRCHER, Eigenvalue analysis of curved sandwich panels loaded in uniaxial
compression..................................................................................................................... 87
MIHAI NEDELCU, An analysis of the effect of holes patterns on coupled instability of
perforated thin-walled members ...................................................................................... 105
FRANC SINUR, PRIMO MOE, KLEMEN REJEC, GAPER LUAR, DARKO BEG,
Imperfection sensitivity of thin plates loaded in shear..................................................... 121

Ro. J. Techn. Sci. Appl. Mechanics, Volume 59, Nos 12, P. 1228, Bucharest, 2014
PART II: Design codification oriented studies
LSZL DUNAI, BALZS KVESDI, DVID WISCHY, BARNABS BZA, Lateral
torsional buckling resistance of castellated beams...................................................................... 139
LEROY GARDNER, KWAN HO LAW, CRAIG BUCHANAN, Unified slenderness limits for
structural steel circular hollow sections ...................................................................................... 153
ULRIKE KUHLMANN, BENJAMIN BRAUN, Evolution of Eurocode 3 amendments to
EN 1993-1-5 for plate buckling....................................................................................... 164
H.H.SNIJDER, L.-G. CAJOT, N. POPA, R.C. SPOORENBERG, Buckling curves for heavy
wide flange steel columns ............................................................................................... 178
ANDREAS TARAS, Probabilistic assessment of the impact of straightness tolerances in en
1090-2 on the stability design of steel columns............................................................... 205
EDITORIAL OF THE SPECIAL ISSUE
ON STABILITY AND NONLINEAR ANALYSIS
OF STEEL STRUCTURES

Stability of Structures, definitions and terminology

Scientifically, buckling is a mathematical instability, leading to a failure


mode. The formal meaning of the notion is found in engineering and sciences,
concerning stability of systems. Broadly speaking, structural stability can be
defined as capacity of a slender structure to recover equilibrium.
Stability is an essential requirement for all structures. Theoretically, for a
structural system, buckling is caused by a bifurcation in the solution to the
equations of static equilibrium. At a certain stage under an increasing load, further
load is able to be sustained in one of two states of equilibrium: an undeformed
state, or a laterally-deformed state. In practice, buckling is characterized by a
sudden failure of a structural member subjected to high compressive stress, where
the actual compressive stress at the point of failure is less than the ultimate
compressive stresses that the material is capable of withstanding. Failure occurs in
a distinct direction compared to the direction of the applied load.
To evaluate the behaviour of a slender structure which might lose its stability
according to the previous definition, three characteristic ranges of the load-
deformation behaviour should be considered:
the pre-critical range , i.e. P (0, Pcr] defining the domain of Structural
stability;
the critical point (bifurcation of equilibrium), P = Pcr;
the post-critical range, i.e. P > Pcr the Structural instability domain.
Since metal structures, steel in particular, are slender, they are most prone to
instability problems; hence the research on steel structures focussed on stability.
Although the stability of bars was first studied over 250 years ago (Euler's
paper was published in 1744), adequate solutions are still not available for many
problems in structural stability. So much has been and is being studied and written
in the field of structural stability, that one may well wonder why, after such
intellectual and financial efforts, there are no definite solutions to these problems.
Numerical facilities and advanced FE codes make it possibly today to calculate
and/or simulate accurately the behaviour of complex structures. However, for
slender structures highly sensitive to buckling, there are still difficulties for a
reliable evaluation of its stability.

Ro. J. Techn. Sci. Appl. Mechanics, Volume 59, Nos 12, P. 38, Bucharest, 2014
4 Dan Dubina, H.H. Snijder, L.S. da Silva 2

This is because determining the load under which a structure collapses due to
the loss of stability still remains one of the most sensitive problems of structural
design. This is due to the following factors (Gioncu, 2005):
a) The loss of stability depends on numerous factors, some of which are very
difficult to control. This is confirmed by a number of recent structure accidents.
Faulty design and execution, overstressing or the use of inadequate materials have
been shown to be mainly responsible for these accidents. It should be noted that
these accidents practically cover the entire range of structures. Today, only a
specialist can carry out stability checks in complete agreement with the actual
behaviour of the structure.
b) Instability occurs in a region with both strong geometrical and material
nonlinearities. For the pre-critical range an extensive literature provides effective
solutions. For the post-critical range, the theoretical background was significantly
developed within the second half of the last century, but only after remarkable
progress in the field of electronic computing equipment, and non-linear analysis
using the Finite Element Method, FEM (e.g. GMNIA- Geometrical and Material
Nonlinear Analysis including Imperfections) and after the development of some
special numerical techniques (e.g. the Arc-Length Method) in the neighbourhood
of the limit point. These developments made it possible to describe correctly the
behaviour of structure, shortly before its failure, and after. However, such analyses
are difficult and costly and they are not accessible for many designers.

Fig. 1 Critical and Post-Critical behaviour.

c) In no other field of structural mechanics, the influence of imperfections


due to the execution is as significant as in the field of instability. In strength
analysis the stress-strain state is determined by means of an idealized scheme of the
structure where neglecting the geometrical and mechanical imperfections leads to
3 Editorial 5

relatively small differences in failure load. In the case of instability on the other
hand, failure loads are influenced to a large extent by imperfections so that they
cannot be neglected.
d) Checking the buckling resistance of structures experimentally is very
difficult, because it is impossible to test the actual structure just until it collapse. In
strength analysis, reduced model tests are used for checking the validity of theoretical
values. In stability analysis, testing on reduced models is irrelevant in most cases,
because a correct modelling of the effect of imperfections is practically impossible.
e) There is a wealth of information available in numerous papers dealing with
the stability and instability of structures, but information available in design codes
and standards is limited, even in Eurocode 3. In this situation structural designers
may commit grave errors in the structural instability checking.

Selected historical milestone references

Structural stability has a long history. One says (Elishakoff, 2000), the first
description of an instability phenomenon goes to the Bible, where the Tower of
Babel lost its stability under its own weight (610 B.C.). According to Godoy (2000)
perhaps the first to investigate structural stability using theoretical tools were the
Greek masters between 400 B.C. and 200 B.C. Aristotle (384 B.C.322B.C.)
employed kinematics concepts to study changes in stationary systems; and
Archimedes (287 B.C.212 B.C.) used geometric methods to assess the stability of
floating bodies.
So, even it is unanimously recognised that mathematically the theory of
stability was initiated by the Swiss mathematician, Euler in 1744 formulated
structural stability in a mathematical way but technically speaking, stability as
structural phenomenon dates from long before Euler.
Heron of Alexandria (about 100 B.C.), in the course of a long dull work on
statics endeavoured to explain why the strength of a piece of wood reduces as its
length increases (Villagio). Leonardo da Vinci (14521519) provided two
empirical rules for the strength of columns in compression. The Jesuit Mersenne
(15881648), in his Reflexiones on the causes of resistance in solids, observed that
iron, copper and other metals, even single bodies, when subject to a force or
weight, curve and bend to the form of an arch before breaking (Benvenuto).
Mersennes conclusions were unexpectedly confirmed by the consistent
programme of experiments conducted by Van Musschenbroek (16921761), the
inventor of testing machines designed to allow systematic variation of experimental
parameters. Van Musschenbroek even proposed a quantitative law for the failure in
compression of a parallelepiped composed of wood. But the greatest achievement
in the period preceding Euler was made undoubtedly by Bernoulli (16541705).
Bernoulli, as distinct from Galileo and Mariotte, ignored the resistance of beams
and instead considered their deflection. He was able to construct the equation of a
6 Dan Dubina, H.H. Snijder, L.S. da Silva 4

flexible bar deformed in the plane considering finite deflection and a nonlinear
(parabolic) dependence between curvature and bending moment. This enabled
Euler to find the well-known todays formula of elastic critical load of a
compression bar. Lagrange (17361813) developed Eulers theory, generalizing it
to columns of variable cross-section, and used it for checking the most stable shape
of compressed columns. He introduced the notion of bifurcation that connects
linearized and fully nonlinear solutions.
Eulers theory found its applications only from the 19th century under the
pressure of practical problems raised by industrial and building development when
most problems of structural stability were basically linear elastic. The twentieth
century has witnessed a great expansion of the stability theory into nonlinear
behaviour, caused either by large defections or by nonlinearity of the constitutive
law of the material. In the second half of this century, dynamic stability, important
especially for non-conservative systems, became reasonably well understood.
A selective review of some milestones of these developments reads as follows:
Young treated the lateral buckling of a column with variable cross-section;
Navier derived the correct differential equation for a thin plate subject to uniform
compressive forces; Kirchhoff proposed an elegant theory for slender rods
experiencing large displacement and small strain; Eulers theory was applied to
thin shells; Fppl and von Krmn derived a system of two equations describing
the large deflection of a thin plate with stresses acting in the middle plane; Reissner
relaxed some of the simplifying assumptions of Fppl and von Karman theory.
At the end of the 19th century there was general agreement that a unified
theory of structural stability, to generalize and give a framework to all previous
results, has to consider instability as a dynamic problem. The development of this
idea really starts with the contributions of Poincar (18541912) who discovered a
general method for dynamic systems involving series proved convergent for all
values of time, and who achieved consistency with the studies by Lyapunov (1857
1912). In simple terms, if all solutions of the dynamical system that start out near
an equilibrium point xe stay near xe forever, then xe is Lyapunov stable. More
strongly, if xe is Lyapunov stable and all solutions that start out near xe converge to
xe, then xe is asymptotically stable. The notion of exponential stability guarantees a
minimal rate of decay, i.e., an estimate of how quickly the solutions converge. The
idea of Lyapunov stability can be extended to infinite-dimensional manifolds,
where it is known as structural stability, which concerns the behaviour of different
but nearby solutions to differential equations.
Based on Lyapunov stability theorems, in 1945 Koiter has published his
Ph.D. thesis describing a general theory for the stability of elastic systems subject
to conservative loadings. His work contained also a rigorous confirmation of the
effect of initial imperfections on the buckling load of axially compressed shells.
Koiters general theory of elastic stability has marked the beginning of the
5 Editorial 7

Imperfection Sensitivity Design Philosophy. In fact, it employed bifurcation theory


in continuum systems. In the new approach, the information given by critical loads
was seen as insufficient, and Koiter employed perturbation theory to develop an
asymptotic analysis that allowed him to follow the post-buckling path in its early
stages (Elishakoff, 2005). While Euler was the first to give a formula for the critical
load of an ideal structure, Koiter was the first to give one for an imperfect structure.
Today, the highly theoretical method of Koiter is numerically implemented
through the Finite Element Method enabling to model complex interactive stability
modes one of the papers in the present volume is dealing with that.

International framework of scientific cooperation


in the Structural Stability domain

Professor Beer from the Technical University of Graz had the idea to
organize in 1971 an International Colloquium in order to compare the ECCS
(European Convention of Constructional Steelwork) approach of buckling curves
of the slender bars in compression, just recently launched in EU countries, with
those applied in Eastern Europe, the United States of America and Japan. This
Colloquium was organized in Paris in 1972, followed in 1974, in London covering
the assembly of structural stability problems. There, Sfintesco as President of
ECCS and Beedle, as Chairman of SSRC (Structural Stability Research Council, USA)
have proposed to enlarge both the topic and geographical areas of this Colloquium
and transform it in a traveling event. A long series of colloquia under the
coordination of the SSRC has started, the last one, the 21st, being organized in Rio
de Janeiro; the next and 22nd one is planned to take place in Timisoara, Romania in
2016. At the 1997 edition, in Nagoya, the general framing topic Stability of Steel
Structures was enlarged and became Stability and Ductility of Steel Structures
SDSS. Previous editions of October 1982 and September 1999 have been
organized by the Politehnica University of Timisoara in cooperation of Romanian
Academy, the Timisoara Branch, through the Committee for Structural Stability.
In parallel with the SSRC series, in October 1992 in Timisoara, another series
of conferences started, dedicated to Coupled Instabilities in Metal Structures. This
CIMS series has a recurrence period of four years: the last and 6th one being
organized in Glasgow in December 2012 after the 5th in 2008 in Sydney, while the
next and 7th CIMS conference is expected to take place in Baltimore in the Autumn
of 2016.
In Europe, the organization offering an integrating framework for the
research activity related to steel structures is ECCS European Convention for
Constructional Steelwork, founded in 1955. Outstanding European scientists and
engineers, from academia, research centers and industry are taking part in the
Technical Committees (TC-s) of ECCS, which are playing an important role in
8 Dan Dubina, H.H. Snijder, L.S. da Silva 6

technical and scientific forums and working groups contributing to developing and
promoting advanced knowledge in the field of steel structures. Among other ECCS
TC-s, there is the TC 8 on Structural Stability. Along the years TC8 has contributed
to the elaboration of European Buckling Curves (1970), the ECCS Manual on
Stability of Structures (1976), Behavior and Design of Plated Structures (1986), the
5th editions starting in 1980 of the Recommendations for Stability of Steel Shells,
and in last two decades to the provisions for Structural Stability Design of
Eurocode 3- Design of Steel Structures (EN 1993-1-1, EN 1993-1-5, EN 1993-1-6).
This Special Issue is in good part the result of ECCS TC8 cooperation,
9 from the 12 papers being authored by the members of this Group.

Present Special Issue on


Stability and nonlinear analysis of steel structures

The topics of the papers included in this volume are diverse enough, tackling
stability problems of steel structures with thin and tick walled bar members, open
and hollow sections, plated structures and curved sandwich panels. There are
theoretical, numerical and experimental approaches and combinations of them used
in solving stability problems. The 12 papers have been framed into two parts:
Part I: Theoretical background, numerical and experimental advanced
studies 7 papers;
Part II: Design codification oriented studies 5 papers.
36 authors from 11 European Countries have contributed with their research
works to this Special Issue of the Romanian Journal of Technical Sciences. We are
expressing our gratitude to all of them.
We are also grateful to the reviewers for the time and effort they spent
evaluating the papers.
Thanks are also due to Dr. Luigi VLADAREANU and Dr. Dan DUMITRU
of Institute of Solid Mechanics of the Romanian Academy, Bucharest, for the
editorial work.
The guest editors hope that this special issue gives an overview of current
research activities contributing to the stability and nonlinear analysis of steel structures.
Timioara, August 2014

Guest Editors:
Dan DUBINA
H.H. (Bert) Snijder, Luis Simes da SILVA
PART I: THEORETICAL BACKGROUND, NUMERICAL
AND EXPERIMENTAL ADVANCED STUDIES

EROSION OF INTERACTIVE BUCKLING LOAD OF


THIN-WALLED STEEL BAR MEMBERS:
CONTRIBUTION OF TIMISOARA SCHOOL

DAN DUBINA1, 2, VIOREL UNGUREANU1, 2

Abstract. The paper presents a summary of the activity and research achievements of
the Romanian researchers of Timisoara School in the field of stability of cold-formed
steel members. Both, fundamental theory and applied instability contributions are
focussed. Post-critical theory of elastic structures, the analysis of stable and unstable
components of bifurcation load, coupling of bifurcations modes (e.g. mod interaction),
erosion of critical load are the topics in which the theoretical contributions of
Timisoara School are significant. Present paper focuses the mode interaction problems
of thin-walled steel bar members only, integrating some relevant results obtained by
the authors through a state-of-art review.

Key words: erosion, critical load, interactive buckling, ECBL approach, bar members.

1. INTRODUCTION

In the case of an ideal structure, the theoretical equilibrium bifurcation point


and corresponding load, Ncr, are obtained at the intersection of the pre-critical
(primary) force-displacement curve with the post-critical (secondary) curve. For a
real structure, affected by a generic imperfection the bifurcation point does not
appear anymore and, instead, the equilibrium limit point is the one characterizing
the ultimate capacity, Nu, of the structure. The difference between Ncr and Nu
represents the Erosion of the Critical Bifurcation Load (ECBL), due to the
imperfections. This model applies in the instability mode interaction. The meaning
of mode interaction inherently refers to the erosion of critical bifurcation load in
case of interaction of two (or more) buckling modes associated with the same, or
nearly the same, critical load. The theoretical and experimental studies and the
contributions of Timisoara School to this topic along more than 30 years are
relevant, being recognised by the scientific community in the field [10].
1
Politehnica University of Timisoara, Civil Engineering Faculty, Department of Steel
Structures and Structural Mechanics, Ioan Curea 1, 300224, Timisoara, Romania
2
The Romanian Academy, Timisoara Branch, Laboratory of Steel Structures, Mihai Viteazu 24,
300223, Timisoara, Romania

Ro. J. Techn. Sci. Appl. Mechanics, Volume 59, Nos 12, P. 9137, Bucharest, 2014
10 Dan Dubina, Viorel Ungureanu 2

As a prove, in October 1982, the First Session of the Third SSRC International
Colloquium of Stability of Steel Structures was organised in Timisoara. Ten years
later, the First International Conferences on Coupled Instabilities in Metal
Structures CIM92, took place in Timisoara on October 10-12, 1992. A number
of 61 contributions prepared by 60 authors of 19 countries have been published in a
Special Issue of Thin-walled Structures journal, with J. Rondal, D. Dubina and
V. Gioncu as Guest Editors [1]. At the second conference, CIMS96, held in Liege,
on September 5-7, 1996, 166 authors from 23 countries presented 62 contributions
published in a volume of 596 pages, edited by the same team [2]; and the series
continues with the next CIMS,2000 in Lisbon, 2004 in Rome, 2008 in Sydney,
2012 in Glasgow, the being planned to be held in 2016 in Baltimore.
In 1997, the series of International Colloquia dedicated to Stability of Steel
Structures promoted by Structural Stability research Council of USA through
travelling Sessions ( e.g. as the one held in 1982 in Timisoara), extended the topic
area and became International Colloquium on Stability and Ductility of Steel
Structures (SDSS). First SDDS was organized in Nagoya, in august 1997; the next,
SDSS99, has been organised in 9-11 of September 1999 in Timisoara, by the
Politehnica University of Timisoara, Technical University of Budapest and Romanian
Academy Timisoara Branch in co-operation with Structural Stability Research
Council (USA) and European Convention for Constructional Steelwork [5] and. In
2016, Politehnica University of Timisoara will organise the next edition of SDSS
colloquium. In parallel with this events, the advanced courses focusing selected
structural stability topics, organized by International Centre of Mechanical Sciences-
CIMS, in Udine, Italy , with the contribution of outstanding teams of outstanding
international lectures , including t representatives of Timisoara School, must be
emphasized i.e.
Coupled Instabilities in Metal Structures. Theory and Practical Aspects [6]
in October 1996;
Light gauge metal structures. Recent advances [7] in June 2002;
Phenomenological and mathematical modelling of coupled instabilities [8]
in October 2003.
Continuing the line, in 2004 a Special Issue of Thin Walled Structures
journal entitled Cold Formed Structures: Recent Research Advances in Central
and Eastern Europe has been published, under the coordination of Professor
Dubina [9].
On the following, this review paper, focuses the mode interaction problems
of thin-walled steel bar members, presenting the theoretical background of ECBL
method Erosion of Critical Buckling Load , and selected results obtained with
this method. ECBL is a creation of Timisoara School of Stability of Steel Structures.
3 Erosion of interactive buckling load of thin-walled steel bar members 11

2. PHENOMENON OF MODE INTERACTION

In the case of an ideal structure, the theoretical equilibrium bifurcation point


and corresponding load, Ncr, are observed at the intersection of the pre-critical
(primary) force-displacement curve with the post-critical (secondary) curve (see
Fig. 1). To evaluate the behaviour of a slender structure, which might loss its
stability, needs for the control by design the three characteristic ranges of load-
deformation:
Pre-critical range, i.e. N(0, Ncr], defining the domain of structural
stability;
Critical point (bifurcation of equilibrium);
Post-critical range, e.g. N > Ncr, the structural instability domain.

Force
bifurcation point (critical)
Ncr

erosion

Nu
post-critical
limit point
ideal
structure pre-critical
actual
structure

geometric imperfection Displacement


0

Fig. 1 Critical and post-critical behaviour.

For a real structure, affected by a generic imperfection, 0, the bifurcation


point does not appear anymore and, instead, the equilibrium limit point is the one
characterizing the ultimate capacity, Nu, of the structure. The difference between
Ncr and Nu represents the Erosion of the Critical Bifurcation Load (ECBL), due to
the imperfections. The model in Fig. 1 can also be applied in case of structures
which might be prone to subsequent buckling modes interacting in the bifurcation
point, i.e.: a primary mode which, if it does not cause the failure of the structure,
play the role in the pre-critical path (e.g. local buckling in case of slender
12 Dan Dubina, Viorel Ungureanu 4

thin-walled members), and the secondary mode which, at the end, that is the one
causing the failure (this will be the post-critical mode). Roughly, this is the
description of instability mode interaction.
The meaning of the mode interaction refers to the erosion of critical
bifurcation load in case of interaction of two (or more) buckling modes associated
with the same, or nearly the same, critical load; it happens when the mode
simultaneity is due to the results of design and/or imperfections. A well-known
example of mode interaction is the coupling of local or distortional buckling with
the overall buckling in the case of thin-walled cold-formed steel members, or the
coupling between local buckling of class 4 web with the lateral-torsional buckling
of plated beam.
In almost all practical cases, the mode interaction, obtained by coupling of a
local instability with an overall one, is a result of design (e.g. calibration of
mechanical and geometrical properties of a member) and has a nonlinear nature:
Coupling by design occurs when the geometric dimensions of structure are
chosen such as two or more buckling modes are simultaneously possible.
For this case, the optimization based on the simultaneous mode design
principle plays a very important role and the attitude of the designer
towards this principle is decisive. This type of coupling is the most
interesting in practice because, even the erosion of critical buckling load is
maximum in the interactive range, the ultimate buckling strength still
remains maximum in this range;
Nonlinearity characterizes the post-buckling behaviour of coupling of
instability modes and is due to design and the presence of the geometrical
imperfections which is indispensable for coupling; this coupling doesnt
exist for ideal structure. For instance, this is the case of the interaction
between flexural buckling and flexural-torsional buckling of some mono-
symmetrical cross-section.
Figure 2 illustrates such a case for a mono-symmetrical T-section in
compression, studied in Timisoara [12], which is prone to the mode interaction
between flexural and flexural-torsional modes. Due to the imperfections the
erosion of critical bifurcation load occurs. The erosion is maximum in the coupling
point vicinity (Fig. 1). For bar members, an interactive slenderness range, in which
sensitivity to imperfections is increased, may be identified. Depending on imperfection
sensitivity, classes of interaction types, characterized by specific levels of erosion
intensity, may be defined.
5 Erosion of interactive buckling load of thin-walled steel bar members 13

u3

Ncr i3
u2
h
i2
flexural buckling b
coupled instability
i1
torsional - flexural buckling b
h u1

Ideal flexural-torsional
Ncr 2 Eth3 buckling
2
l
0.2

tl
=0.816
0.15
Ideal b2
flexural
buckling
b t
0.1

h t
0.05
tests
actual
h/b
1.0 1.2 1.4 1.6 1.8 2.0

Fig. 2 Coupled instability by design: example for T-section [12].

Given a member in compression and assuming two simultaneous buckling


modes which might couple (Fig. 3), the perfect member fails under interactive
critical buckling load, Ncr, while the real capacity of the actual member will be the
ultimate load, Nu. The erosion, , can be expressed as follows:

= 1 N u / N cr , (1)

and

N u = (1 ) N cr . (2)
14 Dan Dubina, Viorel Ungureanu 6

N
nd
2 mode 1st mode
Ncr

Nu

coupled mode

interaction geometrical governing


parameter

Fig. 3 Generic model of two mode interaction.

An erosion factor can be defined and used as a measure of erosion of


Theoretical critical load. Gioncu [13] has ranked in four classes the mode
interaction types in terms of erosion factor, as follows:
class I: weak interaction (WI), 0.1;
class II: moderate interaction (MI), 0.1 < 0.3;
class III: strong interaction (SI), 0.3 < 0.5;
class IV: very strong interaction (VI), > 0.5.
Obviously, an appropriate framing of each mode interaction into a relevant
class is very important because the methods of analysis used for design have to be
different from one class to another. Week or moderate interactions could be
controlled by code-based design procedures, the partial safety coefficients being
able to keep safe those structures; higher interaction classes, particularly SI and VI,
need for more refined examination, in principle using advance numerical methods
and taking into account for relevant imperfection scenarios.
Interaction classes can be associated with erosion levels (Fig. 4).

Nu
N cr ideal actual
weak erosion
1.00

0.75 moderate erosion

0.50 important erosion


very important
erosion
e

Fig. 4 Erosion levels [13].


7 Erosion of interactive buckling load of thin-walled steel bar members 15

In case of thin-walled members, two types of interaction might occur. The


first one is due to multiple local modes, which leads to a so called localized mode,
and gives rise to an unstable post-critical behaviour. The second interaction,
between the localized buckling mode and the overall buckling one, yields to a very
unstable post-critical behaviour, with great erosion due to the imperfections. The
multiple local buckling modes interaction might generate a localized mode which
subsequently can interact with an overall mode, with very destabilizing effects
(Fig. 9, further on). Strong and very strong interactions are the result of this type of
coupled instability. In such a case, very special design methods must be developed.
Currently, such a phenomenon is characteristic for thin-walled columns in
compression. Table 1 qualitatively indicates the erosion levels for mode interaction
classes i.e.
Table 1
Coupled instabilities in bar members [14]
No. Bar member type Instability modes Class of interaction
WI to MI
1. Mono-symmetrical columns F + FT = FFT
0.3
MI
2. Built-up columns F + L = FL
0.1 < 0.3
F + L = FL SI to VI
FT + L = FTL
0.3
(F + FT + L = FFTL)
3. Thin-walled columns
F + D = FD MI to SI
FT + D = FTD
0.3 0.5
(F + FT + D = FFTD)
LT + L = LTL MI
4. Thin-walled beams
LT + D = LTD 0.3
Legend: F = flexural buckling ; FT = flexural-torsional buckling;
L = local buckling ; D = distortional buckling
WI = week interaction ; MI = moderate interaction ; SI = strong interaction;
VI = very strong interaction.

2. THEORETICAL BACKGROUND

2.1. SELECTIVE REVIEW

A number of important problems of structural stability are characterized by


multiple buckling modes associated with the same critical buckling load [15]. Such
problems are used to be known as instability mode interaction. The theoretical
approach of this phenomenon, also addressed as interactive or coupled buckling
involves the general asymptotic theory of instability. The background of structural
theory and post-buckling behaviour of structures is given in books and state-of-art
16 Dan Dubina, Viorel Ungureanu 8

articles by: Hutchinson & Koiter [16], Thompson & Hunt [17, 18], Budiansky [19],
Koiter [20], and Flores & Godoy [21]. The books on bifurcation theory by Chow &
Hale [22], Golubitsky & Schaeffer [23] and Gioncu & Ivan [24] related to the
theory of critical and post-critical behaviour of elastic structures have to be
considered as basic lectures for the readers interested on this topic.
When speaking about the mode interaction, implicitly refers to the erosion of
theoretical critical bifurcation load in case of interaction of two (or more) buckling
modes associated with the same, or nearly the same, critical load; it happens when
the mode simultaneity is due to the results of design and/or imperfections [25]. A
well-known example of such a mode interaction is the one resulting from the
coupling of local or distortional buckling with overall buckling in case of a thin-
walled cold-formed members. In such cases, the critical values corresponding to
global buckling mode are significantly lower than local buckling modes, and their
interaction can be considered within the first non-linear approximation [26].
A comprehensive approach of the problem of elastic interaction between
local and global buckling modes is due to van der Neut [27], who provided the
evidence that the sensitivity to imperfections of thin-walled columns in compression is
maximum into the interactive buckling range, where critical buckling loads
corresponding to local and global modes are closed to each other.
Koiter & Kuiken [28], two years after van der Neut, developed the method
known as method of slowly varying local mode amplitude. In 1976 Koiter has
published his General Theory of Mode Interaction in Stiffened Plates and Shell
Structures [29], followed by the well-known book of Thompson & Hunt, A General
Theory of Elastic Stability [17], in which the theory of interaction between
coincident instability buckling modes is presented. On the same line, fundamental
contributions to the problem of local-overall mode interaction of thin-walled
sections are the studies of Thompson & Lewis [30]. Tvergaard [31,32] presented a
method enabling to evaluate the erosion of ultimate capacity of interaction of
overall mode, in post-buckling range, with plate local buckling mode, which is
stable in post-buckling range, as it was the case of van der Neut problem.
Based on van der Neut principle and applying the Ayrton-Perry equation
[33], Dubina [14] proposed the Erosion of Critical Bifurcation Load (ECBL)
approach, enabling to evaluate the theoretical erosion of critical load into the
interactive buckling range. Later, based on the real behaviour of thin-walled stub
columns and short beams, Ungureanu & Dubina [34,35] used in the interactive
local-overall buckling analysis the sectional plastic mechanism strength instead of
traditional effective section and, the ECBL approach, in order to express the plastic-
elastic interactive buckling of thin-walled cold-formed steel members.
In the last two decades intensive progress in studying the mode interaction
problems was achieved due to the development of specific numerical methods.
Since the late 1980s, the Generalized Beam Theory (GBT) [36, 37, 38, 39, 40, 41]
9 Erosion of interactive buckling load of thin-walled steel bar members 17

has been developed extensively. Particularly connected to the present topic, Camotim &
Dinis [42] performed extended numerical studies, using FEM and GBT, to study
the elastic post-buckling behaviour of cold-formed steel columns affected by mode
interaction phenomena involving distortional buckling, namely local/distortional,
distortional/global (flexural-torsional), local/distortional/global mode interaction
and also sensitivity to imperfections of thin-walled cold-formed steel members.
Alternatively, another approach has been proposed based on conventional
FSM, i.e. CUFSM [43], freely available at the www.ce.jhu.edu/bschafer/cufsm.
The recently developed constrained Finite Strip Method (cFSM) provides a means
to simplify thin-walled member stability solutions through its ability to identify and
decompose mechanically meaningful stability behaviour, notably the formal
separation of local, distortional, and global deformation modes. In this version the
solution has been expanded to allow for general end boundary conditions [44].
Another design method, which can be framed in the class of semi-analytical
methods, is Direct Strength Method [45], which practically replaces the effective
width concept with the effective stress one. The method explicitly incorporates
local or distortional and Euler buckling and does not require calculations of the
effective properties. The procedure is an alternative to effective width method.
Direct Strength Method has been adopted in 2004 as design method in Appendix 1
to the North American Specification for the Design of Cold-Formed Steel Structural
Members [46].
In the last years, very interesting developments based on the fundamental
theoretical works of Koiter [25] have been developed by Garcea et al. [47, 48]. The
asymptotic approach, derived as a finite element implementation of Koiters
nonlinear theory of elastic stability, could be a convenient alternative by providing
an effective and reliable strategy for predicting the initial post-critical behaviour in
both cases of limit or bifurcation points. Its main advantage lies in the possibility of
performing an efficient and reliable imperfection sensitivity analysis, even in cases
of multiple, nearly coincident, buckling loads.

2.2. THE VAN DER NEUT MODEL

As already mentioned, a milestone achievement enabling for both under-


standing and practical characterisation of the local-global mode interaction
problem is the pioneering study conducted by van der Neut [27], which has clearly
demonstrated the erosion of theoretical critical coupling due to imperfections.
In this case, the interaction occurred between the local buckling of the flanges
and flexural buckling of a square box section column; only the flanges have been
considered to be active, while the web role was to connect them. Fig. 5a shows the
buckling curve of the van der Neut column without local or overall imperfections.
For lengths greater than L1 the column fails in overall Euler buckling, i.e.
18 Dan Dubina, Viorel Ungureanu 10

NE = 2 EI / L2 . (3)
For shorter lengths, the local buckling load, i.e.
2
k 2 E t
Ncr , L = 2 (4)
12(1 2 ) d
is reached before Euler buckling takes place (t is the thickness and d is the width of
flanges, is the Poissons ratio and k = 4, the plate buckling coefficient). In the
locally buckled shape, a reduced bending stiffness of the column, given by EI, is
considered, where is the slope of the load-strain diagram of the flange plate in the
post-local buckling range. van der Neut has considered the results of work by
Hemp [49], who demonstrated that is fairly constant over an extended strain
range past the local buckling point and can be taken as = 0.4083 for plates of
which the longitudinal edges are free to pull in. As a result, the reduced overall
buckling load in the post-local buckling range is given by Nu = NE, with
N E = 2 EI / L2 . For column lengths between L1 and L2, the equilibrium at a load
NL is stable if:
2 2 EI
> NL . (5)
1 + L2
N
NL,cr
N NE
1.8
neutr

1.6
al

1.4 w0 / t
stable unsatable 1.2 0
NL 0.0125
2 NE 1.0 0.025
N
+1 E 0.05 erosion
0.8 0.1
0.2
n eu t 0.6 0.4
ra l
0.4
NE
0.2 NL,cr
L2 L0 L1 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
1/ 2
L
2 1/ 2
L0= +1 L1 =0.761 L1 L2= L1 =0.639 L1

Fig. 5 a) The van der Neut curve [27]; b) effect of a local imperfection on the buckling load [27].
Eqn. (5) expresses that the column post-buckling capacity, given by
Engessers double modulus formula, has to be greater than the local buckling load
NL, and results in: L2 < L < L0, with L0 = 0.761L1. Columns with L0 < L < L1 are in a
state of unstable equilibrium once the local buckling load is reached and collapse
explosively (e.g. snap through effect).
In a second step, van der Neut considered a local imperfection affine with the
local buckling mode. In this case, was obtained from a Ritz-Galerkin approximate
solution of the von Karman equations. Fig. 5b displays the non-dimensional
11 Erosion of interactive buckling load of thin-walled steel bar members 19

buckling load N/NL,cr function of NE/NL,cr for different values of w0/t, where w0 is
the local imperfection amplitude and t, is the flange thickness. It is seen that the
local imperfection can cause a severe reduction in column capacity, and that the
effect is most pronounced in the vicinity of the point where NE = NL,cr. For instance,
a reduction (e.g. erosion) of 30% was calculated for w0/t = 0.2. It was also
demonstrated that, in the region where the perfect column displays unstable
collapse, the peak of the load-bar shortening curve gets smoothened out as a result
of the imperfection and the instability almost vanishes for w0/t = 0.2. Van der Neut
[50] also investigated the effect of overall imperfections on the idealised column.
The research concluded that the presence of an overall imperfection (e.g. bar
deflection) has a similar negative effect on the column strength.
At the end, the most important observation of this study is the reduction of N,
due to the initial imperfection of flanges which is most significant when NE = NL,cr.
The ECBL approach, proposed by Dubina [14], presented on the following is based
on that conclusion.

2.3. EROSION OF CRITICAL BIFURCATION LOAD ECBL

To understand better mode interaction problem, let consider the theoretical


elastic buckling modes (bifurcation) characterizing the instability of a thin-walled
member in compression. The local mode could be local buckling (L) or distortional
buckling (D), the lower of NL,cr or ND,cr being considered. Similarly, the overall
mode might be either flexural (F) or flexural-torsional (FT). In Fig. 6, (L) and (F)
modes are assumed in order to identify and qualify the erosion of (L) (F)
interaction. These modes are interacting into the theoretical coupling point (Cth),
while the lowest value N = NL,cr with (F) into the practical coupling point (Cpr),
allowing the theoretical, th, and practical, pr, erosions to be evaluated. In case,
distinction can be made between local buckling strength NL,cr or ND,cr and ultimate
stub column strength, NL,u or ND,u, respectively. The NL,u and ND,u values are
obtained considering the relevant imperfections, while for NL,cr and ND,cr there are
no imperfections taken into account.
(F)
N (L)
Cth

Cpr th
NL,cr
NL,u
pr
N(L)

0
LL,cr Lint,th Lint,pr Length (L)

Fig. 6 Theoretical and practical interaction of two buckling modes: distortional (L) and flexural (F).
20 Dan Dubina, Viorel Ungureanu 12

To introduce the ECBL approach, firstly, the interpretation in terms of


erosion of equation of European buckling curves, expressed in Ayrton-Perry format
has to be done. The Ayrton-Perry equation, for the case of a member in
compression, which is not prone to local buckling, but can undergoes buckling in
post-elastic range, can be written in the form:
2
(1 N )(1 N ) = N ( 0.2) , (6)

where is the non-dimensional slenderness (flexural, F , or flexural-torsional,


FT ). It is easy to show the relation between the imperfection factor, , and
erosion coefficient, [14]. In this case, the erosion of theoretical ultimate capacity
in compression is due to the effect of imperfections and plastic deformations. The
negative sign solution of Eqn. (6), in the point = 1 has to be taken equal with
(1 ) , because it corresponds to the maximum erosion of theoretical critical load
when no local buckling occurs, as shown by Eqn. (7), i.e.
1
N ( = 1, ) = 2 + 0.8 (2 + 0.8)2 4 = 1 , (7)
2
that gives
2
= (8)
0.8(1 )
or
= 0.4 ( 5 + 2 . ) (9)

In this case the erosion can be associated to the plastic-elastic interaction


between the rigid plastic mode (plastic strength) of stub column ( 0.2 ) and the
overall elastic buckling mode of the bar given by Euler formula, as shown in Fig. 7.
N = N / N pl
1

N AYRTON -PERRY

(1) 2
N EULER = 1 /
0.5


0
0 0.2 1 2

Fig. 7 The erosion of bar buckling curve.


13 Erosion of interactive buckling load of thin-walled steel bar members 21

For local-overall mode interaction, the two theoretical simple instability


modes assumed to interact in a thin-walled compression member are: (1) the Euler
2
bar instability mode, N E = NE / N pl , N E = 1 / , and (2) the local instability mode,
N L = NL / N pl (Fig. 8).

N
1
Local instability mode
Maximum erosion

NL
C Emax = N L

Bar instability mode (Euler)


2
N ( , N L , ) = (1 ) N L E N E =1/

Coupling point

C = 1 / N L N L N pl
N erod = N ( , N L , ) =
N cr
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2

Fig. 8 Definition of erosion in local-overall mode interaction via ECBL approach.

In this case, the maximum erosion of critical load, due to both, imperfections
and coupling effect occurs in the coupling point, C ( C = 1 / N L ), where:
N = N / N pl , N pl = A f y , where A is the area of the member cross-section
and fy is the yielding strength;
N L = NL / N pl , where NL is either local buckling mode or distortional
buckling mode;
N E = NE / N pl , the Euler critical buckling load.
The interactive buckling load, N (, N L , ) , pass through this point where the
corresponding value of ultimate buckling load is N (C ) = (1 ) N L . It must be
underlined that N L does not rigorously represents the theoretical local buckling,
but it is assumed to be the lower bound of that, and can be used as reference for
strength of the cross-section corresponding to the local or distortional buckling
mode. It enables to estimate the strength of the stub column and to obtain the
coupling point C. On the other hand, the occurrence of local or distortional
buckling, the stiffness of the member decreases significantly, resulting in a jump of
equilibrium onto the overall buckling path. In this case, if compares Fig. 8 with
Fig. 7, the effect of mode interaction is added to those of plastic deformations and
imperfections, and the reference load for evaluation of erosion is not anymore
N = N / N pl = 1 , but NL / N pl < 1 .
22 Dan Dubina, Viorel Ungureanu 14

When local buckling occurs prior to bar buckling (as it was in the case of van
der Neut problem), then the corresponding solution of Eqn. (10), i.e.:
2
( N L N )(1 N ) = ( 0.2) N (10)
in the coupling point C of Fig. 8 is:
2 2
1 + ( 0.2) + N L 1 1 + ( 0.2) + N 2 4 N 2 =
N= L L
2 2 (11)
2 2
= (1 ) N L

which leads to

2 NL
= . (12)
1 1 0.2 N L

This represents the formula of imperfection factor which should be


introduced in European buckling curves in order to adapt these curves to local
overall buckling. Of course, following this approach, the definition of stub column
(i.e. the lower value in the buckling curve) has to be adapted correspondingly.
When speaking about the erosion of theoretical buckling curve in the
coupling point, distinction should be made between the erosion, , which refers to
the effect of both imperfections and coupling, and the reduced ultimate strength of
member, characterized by the normalized local buckling strength, N L . This
approach applies similarly for both local (L) and distortional (D) buckling modes,
providing they are not interacting each other; the basic Ayrton-Perry formula,
presented by Eqn. (6) does not change.
In case of a thin-walled steel member prone to local buckling, N L = N L,cr
can be approximated by Q = Aeff /A, where A is the area of gross cross-section,
while Aeff is the effective area calculated using effective width method. In case of
distortional buckling, N L = N D,cr = ND,cr / Af y , where ND,cr is the critical value of
distortional buckling.
In order to evaluate the erosion factor, two different methods are possible
to this purpose i.e.: experimental and numerical method, respectively [14].
a) Experimental method. The experimental calibration method requires a
relevant set of experimental values located in a close neighbourhood of the
coupling point, called coupling range. Most often available experimental results
scatter, as a result of unavoidable mechanical and geometrical imperfections.
Consequently, the concerned specimens do not meet the main requirement of
ECBL approach to have reduced member slenderness identical to the one locating
15 Erosion of interactive buckling load of thin-walled steel bar members 23

the coupling point ( C = 1 / N L , see Fig. 8). Even in case of own dimensioned
specimens, sized to be theoretically located in the coupling point, the imperfections
produce an unavoidable scatter of the experimental results and require the work
with a coupling range as well.
The selection of the relevant set of specimens should be performed by
choosing among existing results experimental samples reasonably close to the
instabilities coupling point (in terms of reduced slenderness). This is leading to the
idea of using a coupling range, defined in terms of reduced slenderness as a
vicinity of the coupling point, instead of working strictly in this point. A correct
definition of coupling range limits is therefore of paramount importance for the
selection of a relevant set of specimens. Extensive parametric studies [51] have
indicated as acceptable an unsymmetrical coupling range defined around c with
left limit 1 = 0.85 c and the right limit 2 = 1.075 c . All specimens with a
reduced slenderness comprised between these two limits should be considered as
reasonably close to the coupling point and selected as relevant experimental set.
b) Numerical method. Based on an advanced nonlinear inelastic FEM
analysis and taking into account for the imperfections and cold-forming effect, the
numerical models have to simulate relevant experimental values into the coupling
range. However, the numerical method requires also some experimental results in
order to calibrate the FEM model.
The previous approach can be very easily extended to the case of interactive
local/lateral-torsional buckling of thin-walled beams [52]. Following the same
procedure, the LT imperfection factor can be determined, i.e.:

2LT QLT
LT = . (13)
1 LT 1 0.4 QLT

The new ECBL interactive approach for lateral-torsional buckling of thin-


walled beams is similar to that of EN 1993-1-1, but instead of LT given in EN
1993-1.1 the following value should be used:
2
LT = 0.5[1 + LT (LT 0.4) + LT ] , (14)
with LT calculated from Eqn. (13) in terms of the erosion factor LT.
Finally, it appears easier to evaluate experimentally and/or numerically the
erosion coefficients, , for specific types of cold-formed steel sections and, on this
basis, to calibrate relevant imperfection factors, in order to be implemented in the
EN 1993-1-1. Examples of calibration of imperfection factors are presented in
[14, 34, 35, 5257].
24 Dan Dubina, Viorel Ungureanu 16

3. EXAMPLES OF IMPERFECTION FACTORS CALIBRATION

3.1. PLASTIC-ELASTIC INTERACTIVE BUCKLING


VIA ECBL APPROACH [34,35]

Cold-formed steel sections are traditionally considered with no plastic


capacity, and consequently non-ductile, mainly due to wall slenderness involving
local instability phenomena. However, even they do not have sufficient plastic
rotation capacity to form plastic hinges, they can form local plastic mechanisms.
In case of a thin-walled member multiple local buckling modes may occur
simultaneously under the same critical load. For a long member, multiple local
buckling modes, e.g. m-1, m, m+1, characterized by Lm-1, Lm and Lm+1 half wave-
lengths, respectively may interact each other and give rise to an unstable post-critical
behaviour called localization of the buckling pattern (Fig. 9).
Pcr,m-1 Pcr,m Pcr,m+1 Pcr,l
um-1 um um+1 ul

L + =

Periodical modes Localized modes

P=P/A fy P=P/A fy

Pcr Pcr
periodical mode periodical mode
erosion erosion
Pu Pu

actual behavior localized mode


actual behavior
u/L u/L

Fig. 9 Periodical local modes and localization of buckling patterns in case of flanges
of a plain channel section in compression.
The localized buckling mode is in fact an interactive or coupled mode. This is
a first interaction, which may occur prior the overall buckling mode of the member.
The second interaction, between the localized buckling mode and the overall one is
really dangerous because it is accompanied by a very strong erosion of critical
17 Erosion of interactive buckling load of thin-walled steel bar members 25

bifurcation load. When localization of buckling patterns occurs, the member post-
buckling behaviour is characterized by large local displacements, in the inelastic
range, which produce the plastic folding of walls, and the member, falls into a
plastic mechanism [58].
Starting from this real behaviour of thin-walled stub columns and short
beams, Ungureanu & Dubina [34, 35] used the ECBL approach in order to express
the plastic-elastic interactive buckling of thin-walled members. The main problem
of this approach is to evaluate properly the plastic strength of thin-walled members,
via the local plastic mechanism theory and after, the erosion of critical load into the
plastic-elastic coupling range.
0.7 N EUROCODE3-1.3
U62x62x2.05 ECBL elastic-elastic
0.6 ECBL plastic-elastic
ANSYS 5.4 elasto-plastic
0.5 AISI-1996
Tests
0.4

0.3

0.2

0.1

0.0
0.0 0.3 0.6 0.9 1.2 1.5 1.8
(a)
0.7 EUROCODE3-1.3
N C156x54x30x1.49
ECBL elastic-elastic
0.6 ECBL plastic-elastic
ANSYS 5.4 elasto-plastic
0.5 AISI-1996
Tests
0.4

0.3

0.2

0.1

0.0
0.0 0.4 0.8 1.2 1.6
(b)
Fig. 10 Numerical/Experimental comparison for compression members [34].

Following exactly the same way as for the elastic local-overall interactive
buckling, it results the imperfection factor for the plastic-elastic interactive buckling:
26 Dan Dubina, Viorel Ungureanu 18

2 Qpl
= , (15)
1 1 0.2 Qpl

where
N pl ,m
Qpl = N pl , L = (16)
A fy
and Npl,m is the local plastic mechanism strength.
In case of members in compression, Fig. 10 presents the ECBLpl-el results,
compared with those from FEM elastic-plastic analysis, the ECBL elastic-elastic,
ECBLel-el, and experimental tests [34]. It is easy to observe the quality of ECBLpl-el
results are excellent, particularly in the interactive zone, e.g. 0.4< <1.6.
In case of slender beams, experimental data were used to compare the
ECBLpl-el and ECBLel-el results with those of EN1993-1-3 and AISI-1996 results.
Figure 11 shows again that ECBLpl-el model confirm its accuracy [35].
1.20 M (kNm) LOVELL-SERIA A
ECBL elastic-elastic
U90x36x1.15
ECBL plastic-elastic
1.00
AISI-1996

EUROCODE3-1.3
0.80
Tests

0.60

0.40

0.20

L (mm)
0.00

(a) 0 500 1000 1500 2000 2500 3000 3500

1.6 M (kNm)
LOVELL - SERIA C
C122x25x15x1.15
1.4
ECBL elastic-elastic
1.2 ECBL plastic-elastic
EUROCODE3-1.3
1.0
AISI-1996
0.8 Tests

0.6

0.4

0.2
L (mm)
0.0
0 600 1200 1800 2400 3000 3600
(b)
Fig. 11 Numerical/Experimental comparison for bending members [35].
19 Erosion of interactive buckling load of thin-walled steel bar members 27

The local rigid-plastic model, describes properly the behaviour of thin-walled


short members. This model is consistent with the real phenomenon of stub columns
and short beams failure and is confirmed by test results and advanced elastic-
plastic FEM analysis. The plastic-elastic interactive model naturally describes the
phenomenon of the interactive buckling of thin-walled members. The ECBL
plastic-elastic interactive approach, based on the erosion theory of coupled
bifurcation, is much more rigorous and understandable than the semi-empirical
methods used for the buckling curves in existing design codes.

3.2. EROSION OF BUCKLING STRENGTH DUE TO THE INFLUENCE


OF THE SHAPE OF SECTIONAL GEOMETRICAL IMPERFECTIONS [53]

Based on numerical simulations and applying the ECBL approach, Dubina &
Ungureanu [53] have systematically studied the influence of size and shape of
sectional geometrical imperfections on the ultimate buckling strength of plain and
lipped channel sections, both in compression and bending, in order to evaluate the
erosion of theoretical strength when sectional and overall buckling modes interact.
Fig. 12 explains the erosion phenomenon applied to this problem [14].
The following notations were used:
N = N Npl , where N is the ultimate strength of the member; Npl is
corresponding full plastic strength;
N L,th = NL,th N pl , with NL,th the ultimate theoretical stub column strength;
N L = NL N pl , NL being the ultimate strength of imperfect stub column;
= NL / Ncr , the reduced slenderness of the member.

1 c,th = theoretical erosion due to


N 2
(Euler)
coupling effect only;
L = actual erosion due to local
1 imperfections only;
c = actual erosion due to coupling
M effect and global imperfection;
N L,th = actual total erosion due to both
c,th L coupling and imperfections.
NL
N (c,th, )
c
N (, )

1 1 1
N L,th NL

Fig. 12 The interactive buckling model based on the ECBL theory [53].
28 Dan Dubina, Viorel Ungureanu 20

Maximum erosion of theoretical interactive buckling strength is calculated


in regard with the theoretical interaction point, M (int = 1 / N L,th ) , and is:

= N L,th - N ( = 1 / N L ) . (16)

The total erosion can be associated with the (LT) imperfection factor used
in European buckling curves for members in compression (bending), by means of
ECBL formula:
Compression Bending

2 NL 2LT ML (17)
= LT = .
1 1 0.2 N L 1 LT 1 0.4 M L

The N and M values can be computed for perfect and imperfect shapes of
both, cross-section and member. Therefore, the erosion can be evaluated for
different imperfection cases. If no imperfections, the evidence of interactive
buckling effect only will be observed. Further, the values of (LT) imperfection
sensitivity factor used in European buckling curves have been evaluated for all
these imperfection shapes. Tables 2 and 3 show the main results of this study [53].

Table 2
imperfection sensitivity factor for members in compression [53]

Plain channel 96361.5 Lipped channel 9636121.5


Buck- Buck-
Shape of Imperfection Shape of Imperfection
ling ling
imperf. mode imperf. mode
curve curve
- local buckling - local buckling
PL1 (symmetric 0.450 0.322 b LL1 (symmetric 0.286 0.109 ao
sine shape) sine shape)
- local buckling - local buckling
PL2 LL2
0.442 0.304 b 0.283 0.105 ao
(asymmetric (asymmetric
sine shape) sine shape)
- distortional - distortional
buckling PD3 buckling LD3
0.466 0.354 c 0.492 0.461 c
(symmetric sine (symmetric sine
shape) shape)
- distortional - distortional
buckling PD4 buckling LD4
0.471 0.365 c 0.404 0.265 b
(asymmetric (asymmetric
sine shape) sine shape)
21 Erosion of interactive buckling load of thin-walled steel bar members 29

Table 3
LT imperfection sensitivity factor for members in bending [53]
Plain channel 96361.5 Lipped channel 9636121.5
Shape Buck- Buck-
Imperfection Shape of Imperfection
of LT LT ling LT LT ling
mode imperf. mode
imperf. curve curve
- distortional - distortional
buckling PD2 buckling LD1
(the imperf. is 0.311 0.140 a (the imperf. is 0.355 0.292 b
constant over constant over the
the length) length)
- distortional - distortional
buckling PD7 buckling LD4
0.312 0.142 a 0.411 0.422 c
(asymmetric (asymmetric sine
sine shape) shape)

The appropriate identification and selection of imperfection shape and size


associated to the relevant instability mode is crucial for ultimate strength analysis.
In a two-mode interacting buckling (e.g. local-overall interaction) different shapes
of local-sectional imperfections have different effects on the ultimate strength of
the member. The values of imperfection factor prove the higher sensitivity of
distortional-overall interactive buckling to sectional imperfections. This fact can be
explained by the lower post-critical strength reserve of distortional mode if coupled
with local one.

3.3. DISTORTIONAL-OVERALL MODE INTERACTION OF PERFORATED


PALLET RACK UPRIGHTS. IMPERFECTION SENSITIVITY ANALYSIS

The sections currently used in pallet rack uprights are particularly prone to
distortional-overall interaction. An extensive experimental study on pallet rack
uprights in compression has been carried out at the Politehnica University of
Timisoara on the aim to observe the erosion of theoretical buckling load due to
both coupling effect and imperfections for this type of interaction. The experimental
program was extensively presented in [54].
Two cross-sections of the same typology but different sizes, RS1253.2 and
RS952.6, have been considered, of perforated-to-brut cross-section ratios (AN/AB)
of 0.806 and 0.760, respectively. Their brut and perforated (i.e. net) sections are
shown in Fig. 13 together with the perforations details. The pitch is 50mm for both
studied sections.
Both perforated and unperforated section specimens have been tested, of
calibrated lengths for: stub columns (s); upright member specimens for distortional
buckling (u); specimens of lengths equal with the half-wave length for distortional
buckling (d); specimens of lengths corresponding to interactive buckling range (c).
30 Dan Dubina, Viorel Ungureanu 22

Fig. 13 a) Brut and perforated specimen cross-section; b) perforation details [54].


Table 4 presents the failure modes for each type of the tested specimen/
section. The following notations have been used: S Squash, DS symmetrical
distortional buckling, FT flexural-torsional buckling, F flexural buckling
Additional experimental tests have been done in order to determine the mechanical
properties of the material. A set of samples were tested from the base material.
Additional series of tests on coupons cut over the cross-section of specimens
without perforations was done for both types of sections to determine the increase
of yield strength, ultimate tensile strength and residual stresses. In what concerns
the geometric imperfections, all tested specimens were measured. Two types of
imperfections were recorded, i.e. (a) sectional and (b) global.
Table 4
Failure modes for tested sections
Section RS952.6 RS952.6 RS1253.2 RS1253.2
Test type brut perforated brut perforated
Stub (s) S S/DS DS DS
Distortional (d) DS DS DS DS
Upright (u) F or FT F or FT DS DS
DS+F or DS+F or DS+F or DS+F or
Interactive buckling (c)
DS+FT DS+FT DS+FT DS+FT

Advanced numerical models (i.e. GMNIA) have been applied to simulate the
behaviour of studied sections, using the commercial FE program ABAQUS/CAE.
The numerical models were calibrated to replicate the physical experimental tests.
It must be underlined that for all considered numerical models, the failure modes
were in accordance with the failure modes observed in experimental tests (see Fig. 14).
The calibrated numerical models were validated against experimental tests for all
tested sets of profiles. Table 5 presents the values of ultimate load from numerical
simulations and the experimental ones for all types of members ((s), (u), (d), (c)),
for both RS1253.2 and RS952.6 cross-sections, with and without perforations.
For details see [55, 56].
23 Erosion of interactive buckling load of thin-walled steel bar members 31

On the following the numerical investigations on the sensitivity to


imperfections of pallet rack sections in compression, having the member length
equal to the interactive buckling length, using ECBL approach are summarized [57,
59]. On this purpose, FE analyses were performed to simulate the influence of
different types of imperfections in the coupling point. Because the interest is to
observe the erosion of critical bifurcation load, this time, the ECBL approach is
applied considering the distortional critical load, obtained for the relevant section
by an eigen buckling analysis, in interaction with Euler buckling of the cor-
responding bar member

RSNd RSNu
RSBc952.6
RSBs 1253.2 1253.2 952.6
Fig. 14 Failure modes: Experimental vs. FE models [55, 56].

Table 5
Ultimate load [kN]: experimental vs. FEM [55, 56]
RSBs1253.2 RSNs1253.2 RSBs952.6 RSNs952.6
EXP FEM EXP FEM EXP FEM EXP FEM
487.05 486.13 411.02 422.98 338.88 335.15 274.33 272.01
RSBd1253.2 RSNd1253.2 RSBd952.6 RSNd952.6
EXP FEM EXP FEM EXP FEM EXP FEM
440.79 440.78 394.62 397.04 325.10 331.05 262.67 255.47
RSBu1253.2 RSNu1253.2 RSBu952.6 RSNu952.6
EXP FEM EXP FEM EXP FEM EXP FEM
386.72 384.40 347.26 344.00 279.65 285.96 223.33 231.89
RSBc1253.2 RSBc1253.2 RSBc952.6 RSBc952.6
EXP FEM EXP FEM EXP FEM EXP FEM
317.89 316.67 293.62 292.9 220.29 220.26 168.88 177.11
(s) Stub columns; (d) Specimens of lengths equal with the half-wave length of distortional buckling;
(u) Upright member specimens; (c) Specimens of lengths corresponding to interactive buckling
range. N/B perforated/brut
32 Dan Dubina, Viorel Ungureanu 24

Table 6 shows the reference values for critical and ultimate sectional loads
obtained numerically and experimentally for the studied sections. Table 7 presents
the lengths corresponding to the theoretical interactive buckling loads determined
via the ECBL approach, in the interactive buckling point for each section [57, 59].

Table 6
Sectional capacity and distortional buckling load [57]
Section RSN1253.2 RSN952.6
Length [mm] 600 500
Distortional buckling load*
370.48 340.78
(Ncr,D) [kN]
Distortional ultimate load**
388.35 ---
(ND,u) [kN]
Stub ultimate load***
407.79 279.27
(NS,u) [kN]
Squash load****
480.94 286.72
(Npl) [kN]
* distortional buckling load determined using LBA; ** experimental failure load corresponding to
distortional specimens mean values; *** experimental failure load corresponding cu stub
column specimens mean values; **** Npl=A.fy

Table 7
Lengths corresponding to the theoretical interactive buckling [57]

Profile Ncr,D [kN] Npl [kN] ND Coupling length [mm]


RSN125 370.48 480.94 0.770 2559
RSN95 340.78 286.72 1.000 1667

It can be observed that for RS95N cross-sections, the critical load


corresponding to distortional buckling is greater than the cross-section squash load.
In this case the N D value has to be limited to 1.00. Based on this limitation for
RS95 section, with and without perforation, there is no classical interactive buckling,
but we could speak about a local plastic elastic global buckling interaction.
On the following, an imperfection sensitivity study was conducted in order to
identify the most critical imperfection or combination of imperfections.
Fig. 15 shows the geometrical imperfections, considered in the analysis,
i.e. distortional (d ), flexural about the minor axis (f ), and coupling of these two
(f d ). Also, load eccentricities, located on the axis of symmetry, were taken into
consideration.

CG CG y CG y
f+ y
CG z
z z Load Ecc. z

d+ Ecc. y

Fig. 15 Example of considered simple imperfections (f and d) [57, 59].


25 Erosion of interactive buckling load of thin-walled steel bar members 33

In case of flexural-torsional buckling (FT), both initial deflection and initial


twisting imperfection (ft) were considered together, according to Australian
Standard AS4100 [60]. Due to the fact that the global flexural buckling mode about
the minor axis has the minimum value for the studied sections the global
imperfection considered for coupling was considered a global bow imperfection.
The imperfections used for this study were: distortional symmetric imper-
fection (ds), distortional asymmetric imperfection (da) (only for RSN1253.2 section),
flexural bow imperfection about the minor inertia axis (f), loading eccentricities on
both axes (independent and coupled EY, EZ, EY-EZ) and flexural-torsional
imperfection (FT). The distortional imperfection, symmetric and asymmetric, was
scaled to 0.5t, 1.0t and 1.5t, the flexural bow imperfection was scaled to L/750,
L/1000 and L/1500, while the flexural-torsional imperfection was considered in
accordance with the provisions of Australian design code [3.60]. The loading
eccentricities were varied on both sectional axes, with 2 mm, 4 mm, 6 mm,
independently (e.g. EZ-4 means 4 mm eccentricity about z-axis) and together, as
an oblique eccentricity (e.g. EY-EZ4 means + 4 mm eccentricity about y-axis and
+4 mm about z-axis).
Table 8 presents the considered simple imperfections, sectional, global and
loading eccentricities for RSN1253.2 section together with erosion coefficient
and imperfection factors for simple imperfections.
In Table 8 can be easily observed that, for simple imperfections, symmetric
distortion imperfection and major axis eccentricities give higher values for erosion
coefficient than those corresponding to flexural and flexural-torsional imperfections.
Table 9 presents the coupled imperfections considered for the RSN1253.2
section, i.e. f L/750, ds 0.5t; f L/750, ds 1.5t; f L/1500, ds 0.5t and f
L/1500, ds 1.5t, combinations coupled with various types of eccentricities. It is
easy to observe that the combination (f L/750, ds 1.5t) of imperfections is the
most critical one. However, statistically is not recommended to combine all
imperfections to cumulate their negative effects, because their random compensation.

Table 8
erosion coefficients and imperfection factors for simple imperfections
RSN1253.2 RSN1253.2
Imperfection Imperfection

ds 0.5 t 0.236 0.078 EZ -6 0.313 0.152
ds 1.0 t 0.339 0.185 EZ -4 0.272 0.108
ds 1.5 t 0.398 0.280 EZ -2 0.210 0.059
da 0.5 t 0.152 0.029 EZ +2 0.216 0.063
da 1.0 t 0.245 0.085 EZ +4 0.255 0.093
da 1.5 t 0.321 0.162 EZ +6 0.285 0.121
f L/750 0.240 0.081 EY-EZ 0 0.157 0.031
f L/1000 0.216 0.063 EY-EZ +6 0.321 0.162
f L/1500 0.181 0.043 EY-EZ +4 0.276 0.112
ft 0.240 0.081 EY-EZ +2 0.215 0.063
EY +2 0.169 0.037 EY-EZ -2 0.223 0.068
EY +4 0.196 0.051 EY-EZ -4 0.270 0.106
EY +6 0.224 0.069 EY-EZ -6 0.307 0.145
34 Dan Dubina, Viorel Ungureanu 26

Table 9
erosion coefficients and imperfection factors for coupled imperfections

Imperfection
f L/750, ds 0.5t f L/750, ds 1.5t f L/1500, ds 0.5t f L/1500, ds 1.5t
EY 2 0.339 0.185 0.440 0.368 0.302 0.139 0.422 0.328
EY 4 0.342 0.189 0.442 0.373 0.305 0.142 0.423 0.330
EY 6 0.346 0.195 0.443 0.375 0.310 0.148 0.425 0.334
EZ 6 0.425 0.334 0.493 0.510 0.411 0.305 0.483 0.480
EZ 4 0.404 0.292 0.479 0.469 0.384 0.255 0.467 0.436
EZ 2 0.376 0.241 0.461 0.420 0.350 0.201 0.447 0.385
EZ -2 0.279 0.115 0.413 0.309 0.174 0.039 0.387 0.260
EZ -4 0.194 0.050 0.374 0.238 0.228 0.072 0.326 0.168
EZ -6 0.240 0.081 0.276 0.112 0.264 0.101 0.261 0.098
EY-EZ 0 0.240 0.081 0.440 0.368 0.301 0.138 0.421 0.326
EY-EZ 6 0.430 0.345 0.495 0.517 0.414 0.311 0.485 0.486
EY-EZ 4 0.406 0.295 0.480 0.472 0.386 0.258 0.467 0.436
EY-EZ 2 0.377 0.243 0.462 0.422 0.351 0.202 0.447 0.385
EY-EZ -2 0.280 0.116 0.413 0.309 0.182 0.043 0.387 0.260
EY-EZ -4 0.218 0.065 0.376 0.241 0.247 0.086 0.330 0.173
EY-EZ -6 0.271 0.107 0.298 0.135 0.289 0.125 0.285 0.121

A precise framing for coupled instabilities is very important in order to


choose a suitable design strategy. For weak and moderate interaction class, simple
design methods based on safety coefficients can be used. In case of strong and very
strong interaction, special design methods must be developed [14].
It can be observed that for the case of RSN1253.2 pallet rack section, the
computed erosion can classify the section into medium up to very strong
interaction, depending on the considered imperfection.

4. CONCLUDING REMARKS

The main aim of this chapter was to provide evidences that the activity in the
field of structural stability, particularly focussing the mode interaction problems,
developed by the Timisoara researchers can be characterised as an activity of a
school. Among the different subjects which have been subjects of theoretical,
experimental and numerical investigation of the school in connection with that
topic, those referring to coupled bifurcations, erosion of critical bifurcation load
and ultimate post-critical strength are, in our opinion the most significant, leading
to the so called ECBL approach, actually known as an available procedure
enabling to calibrate buckling curves for mode interaction problems.

Received on July 16, 2014


27 Erosion of interactive buckling load of thin-walled steel bar members 35

REFERENCES

1. RONDAL, J., DUBINA, D., GIONCU, V. (guest editors), Thin-walled Structures: Coupled
Instabilities in Metal Structures, Thin-walled Structures, 19, 24, and 20, 14, 1994.
2. RONDAL, J., DUBINA, D., GIONCU, V. (eds.), Coupled Instabilities in Metal Structures - CIMS96,
Imperial College Press, London 1996.
3. DUBINA, D., UNGUREANU, V. (eds.), Proceedings of the 6th International Conference on Thin-
Walled Structures: Recent Research Advances and Trends, Volume 1 + 2, 57 September
2011, Timisoara, Romania, ECCS, 2011.
4. DUBINA, D., UNGUREANU, V. (guest editors), Recent Research Advances on Thin-Walled
Structures, Special Issue, Thin-Walled Structures, 61, December 2012.
5. DUBINA, D., IVANYI, M. (eds.), Stability and Ductility of Steel Structures, Proceedings of the 6th
International Colloquium, First Session SDSS99, Timisoara, Romania, 9-11 September
1999, Elsevier Science Ltd., 1999.
6. RONDAL, J. (ed.), Coupled Instabilities in Metal Structures. Theory and Practical Aspects, CISM
Courses and Lectures No. 379, Springer Verlag, Wien, New York, 1998.
7. RONDAL, J., DUBINA, D. (eds.), Light gauge metal structures. Recent advances, CIMS Udine
Courses and Lectures No. 455, Springer Verlag, Wien, New York, 2005.
8. PIGNATARO, M., GIONCU, V. (eds.), Phenomenological and mathematical modelling of coupled
instabilities, CIMS Udine Courses and Lectures No. 470, Springer Verlag, Wien, New
York, 2005.
9. DUBINA, D. (guest editor), Cold Formed Structures: Recent research advances in Central and
Eastern Europe, Special Issue, Thin-Walled Structures, 42, 2, 2004.
10. MATEESCU, D., GIONCU, V., DUBINA, D., Timisoara Steel Structures Stability Research School:
relevant contributions, Journal of Constructional Steel Research. 55, 13, pp. 343354, 2000.
11. DUBINA, D., UNGUREANU, V., Review and Current Trends in Stability of Structures. Statics,
Dynamics and Stability of Structures, Vol. 3 (eds. K. Kowal-Michalska & R.J. Mania), Chapter 1:
Mode interaction buckling in thin-walled bar members, pp. 1968, TU Lodz Press, Poland, 2013.
12. GIONCU, V., BALUT, N., DUBINA, D., MOLDOVAN, A., PACOSTE, C., Coupled instabilities in
mono-symmetrical steel compression members, Journal of Constructional Steel Research, 21,
13, pp. 7195, 1992.
13. GIONCU, V., General Theory of Coupled Instability, Thin-Walled Structures, Special Issue on
Coupled Instability in Metal Structures, 19, 24, pp. 81128, 1994.
14. DUBINA, D., The ECBL approach for interactive buckling of thin-walled steel members, Steel &
Composite Structures, 1, 1, pp. 7596, 2001.
15. OLHOFF, N., SEYRANIAN, A.P., Bifurcation and post-buckling analysis of bimodal optimum
columns, International Journal of Solids and Structures, 45, pp. 39673995, 2008.
16. HUTCHINSON, J.W., KOITER, W.T., Post-buckling theory, Applied Mechanics Reviews, 23,
pp. 13531366, 1970.
17. THOMPSON, J.M.T., HUNT, G.W., A General Theory of Elastic Stability, Wiley, London, 1973.
18. THOMPSON, J.M.T., HUNT, G.W., Elastic Instability Phenomena, Wiley, London, 1984.
19. BUDIANSKY, B., Theory of buckling and post-buckling behavior of elastic structures, Advances
in Applied Mechanics (ed. C.-S. Yih), 14, pp. 165, 1974.
20. FLORES, F.G., GODOY, L.A., Elastic post-buckling analysis via finite element and perturbation
techniques. Part 1: Formulation, International Journal of Numerical Methods in Engineering,
33, pp. 17751794, 1992.
21. KOITER, W.T., Current trends in the theory of buckling, Proceedings of IUTAM Symposium.
Buckling of Structures (ed. B. Budiansky), June 1724, Cambridge, MA/USA, Springer,
Berlin, pp. 116, 1976.
22. CHOW, C.-N., HALE, J.K., Methods of Bifurcation Theory, Grundlehren der Mathematischen
Wissenschaften, 251, Springer, New York, 1982.
36 Dan Dubina, Viorel Ungureanu 28

23. GOLUBITSKY, M., SCHAEFFER, D., Singularities and Groups in Bifurcation Theory, Vol. I,
Springer, New York, 1985.
24. GIONCU, V., IVAN, M., Teoria comportarii critice si postcritice a structurilor elastic (in
Romanian), Edit. Academiei Romane, Bucharest, Romania, 1984.
25. KOITER, W.T., On the stability of elastic equilibrium, Thesis, Delft, English Translation, 1945,
NASA TT-F10, 883 (1967) and AFFDL TR70-25 (1970).
26. BYSKOV, E., Elastic buckling problem with infinitely many local modes, Mechanics of
Structures and Machines, 15, 4, pp. 413435, 19871988.
27. Van der NEUT, A., The interaction of local buckling and column failure of thin-walled
compression members, Proc. of the 12th Int. Congress on Applied Mechanics (eds. M.K.
Heteny, W.G. Vincenti), Springer-Verlag, Berlin, 1969, pp. 389-399.
28. KOITER, W.T., KUIKEN, G.D.C., The interaction between local buckling and overall buckling
on the behavior of built-up columns, Technical Report Rept. WTHD-23, Delft University of
Technology, The Netherlands, 1971.
29. KOITER, W.T, General Theory of Mode Interaction in Stiffened Plates and Shell Structures,
Technical Report Rept. WTHD-91, Delft University of Technology, The Nederland, 1976.
30. THOMPSON, J.M.T., LEWIS, G.M., On the optimum design of thin walled compression
members, J. Mech. Phys. Solids, 20, 2, pp. 101109, 1972.
31. TVEGAARD, V., Imperfection-sensitivity of a wide integrally stiffened panel under compression,
International Journal of Solids and Structures, 9, 1, pp. 177192, 1973.
32. TVEGAARD, V., Influence of post buckling behavior on optimum design of stiffened panels,
International Journal of Solids and Structures, 9, 12, pp. 15191534, 1973.
33. RONDAL, J., MAQUOI, R., Formulation dAyrton-Perry pour le flambement des barres
mtalliques, Construction Mtallique, 4, pp. 4153, 1979.
34. UNGUREANU, V., DUBINA, D., Recent research advances on ECBL approach. Part I: Plastic-
elastic interactive buckling of cold-formed steel sections, Thin Walled Structures, 42, 2,
pp. 177194, 2004.
35. UNGUREANU, V., DUBINA, D., Post-elastic strength and ductility of cold-formed steel
sections, Proc. of the Fourth International Conference on Thin-Walled Structures,
Loughborough, UK, 22-24 June 2004, pp. 283290.
36. SCHARDT, R., Verallgemeinerte Technische Biegetheorie, Springer-Verlag, Berlin, 1989.
37. DAVIES, J.M., LEACH, P., First-order generalised beam theory, Journal of Constructional Steel
Research, 31, 23, pp. 187220, 1994.
38. DAVIES, J.M., LEACH, P., HEINZ, D., Second-order generalised beam theory, Journal of
Constructional Steel Research, 31, 23, pp. 221241, 1994.
39. SILVESTRE, N., CAMOTIM, D., First-order generalised beam theory for arbitrary orthotropic
materials, Thin-Walled Structures, 40, 9, pp. 755789, 2002.
40. SILVESTRE, N., CAMOTIM, D., Second-order generalised beam theory for arbitrary
orthotropic materials, Thin-Walled Structures, 40, 9, pp. 791820, 2002.
41. SILVESTRE, N., CAMOTIM, D., Nonlinear Generalized Beam Theory for Cold-formed Steel
Members, Int. Journal of Structural Stability and Dynamics, 3, 4, pp. 461490, 2003.
42. CAMOTIM, D., DINIS, P.B., Coupled instabilities with distortional buckling in cold-formed steel
lipped channel columns, Thin-Walled Structures, 49, pp. 562575, 2011.
43. SCHAFER, B.W., DNY, S., Buckling analysis of cold-formed steel members using CUFSM:
Conventional and constrained finite strip methods, Proc. of the 18th International Specialty
Conference on Cold-Formed Steel Structures: Recent Research and Developments in Cold-
Formed Steel Design and Construction, Orlando, Florida, October 2006. pp. 3954.
44. LI, Z., SCHAFER, B.W., Buckling analysis of cold-formed steel members with general boundary
conditions using CUFSM: conventional and constrained finite strip methods, Proc. of the Int.
Specialty Conference on Cold-Formed Steel Structures, St. Louis, MO, November, 2010.
29 Erosion of interactive buckling load of thin-walled steel bar members 37

45. SCHAFER, B.W., Designing cold-formed steel using the Direct Strength Method, Proc. of the
18th International Specialty Conference on Cold-Formed Steel Structures, October 26-27,
Orlando, Florida, USA, 2006.
46. *** Supplement 2004 to the North American Specification for the Design of Cold-Formed Steel
Structural Members, 2001, Edition: Appendix 1 Design of Cold-Formed Steel Structural
Members Using Direct Strength Method. American Iron and Steel Institute, Washington,
D.C., SG05-1, 2004.
47. GARCEA, G., MADEO, A., ZAGARI, G., CASCIARO, R., Asymptotic post-buckling FEM
analysis using corotational formulation, Int. J. Solids and Structures, 49, pp. 377397, 2009.
48. GARCEA, G., MANDEO, A., CASCIARO, R., The implicit corotational method and its use in
the derivation of nonlinear structural models for beams and plates, Journal of Mechanics of
Materials and Structures, 7, 6, pp. 509-538, 2012.
49. HEMP, W.S., The Theory of Flat Panels Buckled in Compression, Reports and Memoranda,
Aeronautical Research Council, No. 2178, 1945.
50. van der NEUT, A., The Sensitivity of Thin-walled Compression Members to Column Axis
Imperfection, Int. Journal of Solids and Structures, 9, pp. 9991011, 1973.
51. GEORGESCU, M., Instability problems by Thin-walled Cold-formed Steel Members, Politehnica
University of Timisoara, PhD Thesis, 1998.
52. DUBINA, D., UNGUREANU, V., Lateral-Torsional and Local Interactive Buckling of Thin-
Walled Cold-Formed Beams, Rev. Roum. Sci. Techn. Mc. Appl., 42, 56, pp. 467481, 1997.
53. DUBINA, D., UNGUREANU, V., Effect of imperfections on numerical simulation of instability
behaviour of cold-formed steel members, Thin Walled Structures, 40, 3, pp. 239262, 2002.
54. CRISAN, A., UNGUREANU, V., DUBINA, D., Behaviour of cold-formed steel perforated
sections in compression. Part 1 Experimental investigations, Thin Walled Structures, 61,
pp. 8696, 2012.
55. CRISAN, A., UNGUREANU, V., DUBINA, D., Behaviour of cold-formed steel perforated
sections in compression. Part 2 Numerical investigations and design considerations, Thin
Walled Structures, 61, pp. 97105, 2012.
56. DUBINA, D., UNGUREANU, V., CRISAN, A., Experimental Evidence of Erosion of Critical
Load in Interactive Buckling, Journal of Structural Engineering ASCE, 139, 5, pp. 705716,
2013.
57. UNGUREANU, V., DUBINA, D., Sensitivity to Imperfections of Perforated Pallet Rack Sections,
Mechanics and Mechanical Engineering, 17, 2, 209222, 2013.
58. UNGUREANU, V., KOTEKO, M., MANIA, R.J., DUBINA, D., Plastic mechanisms database
for thin-walled cold-formed steel members in compression and bending, Thin Walled
Structures, 48, 1011, pp. 818826, 2010.
59. DUBINA, D., UNGUREANU, V., Instability mode interaction: From Van Der Neut model to
ECBL approach. Thin-Walled Structures, 81, AS4100-1990, pp. 3949, 2014.
60. *** Australian Standard: Steel Structures, Homebush, Australia, 1990.
A NUMERICAL ASYMPTOTIC FORMULATION
FOR THE POST-BUCKLING ANALYSIS OF STRUCTURES
IN CASE OF COUPLED INSTABILITY

GIOVANNI GARCEA, ANTONIO BILOTTA, ANTONIO MADEO,


GIUSEPPE ZAGARI, RAFFAELE CASCIARO*

Abstract. The analysis of slender structures, characterized by complex buckling and


postbuckling phenomena and by a strong imperfection sensitivity, is heavily penalized
by the lack of adequate computational tools. Standard incremental iterative approaches are
computationally expensive and unaffordable, while FEM implementation of the
Koiter method is a convenient alternative. The analysis is very fast, its computational
burden is of the same order as a linearized buckling load evaluation and the
simulation of different imperfections costs only a fraction of that needed to
characterize the perfect structure. The main objective of the present work is to show
that finite element implementations of the Koiter method can be both accurate and
reliable and to highlight the aspects that require further investigation.

Key words: Koiter asymptotic analysis, coupled instability, FE method.

1. INTRODUCTION

A global evaluation of the structural collapse safety of slender elastic


structures should consider all possible loadings, including the deviations due to
load imperfections and geometrical defects. Standard path-following approaches,
aimed at recovering the equilibrium path for a single loading case and assigned
imperfections, are not suitable for this purpose. In fact in order to perform a
reliable structural safety assessment the nonlinear analysis should be performed
with respect to all possible imperfection shapes. The consequent computational
burden can be very high particularly if no reliable information about the worst
imperfection shapes is available.
The asymptotic approach, derived as a finite element implementation [13, 5,
10, 14, 23, 24, 3239] of the Koiter nonlinear theory of elastic stability [28], can be
a convenient alternative as it provides an effective and reliable strategy for
predicting the initial post-critical behavior in both cases of limit or bifurcation
points [31, 30, 6]. The implementation of the asymptotic approach as a computational
tool is quite easy and its total computational burden remains of the order as that
*
University of Calabria, Arcavacata di Rende, CS, Italy
2 A numerical asymptotic formulation for post-buckling analysis in case of coupled instability 39

required by a standard linearized stability analysis. It provides the initial post-


buckling behavior of the structure, including modal interactions and jumping-after-
bifurcation phenomena. Moreover, once the analysis has been performed, the
presence of small loading imperfections or geometrical defects can be taken into
account in a post-processing phase with a negligible computational extra-cost, so
allowing an inexpensive imperfection sensitivity analysis [30, 7]. It is also possible
to extract information about the worst imperfection shapes [34, 8], and it can be
used to improve the imperfection sensitivity analysis or for driving more detailed
investigations through specialized path following analyses (see [6, 8] and references
therein). From this point of view the method could be an effective tool for the
evaluation of the buckling curves used in European codes [11, 13, 12]. The
asymptotic analysis can provide a very accurate recovery of the equilibrium path,
as it is confirmed by numerical testing and theoretical investigations [4] but requires
great care in both the mechanical modeling [17, 18] and its finite element
implementation. As it will be shown in the paper accuracy cannot be obtained by
an inappropriate finite element interpolation due to the occurrence of interpolation
locking phenomena in the evaluation of the energy variation terms used to
reconstruct the post-critical behavior [31, 6]; by an inappropriate format used in the
control variables that can produce extrapolation locking phenomena [20, 15]) or by
the use of non-objective structural models [17, 18].
The paper is organized as follows: section 2 presents the asymptotic method,
section 3 discusses all the aspects regarding the FEM implementation and the
accuracy, section 4 gives some numerical results showing the potentialities of the
method and finally section 5 summarizes the discussion and suggests possible
extensions.

2. THE ASYMPTOTIC METHOD

In this section an asymptotic algorithm capable of treating single or multiple,


also not coincident, bifurcations and of considering the effects of a nonlinear pre-
critical behavior is presented. Further details can be found in [3138].

2.1. A LYAPUNOV-SCHMIDT-KOITER ASYMPTOTIC METHOD

A brief overview of the FEM implementation of Koiters asymptotic


approach is presented here, for the convenience of the reader and to summarize the
main notation and equations involved. Further details can be found in [1831].
We consider a slender hyperelastic structure subjected to conservative loads
increasing with an amplifier factor. The equilibrium is expressed by the virtual
work equation:
40 G. Garcea, A. Bilotta, A. Madeo, G. Zagari, R. Casciaro 3

'[u ] u p u = 0, u J , (1)

where u U is the field of configuration variables, [u ] denotes the strain energy,


is the tangent space of U at u and a prime is used for expressing the Frchet
derivative with respect to u. We assume that U will be a linear manifold so that its
tangent space will be independent from u. Eq. (1) defines a curve in the (u, )
space, the equilibrium path of the structure, that can be composed of several
branches. We are usually interested in the branch starting from an initial known
equilibrium point u0, 0 and without any loss of generality we can assume u0 = 0,
0 = 0. It is worth mentioning that a mixed format is generally convenient to avoid
the so called nonlinear locking phenomena [20, 15, 22], so configuration u usually
collects both displacement and stress fields.
The asymptotic method is based on an expansion of the potential energy, in
terms of load factor and buckling mode amplitudes i , which is characterized by
fourth-order accuracy. It provides an approximation of the equilibrium path by
performing the following steps:
1. The fundamental path is obtained as a linear extrapolation, from a known
equilibrium configuration:
u f [ ] = u, (2)

where u is the tangent {0; 0}, obtained as a solution of the linear equation
0u u = p u , u J (3)

and an index denotes the point along u f for which the quantities are evaluated, that
is 0 u f [ 0 ] .
2. A cluster of buckling loads {0m} and associated buckling modes
(  ... m ) are defined along u f [ ] by the critical condition

u f [ i ]i u = 0, u J , (4)

Buckling loads are considered to be sufficiently close to each other to allow


the following linearization

bi u + ( i b )bu i u = 0, u J , (5)

b being an appropriate reference value of (e.g. the first of i or their mean


value). Normalizing, we obtain bui j = ij , where ij is Kronekers symbol.
4 A numerical asymptotic formulation for post-buckling analysis in case of coupled instability 41

3. The tangent space is decomposed into the tangent V { = i  i } and


orthogonal W {w : b u iw = 0} subspaces so that J = V W . Making and 0 =
and  0 = u, the asymptotic approximation for the required path is defined by the
expansion
m m
(6)
 i + w
1
u [ , k ] i j ij ,
i =0 2 i =0

where w ij are quadratic corrections introduced to satisfy the projection of eq. (1)
onto W and obtained by the linear orthogonal equations
(7)
bw ij w = b  i  j w , w ij w W ,

where, because of the orthogonality condition, w 0i = 0.


4. The following energy terms are computed for i, j, k = 1m:
1 1
k [ ] = 2b u 2  k + 2 ( 3b )b u 3 k
2 6

Aijk = b  i  j  k

Bijhk = b  i  j  h  k b (w ij w hk + w ihw jk + w ik w jh ) (8 )

B00 jk = b u 2  i  k b w 00w ik

B0ijk = b u i  j  k

Cik = b w 00w ik ,
where the implicit imperfection factors are defined by the 4th order expansion of
the unbalanced work on the fundamental (i.e. k [ ] = ( p [ u ])  k ).
5. The equilibrium path is obtained by satisfying the projection of the
equilibrium equation (1) onto V. According to eqs. (7) and (8), we have
b m
1 m
( k ) k b
2 i Cik +
2 i , j =1
i j Aijk +
i =1

m m
1 1
( b ) B
( b ) i j B0ijk +
2
+ i 00 ik + (9)
2 i =1 2 i , j =1

1 m
+
6 i , j , k =1
i j h Bijhk + k [ ] = 0, k = 1m.
42 G. Garcea, A. Bilotta, A. Madeo, G. Zagari, R. Casciaro 5

Equation (9) corresponds to a highly nonlinear system in the m + 1 unknowns


i and can be solved using a standard path-following strategy. It provides the
initial post-buckling behavior of the structure, including modal interactions and
jumping-after-bifurcation phenomena.

2.2. IMPERFECTION SENSITIVITY ANALYSIS

When analyzing a structure, it is difficult to characterize its geometry and


loads exactly, for the presence of a random distribution of small external
imperfections. This circumstance, also if the general behavior of the structure is
preserved, changes some aspects of its response and often causes a reduction in the
carrying capacity.
In the proposed asymptotic algorithm the presence of small additional imper-
fections expressed by a load q q [ ] and/or an initial displacement q u affect Eq. (9)
only with the imperfection term k [ ] that becomes [31, 30, 6]

1 1
k = 2c u 2  k + 2 ( 3c )c u 3 k + k1 [ ] + kg [ ] , (10)
2 6
with

k1 [ ] = q [ ]  k , kg [ ] = c uu
 k . (11)

The aim of the imperfection sensitivity analysis is to link the presence of


geometrical and load imperfections to the reduction in the limit load. For structures
presenting coupled buckling even a small imperfection in loading or geometry can
mean a marked reduction in collapse load with respect to the bifurcation load [11,
13, 12]. So an effective safety analysis should include an investigation of all
possible imperfection shapes and sizes to recover, albeit in a statistical sense, the
worst case.
The asymptotic approach provides a powerful tool for performing this
extensive investigation. In fact, the analysis for a different imperfection only needs
to update the imperfection factors kg [ ] and kl [ ] through Eq. (11) and to solve
once more the nonlinear system (9). Even if this system, collecting all the nonlinear
parts of the original problem, proves to be highly nonlinear and some care has to be
taken in treating the occurrence of multiple singularities, its solution through a
path-following process is relatively simple because of the small number of
unknowns involved.
However, exhaustive results can only be obtained in a statistical context
linking the distribution probability of the imperfection to that of the load. An
effective imperfection sensitivity analysis can be performed by a Monte-Carlo
6 A numerical asymptotic formulation for post-buckling analysis in case of coupled instability 43

statistical technique, where both the magnitude and the form of the imperfections
are treated as random variables. The analysis is then performed by taking the
additional imperfection factors in the form

(
kl [ ] + kg [ ] = q [ ] c uu )
 k = k , (12)

and producing a random sequence of imperfection vectors = {1 , 2 ... m ,} ,


modeling possible small deviations in the loads and in the geometry of the
structure, and repeating a path-following solution of (9) for each of these. By a
statistical treatment of the obtained results we obtain the probability distribution
function for the limit load multiplier and all the other useful statistical information.
This solution process, can be considered as a standard approach for imperfection
sensitivity analysis [8]. The number of repetitions needed to obtain statistically
reliable results increases (quite) exponentially with the number of the buckling
modes and for large m can become very expensive. The imperfection sensitivity
analysis can however be performed in a simple and efficient way when a criterion
for defining the (few) significant imperfection forms is available [34].

2.3. ATTRACTIVE PATH THEORY

A large number of different imperfections (up to several thousands) has to be


considered to obtain statistically significant results, so, while the analysis for a
single imperfection can be considered an easy task, the entire solution process
performed proves to be computationally expensive, especially when a large number
of coupled buckling modes have to be considered. We can, however, noticeably
reduce the computational effort by exploiting information given by the knowledge
of the complete set of attractive radial paths
i = ti* , i = 1... m, t \ (13)
which are local minimizers for the cubic form
1 m m
b =
2 i , j , h =1
Aijhi*i* h* = min
N, * *
i i =1 (14)
(k* ) i =1

or for the quartic form


1 m m
b =
3 i , j =1
Bijhk i*i* h* k* = min
N,
k*

i =1
* *
i i =1 (15)

on the unit hypersphere.


Attractive paths theory [8, 25, 26, 29, 34] can actually provide a helpful tool
for driving the analysis and reducing its total cost. In fact, it suggests that each
imperfect path obtained from the solution of (9) will be attracted by one of the
44 G. Garcea, A. Bilotta, A. Madeo, G. Zagari, R. Casciaro 7

minimizing radial directions * (see Fig. 7 in the numerical results section). Then,
an evaluation for the limit load associated to the single imperfection vector can
be obtained by performing a series of different monomodal analyses, one for each
minimum radial path (13), and then taking the smallest value obtained for the limit
load within all directions. The single monomodal analysis is quite quick, so a large
number of different imperfections can be investigated rapidly with results, in terms
of limit load distribution, equivalent to that provided by a full analysis [8].
Furthermore, it is worth mentioning that, once the worst imperfection shapes
have already been obtained from an imperfection sensitivity analysis, a detailed
investigation can be performed through a specialized path-following analysis,
taking into account these imperfections [27, 22].

3. ACCURACY AND EFFICIENCY IN FINITE ELEMENT


IMPLEMENTATION OF THE ASYMPTOTIC ANALYSIS

In the following we present a numerical formulation of the method suitable


for implementation using finite elements and discuss some aspects that are crucial
to achieve accuracy and efficiency.

3.1. FE IMPLEMENTATION OF ASYMPTOTIC METHOD

Applying a FE interpolation u = Lu , L being the interpolation operator and


u the vector collecting the discrete displacement and stress parameters, the
asymptotic analysis requires the following steps:
i. The fundamental path is obtained introducing the linear extrapolation
u f [ ] = u0 + u , (15)

where u is the initial path tangent, solution of the linear vectorial equation
K 0 u = p , (17)
K0 being the stiffness matrix evaluated for u = u0 and p the unitary load vector,
defined by the energy equivalencies
uT K [u ] u = [u ] u u , uT p = p u.
The solution of linear system (17) requires a standard factorization of K0.
ii. A cluster of buckling loads i , i = 1...m, and associated buckling modes
i are obtained along u f [ ] exploiting the critical condition

K [ i ]i = 0, K [ ] = K [ u0 + u ]. (18)
8 A numerical asymptotic formulation for post-buckling analysis in case of coupled instability 45

This corresponds to a nonlinear eigenvalue problem which can be linearized


and solved using standard algorithms and exploiting matrix K 01 , already available
from the previous step, to perform the iterations [7].
iii. Letting c be an appropriate reference value for the cluster, e.g. the
smallest of i or their mean value, the asymptotic approximation for the required
path is defined by the expansion
1
u [ , k ] = ub +
m m
 +
i =0 i i
i jw ij , (19)
2 i , j =0

where the quadratic corrections w ij W are obtained by the linear orthogonal


equations
w T ( K c wij + pij ) = 0, w W (20)

with Kc = K [uf c] and vectors pij defined as a function of modes i and i = 0 m


obtained by an element-by-element assembling process using the energy equivalence
w T pi, j = c wi j . (21)

The solution of linear system (20) can be conveniently obtained, as described


in [7, 6], through a Modified Newton-like iteration scheme exploiting K 01 as
iteration matrix.
iv. The energy terms in (9) being scalar quantities are evaluated as a sum, at
the element level, of the integrals of known functions.
v. The equilibrium path is obtained by solving the algebraic nonlinear system
of m equations in the m + 1 variables 0 , 1 ... m defined in Eq. (9) using a path-
following algorithm. Because of the small dimensions of the system, this can be
obtained very quickly using standard or even specialized variants of the arc-length
scheme.
The actual implementation of the asymptotic approach as a computational
tool is therefore quite easy in practice and its total computational burden, which is
mainly involved in the factorization of matrix K0, remains of the order of that required
by a standard linearized stability analysis. It provides the initial post-buckling
behavior of the structure, including modal interactions and jumping-after-bifurcation
phenomena. Moreover, once the preprocessing phase of the analysis has been
performed (steps i to iv), the presence of small loading imperfections or geometrical
defects can be taken into account in the postprocessing phase (step v), by adding
some, easily computed, additional imperfection terms in the expression of k[],
with a negligible computational extra-cost, so allowing an inexpensive imperfection
sensitivity analysis [30, 15].
Imperfection sensitivity analysis can be simplified by evaluating all the
minimum directions of the cubic (14) and quartic forms (14) to obtain the worst
46 G. Garcea, A. Bilotta, A. Madeo, G. Zagari, R. Casciaro 9

imperfection directions as discussed in subsection 2.2.1. Today this is not a


completely solved problem and it can also be expensive. Nevertheless, relative
minimum solutions can be (quite easily) obtained by using the iterative scheme
suggested in [34]. Furthermore for the case of symmetric buckling, problem (15)
can be transformed into a non-convex Quadratic Problem subject to linear
constraints and solved using the strategy presented in [8].

3.2. ON THE ACCURACY OF THE ASYMPTOTIC FORMULATION

The method, as will be shown in the numerical results section, is potentially


capable of furnishing accurate results if a series of modelling and implementation
aspects are carefully tuned. In the following we quickly present some of the
sources of inaccuracy referring readers to the references for a deeper discussion.

3.2.1. Interpolation locking


In the asymptotic algorithm a locking phenomenon related to the dis-
cretization process can arise from the evaluation of the fourth-order terms

Bijhk = c i  j  h  k c (wij whk + wihw jk + wik w jh ),

that define the initial curvature of the post-buckling path. The coefficients Bijhk are
obtained as the difference between two quantities derived from the fourth and
second variations. In compatible formulations the single term of this difference is,
usually, very large while the difference is small. The discretization error on the
single term could in this case be greater than the small results in their difference.
Obviously, the numerical response given by the asymptotic algorithm in this case is
completely unreliable.
The size of the error produced by this locking pathology depends on the finite
element interpolation functions and decreases for an appropriate balancing of the
polynomial functions used to describe each displacement component. The phenomenon
is particular evident for beam and plate structures where the buckling modes  i usually
contain only flexural displacement components while wij only in plane or axial ones.
The locking is sanitized when a mixed finite element is used [31, 6].
Figure 1, which refers to a planar Euler rod case reports numerical results for
the post-buckling factor b = B1111 obtained for different values of the ratio EAL2/EJ
between the axial and the flexural stiffness, by using an element called HC [31]
that uses the same quadratic spline functions for both the transversal and the axial
components and standard beam elements (linear and cubic interpolation for the
axial and transversal displacements, respectively).
10 A numerical asymptotic formulation for post-buckling analysis in case of coupled instability 47

Note that, for EAL2/EJ = 1.2105, 20 HC elements are sufficient to contain the
error in b at under 1% while standard discretizations do not yield reliable results
even using a large number of elements. A mixed finite element completely sanitizes
this pathological phenomenon.

3.2.2. Extrapolation locking


Mixed or compatible formats, while completely equivalent in principle,
behave very differently when implemented in asymptotic but also in path-following
solution strategies. This is an important, even if frequently misunderstood, point in
practical computations which has been widely discussed in [21, 20, 15, 16, 6]. By
referring readers to these papers for more details, we only recall here that both
numerical strategies need function and its Hessian K [ u] to be appropriately
smooth in its controlling variables. In path-following analysis, this ensures a fast
convergence of the Newton iterative process; in asymptotic analysis, it implies that
the higher-order energy term neglected in the Taylor expansion be really irrelevant,
allowing an accurate recovery of the equilibrium path. We know that the
smoothness of a nonlinear function strictly depends on the choice of the set of its
control variables, that is on the format of its description, and can change noticeably
when referring to another, even corresponding, set. As a consequence, the mixed
and compatible format, even if referring to the same problem, can be characterized
by a different smoothness and so they behave differently in practice, when used
within a numerical solution process. Actually, the compatible format is particularly
sensitive to what we call extrapolation locking in [20, 15] which can produce a loss
in convergence when used in path-following analyses, or unacceptable errors in the
path recovery in the asymptotic case. These inconveniences are easily avoided by
changing to a mixed format.
2 3 2 5
HC (EAl /EJ = 1.2x10 ) HC (EAl /EJ = 1.2x10 )
standard standard
2

1000 8
6
4

100 8
6
4

10 8
6
4

18
6
4

2
exact value = 0.25
0.1
0 10 20 30
number of elements

Fig. 1 Locking in the Euler case [30].


48 G. Garcea, A. Bilotta, A. Madeo, G. Zagari, R. Casciaro 11

3.2.3. Objective structural model

The asymptotic analysis makes great use of information attained from a


fourth-order expansion of the strain energy and then requires a fourth-order
accuracy be guaranteed in the structural modeling. Small inaccuracies, deriving
from geometrical incoherencies in the higher-order terms of the expansion of the
kinematical laws or in its finite element representation, significantly affect the
accuracy of the solution and can make it unreliable. Structural models not affected
by rigid body motions or by changes in observer are then required. This aspect is
more important with respect to the path-following case where only the first
variation needs to be correctly evaluated. With this aim the Implicit Corotational
Method (ICM) [17, 18] has been proposed as a tool to obtain geometrically exact
nonlinear models for structural elements, such as beams or shells, undergoing finite
rotations and small strains starting from the solutions for the 3D Cauchy continuum
used in the corresponding linear modeling. The main idea is to associate a
corotational frame to each point of the 3D continuum so allowing the motion in the
neighbor of the point to be split in a pure stretch followed by a pure rotation,
according to the decomposition theorem. It is possible to show how, using the
small strain hypothesis and rotation algebra, the linear stress and linear strain fields,
when viewed in this corotational frame, can provide accurate approximations for the
Biot nonlinear stress and strain fields. Once the corotational rotation is appropriately
defined, the local statics and kinematics of the model are recovered from the linear
solution as a function of the stress/displacement resultants. Stress and strain fields
are then introduced within a mixed variational principle in order to obtain the
constitutive laws directly in terms of stress/strain resultants. This completes the
ICM definition of the nonlinear model.
The nonlinear model so obtained retains all the details of the 3D linear
solution, including torsion/shear warping, while its objectivity is ensured implicitly.
Furthermore, the use of the mixed formulation and the greater accuracy with which
the ICM recovers the stress field, allows an accurate description of the constitutive
laws in terms of resultants. ICM does not require any ad-hoc assumption about the
structural model at hand, nor depends on any particular parametrization of the
rotation tensor, but actually behaves as a black-box tool able to translate known
linear models into the corresponding nonlinear ones. Moreover, the direct use of a
mixed (stress/strain) description provides an automatic and implicitly coherent
methodology for generating models free of the nonlinear locking effects previously
discussed, in a format directly suitable for use in FEM implementations. The
method was applied in [17] to derive 3D beam and plate nonlinear models starting
from the Saint Vennt rod and Kirchhoff and Mindlin-Reissner plate linear theories,
respectively.
12 A numerical asymptotic formulation for post-buckling analysis in case of coupled instability 49

4. NUMERICAL RESULTS

Some results regarding the analysis of both 3D beams and plates are reported
and compared with particular reference to accuracy as previously discussed. In the
monomodal buckling tests, to compare the accuracy with known solutions, the
following quantities, defining the postcritical tangent and curvature to the bifurcated
path, have been introduced
1 A111 B1111 + 3b2 B0111 + 3b2 B0011
b = , b = .
2 A011 3 A011
The results are compared with known analytical solutions (see [6]) and with
the ones obtained using the LC (Complete Lagrangian) and LS (Simplified
Lagrangian) technical plate models (see [15, 22] for a discussion on these models)
already implemented in the code named KASP. An independent analysis has also
been made using the commercial code ABAQUS.

4.1. THE INFLUENCE OF THE STRUCTURAL MODEL


The test refers to the Euler beam shown in Fig. 2. The beam is analyzed
forcing the buckling to have in-plane or out-of-plane components only. Despite its
simplicity, when analyzed with an asymptotic approach, the problem is taxing with
regard to the accuracy of the structural model and its FEM discretization [18]. In
Fig. 2 the values of the buckling loads and post-critical curvatures are compared
with the values obtained by using the Antman beam model and exact interpolation
functions [6]. The ICM model recovers the analytical solution for sufficiently fine
grids exactly. The LC and LS models provide a correct answer for the buckling
load, but have a different post-buckling behavior in the in-plane or out-of-plane
analysis: LC agrees perfectly with the exact solution in the in-plane case, whereas
LS provides the wrong result  / B = 2, which is eight times greater; conversely,
LS behaves better in the out-of-plane case, by providing the approximation  / B = 0,
while LC gives a completely erroneous unstable postbuckling curvature  / B = 0.75.
The resulting paths in Fig. 3 show a good agreement with those computed by path-
following analyses.
Table 1
Out plane In plane
N. elm. LC LS ICM LC LS ICM 2D Beam(*)
16 9.901 9.901 9.901 9.918 9.918 9.918
b 32 9.877 9.877 9.877 9.870 9.870 9.870 9.870
64 9.872 9.872 9.871 9.867 9.870 9.870
16 -0.354 0.020 0.145 0.166 1.03 1.03
b 32 -0.375 0.000 0.125 0.126 1.00 1.00 9.870
2b 64 -0.375 0.000 0.125 0.125 1.00 1.00
50 G. Garcea, A. Bilotta, A. Madeo, G. Zagari, R. Casciaro 13

Fig. 2 Euler beam: problem description and buckling


and post-buckling parameters [31].

Fig. 3 Euler beam: out-plane and in-plane equilibrium path [31].

4.2. TEST WITH NONLINEAR PRECRITICAL BEHAVIOUR

The test in Fig. 4 is relative to a structure characterized by a highly nonlinear


pre-critical behavior. The first two buckling loads are equal to 1 = 4.52 and
2 = 7.11, while the limit load is almost an half of the minimum buckling value and
is evaluated exactly as can be observed by the comparison with the asymptotic and
path-following (ABAQUS) curve denoted respectively CR4 and SR8. It is worth of
noting that only the implicit imperfection acts on the structure.
14 A numerical asymptotic formulation for post-buckling analysis in case of coupled instability 51

2.4

1.8

1.2

0.6

S8R t=12.70
CR4 t=12.70
0
-0.075 -0.05 -0.025 0
va / L

Fig. 4 Hinged cylindrical shell [41].

4.3. MULTIMODAL BUCKLING AND ATTRACTIVE PATH

The first test is the thin-walled beam in Fig. 5 modeled as a plate assemblage.

Fig. 5 T beam: problem description, buckling modes and equilibrium paths [31, 6].

The model is that proposed in [17, 18] on the basis of the ICM and is denoted
as MP in the results. The results are compared with those of an ABAQUS analysis
using a path-following approach and of the technical plate models [15]. The greater
accuracy of the objective structural model is evident in Fig. 5 where the equilibrium
paths are depicted.
52 G. Garcea, A. Bilotta, A. Madeo, G. Zagari, R. Casciaro 15

1=169.99 2=244.57 3=318.40

4=322.23 11=526.70 12=574.92

Fig. 6 C-shaped cantilever beam [41].

Fig. 7 Geodetic dome: modal interaction between 20 critical modes [8].


16 A numerical asymptotic formulation for post-buckling analysis in case of coupled instability 53

The second case regards a C-shaped cantilever beam subjected to a single


force at the free end reported in Fig. 6. In this case the strong modal interaction
between non near buckling loads also produce a highly unstable behaviour as
shown by the equilibrium path.
Finally the last test regards the Geodetic Dome modeled through a 3D truss
as proposed in [8]. In this case many locally coincident buckling modes interact
and the structure exhibits a very marked unstable behavior. In Fig. 7, in the modal
space k different equilibrium paths, clearly converging along only one of the
minimum directions, are reported. The test shows how it is possible to perform the
sensitivity analysis in a simplified way along the predetermined quartic form
minimum directions.

5. CONCLUSION

The paper presented the numerical implementation of the Koiter asymptotic


approach to evaluate the buckling and postbuckling behaviour of geometrically
nonlinear structures in the case of multiple coincident buckling loads and random
external imperfections. Standard techniques, based on repeated path-following
analyses, are useful for a thorough investigation of the structural behaviour with a
single imperfection shape, but cannot be considered effective tools to predict the
safety factor for geometrically nonlinear problems. The asymptotic method, instead,
appears to be an attractive alternative as it allows a reliable analysis with computational
costs similar to those required by a standard load buckling prediction, while
subsequent analyses for different imperfections are inexpensive.
The method furnishes accurate results and also information about the worst
imperfection shape if a series of modeling and implementation aspects are carefully
tuned. In particular it was shown how the effects of the use of geometrically exact
structural models and their coherent finite element implementation are very
relevant, while a mixed formulation eliminates both interpolation and extrapolation
locking phenomena.

Received on July 16, 2014

REFERENCES

1. Arbocz, J., Hol, J. M. A. M., Koiters stability theory in a computer-aided engineering (CAE)
environment, Int. J. Solids Struct, 26, pp. 945973, 1990.
2. Barbero, E. J., Godoy, L. A., Raftoyiannis I. G., Finite elements for three-mode interaction in
buckling analysis, Int. J. Numer. Meth. Eng, 39, pp. 469488, 1996.
3. Boutyour, E. H., Zahrouni, H., Potier-Ferry, M., Boudi, M., Asymptotic-numerical method for
buckling analysis of shell structures with large rotations, Journal of Computational and
Applied Mathematics, 168, 12, pp. 7785, 2004.
54 G. Garcea, A. Bilotta, A. Madeo, G. Zagari, R. Casciaro 17

4. Brezzi, F., Cornalba, M., Di Carlo, A., How to get around a simple quadratic fold, Numer. Math,
48, pp. 417427, 1986.
5. Budiansky, B., Theory of buckling and postbuckling of elastic structures, Advances in Applied
Mechanics, 14, Academic Press, New York, 1974 .
6. Casciaro R., Computational asymptotic post-buckling analysis of slender elastic structures, CISM
Courses and Lectures No. 470, SpringerWien, New York, 2005.
7 Casciaro, R., Garcea, G., Attanasio, G., Giordano, F., Perturbation approach to elastic postbuckling
analysis, Comput. Struct, 66, 585595, 1998.
8. Casciaro R., Mancusi G., Imperfection sensitivity due to coupled local instability: a nonconvex QP
solution algorithm, Int. J. Numer. Meth. Eng, 67, pp. 815840, 2006.
9. Casciaro R., Salerno G., Lanzo A. D., Finite Element Asymptotic Analysis of Slender Elastic
Structures: a Simple Approach, Int. J. Numer. Meth. Eng, 35, pp. 13971426, 1992.
10. Chen, H., Virgin, L. N., Finite element analysis of post-buckling dynamics in plates, Part I: An
asymptotic approach, Int. J. Solids Struct, 43, pp. 39834007, 2006.
11. Dubina, D., The ECBL approach for interactive buckling of thin-walled steel members, Steel and
Composite Structures, 1, 1, pp. 7596, 2001.
12. Dubina, D., Ungureanu, V., Instability mode interaction: From Van Der Neut model to ECBL
approach, Thin-Walled Structures, Article in Press, 2013.
13. Dubina, D., Ungureanu, V., Effect of imperfections on numerical simulation of instability
behaviour of cold-formed steel members, Thin-Walled Structures, 40, 3, pp. 239262, 2002.
14. Flores, F. G., Godoy, L. A., Elastic postbuckling analysis via finite element and perturbation
techniques. Part 1: Formulation, Int. J. Numer. Meth. Eng., 33, pp. 17751794, 1992.
15. Garcea, G., Mixed formulation in Koiter analysis of thin-walled beam, Comput. Methods Appl.
Mech. Eng, 190, pp. 33693399, 2001.
16. Garcea, G., Formica, G., Casciaro, R., A numerical analysis of infinitesimal mechanisms, Int. J. Numer.
Meth. Eng., 62, pp. 9791012, 2005.
17. Garcea, G., Madeo, A., Casciaro, R., The implicit corotational method and its use in the
derivation of nonlinear structural models for beams and plates, J. Mech. Mater. Struct, 7, 6,
pp. 509538, 2012.
18. Garcea, G., Madeo, A., Casciaro, R., Nonlinear FEM analysis for beams and plate assemblages
based on the Implicit Corotational Method, J. Mech. Mater. Struct., 7,6, pp. 539574, 2012.
19. Garcea, G., Madeo, A., Zagari, G., Casciaro, R., Asymptotic post-buckling FEM analysis using a
corotational formulation, Int. J. Solids Struct., 46, 2, pp. 523532, 2009.
20. Garcea, G., Salerno, G., Casciaro, R., Extrapolation locking and its sanitization in Koiter
asymptotic analysis, Comput. Methods Appl. Mech. Eng., 180, 12, pp. 137167, 1999.
21. Garcea, G., Trunfio, G. A., Casciaro, R., Mixed formulation and locking in path-following
nonlinear analysis, Comput. Methods Appl. Mech. Eng., 165, 14, pp. 247272, 1998.
22. Garcea, G., G.A. Trunfio, Casciaro, R., Path-following analysis of thin-walled structures and
comparison with asymptotic post-critical solutions, ComputInt. J. Numer. Meth. Eng., 55,
pp. 73100, 2002.
23. Godoy, L. A., Banchio, E. G., Singular perturbations for sensitivity analysis in symmetric
bifurcation buckling, Int. J. Numer. Meth. Eng., 52, pp. 14651485, 2001.
24. Goltermann, P., Mllmann, H., Interactive buckling in thin-walled beams I. Theory, Int. J. Solids
Struct., 25, pp. 715749, 1989.
25. Ho, D., The influence of imperfections on systems with coincident buckling loads, Int. J. Non-Linear
Mech., 7, pp. 311321, 1972.
26. Ho, D., Buckling load of non-linear systems with multiple eigenvalues, Int. J. Solids Struct., 10,
pp. 13151330, 1974.
27. Liang, K., Abdalla, M., Gurdal, Z., A Koiter-Newton approach for nonlinear structural analysis,
Int. J. Numer. Meth. Eng., 96, 12, pp. 763786, 2013.
28. Koiter, W. T., On the stability of elastic equilibrium, Thesis, Delft, 1945. English transl. NASA
TT-F10, 883, 1967 and AFFDL-TR70-25, 1970.
18 A numerical asymptotic formulation for post-buckling analysis in case of coupled instability 55

29. Koiter, W. T., Some properties of (completely) symmetric multilinear forms with an application to
Hos theorem for multi-mode buckling, Technologic University of Delft. Lab. Rep., 587, 1976.
30. Lanzo, A. D., Garcea, G., Koiter analysis of thin-walled structures by a finite element approach,
Int. J. Numer. Meth. Eng., 39, pp. 30073031, 1996.
31. Lanzo, A. D., Garcea, G., Casciaro, R., Koiter post-buckling analysis of elastic plates, Int. J. Numer.
Meth. Eng., 38, pp. 23252345, 1995.
32. Peek, R., Kheyrkhahan, M., Postbuckling behavior and imperfection sensitivity of elastic
structures by the Lyapunov-Schmidt-Koiter approach, Comput. Methods Appl. Mech. Eng.,
108, 34, pp. 261279, 1993.
33. Poulsen, P.N., Damkilde, L., Direct determination of asymptotic structural postbuckling behavior
by the finite element method, Int. J. Numer. Meth. Eng., 42, 4, pp. 685702, 1998.
34. Salerno, G., Casciaro, R., Mode jumping and attrative paths in multimode elastic buckling, Int. J.
Numer. Meth. Eng., 40, pp. 833861, 1997.
35. Schafer, B.W., Graham-Brady, L., Sthocastic post-buckling of frames using Koiters methods, Int.
J. Struct. Stab. Dyn., 6, 3, pp. 333358, 2006.
36. Silvestre, N., Camotim, D., Asymptotic-Numerical Method to Analyze the Postbuckling Behavior,
imperfection-Sensitivity, and Mode Interaction in Frames, Journal of Enginering Mechanics,
131, 6, pp. 617632, 2005.
37. Sridharan, S., Doubly symmetric interactive buckling of plate structures, Journal J. Solids Struct.,
19, pp. 625641, 1983.
38. Van Erp, G. M., Menken, C. M., Initial post-buckling analysis with the spline finite-strip method,
Comput. Struct., 40, pp. 11931201, 1991.
39. Wu, B., Wang, Z., A perturbation method for the determination of the buckling strength of
imperfection-sensitive structures, Comput. Methods Appl. Mech. Eng., 145, 34, pp. 203215,
1997.
40. Barbero, E.J. Madeo, A., Zagari, G., Zinno, R., Zucco, G., Koiter asymptotic analysis of folded
laminated composite plates, Composites Part B: Eng., 61, pp. 267274, 2014.
41. Zagari, G., Madeo, A., Casciaro, R., de Miranda, S., Ubertini, F., Koiter analysis of folded structures
using a corotational approach, Int. J. of Solids and Structures, 50, 5, pp. 755765, 2013.
EFFECT OF DISTORTION ON THE STRUCTURAL BEHAVIOUR
OF THIN-WALLED STEEL REGULAR POLYGONAL TUBES

RODRIGO GONALVES1, DINAR CAMOTIM2

Abstract. This paper addresses the effect of cross-section distortion on the structural
behaviour of thin-walled tubes with single-cell regular convex polygonal cross-
sections (RCPS) and provides an in-depth view on the underlying mechanical aspects.
In particular, the first-order, buckling (bifurcation) under uniform compression and
undamped free vibration behaviours are characterised using the modal decomposition
features and computational efficiency of a GBT (Generalised Beam Theory) specialization
for RCPS recently developed by the authors [1]. Several analytical and illustrative
numerical results are presented and discussed within the paper.

Key words: thin-walled members, regular polygonal cross-sections, cross-section


distortion, Generalised Beam Theory (GBT), buckling behaviour, vibration behaviour.

1. INTRODUCTION

It is a well-known fact that cross-section distortion, which comprises both in-


plane and out-of-plane (warping) displacements, can influence significantly the
structural behaviour of open section thin-walled members. This phenomenon has
been investigated rather intensively in the recent past and it may be argued that it is
currently quite well understood, particularly for cold-formed steel lipped channel,
zed, hat or rack section members. Much less known are the effects of distortion
in thin-walled tubes with single-cell regular (equiangular and equilateral) convex
polygonal cross-sections (RCPS), which are widely employed in slender structures
such as lighting posts and telecommunication towers. The first papers on this
subject were published only in the last year [13], where it was shown that, besides
the well-known global and local (plate-type) deformation modes, the distortional
modes also play a key role in the structural response of RCPS tubes, even in the
case of geometrically linear problems.
This paper addresses specifically the influence of cross-section distortion on
the structural behaviour of RCPS tubes and aims at providing a general and broad
1
UNIC, Faculdade de Cincias e Tecnologia, Department of Civil Engineering, NOVA
University Lisbon, 2829-516 Caparica, Portugal
2
ICIST, Instituto Superior Tcnico, Department of Civil Engineering, Architecture and
Georesources, University of Lisbon, Av. Rovisco Pais, 1049-001 Lisbon, Portugal
2 Effect of distortion on the structural behaviour of thinwalled steel regular polygonal tubes 57

perspective on the subject. The fundamental findings of the previous work are
summarised and new results are reported, which further help characterising the
distortional modes and also their relevance for the adequate assessment of the
RCPS tube first-order, buckling and vibration behaviours. The Generalised Beam
Theory (GBT) specialisation for RCPS proposed in [1] is employed, leading to the
identification of a well-defined set of fully uncoupled cross-section distortional
deformation modes. This approach also makes it possible to derive analytical or
semi-analytical formulae that provide in-depth information concerning the structural
behaviour of RCPS tubes.
The outline of the paper is as follows. Section 2 presents a brief overview of
the GBT specialisation for RCPS and explores the features of the resulting orthogonal
distortional deformation modes. Each of the subsequent sections focuses on a
specific type of analysis, namely first-order (Section 3), linearised buckling under
uniform compression (Section 4) and undamped free vibration (Section 5).
Although GBT-based finite elements may be always employed to obtain numerical
results, namely for arbitrary loading and/or boundary conditions, attention is
devoted to analytical or semi-analytical solutions, which generally enable a better
grasp of the underlying mechanical aspects. The paper closes in Section 6, with
some concluding remarks.
The cross-section geometric parameters employed in this paper are indicated
in Fig. 1(a), together with the non-dimensional parameters and relations
L r b
1 = , 2 = , = 22 sin , (1)
r t t n
where L is the tube length. The material parameters are E (Youngs modulus),
G (shear modulus), (Poissons ratio) and (volumetric mass density).

Fig. 1 a) Geometry and notation for RCPS;


b) wall mid-surface local coordinate systems.
58 Rodrigo Gonalves, Dinar Camotim 3

2. GBT FOR REGULAR CONVEX POLYGONAL TUBES

Following the notation of [4] and using the wall mid-surface local coordinate
systems shown in Fig. 1(b), the GBT displacement field for each wall is given by
D
u ( x, y ) = u k ( y ) k , x ( x),
U x u ( x, y ) zw, x ( x, y ) k =1

U = v( x, y ) zw ( x, y ) ,
D

y ,y v( x, y ) = vk ( y ) k ( x), (2)
k =1
U z w( x, y ) D
w( x, y ) = wk ( y ) k ( x),
k =1

where: (i) the comma indicates a partial differentiation, (ii) u, v, w are the mid-
surface displacement components along x, y, z, respectively, (iii) the bars identify
the associated deformation mode components, (iv) D is the number of deformation
modes and (v) k are their amplitude functions along the length of the beam (the
problem unknowns).
The cross-section deformation modes (i.e., functions u k , vk , wk ) are obtained
from the GBT cross-section analysis, which consists essentially of defining an
initial set of modes and, then, sequentially solving a set of eigenvalue problems
that partially uncouple the differential equilibrium equation system. A hierarchic
set of deformation modes is then retrieved, which includes the classic prismatic
beam theory modes (axial extension, bending about principal axes and torsion
about the shear centre) and also the so-called local, distortional, shear and
transverse extension deformation modes. This work focuses on the natural
Vlasov warping modes, obtained under the assumption of null membrane (i) shear
strains (Vlasovs hypothesis) and (ii) transverse extensions (i.e., the walls are
deemed inextensible in the cross-section plane). Then, in each wall, the vk functions
are constant and the u k (warping) functions are linear. An initial base for these
modes is obtained by (i) imposing unit warping displacements at each wall junction
(cross-section natural node), (ii) calculating the vk functions that ensure null
membrane shear strains and, finally, (iii) obtaining the wk functions by analysing
the cross-section as a plane frame subjected to imposed vk displacements.
RCPS constitute a rather distinctive special case amongst cross-section
geometries. Indeed, these cross-sections exhibit rotational symmetry of order equal
to the number of walls and nodes (n), a feature that is at the root of some
remarkable peculiar features. In particular, the deformation mode configurations
4 Effect of distortion on the structural behaviour of thinwalled steel regular polygonal tubes 59

obtained by imposing a single unit nodal warping displacement constitute rotations


of each other. This implies that the resulting GBT modal matrices, given in Annex B,
are symmetric and circulant and, therefore, may be diagonalised by means of
QtAQ, where A is a GBT modal matrix and Q is a matrix whose columns are given
by vectors a(l) and b(l), whose components read [1]

2jl n
a (jl ) = cos , l = 0,..., ,
n 2 (3)
2jl n 1
b (j l ) = sin , l = 1,..., ,
n 2
where [k] designates the largest integer not exceeding k. Each vector corresponds to a
specific orthogonal warping function, where component j contains the warping at node j.
In particular:
(i) For l = 0, one has a (j0 ) = 1 , which corresponds to constant warping at the
cross-section, i.e., to the classic axial extension mode.
(ii) For each l = 1,, (n1)/2, two deformation modes are obtained, associated
with warping functions having l full cycles around the cross-section.
Bending about orthogonal axes corresponds to l =1, i.e., warping functions
with one full cycle and neutral axes rotated by /2. The subsequent defor-
mation mode pairs are termed distortional, since cross-section in-plane
displacements of the natural nodes are involved (besides warping). As
discussed in [1], the two diagonal components of QtAQ corresponding to
each l are identical. Furthermore, it can be shown that a combination of
the mode pair of the form
a (jl ) cos + b (j l ) sin , (4)

where is the rotation angle in the two-dimensional mode space, does not
change the matrix diagonal component. The fact that the stiffness properties
of the deformation mode pairs (the GBT matrix components) are invariant
with respect to constitutes an important generalisation of the bending
behaviour of RCPS tubes, where all central axes are principal bending
axes bending may now be viewed as just the particular case of l = 1.
Finally note that, in order to obtain bending deformation modes associated
with unit curvatures, the corresponding warping functions must be
multiplied by r (this was not done in the present paper).
(iii) For l = n/2, which only applies to cross-sections with even n, a single
distortional mode is obtained, which exhibits alternating positive and
negative warping displacements at consecutive nodes.
60 Rodrigo Gonalves, Dinar Camotim 5

Since the orthogonal RCPS warping functions are already known, it is a quite
straightforward task to calculate the complete deformation mode shapes and the
corresponding GBT matrix diagonal components Annex A provides a set of
expressions that can be employed for this purpose. In [1], some analytical formulae
are given and all matrix components are shown in a graphical format, normalised
with respect to the values for circular tubes.
For illustrative purposes, Table 1 provides the distortional mode matrix
diagonal components, in a non-dimensional form, for n = 4-8, 2=100 and = 0.3
(the parameters and appearing in the table are discussed in the next section).
Furthermore, Fig. 2 shows the shapes of the natural Vlasov warping modes and the
associated warping functions, as well as their sinusoidal counterparts, for n = 6.
The analytical expressions providing the GBT matrix components are also shown
in the figure they were calculated using the expressions given in Annex B. As
previously explained, this cross-section has 6 warping modes: axial extension (l = 0),
the bending mode pair (l = 1), one distortional mode pair (l = 2) and a single
distortional mode (l = 3).
Fig. 3(a) makes it possible to visualise the shapes of the distortional
deformation modes for cross-sections with n = 48. Note that, as already
mentioned, the distortional modes appear in pairs, with the exception of the l = n/2
mode for even n. Finally, Fig. 3b shows the effect of the rotation on the
deformation mode shape, for the particular case of the l = 3 distortional pair of a
cross-section with n = 20. The deformation modes obtained by means of Eqs. (3)
correspond to = 0 and = 90.
Table 1
Distortional mode matrix diagonal components
and exponential solution parameters

n l C/Er2 (103) Br2/E (106) D/E (106) r (103) r (103)


4 2 18.86 12.43 4.97 113.6 113.0
5 2 11.67 18.59 7.57 141.8 140.7
2 15.00 26.37 12.09 145.5 144.1
6
3 20.00 316.48 63.30 252.4 249.2
2 18.00 30.78 15.02 144.5 143.1
7
3 11.13 390.85 72.40 308.7 303.4
2 20.41 33.47 16.90 143.0 141.6
8 3 13.20 603.60 118.95 330.4 323.5
4 20.42 3121.6 365.71 447.2 437.0
6 Effect of distortion on the structural behaviour of thinwalled steel regular polygonal tubes 61

Fig. 2 Hexagonal cross-section: shapes of the natural Vlasov warping modes and associated matrix
components (the mode shapes are depicted assuming a linear amplitude function k(x) = x).

Fig. 3 a) Distortional deformation modes for n = 48; b) distortional deformation mode


configurations for n = 20 and l = 3, as a function of the rotation .
62 Rodrigo Gonalves, Dinar Camotim 7

3. FIRST-ORDER BEHAVIOUR

This section addresses the influence of the distortional modes on the first-
order (linear) behaviour of RCPS tubes. First, consider the homogeneous form of
the GBT equilibrium differential equation system for RCPS, which is uncoupled
and reads, for mode k,
Ckk k , xxxx Dkk k , xx + Bkk k = 0, (5)
where Dkk and Bkk are null for the axial extension and bending deformation modes.
In long tubes, the general solution for the distortional modes is given by [5]
k = e x ( A1 sin x + A2 cos x ),
Bkk D Bkk D (6)
= + kk , = kk ,
4Ckk 4Ckk 4Ckk 4C kk

where is the exponential decay and (herein assumed to be real) is the frequency
of the sinusoid. The factors / and / provide a measure of the influence length
of the deformation mode note that, at x = /, the exponential term is 4.32% and
/ corresponds to the half-wavelength. For the cross-sections indicated in Table 1,
the influence length varies between about 30r and 7r, decreasing as l increases i.e.,
the higher order deformation modes have a smaller influence length.
If at x = 0 (i) a diaphragm, restraining only the displacements along z, is
introduced and (ii) a distortional-like stress distribution is applied, i.e.,
D
xx ( y ) = k uk ( y ) , (7)
k =4

where k is the stress amplitude of mode k (with k 4, i.e., l 2), one obtains, for
each deformation mode, the constants A1 = k/2E and A2= 0. For illustrative
purposes, Fig. 4(a) plots the function e x sin x for the distortional modes listed in
Table 1, with respect to the normalised coordinate x/r it is clearly shown that the
influence length decreases with l (as already concluded) and also with n.
Consider now simply supported beams of length L and acted by sinusoidal
lateral loads. In this case, the analytical solution is given by [1]
x qk
k = k sin , k = 4 , qk = q y vk + qz wk , (8)
L 2
C kk + Dkk + Bkk
L4 L2
where q y , q z are the components of the distributed load along the local axes and
vk , wk are the modal displacement components at the cross-section point of load
application.
8 Effect of distortion on the structural behaviour of thinwalled steel regular polygonal tubes 63

Fig. 4 a) Exponential and b) sinusoidal solutions for the distortional modes


indicated in Table 1 (2 = 100, = 0.3).

This solution shows that, as L increases, the amplitude k increases and tends
asymptotically to qk / Bkk . For the bending modes, one has Bkk= 0 and, therefore,
the amplitude grows unboundedly with L.
The amplitude k may be written in a non-dimensional format, using the
solution corresponding to an infinite beam span, which leads to
k 1
= 4 . (9)
k Ckk 2
+ D + 1
14 Bkk r 4 12 Bkk r 2
kk

Figure 4(b) plots the results obtained with the above expression, for the values
shown in Table 1. Note that the solution for L = is approached more rapidly as
n and l increase.
It is worth mentioning that a set of numerical examples involving distortional,
torsional, bending and local deformation modes has been presented and discussed
in [1]. In all of those examples, the GBT-based solutions were compared with the
results provided by shell finite element models and an excellent agreement was
invariably found.
64 Rodrigo Gonalves, Dinar Camotim 9

4. BUCKLING BEHAVIOUR

This section presents the fundamental aspects of the distortional buckling


(bifurcation) behaviour of RCPS tubes subjected to uniform compression. A more
general discussion of the buckling behaviour of RCPS, including multi-mode
interactions and the effects of bending or torsion, can be found in [2, 3].
For the classic benchmark problem of simply supported members under uniform
compression, sinusoidal amplitude functions correspond to the exact solution and,
therefore, one has, for each individual deformation mode k, (negative means
compression)
1 a 2 2 C kk 12 2
( b ) k = 2 + Dkk + r Bkk ,
1 r a 2 2
2
X kk
(10)
Dkk + 2 C kk Bkk 1 4C
( cr ) k = , = 4 4 kk ,
X kk a cr r Bkk
where: (i) b is the bifurcation stress, (ii) cr is the critical stress (the lowest b),
(iii) a is the buckling mode longitudinal half-wave number, (iv) (1/a)cr is the
buckling mode normalised half-wavelength corresponding to cr and (v) Xkk are the
components of the GBT geometric modal matrix for uniform compression (see
Annex B).
It can be shown that the minimum distortional critical buckling stress always
corresponds to l = 2, i.e., to the first distortional deformation mode pair (or to the
single distortional mode if n = 4). Moreover, the critical stresses associated with
each l decrease as n increases and approach asymptotically the values cor-
responding to circular tubes for instance, for l = 2 and 2 = 75, the difference
between the critical buckling stresses concerning columns exhibiting RCPS and
circular hollow sections is below 4% for n 9.
For the cross-sections listed in Table 1 (2 = 100, = 0.3), the (1/a)cr values
vary between 19.6 (n = 4 and l = 2) and 5 (n = 8 and l = 4) and, essentially,
decrease when l and n increase. Note that these half wave-lengths are significantly
higher than those associated with local buckling, which are of the order of the wall
width b, i.e., (1/a)cr equal to (i) 1.4 for n = 4, (ii) 1.0 for n = 6 and (iii) 0.62 for n = 8.
The graph in Fig. 5(a) shows the parameter range values for which the critical
buckling stress is either distortional or local [2]. These values were determined for
= 0.3 and assuming that no mode interaction occurs. It is concluded that
distortional buckling is critical for the lower 2 values but, as n increases, the
critical buckling mode transition occurs for increasingly higher 2 values.
As show in [2], the true distortional critical modes involve participations of
local and shear deformation modes. The graph in Fig. 5(b) shows the variation of the
distortional critical buckling stresses with n, determined by means of GBT analyses
that include the following deformation mode sets: (i) Vlasov (distortional), (ii) Vlasov +
shear and (iii) local + Vlasov + shear. It is observed that the shear modes play a
10 Effect of distortion on the structural behaviour of thinwalled steel regular polygonal tubes 65

significant role, particularly for low 2 values. The local modes are also relevant,
but only for low n.
Finally, Fig. 5(c) presents an illustrative numerical example: it plots the variation
of the buckling stresses with 1/a, for the particular case defined by n = 10,
r = 100 mm, t = 4 mm (2 = 25) and =0.3, which corresponds to almost coincident
local and distortional

Fig. 5 Buckling of simply supported RCPS tubes under uniform compression ( = 0.3):
a) parameter ranges corresponding to critical local and distortional buckling;
b) influence of the local and shear modes on distortional buckling; c) illustrative example.
66 Rodrigo Gonalves, Dinar Camotim 11

critical buckling stresses. This figure includes single mode curves and also
curves obtained from analyses with various deformation mode sets, making it
possible to conclude that the buckling mode nature changes with 1/a in the
following manner:
(i) For 1/a < 1.6, local buckling governs (curve L in Fig. 5(c)). However,
the distortional and shear deformation modes participate in the ascending
branch of the curve, near the buckling mode transition zone.
(ii) For 1.6 < 1/a < 4.0, the second distortional mode pair (l = 3 curve D2)
is critical, with significant participations from both the shear and local
deformation modes.
(iii) For 4.0 < 1/a < 13, the first distortional mode pair (l = 2 curve D1)
governs and, as already concluded from Fig. 5(b), the shear modes have a
significant influence, particularly in the descending branch.
(iv) For 1/a > 13, the buckling mode involves global bending (curve B)
and the shear modes have a small participation up to 1/a = 20.

5. VIBRATION BEHAVIOUR

Although GBT is mostly employed for buckling analyses, formulations have


also been developed to analyse the vibration behaviour of thin-walled members
(see, e.g., [68]). In this section, GBT analyses are used to investigate the
undamped free vibration behaviour of RCPS tubes. As in the previous section, only
simply supported members are dealt with, which means that sinusoidal amplitude
functions constitute exact solutions and, therefore, the single mode solution is
given by
a 2 2 C kk 2
+ Dkk + 2 1 2 r 2 Bkk
r2 2
a
2k = 1 ,
1 2
2
(11)
Qkk + 2 2 r Rkk
a
(k ) min = Bkk / Rkk ,

where (i) is the natural angular frequency, (ii) a is the vibration mode
longitudinal half-wave number and (iii) the expressions for the mass matrices Q
and R are given in Annex B. These formulae show that, since Bkk and Dkk are non-
null for each distortional mode, the corresponding frequency always decreases with
1/a and the minimum is attained at 1/a = . Moreover, as in the case of buckling
under uniform compression, it can be shown that the minimum distortional frequency
always corresponds to l = 2 and, as n increases, it approaches the solution for
circular tubes.
12 Effect of distortion on the structural behaviour of thinwalled steel regular polygonal tubes 67

Figure 6 shows the variation of the fundamental frequency with 1/a, for the
particular cases of n = 6, 2 = 25, 100 and = 0.3. The frequencies are normalised
with respect to the fundamental frequency of a simply supported rectangular plate
of infinite length, width b and thickness t, which is given by

Df
f0 = , (12)
2b 2 t

where Df = Et3/12(12) is the plate bending stiffness and is the volumetric mass
density. The plots show single mode solutions and results obtained by means of
analyses including all the deformation modes, namely the shear and local modes.
The observation of these results prompts the following remarks:
(i) The fundamental vibration mode nature changes with 1/a as follows:
(i1) local for low 1/a (curves L), (i2) first distortional mode pair (l = 2
curves D1) for intermediate 1/a and (i3) global bending for high 1/a
(curves B).
(ii) With the exception of the transition zone between the local and distortional
vibration modes, the individual mode solutions provide accurate solutions.
(iii) Although this is not shown in Fig. 6, in the local-distortional transition
zone the vibration mode is affected by local-distortional interaction and
also by the shear modes. The maximum influence of the shear modes is
about 7% and occurs for 2 = 25 and 1/a = 4, which corresponds to the
initial stage of the transition zone. For 2 = 100, the influence of the shear
modes drops to a maximum of only 1.7%.

Fig. 6 Normalised fundamental frequencies for simply supported RCPS tubes


(n = 6, 2 = 25, 100, = 0.3)
68 Rodrigo Gonalves, Dinar Camotim 13

6. CONCLUDING REMARKS

This paper addressed the effect of cross-section distortion on the: (i) first-
order, (ii) buckling (bifurcation) under uniform compression and (iii) undamped
free vibration behaviour of thin-walled RCPS tubes. In particular, attention is
called to the following findings of the work carried out:
(i) Duplicate solutions are obtained for cross-sections with more than four
walls, due to the cross-section rotational symmetry. This implies that the
stiffness properties associated with the duplicate deformation modes are
invariant upon a rotation in the two-dimensional mode space.
(ii) The GBT specialisation for RCPS makes it possible to identify a set of
uncoupled and hierarchic distortional deformation modes. The exponential
solutions show that the higher-order distortional modes decay more rapidly
and, therefore, have a smaller influence length.
(iii) The distortional deformation modes play a significant role in the
buckling and vibration behaviour of RCPS tubes. It was shown that, for
some parameter ranges, the distortional modes correspond to the critical
modes/fundamental frequencies.
(iv) It was also shown that local/distortional/shear interaction is relevant,
particularly in mode transition zones.

ANNEX A: CROSS-SECTION ANALYSIS FOR RCPS

In a circulant matrix, all rows correspond to cyclic permutations of a single


row vector, with an offset equal to the row index. The generating row vector is
herein defined as the first row of the matrix. For instance, for a given 44 matrix
A, one has
A = ( A11 , A12 , A13 , A14 ) (A.1)

and, if the matrix is symmetric, one has A12 = A14. Due to the rotational symmetry
of RCPS, the GBT cross-section analysis involves several circulant matrices, which
lead to rather simple expressions.
Fig. A1 shows, for the particular case of n = 5, (a) the node/wall numbering
and local axes, and (b) the statically determinate system adopted to perform the
usual GBT cross-section analysis and the convention for positive nodal moments.
Let (Ux)ij, Vij, ij, Mij be nn matrices, whose components contain, respectively,
(i) the u nodal displacements, (ii) the wall v displacements, (iii) the converging
wall relative rotations and (iv) the nodal moments. In these matrices, j identifies the
14 Effect of distortion on the structural behaviour of thinwalled steel regular polygonal tubes 69

deformation mode and i concerns the node/wall number. As discussed in Section 2,


the orthogonal warping deformation modes are already given by Ux = Q and, from
the null membrane shear strain assumption, the v displacements are obtained as
1~ ~
V = VQ, V = (1, 1, 0, ..., 0), (A.2)
b
~
where the auxiliary matrix V is circulant. The complete cross-section in-plane
configuration is determined by imposing v displacements in the static system,
using the flexibility/force method, which is convenient in this case (the degree of
static indeterminacy is lower than the degree of kinematic indeterminacy). Simple
geometric considerations make it possible to conclude that the imposed relative
rotations in the pin-jointed frame are given by
1 ~ ~
= V , = ( 2 cos 1, 1, 0, ..., 0, 1, 2 cos + 1), (A.3)
b sin
where = 2/n is the angle between two adjacent walls.

Fig. A1 GBT cross-section analysis for n = 5: a) node/wall numbering and local axes; b) statically
determinate pin-jointed frame and convention for positive nodal moments

The matrix equation to be solved is then


b ~
FM = , (A.4)
6D f
~
where the normalised flexibility matrix F is symmetric and circulant, with
~
F = (4, 1, 0, ..., 0, 1). (A.5)
In order to solve the system (A.4), one may diagonalize the symmetric circulant matrix
through QtFQ=diag( Fi ), where Fi is the ith diagonal component (instead of directly
~
inverting F ). This leads to
6D f 1 t ~~
M= Q diag Q VQ.
(A.6)
b sin
3
Fi
70 Rodrigo Gonalves, Dinar Camotim 15

The w displacements associated with each deformation mode are obtained


from the integration of the nodal moments of the corresponding column of M.
In [1], analytical formulae for matrix CM are provided and all matrix
components are shown in a graphical format. Using the former expressions, it is
now possible to obtain the (diagonal) components of matrix B from the rather
simple formula

(M + M (2j +1) i + M ji M ( j +1) i ),


b n
BiiB = 2
ji (A.7)
3D f j =1

where j + 1=1 if j = n.

ANNEX B: GBT MODAL MATRICES

Assuming null membrane shear strains and null membrane transverse


extensions, the classic GBT modal stiffness matrices read
Bij = BijB = D f wi , yy w j , yy dy,
S

C ij = C M
ij + C ijB = (Et u i u j + D f wi w j ) dy,
S

D = D1 D 2 D , t
2 (B.1)
3
(D1 )ij = (D1 )ijB = Gt wi , y w j , y dy,
3 S

(D2 )ij = (D2 )ijB = D f wi , yy w j dy,


S
where: (i) M and B designate membrane and bending terms, (ii) i, j = 1, ..., D, (iii) S
is the cross-section mid-line, (iv) t is the wall thickness, (v) E and G are Young's
and shear moduli and (vi) is Poissons ratio. For buckling analyses of uniformly
compressed members, the geometric matrix reads
X ij = t (vi v j + wi w j ) dy. (B.2)
S

Finally, for vibration analyses, the mass matrices Q and R, which involve
translational and rotational terms, are given by
t3
Qij = t u i u j + wi w j dy,
S 12
(B.3)

Rij = t (vi v j + wi w j ) + wi , y w j , y dy ,
t3
S 3
where is the volumetric mass density.

Received on: July 16, 2014


16 Effect of distortion on the structural behaviour of thinwalled steel regular polygonal tubes 71

REFERENCES

1. GONALVES, R., CAMOTIM, D., On the behaviour of thin-walled steel regular polygonal tubular
members, Thin-Walled Structures, 62, pp. 191205, 2013.
2. GONALVES, R., CAMOTIM, D., Elastic buckling of uniformly compressed thin-walled regular
polygonal tubes, Thin-Walled Structures, 71, pp. 3545, 2013.
3. GONALVES, R., CAMOTIM, D., Buckling behaviour of thin-walled regular polygonal tubes
subjected to bending or torsion, Thin-Walled Structures, 73, pp. 18597, 2013.
4. GONALVES, R., RITTO-CORRA, M., CAMOTIM, D., A new approach to the calculation of
cross-section deformation modes in the framework of Generalized Beam Theory, Computational
Mechanics, 46, 5, pp. 75981, 2010.
5. SCHARDT, R., Verallgemeinerte Technische Biegetheorie, Springer-Verlag, Berlin, 1989.
6. SCHARDT, R., HEINZ, D., Vibrations of thin-walled prismatic structures under simultaneous
static load using Generalized Beam Theory, in: Structural Dynamics, eds. W. Kratzig et al.,
Balkema, Rotterdam, 1991, pp. 921927.
7. SILVESTRE, N., CAMOTIM, D., GBT-based local and global vibration analysis of loaded
composite open-section thin-walled members, International Journal of Structural Stability and
Dynamics, 6, 1, pp. 129, 2006.
8. BEBIANO, R., SILVESTRE, N., CAMOTIM, D., Local and global vibration of thin-walled
members subjected to compression and non-uniform bending, Journal of Sound and
Vibration, 315, 3, pp. 509535, 2008.
AXIAL IMPACT OF OPEN-SECTION TWCF COLUMNS
EXPERIMENTAL STUDY

MARIA KOTEKO1, ARTUR MODAWA1*, MARCIN JANKOWSKI2**

Abstract. The paper is devoted to the results of experimental study into the crushing
behaviour of TWCF open section columns subjected to axial impact. Steel channel
and top-hat section was under investigation. The paper contains a results summary of
quasi-dynamic and dynamic impact tests performed on about 100 of those sections of
different dimensions subjected to axial load of different velocities. Experimental
quasi-dynamic tests were conducted on the testing machine with different loading
velocities up to 600 mm/min. The impact tests were performed on the drop hammer
rig with the impact energy up to 5 kJ and impact velocity up to 10 m/s. An influence
of the column initial length and impact velocity on the crushing behaviour (failure
mode) was investigated. Particularly, the critical length of the transition from
progressive buckling to global bending failure mode and its dependence on section
dimensions and the impact velocity was under investigation. Experimental results
were compared with the results of the analytical calculations of critical (transition)
buckling length based on the simplified analytical theoretical models. The results are
presented in load-time and load-shortening diagrams and failure patterns. Some
conclusions concerning the determination of critical buckling length, applicability of
the theoretical models applied and an influence of the impact velocity upon the critical
buckling length and final mode of failure related to the energy absorption capability of
columns are derived.
Key words: thin-walled columns, open sections, impact, experiment

1. INTRODUCTION

Load-carrying capacity and energy absorption of TWCF subjected to


dynamic loading is an important problem in two areas of engineering applications:
first energy absorption of members acting as energy absorbers during accidental
collision of automotive or rail vehicles, secondly load-carrying capacity of
structural members subjected to seismic loads. The term dynamic crushing is
used in the present paper in the sense of progressive crushing of a structure
subjected to the impulse of an applied load.
An effective design of energy absorbing TWCF structural member demands
to determine a load-deformation relation of the member under impact load. In the

1
Ld University of Technology, Department of Strength of Materials, 90-924 d,
Stefanowskiego 1/15, Poland
2
Design & Research Center OKB, 95-006 Brjce, Bukowiec, Rokiciska 108/110, Poland
2 Axial impact of open-section TWCF columns 73

case of TWCF members subjected to axial compression, a substantial issue is such


a design, which promotes a progressive buckling mechanism [1]. However,
relatively long thin-walled columns (particularly of open sections) can exhibit
additional buckling modes, similar to the Euler buckling, leading to poor energy
absorption. Thus, a threshold conditions between progressive buckling and global
bending are of crucial importance.
Dynamic buckling response of thin-walled structural members subjected to
impact, axial compressive load has been investigated extensively by many
researchers. Particularly, closed-section members (tubes) have been investigated
since early 60ties of XXth Century. One should mention here Alexander solution for
a circular tube and Wierzbicki and Abramowicz works, concerning tubes of
prismatic cross-section, quoted by Jones [1]. Those solutions were based on the
rigid-perfectly plastic material model and used plastic mechanism approach. They
took into account an influence of the strain rate upon the initial material yield
stress, but neglected inertia effects. In the contrary to closed-section members, few
works have been devoted to the same problem concerning open-section columns.
Among them, Koteko et al. [2, 3] published the solution for top-hat section
column based on the yield line approach [4], taking into account the strain rate
effect. The solution was validated by quasi-dynamic tests, which will be discussed
in the next paragraph. Results of experimental investigation of top hat section
columns, as well closed section tubes subjected to axial impact were published by
Langseth et al. [5, 6] and Hsu& Jones [7]. Langseth [6] investigated inertia effect
and an influence of impact velocity upon the crushing behaviour of aluminium
tubular members.
In recent years some research have been carried out into the transition from
progressive local buckling to global bending failure mode in thin-walled columns
(bars) under axial impact. Such a transition is observed in the stage of collapse of
relatively long bars and at impact velocities higher than a certain critical one.
Results, concerning impact behaviour of long, circular and prismatic steel
tubes, subjected to dynamic, axial compressive load were bublished by Alves and
Karagiosova [8]. They investigated an influence of impact velocity on the crushing
behaviour of tubes. The problem of transition from progressive buckling to global
bending of circular tubes subjected to axial impact compressive load was also
investigated by Alves and Karagiosova [9, 10], as well as Jensen et al. [11]. Alves
and Karagiosova [10] proposed an analytical model, taking into account inertia
effects, which will be discussed in details in 3 paragraph.
Teramoto and Alves [12] conducted research into the same problem for open,
channel-section columns subjected to axial impact compressive force. They published
results of both experimental tests and FE numerical simulations. They did not find
clear condition of the transition from progressive local buckling to global bending.
Determination of limit or critical values of bars length and impact
velocity is still an open question. Since a transition from local progressive buckling
74 Maria Koteko, Artur Modawa, Marcin Jankowski 3

to global bending leads to a significant decrease of energy absorption capability,


the determination of a threshold criterion between those two failure modes is of
crucial importance.

2. QUASI- DYNAMIC BEHAVIOUR

Preliminary static and quasi-dynamic tests have been carried out on relatively
short top-hat section and plain channel section columns subject to uniform
compression. These tests were conducted in order to plan a program for impact
tests and to identify crushing behaviour of open-section columns at relatively low
impact velocities. Results of those research was reported by Koteo&Mania [3].
Specimens were made of steel sheets and folded in special folding machine.
Dimensions of specimens and material parameters are shown in Table 1. The
experiment was conducted on the testing machine Instron of loading range 20 kN.
Both compressive force and column deformation (displacement of the upper
crosshead beam of the testing machine) was recorded using the integrated,
computer aided measurement system of the testing machine.
Table 1
Section Material parameters Dimensions
Top hat Y = 165 MPa
a = 30 mm
ult = 280 MPa
b = 30 mm
E = 192000 MPa
b w = 8 mm
t Et = 2000 MPa
wall thickness t = 0.6 mm
q=5
w a column length l = 150 mm
D = 40.4 [1/s]
Plain channel Y= 286 MPa
a = 75 mm
ult = 315 MPa
b = 37.5 mm
b E = 194000 MPa
t wall thickness t = 1 mm
Et = 3000 MPa
column length l = 375 mm
a q, D as above

Notation:
E Young modulus
Et tangent modulus
0 Yield stress
A area of the cross-section
I second moment of area of the cross-section
density of the rod material
m mass per unit length
L rod length
G impactor weight
4 Axial impact of open-section TWCF columns 75

Figure 1 shows failure modes observed in static and quasi-dynamic tests. Failure
modes of top hat sections are significantly different under static and quasi-
dynamic load, while failure modes of channel sections did not differ.

a) b)

c) d)

Fig. 1 Exemplary failure patterns of tested columns: a) top hat static;


b) top hat v0 = 600 mm/min; c) plain channel v0 = 300 mm/min; d) plain channel v0 = 600 mm/min.

18
vo=
16
5 mm/min
14 300 mm/min
compression force [kN]

400 mm/min
12 600 mm/min

10
8
6
4
2
0
0 2 4 6 8
shortening [mm]
Fig. 2 Experimental load-shortening diagrams of plain channel sections (quasi-dynamic test).
76 Maria Koteko, Artur Modawa, Marcin Jankowski 5

Experimental quasi-static results show a minor increase of ultimate load due


to the increase of loading velocity. Also the failure curves are not significantly
influenced by the latter, as shown in Fig. 2 (load-shortening diagrams). That indicates a
necessity of actual dynamic crushing tests at much higher impact velocities.

3. ANALYTICAL MODELS

Experiments carried out on axially impacted circular tubes by Alves and


Karagiosova [9] and channel section columns by Teramoto [12] show, that the
critical buckling length depends on initial impact velocity and material parameters.
Research into analytical solutions of the problem was carried out by Alves and
Karagiosova [10], who proposed three theoretical models of dynamic buckling of
circular tubes subjected to axial compressive force. There are three different
theoretical rod models proposed, namely; the Euler buckling model (Fig. 3a), the
three-link model (Fig. 3b) and the two phase deformation model taking into
account inertia effects and allowing one to analyze an influence of initial velocity
on the critical buckling length (Fig. 3c). Generally, it is a model of incompressible
rod with uniformly distributed mass m = M/L, subjected to the constant axial fully
plastic force P = A0. Since numerous experimental results show, that plastic
deformation occurs first near the column ends, in the rod models these deformations
are taken into account as elasto-plastic bending moments. The models have been
derived for the circular tube. Two first models were adopted by Koteko and
Mania [13] for the case of an arbitrary open section column, who presented results
of preliminary impact tests of open section members. The last model is also
adopted in this work by the authors for an arbitrary case.
a) b) c)
P
c1

l/2

B

l/2

c1 C

Fig 3 Collapse modes of rod models: a) Euler buckling mode;


b) three-link buckling mode; c) two phase deformation model.
6 Axial impact of open-section TWCF columns 77

In the first case (Fig. 1a) of Euler buckling mode, the equation of motion of
each part of the rod (AB and BC), for an arbitrary cross-section, takes the
following form:
12 6E I

 r2 . = 0,
2 0
(1)
l Al

4 E Et
where: E r = reduced modulus for bi-linear material [14].
( E + Et ) 2

The critical buckling length LcrE (Table 2) is an eigenvalue of the equation (1).

Table 2
No Critical buckling length
Euler dynamic buckling mode:
1 6 Er I
LcrE =
A 0
Three-link first dynamic buckling mode:
2 Er I
L1 cr = 3
A 0
Three-link dynamic second buckling mode:
3 15 E r I
L2 cr =
A 0

In the second case (Fig. 1b) one can derive four equations of motion (two for
each buckling mode) with unknowns 1 and 2 . If the solution of this simultaneous
equation is predicted in the following way

1 = D1 sin 2 t + ( ) (
2 = D2 sin 2 t + ) (2)

the roots of characteristic equation for non-trivial solution are as follows


54 5.54
12 = 3
( 6c PL ) ; 22 = (10c PL ) , (3)
15mL 15mL3
6 Er I
where c = ; = r / E . The roots of (3) determine two critical
(1 + )
buckling lengths for dynamic buckling L1cr and L2cr, corresponding to the modes
shown in Fig. 1b, given in Table 2.
78 Maria Koteko, Artur Modawa, Marcin Jankowski 7

Fig. 4 Time velocity diagram.

The third model takes into account inertia effects. The flexure equation of
motion takes form:
54 w
 =
w
mL 2
(L A 0 10c ) L
(4)

with the solution of a hyperbolic function w = wo cosh (t), where


54 ( LA 0 10c )
2 = , (4a)
mL 3
being an eigenvalue of eq. (4).
The equation of linear motion of the upper end of the rod takes the form:
 +  .
u = 2 L (5)
L
Assuming a relation w = , we can derive the velocity of the upper end of
3
the rod u as follows:
9w 02
v1 = sinh ( t ) . (6)

On the other hand, the equation of the motion of the impactor is as follows:
Gv = A 0 (7)
and, subsequently the velocity of the impactor amounts:
A
v = v0 0t . (8)
G
Comparing the velocity of the impactor (8) with the velocity of the impacted
end of the rod (9), we obtain a transcendent equation:
A 9 w 02
v0 0 t = sinh ( t ) , (9)
G L
8 Axial impact of open-section TWCF columns 79

with the root (tcr), a critical instant, at which a transition from progressive local
buckling to global bending takes place (Fig. 4). A corresponding critical velocity is
as follows:
A
v cr = v 0 0 t cr . (10)
G
The values of tcr and vcr , corresponding to relations (9) and (10) are given in
Table 5.

4. DYNAMIC IMPACT TESTS

4.1. TESTING PROGRAM

The testing program for impact tests was planned in the way, which allowed
one to observe the transition from local progressive buckling to global bending
failure mode at certain critical lengths and velocities. Thus, dimensions of
specimens to be tested were determined in order to assure local buckling under
static load. Omn the other extreme, lengths of specimens were taken to be slightly
less and slightly greater than critical values, calculated on the basis of theoretical
models discussed in paragraph 3.
The dimensions of specimens (columns), which have been tested are shown
in Table 3. Two series of columns have been tested: series 1 plain channel and
series 2 top hat section. Impact tests were carried out on steel specimens of
different a/b ratio and different length from 250 to 500 [mm]. Wall thickness of all
columns was t = 1 mm.
The specimens were made of steel sheets. Standard tensile tests were
performed on coupons cut from the sheet (row material). Material properties
determined were as follows:
Yield stress 0 = 201.5 MPa
Young modulus E = 1.75105 MPa
tangent modulus Et = 2000 MPa.
Similarly to the specimens made for quasi-dynamic tests, steel sheets were
folded in folding machine. All rolling radii were of the same magnitude r = 2.5 mm.
All specimens were clamped in two parallel plates (situated in grooves made
by milling machine) and spot welded , as shown in Fig. 7.
Table 4 shows main parameters and critical lengths of specimens, calculated
according to two first theoretical models (pos. 13 Table 2) for sections under
investigation.
Alltogether about 90 impact tests were carried out at different impact
velocities from 4 to 10 [m/s] and impact energy up to 4.5 kJ.
80 Maria Koteko, Artur Modawa, Marcin Jankowski 9

1 2

Fig. 5 Tested sections: 1 plain channel, 2 top hat.

Table 3
Symbol a b w Length L
[mm] [mm] [mm] [mm]
A1 60 60 - 250
A1 60 60 - 300
A1 60 60 - 400
A1 60 60 - 500
B1 90 45 - 200
B1 90 45 - 300
B1 45 90 - 350/400
D1 60 45 - 250
D1 60 45 - 300
D1 60 45 - 400
D1 60 45 - 500
D2 60 45 15 250
D2 60 45 15 300
D2 60 45 15 400
D2 60 45 15 500

Table 4
a b w A A0
Symbol LcrE Lcr1 Lcr2
mm mm mm mm2 N
A1 60 60 - 187.8 37841.7 291 356 460
B1 90 45 - 187.8 37841.7 215 263 339
D1 60 45 - 157.8 31796.7 220 269 348
D2 60 45 15 195.6 39413.4 278 341 440

4.2. EXPERIMENTAL STAND

Experiments were carried out on the drop hammer rig (Fig. 6), which
consists of the impactor of changeable mass (2) , hoisting mechanism and release
mechanism of the hammer (4,5) as well as additional springs (3), which allow one
to increase the energy of impact. The springs are driven by the separate driving
system. All mechanisms are controlled by a special computer program. The
measuring system consists of piezoelectric force transducer, situated at the bottom
of the specimen (1), displacement transducer and accelerometer.
10 Axial impact of open-section TWCF columns 81

5 4
3

Fig. 6 Drop hammer rig general view.

Fig. 7 Drop hammer rig: specimens lay-out.

4.3. TEST RESULTS

As was mentioned above, altogether about 90 impact tests were conducted


under different impact velocities from 4.5 to 10 [m/s]. Results of tests carried out
on the specimens of the same dimensions and under the same testing conditions
were repeatable.
The summary of the results of tests is given in Table 5. Theoretical values of
critical instant and corresponding critical velocity calculated according to the
model III par. 3, relations (9, 10) are shown in last two columns.
The results of the tests were partially in agreement with analytical models I and II.
For 14 of the examined cross-sections the observed failure mode was in good
82 Maria Koteko, Artur Modawa, Marcin Jankowski 11

agreement with the theoretical one. In 9 cases of the tested cross-sections a


transition from local progressive dynamic buckling (PDB) to global bending failure
mode (GB) was observed for columns of lengths, which corresponded to intervals
of critical lengths coming from analytical models I and II. It concerned plain
channel sections.
In the case of top hat sections only the progressive buckling failure mode was
observed, additionally, in some cases, an interaction of progressive buckling and
flexural-torsional buckling was recorded.
In the case of IIIrd analytical model, the calculated critical time intervals were
larger than the actual total time of impact (Fig. 9).
Table 6 shows failure patterns of all examined columns, corresponding to
Table 5. In Figs. 8 and 9 exemplary failure modes are presented in details. Fig. 8
shows a difference in failure patterns of three columns of the same dimensions of
the cross-section and length, subjected to different impact velocities. As one can
notice, the difference is significant. In Fig. 9 the transition from progressive
dynamic local buckling to global bending failure mode is presented for tewo
columns of the same cros-section dimensions and different length: L < LcrE (Fig. 9a)
and Lcr1 < L< Lcr2 (Fig. 9b). Failure mode in Fig. 9b corresponds to II. theoretical
model, namely three-link model.
Table 5
Failure
Dimensions Energy/velocity Notes tcr* Vcr*
Sp. No Symbol mode
a / b /w / L [mm] [kJ] / [m/s] - - [s] [m/s]
1 A1 60/60/-/250 2/8.94 PDB(*) L<LcrE
2 A1 60/60/-/250 1/6.32 PDB L<LcrE
3 A1 60/60/-/250 0.5/4.47 PDB L<LcrE
4 A1 60/60/-/300 1/6.32 PDB LcrE<L<Lcr1 0.20 6.16
5 A1 60/60/-/300 0.5/4.47 PDB LcrE<L<Lcr1 0.19 4.32
6 A1 60/60/-/400 1/6.32 PDB LcrE<L<Lcr2 0.19 6.18
7 A1 60/60/-/400 0.5/4.47 GB(**) LcrE<L<Lcr2 0.18 4.33
8 A1 60/60/-/500 1/6.32 PDB L>Lcr2 0.21 6.16
9 B1 90/45/-/200 1.5/7.75 PDB L<LcrE
10 B1 90/45/-/300 1.5/7.75 GB Lcr1<L<Lcr2 0.13 7.65
11 B1 90/45/-/400 1.5/7.75 GB L>Lcr2 0.16 7.63
12 D1 60/45/15 -/250 1.5/7.75 GB LcrE<L<Lcr1 0.13 7.65
14 D1 60/45/-/250 1.5/7.75 PDB L<LcrE
15 D1 60/45/-/250 1.5/7.75 GB LcrE<L<Lcr1 0.13 7.66
16 D1 60/45/-/300 1.5/7.75 GB Lcr1<L<Lcr2 0.14 7.66
17 D1 60/45/-/300 1/6.32 GB Lcr1<L<Lcr2 0.13 6.24
19 D1 60/45/-/400 1.5/7.75 GB L>Lcr2 0.16 7.65
20 D1 60/45/-/400 1/6.32 PDB L>Lcr2 0.16 6.22
21 D1 60/45/-/500 1/6.32 GB L>Lcr2 0.2 6.119
22 D2 60/45/15/250 1/6.32 PDB L<Lcr
23 D2 60/45/15/300 1/6.32 PDB LcrE<L<Lcr1 0.18 6.18
24 D2 60/45/15/400 1/6.32 PDB Lcr1<L<Lcr2 0.18 6.18
25 D2 60/45/15/500 1/6.32 PDB L>Lcr2 0.21 6.16
13 D2 60/45/-/250 1.5/7.75 PDB L<LcrE 0.38 7.44
18 D2 60/45/15/400 2.2/9.38 PDB Lcr1<L<Lcr2 0.18 9.24
(*)
progressive dynamic local buckling, (**) global bending
12 Axial impact of open-section TWCF columns 83

Table 6

1 2 3 4 5

6 7 8 9 10

11 12 14 15 16

17 19 20 21 22

23 24 25 13 18

a) b) c)

Fig. 8 Column A1, L= 250 mm: a) v0 = 4.47 m/s; b) v0 = 6.32 m/s; c) v0 = 8.94 m/s.
84 Maria Koteko, Artur Modawa, Marcin Jankowski 13

a) b)

Fig. 9 Column B1, v0 = 7.75 m/s: a) L = 200 mm (No 9); b) L = 300 mm (No 10).

a) b)
F[N]

u[m]

Fig. 10 Load-shortening diagrams: column B1, v0 = 7.75 m/s: a) L = 200 mm (No 9);
b) L = 300 mm (No 10).

0.015 s
tcr = 0.48 s

Fig. 11 Column D2, L = 500 mm, v0 = 6.32 m/s (25).


14 Axial impact of open-section TWCF columns 85

Load-shortening diagrams for columns shown in Fig. 9a and 9b are presented


in are presented in Figs. 10a and 10b, respectively. The diagrams indicate, that a
global bending failure mode of the column of length L > Lcr leads to a poorer
energy absorption capacity. An exemplary load-time for top-hat section column is
shown in Fig. 11a, while Fig. 11b presents the typical failure mode for the longest
top-hat section column, for which local dynamic buckling was observed.

5. CONCLUSIONS

The rod models proposed by Alves and Karagiosova [10] occurred to be


not entirely adequate for open section members. However, the critical buckling
length for dynamic buckling, calculated on the basis of the I. and II. model, was in
relatively good agreement with the results of experiments in several cases of plain
channels. Thus, those simple analytical models may be useful in prediction of
crushing behaviour of channel section columns, particularly at the preliminary
stage of design process.
In nearly all examined cases, critical time intervals, calculated according to
IIIrd two-phase theoretical model , taking into account inertia effects, for which a
transition should take place, were larger than the total actual time of impact.
In the case of top hat sections only the progressive buckling failure mode was
observed. It is probably induced by the edge stiffeners, which stabilize the column
during impact.
More clear indications of the threshold criterion (or criteria) of the transition
between progressive buckling and global bending can be obtained from further
research which would take into account larger diversity of impact velocities and
dimensions. Also impact tests at different impact mass have to be performed in
order to investigate inertia effects.
Simultaneously with experiments reported in this paper, numerical FE
simulations were carried out and are in progress. Results of those numerical experiments,
which will be presented in a separate publication will deliver a more clear
understanding into the phenomenon of transition between progressive buckling and
global bending of open section members subjected to axial, impact compressive force.
Received on: July 16, 2014

REFERENCES

1. JONES, N. Structural Impact, Cambridge University Press, 2003.


2. KOTEKO,M, MANIA, R., KOAKOWSKI, Z., Dynamic crushing of thin-walled profiles, Proc.
of ICTWS Conference Thin-walled Structures-Recent Innovations and Developments, ed.
by M. Mahendran, Queensland University of Technology, Brisbane, June 2008, pp. 687693.
86 Maria Koteko, Artur Modawa, Marcin Jankowski 15

3 KOTEKO M., MANIA R.J.: Quasi-static and dynamic axial crushing of TWCF open-section
members, Int. Journal of Thin-Walled Strc., 61, pp. 115120, 2012.
4. KOTEKO M. et al., Plastic mechanisms database for thin-walled cold-formed steel members in
compression and bending, Thin-Walled Structures, 48, pp. 818826, 2010.
5. LANGSETH M. et al., Crash behaviour of thin-walled aluminium members, Thin -Walled Struct., 32,
pp. 127150, 1998.
6. LANGSETH M. et al., Crashworthiness of aluminium extrusions. Validattion of numerical
simulation. Effect of mass ratio and impact velocity, Int. J. Impact Eng., 22, pp. 829854, 1999.
7. HSU S.S., JONES N., Dynamic axial crushing of circular and square stainless steel tubes.
Structures under shock and impact, VII, pod red. Jonesa i in. WIT Press, 2001, pp. 169178.
8. ALVES M. KARAGIOSOVA D., Influence of the Arial impast velocity on the buckling behaviour
of circular cylindrical shells, Proc. of 9th National Congress on Theoretical and Applied
Mechanics, Demetra Ltd., Bulgaria, 2001, pp. 388393.
9. ALVES M. KARAGIOSOVA D., Transition from progressive buckling to global bending of
circular shells under axial impact Part I: Experimental and numerical observations, Int.
J. of Solids and Struct., 41, pp. 15651580, 2004.
10. ALVES M. KARAGIOSOVA D., Transition from progressive buckling to global bending of
circular shells under axial impact Part II: Mathematical model, Int. J. of Solids and Struct.,
41, 2004, pp. 15811604.
11. JENSEN O. et al., Transition between progressive and global buckling of aluminium Extrusions,
in: Structures under shock and impact, VII, Ed. by N. Jones, WIT Press, Southhampton,
Boston, 2002, pp. 269277.
12. TERAMOTO S.S., ALVES M., Buckling transition of axially impacted open shells, Int. J. of Impact
Eng., 30, 2004, pp. 124126.
13. KOTEKO M., MANIA R.J., Investigation into different collapse modes of twcf open section
members under axial impact, Proc. of ICTWS2012, Glasgow , pp.391399, 2012.
14. WIERZBICKI T., SINMAO N.W., A simplified model of Brazier effect in plastik bending of
tubes, Int. Journal of Pressure Vessels, 71, 1, pp. 1928, 1997.
EIGENVALUE ANALYSIS OF CURVED SANDWICH PANELS
LOADED IN UNIAXIAL COMPRESSION

JOO PEDRO MARTINS, L. SIMES DA SILVA,


LILIANA MARQUES, MARTIN PIRCHER*

Abstract. An energy formulation of a simply supported cylindrically curved sandwich


panel loaded in compression is introduced and the corresponding potential energy
function is evaluated. The mechanical model of a cylindrically curved sandwich panel
comprises three interacting buckling modes, corresponding to nine degrees of freedom
comprising qs, qx and qy components of local snake (m, n) and overall (k, l) modes, the
related local hourglass (m, n) mode, Finally, closed-form solutions are presented for
the global and local buckling modes.
Key words: curved panels, sandwich panels, composite structures.

1. INTRODUCTION

Sandwich structures exhibit a potentially unstable post-buckling behaviour


due to coupling of stable buckling modes [1]. Hunt et al. [2] have shown that
realistic combinations of face materials with significantly less stiff core materials
lead to unstable post-buckling response in the case of sandwich struts. This effect is
even more pronounced in the case of core materials with orthotropic properties,
such as honeycomb cores [3].
The characterization of the post-buckling behaviour of sandwich panels
requires the identification of the relevant buckling modes and the corresponding
elastic critical loads.
Sandwich panels can be nowadays produced with arbitrary shapes. The
construction sector increasingly uses curved panels in structural applications,
clearly extending the classical range of application of cylindrical sandwich panels
as shells of revolution in the aeronautical industry.
This paper presents an energy formulation based on the Rayleigh-Ritz
method for the calculation of the elastic critical loads of cylindrically curved
sandwich panels. Closed-form solutions are obtained for the global and local
buckling modes.

2. MODAL DESCRIPTION

Figure 1 illustrates a typical curved sandwich panel loaded in uniaxial


compression with simply-supported edges. Figure 1b depicts the particular case of a
*
University of Coimbra, Department of Civil Engineering, ISISE, Portugal
88 Joo Pedro Martins , L. Simes da Silva , Liliana Marques , Martin Pircher 2

flat panel henceforth denoted a sandwich plate. A sandwich panel loaded in


compression exhibits two distinct sets of buckling modes: snake and hourglass
modes, assumed to vary sinusoidally along the length of the panel [4], typically
illustrated in Fig. 2.
(i) Snake mode. The snake mode is defined as the combination of three
independent degrees of freedom, [4], allowing complete flexibility of mode form
between that of pure shear to those of pure bending in each direction with no shear,
as represented in expressions (1) to (3).

z (w) a
z (w) a

y (v)
y (v)
Flanges x (u) R
Flanges x (u)
Core
t
c Core t
t c
t

b b

a) b)
P
c

a
c)
Fig. 1 Simply supported sandwich panel: flat (R = ) and curved panel.

Snake mode

Hourglass mode

Fig. 2 Local buckling modes of sandwich panel.

a ix jy
w = qs sin sin (1)
i a b
3 Eigenvalue analysis of curved sandwich panels 89

ix jy
u = zq x cos sin (2)
a b
a j ix j y
v = zq y sin cos . (3)
b i a b
qs represents the amplitude of the pure shear component of the snake mode. There
is bending in each flange over its thickness but no bending action in the section as a
whole. The other two displacement fields are introduced to allow for bending
strains to develop, consisting of an axial variation in the angle of tilt of each plane
section in the x- (qx) and y-direction (qy) [1]. Depending on the number of half-
waves, the snake mode may appear as a local (large numbers of half wavelength m, n)
or as a global mode. In the latter case it is denoted overall mode (small numbers of
half wavelength k, l).
(ii) Hourglass mode. The symmetrical nature of the hourglass mode requires
that there must be zero transverse displacements at the middle plane of the sandwich
panel, as well as no longitudinal displacements. Thus, the shear strain must be zero
at the centre line and maximum at the interface between the core and the flanges. If
a linear variation in shear angle is assumed, such that it matches the angle of
inclination of the flange at each interface, the hourglass mode is described by [1]:
2 z a ix jy
w = qh sin sin . (4)
c i a b
The model includes three interacting buckling modes, corresponding to nine
degrees of freedom comprising qs, qx and qy components of local snake (m, n) and
overall (k, l) modes, the related local hourglass (m, n) mode, and constant and
variable end shortening. Furthermore, the longitudinal wave number m is assumed
to be even [1]. Specifically the degrees of freedom are as follows:
a1 shear component of snake mode (i, j) = (m, n)
a2 corresponding hourglass mode (i, j) = (m, n)
a3 shear component of overall mode (i, j) = (k, l)
A4 constant end-shortening
a5 tilt component of snake mode in x-direction (i, j) = (m, n)
a6 tilt component of overall mode in x-direction (i, j) = (k, l)
a7 tilt component of snake mode in y-direction (i, j) = (m, n)
a8 tilt component of overall mode in y-direction (i, j) = (k, l)
A9 variable end-shortening.

3. STRAIN-DISPLACEMENT RELATIONS

As the faces and core materials are different it is necessary to consider


separately the stress states in both faces and core. Because of the assumption that
90 Joo Pedro Martins , L. Simes da Silva , Liliana Marques , Martin Pircher 4

the faces behave as thin plates or shells, the direct stress and shear strains related to
z-direction are taken as zero (5) and only the strains in the xy-plane are relevant
((6) to (8), where z* denotes a local coordinate system with its origin at the centre
line of each flange). The strain displacement relations for the core correspond to a
three dimensional stress state ((9) to (14))
z = xz = yz = 0. (5)

It is further assumed that the panels are in the range of shallow shells and
DMV (Donnell-Mushtari-Vlasov) nonlinear shell theory is applicable. This theory
assumes that the shell shows infinitesimal deformations and moderate rotations.
Also the intrinsic geometry of a shallow shell is identical to the geometry of a
plane of its projection. This actually represents the first basic assumption of the
theory of shallow shells [5]. The second assumption of shallow shells theory is that
the effect of transverse shear forces on the in plane equilibrium equations is
negligible and the influence of the deflections, w, predominates over the influence
of the in plane displacements u and v in the bending response of the shell.
Furthermore, in order to simplify the strain displacements relations a mean
radius of curvature is considered (instead of considering one for the core and two
other for the flanges) which can lead to errors when computing critical stresses.
Nevertheless, this simplification is in line with the geometrical definition of
shallow shells.
The next set of equations reflects these assumptions for a sandwich panel.
(i) Flanges
2
u 1 a 1 w 2 w
x =
x a
0

2 x
dx z *

x 2
, (6)

v w * 2 w
y = + z , (7)
y R y 2

u v 2 w
xy = + 2 z* . (8)
y x xy

(ii) Core
2
u 1 a 1 w
x =
x a 0
dx,
2 x
(9)

v w
y = + , (10)
y R
5 Eigenvalue analysis of curved sandwich panels 91

w
z = , (11)
z
u v
xy = + , (12)
y x

u w
xz = + , (13)
z x
v w
yz = + . (14)
z y

4. STRAIN ENERGY EXPRESSIONS

The general strain energy expression is given by (15)


1
U=
2 V
( x x + y y + z z + x x + y y + z z ) dV . (15)

For the faces, introducing Hookes law in expression (15) leads to, after some
reworking,

Ef
( )
a b t/2
Vf = 2
x + 2y + 2 x y dxdydz * +
2 0 0 t / 2
(16)
Gf a b t/2
+
2
0 0 t / 2
2xy dxdydz *

where
Ef
Ef = . (17)
1 2f

The strain energy of the faces may be split into two contributions: pure
bending in the faces (Vfb) and membrane action in the faces (Vfm).
Similarly, for the core the strain energy is given by:
Ec a b c/2
c(1) ( 2x + 2y + 2z ) + c(2) ( x y + x z + y z )
Vc =
2
0 0 c / 2
(18)
G
( )
a b c/2
dxdydz + c 2
xy + 2
xz + 2
yz dxdydz ,
2 0 0 c / 2
92 Joo Pedro Martins , L. Simes da Silva , Liliana Marques , Martin Pircher 6

where, for simplicity:

Ec
Ec = , (19)
1 c2

(1)
=
(1 c )
2
, (20)
c
1 2 c

1 c
c( 2 ) = c . (21)
1 2 c

5. TOTAL POTENTIAL ENERGY FUNCTION

The total potential function (22) is obtained by adding to the two components
of the strain energy (U), the strain energy due to the end-beam of stiffness Kb/a
(23) and the work done by the load P (24)

V = V f + Vc + Vb WDL, (22)

K b. b y 3
Vb =
a 0
A 9 a sin 2 dy = K b A 92 ab,
b 16
(23)

b 1
WDL = P u x = a dx = P A 4 + A 9 ab. (24)
0
2

Introducing the last term of the strain displacement relations ((6) to (8)) into
(16) gives, after summing the contributions from both flanges, the strain energy of
flange bending of the panel:

t 3 4 E f 2 k 2 l 4 a 2 l2
Vbf , panel = Vbf , plate = a + + 2 +
12 4 a 2 b 4 k 2 b 2
3 f

(25)
n 2 G f 2 2
(a12 + a12 ) m2 n4a2
+
a2 b4m2 + 2 f
+
2
b b 2
( (
) )
a 3 l + a12 + a 22 n 2 ab .

Similarly, introducing the first terms of the strain displacement relations ((6)
to (8)) into (16) yields the flange membrane energy of the panel
7 Eigenvalue analysis of curved sandwich panels 93

1 a2 1 1
Vmf , panel = Vmf , plate + E f t 2 2 ( a12 + a22 ) + 2 a32 +
2 2R m k
+v f
1 c2
R 2
a
k l
(
a2 a5 2 1 ( 1) 1 ( 1)
k
)(
l
) (26)
4 + 2 ( 1)l 2n 2 4a3 ( 2 A9 A4 ( l 2 4 ) )
a33 2 ( a 2
+ a 2
) a +
2 ( l 2 4 )
1 2 3
3 l 4n 2
ab
where

1 3 1
V mf , plate =E f t 2 A 42 + 2 A 4 A 9 + A 92 2 A 4 ( a 12 + a 22 + a 32 )
2 4 2
4 4 + K nl
a 1 + a 2 + a 3 + 6a 1 a 2 +
4 4 2 2
3
A 9 ( ( a 12 + a 22 ) K n + a 32 K l ) + 4 3
64
(a1 a 3 + a 2 a 3 )
2 2 2 2

(27)
2 2 c
n2
4n l 2
2 (
1 ( 1)
k
) (1 ( 1) )l 1 a l
a l a 1a 2 a 6 + f b 2 k 2 a 1a 2 a 8 +

2
c 2 b 2 a2 n4 2 l4 2
+ 2 (m a 5 + k a 6 ) + 2 2 a 7 + 2 a 8 + ( n 2 a 5 a 7 + l 2 a 6 a 8 )
2 2 2 2
f
2b 2a 2b m k
a 2

1 2 c
(l ) ab.
2
(a 6 + a 8 ) + n 2 (a 5 + a 7 )
2 2
ab + G f t 2 A 92 + 2
2 3b 2b

Equation (26) highlights the specific contributions from curvature, compared


to a flat panel. The strain energy of the core results from (18), introducing the
strain displacement relations (9) to (14)

a2 1 2 1 2 1 2
V c , panel = V c, plate + E c c v c(1) a1 + a 2 + 2 a 3 +
8R 2 m 2 3 k
1 a c a
( )( )
2 2
a 2 a 5 2 1 ( 1) 1 ( 1)
k l
+v c(2) a 1a 2 +
R 2m c 2
24 k l (28)
2 + ( 1) a 3 ( 2 A 9 A 4 ( l 2 4) )
l
n2
a 33 ( 3a 2
+ a 2
) a + .
6 ( l 2 4n 2 ) 2 ( l 2 4)
1 2 3
6

ab
where
94 Joo Pedro Martins , L. Simes da Silva , Liliana Marques , Martin Pircher 8

1 3 a2a2 1 1
Vc , plate = E c c v c(1) A42 + A4 A9 + A92 + 2 22 2 A4 a12 + a 22 + a 32
2 8 c m 4 3

4 1 4 4 2 2
1 2 1 2 3 a1 + 5 a 2 + a 3 + 2 a1 a 2 +
A9 a1 + a 2 K n + a 32 K l + 4
2 3 128 + 4 + K nl a 2 a 2 + 1 a 2 a 2
1 3 2 3
3 3
(1) 1
vc a1a 2 a 6 +
( )( )
2
1 n a l
2 c 2 2 1 ( 1) 1 ( 1)
k l
+
3 4n l (2) a l
+ vc aa a
2 2 1 2 8
b k
2
c 2 b2 a2 n4 2 l4 2
+
4 3b (
2 2 2 2
)
2 m a5 + k a 6 + 2 2 a 7 + 2 a8 + (29)
a b m k
2
c 2
+ vc(2)
4 6 b (n a a
2
5 7 + l 2 a 6 a8 )
ab +


a2 2
1 c 2
+ 2 G c c 2 A92 +
2 6b 4 3b
2
(
l ( a 6 + a8 ) + n ( a5 + a 7 ) +
2 2
)
n2 1 2
2 ( a1 a 7 ) + a 2 +
2
2
1 1 2 a m 3
( a1 a5 ) + a 2 + ( a 3 a 6 ) + 2
2 2
+ ab,
4 3 b l2
+ ( a 3 a8 ) 2

k2

and (see Annex A)

3 / 8 if n = 1 3 / 8 if l = 1 2 if n = l
Kn = , Kl = and K nl =
/ 4 if n > 1 / 4 if l > 1 0 if n l (30)
n l.
The degrees of freedom associated with the total end shortening can be
eliminated from further consideration by using the corresponding equilibrium
equations and solving them simultaneously with respect to A4 and A9.


A4 =
2E f t +
P
E c cvc(1)
+
8
( )
3 ( 4 K n ) plate a12 +
9 Eigenvalue analysis of curved sandwich panels 95

2
(
+ 3 ( 4Kn ) plate
2
3
) 2
(
1 plate a2 + 3 ( 4Kl ) plate a3

)
(31)
a
(
2 1 ( 1)
R
k
)(l panel
)
1 ( 1) 2 + 3 panel a3 ,
k l

2 2
( 4 K n ) a1 + ( 4 K n )
2
3 1 3 plate a2 +
A9 = plate +
4
+ ( 4 K l ) a32 (32)

6a
R
(
+ 2 1 ( 1)
k
)(
1 ( 1)
l
) panel a3 ,

where

8 E f tv f + E c cv c( 2 )
panel = , (33)
2 E f t + E c cv c(1)

panel =
( ( )
b 2 8 E f tv f l 2 3 + E c cv c( 2 ) l 2 )
( )
( 3
) ( , (34)
k 2 l l 2 4 2a 2 2 2G f t + Gc c + b 2 3K b + 2 E f t + E c cvc(1)
2
)

E c cv c(1)
plate = , (35)
2 E f t + E c cv c(1)

plate =
(
b 2 2 E f t + E c cv c(1)
.
) (36)
2 2 3 2
2
(
2a (2G f t + Gc c ) + b 3K b + 2 E f t + E c cv c
(1)
)
Introducing equations (31) and (32) into (22) (the latter divided by a.b) gives
the reduced form of the total potential energy function of the panel in seven
degrees of freedom expanded about a point on the fundamental path, denoted by F,
written in general form.
The coefficients VijF are given in Annex B.
1 1 F 2 1 F 2 1 F 2 1 F 2 1 F 2
V panel = V11F a12 + V22 a2 + V33 a3 + V55 a5 + V66 a6 + V77 a7 +
2 2 2 2 2 2
1
+ V88F a82 + V12F a1a2 + V15F a1a5 + V17F a1a7 + V25
F
a2 a5 + V57F a5a7 + V36F a3a6 +
2
96 Joo Pedro Martins , L. Simes da Silva , Liliana Marques , Martin Pircher 10

+V38F a3a8 + V68F a6 a8 + V333


F 3 F
a3 + V126 F
a1a2 a6 + V128 F 2
a1a2 a8 + V113 a1 a3 +
F 2 1 F 4 1 F 4 1 F 4 1 F 2 2 (37)
+V223 a2 a3 + V1111a1 + V2222 a2 + V3333a3 + V1122 a1 a2 +
24 24 24 4
1 F 2 2 1 F 2 2 1
4 4 2
(
+ V1133a1 a3 + V2233a2 a3 + P P F V11' F a12 + V22 )(
'F 2 'F 2
a2 + V33 a2 . )
The equilibrium equations are obtained by stetting to zero first-order
derivatives of the total potential function [7]:
V
= 0. (38)
ai
It can be seen that the fundamental path (pre-buckling solution) is non trivial
with respect to the chosen degrees of freedom, yielding out of plane deformations
along the fundamental path. However, for a flat panel (R tends to infinity) this
behavioural feature disappears and the fundamental path becomes trivial (ai = 0).

6. CRITICAL LOADS

The critical loads are obtained by setting to zero the determinant of the
second derivatives with respect to each of the degrees of freedom ai in turn,
evaluated along the fundamental path,

V11F V12F 0 V15F 0 V17F 0


V12F V22F 0 V25F 0 0 0
0 0 V33F 0 V36F 0 V38F
VsF,ij = V15F V25F 0 V55F 0 V57F 0 = 0. (39)
0 0 V36F 0 V66F 0 V68F
V17F 0 0 V57F 0 VsF,77 0
0 0 V38F 0 V68F 0 V88F

Solving equation (39) yields the three critical loads corresponding to the three
buckling modes: snake, hourglass and overall
1 2 m2 a 2n 4 n2 2 a2
P sC, panel = E ft3 + + + E t +
2b 4 m 2 b 2 2 R 2
f
3 2a
2
m2
(40)
a 2n 2 4
+G c c 1 + 2 2 2 V s ,
b m
11 Eigenvalue analysis of curved sandwich panels 97

1 m2 a 2n 4 n2 2 a2
P hC, panel = 2 E f t 3 + + +
2R 2 E f t +
3 2a 2b 4 m 2 b 2 m2
2

(41)
1
1 a 2n 2 4 a 2 (1) 4 2
+ G c c 1 + 2 2 +
2 E c v c + 2 V h 1 plate ,
3 b m cm 2
3

1 k2 a 2l 4 l 2 a 2l 2
PoC, panel = 2 E f t 3 2 + 4 2 + 2 + G cc 1 + 2 2 +
3 2a 2b k b b k
1
( (1) a
) 4a 2
( ) (1 ( 1) )
2 2 2
1 ( 1)
k l
+ 2E f t + E c cv c (42)
R
2 2
k 2
R
6 2

E ccv c(2) + 4E f t ( v f 2) E ccv c(2)l 2 4E f tv f ( l 2 6) 4


panel + 3 + 2 V o ,
l ( l 2 4)
panel

2
k 4l 2

where

1
Vs = [V172V55 2V15V17V57 + V152V77 + V252V77 +
2 ( V + V55V77 )
2

(
57

+ ( V172V55 + 2V15V17V57 V152V77 V252V77 ) 4 ( V572 + V55V77 )


2
(43)
(V172V252 2V12V17V25V57 + V122V572 2V12V15V25V77 V122V55V77 ) ) ,
1/ 2

1
Vh = [V172V55 2V15V17V57 + V152V77 + V252V77
2 ( V + V55V77 )
2

(
57

( V172V55 + 2V15V17V57 V152V77 V252V77 ) 4 ( V572 + V55V77 )


2
(44)
(V172V252 2V12V17V25V57 + V122V572 2V12V15V25V77 V122V55V77 ) ) ,
1/ 2

2 2
V38F V66F 2V36F V38F V68F + V36F V88F
Vo = 2
. (45)
V68F V66F V88F

7. RESULTS AND DISCUSSION

Although the expressions for the buckling loads of a sandwich panel were
obtained analytically with closed form solution, they are complex, making it
difficult to establish how these critical loads vary with the different parameters.
98 Joo Pedro Martins , L. Simes da Silva , Liliana Marques , Martin Pircher 12

Nevertheless, because m and k figure always in the denominator as well as the


numerator, the sandwich panel will yield minima for certain values of m and k. If
only the first terms of the expressions are considered it can be seen that the
transverse wavenumbers n and l always figure in the numerator only, suggesting
that the minimum value is obtained for n and l = 1, as can be easily checked
numerically (Fig. 3b) for example).
To exemplify the formulation developed in this study and to illustrate the
behaviour of curved sandwich panels loaded in compression some typical examples
were selected from the literature ([1] and [5]). This component is longitudinally
compressed, simply supported plate with aspect ratio 2, with length a = 508 mm,
width b = 254 mm, flange thickness t = 0.508 mm, core thickness c = 5.08 mm,
flange Youngs modulus Ef = 68 947.57 N/mm2, flange Poissons ratio vf = 0.3,
core Youngs modulus Ec = 198.57 N/mm2, core shear modulus Gc = 82.74 N/mm2
and core Poissons ratio vc = 0.2. Here, to enhance the interactive effect, by
bringing closer together the critical loads for the overall and hourglass modes, it is
considered a core thickness of c = 15.24 mm [1]. In this example a similar panel is
calculated having a radius of curvature equal to R = 1000 mm.
The variation of the critical loads are shown plotted against longitudinal (m)
and transverse (n) wavenumbers in Fig. 3, where Ps denotes the critical load for the
snake mode and Ph denotes the critical load for the hourglass mode. As discussed
before, both modes exhibit a monotonically increasing variation with the transverse
wavenumber n.
Longitudinally, and assuming that n and l are equal to 1, the software
Mathematica [8] was used to compute the minimum critical loads (Fig. 3a) and the
corresponding wavenumbers (Table 1).
It is important to highlight that in contrast to what happens to the sandwich
plate, the minimum values of the critical loads depend on the radius of curvature. It
is possible to prove that whenever R tends to infinite, m tends to 4.
The variation of the critical loads with width b, plotted in Fig. 4a, b at constant
values of m, n and R, which correspond to the minimum critical load for the snake
mode (m = 4, n = 1 and R = infinite; m = 70, n = 1 and R = 1000) and for the
hourglass mode (m = 6, n = 1 and R = infinite; m = 70, n = 1 and R = 1000). It
noticeable that while the variation of snake mode is quite abrupt for low values of b,
the hourglass mode is hardly sensitive to variations of width, reflecting the local
nature of its corresponding wavelength, already small compared to the length of the
panel [1].

Table 1
Minimum values of critical loads (Ps and Ph)
Snake mode Hourglass mode
Radius of
Longitudinal Longitudinal
curvature Critical Load Critical Load
wavenumber, m wavenumber, m
infinite 1157.53 N 4 1070.26 N 70
1000 1244.26 N 6 1070.77 N 70
13 Eigenvalue analysis of curved sandwich panels 99

6000

5000

Ps n1

Ph n1
600000

500000

4000 400000

CriticalLoad
CriticalLoad


3000 300000

2000 200000
Ps m4
1000 100000
Ph m4

0 0
0 20 40 60 80 100 0 10 20 30 40 50
Wavenumber , m Wavenumber , n
a) b)
Fig. 3 Critical load variations with longitudinal


and transverse wavenumbers (R = 1000 mm).

2500
Ps m4;n1
Ph m70;n1
2500 Ps m6;n1
Ph m70;n1

2000 2000
CriticalLoad

CriticalLoad

1500 1500

1000 1000

500
100 150 200 300 500 700 1000 1500
Width , b Rinfinite
a)
500


100 150 200 300 500 700 1000 1500
Width , b R1000 mm
b)
Fig. 4 Critical load variations with width, b.

The conclusion on the effect of curvature in the critical behaviour of the


sandwich panel is immediately drawn from Fig. 5: the bigger the curvatures (low
values of R) the higher the critical loads are.

2500
Ps m4;n1

Ph m70;n1
2500
Ps m6;n1
Ph m70;n1

2000 2000
CriticalLoad
CriticalLoad

1500 1500

1000 1000

500
100 200 500 1000 2000 50001 104
Radius of curvature , R b254mm
a)
500
100 200 500 1000 2000 50001 104

b)

Radius of curvature , R b254mm

Fig. 5 Critical load variations with width radius of curvature, R.


100 Joo Pedro Martins , L. Simes da Silva , Liliana Marques , Martin Pircher 14

8. CONCLUSIONS

The energy formulation presented in this paper provides closed-form


solutions for the critical loads of curved sandwich panels. Despite the complexity
of the obtained expressions, they are easily programmed in a spreadsheet. Furthermore,
the formulation already includes the nonlinear terms required for a post-buckling
evaluation, an issue that is currently actively being pursued by the authors.

Acknowledgements. Financial support from the Portuguese Ministry of Science and Higher
Education (Ministrio da Cincia e Ensino Superior) under contract grant SFRH / BD / 70424 / 2010
is gratefully acknowledged.

ANNEX A

In the membrane strain energy of the flanges, Kn, Kl and Knl (30) correspond
to simplifications of the following general expressions

2n ( 1 + n 2 ) + sin ( 2n ) 2l ( 1 + l 2 ) + sin ( 2l )
Kn = ; Kl = (46)
8n ( 1 + n 2 ) 8l ( 1 + l 2 )

K nl =
(
sin 2 (l n )
.
) (47)
(l n )
It is noted that equation (42) represents an indeterminate expression for n or l =1
that is solved using a limit approximation (48). In all other possible cases (l >1) the
value of Kl and Kn is easily computed and is equal to 1/4.

2l ( 1 + l 2 ) + sin ( 2l ) 3 (48)
lim =
l 1 8l ( 1 + l )2
8
2l ( 1 + l 2 ) + sin ( 2l )
lim = .
l Int [ >1] 8l ( 1 + l 2 ) 4

Equation (47) is also indeterminate for l = n. Again a solution is found using


a limit approximation, and in all cases it should be noted that n is always greater
than l. It is concluded (from (49)) that whenever n = l Knl is equal to 2; whenever
n > l Knl is equal to 0
15 Eigenvalue analysis of curved sandwich panels 101

sin (2 (l n ))
lim =2
l n (l n )
. (49)
sin (2 (l n ))
n > 1 : lim =0
l 1 (l n )

ANNEX B

(i) Sandwich plate Vp,ijF terms


1 2 a 2n2 1 4 n2
V pF,11 = Gc c1 + 2 2 + Gf t3 2
6
4 b m b
(50)
1 m2 a2n4 n2 1
4 E f t 3 2 + 4 2 + 2 v f 2 P F ,
12 2a 2b m b 4

1 2 a 2n2 1 n2
V pF, 22 = Gc c1 + 2 2 + 4 G f t 3 2
12 b m 6 b
(51)
3 m n2
2
1 4 a2n4 a 2 (1)
E f t 2 + 4 2 + 2 v f + E c vc
12 2a 2b m b c m2

1 2
2 P F 1 plate
4 3 ,
1 2 a 2l 2 1 4 l2
V pF,33 = Gc c1 + 2 2 + Gf t3 2
6
4 b k b
(52)
1 4 k2 a 2l 4 l2 1
E f t 3 2 + 4 2 + 2 v f 2 P F ,
12 2a 2b k b 4

m2 1 (1)
2 a 2 E f t + 6 E c cvc +
1 c 2
V pF,55 = 2Gc c + , (53)
4 2 2 n 2
1
+ 2 G f t + Gc c
b 6
102 Joo Pedro Martins , L. Simes da Silva , Liliana Marques , Martin Pircher 16

k2 1 (1)
2 2 2
E f t + E c cvc +
1 c a 6
V pF, 66 = 2Gc c + , (54)
4 2 2 + l G t + 1 G c
2

2 f c
b 6
2
1 2 a 2 n 2 c 2 a 2 n 2 1
V F
= Gc c 2 2 + n 2 2 E f t + E c cvc(1) +
p , 77
4 6
b m 2 2 b b m
(55)
1
+ G f t + Gc c ,
6

2
1 2 a 2 l 2 c 2 a 2 l 2 1
V F
= Gc c 2 2 + l 2 2 E f t + E c cvc(1) +
p ,88
4 6
b k 2 2 b b k
(56)
1
+ G f t + Gc c ,
6
1
V pF,15 = V pF,36 = 2 Gc c , (57)
4
1 2 a2n2
V pF,17 = Gc c 2 2 , (58)
4 b m
1 a 2l 2
V pF,38 = 2 Gc c 2 2 , (59)
4 b k
2
c 2 1 1
F
= n E tv + E cv ( 2) + G f t + Gc c ,
2 2 b f f 6 c c
V p , 57 (60)
6
2
c 2 1 1
F
= l E tv + E cv ( 2) + G f t + Gc c .
2 2 b f f 6 c c
V p , 68 (61)
6
(ii) Sandwich panel VijF terms
1 a2
. V11F = V pF,11 + E ft , (62)
2R 2 m2
17 Eigenvalue analysis of curved sandwich panels 103

1 a2
V12F = E c v c( 2 ) 2 , (63)
2R m
1 a2
V 22F = V pF, 22 + E ft , (64)
2R 2 m2

( ) ( ) (1 ( 1) )
2
1 (1) a a2
1 ( 1)
k 2 l 2
V33F = V pF,33 + 2 E f t + E c cv c
4R 2
k 2
R
4 2


panel
(
E c cv c( 2 ) + 4 E f t v f 2)+ 3
(
E c cv c( 2 ) l 2 4 E f tv f l 2 6 ) , (65)
k 4l 2
panel
(
l l2 4 )
2

V55F = V pF,55 (66)

V66F = V pF,66 (67)

V66F = V pF,77 (68)

V88F = V pF,88 (69)

V15F = V36F = V pF,15 (70)

V17F = V pF,17 (71)

V25F =
1
24 R
(
c 2 6 E f tv f + E c cvc( 2) , ) (72)

V38F = V pF,38 (73)

V57F = V pF,57 (74)

V68F = V pF,68 . (75)

Received on March 14, 2014

REFERENCES

1. SIMOES DA SILVA, L., Interactive Bending and Buckling in Sandwich Structures, PhD Thesis,
Imperial College, 1989.
2. HUNT, G.W., SIMOES DA SILVA, L. and MANZOCCHI, G.M.E., Interactive Buckling in
Sandwich Structures, Proceedings of the Royal Society of London, A 417, pp. 155177, 1988.
104 Joo Pedro Martins , L. Simes da Silva , Liliana Marques , Martin Pircher 18

3. SIMOES DA SILVA, L. and HUNT, G.W., Interactive Buckling in Sandwich Structures with Core
Orthotropy, Mechanics of Structures and Machines, 18, 3, pp. 6180, 1990.
4. ALLEN, H. G., Analysis and Design of Structural Sandwich Panels, Pergammon Press, 109, 6,
14601471, 1969.
5. BRUSH, D. O. and ALMROTH, B. O., Buckling of Bars, Plates and Shells, McGraw Hill, New
York, 1975.
6. REDDY, J. N., Mechanics of Laminated composite plates and shells Theory and analysis, CRC
Press, Florida, 2004.
7. THOMPSON, J. M. T. and HUNT, G. W., A general theory of elastic stability, Wiley, 1973,
London.
8. *** WOLFRAM MATHEMATICA 8 (version 8.0.0).
AN ANALYSIS OF THE EFFECT OF HOLES PATTERNS
ON COUPLED INSTABILITY
OF PERFORATED THIN-WALLED MEMBERS

MIHAI NEDELCU*1

Abstract. The buckling behaviour of vertical steel pallet rack columns is greatly
influenced by the distribution of perforations punched continuously along their length.
To study the influence of the holes patterns on the elastic coupled instabilities of thin-
walled members, this paper uses an original modal identification method (based on
the Generalised Beam Theory) which is able to provide the pure buckling modes
participation (of global, distortional, local nature) in a general eigenvalue buckling
mode given by the shell Finite Element Analysis.

Key words: rack columns, perforated thin-walled member, interactive buckling, mode
decomposition, shell finite element analysis, generalised beam theory.

1. INTRODUCTION

The perforated thin-walled members are often used in civil and mechanical
engineering as highly efficient load bearing components. A special case is
represented by the cold-formed steel pallet rack columns, and the assessment of
their resistance capacity is a particularly difficult problem in structural engineering.
The perforations are punched continuously along their length to enable horizontal
storage rack shelving to be clipped into position at arbitrary levels (Fig. 1) and
these perforations decrease the rack columns axial capacity. Also, their typical
slenderness leads to a failure mechanism usually governed by coupled instabilities
(or interactive buckling). To study these complex phenomena one usually starts
with the fundamental (pure) buckling types, widely accepted as: Global (rigid-body
behaviour of the cross-section in its plane, yielding to exural or flexuraltorsional
buckling), Distortional (relative transversal displacements between the cross-
sectional corners) and Local (only local plate deformations) buckling (GDL). Each
buckling type has its characteristic post-buckling behaviour and strength reserve;
consequently, the derivation of the pure modes participation to a general buckling
mode is a crucial step in assessing the real design resistance of the member.
Nowadays there are a number of specialized methods capable to provide the
modal identification and participation, the most famous being the Generalised
*
Technical University of Cluj-Napoca, Faculty of Civil Engineering, Romania1
106 Mihai Nedelcu 2

Beam Theory (GBT) [1, 2] and the Constrained Finite Strip Method (cFSM) [3].
The only problem is that they present limitations when dealing with arbitrary cross-
sections and boundary/loading conditions. Even if recently, solutions were developed
for many cases of loading and boundary conditions (e.g. [4, 5]), publicly available
codes are not yet released (the available codes as GBTUL [6] and CUFSM [7] can
handle nowadays only bar classical loading and boundary conditions).
As for perforated members, the specialised methods usually take into account
the effect of the holes by introducing the concept of reduced thickness of the
perforated strip ([8, 9]). Even if satisfactory results have been reported, this
procedure could be criticized for using an unperforated model, disregarding the
stress concentrations around the holes. It would be also impossible to apply this
procedure for the case of uneven distribution or size of the holes.

Fig. 1 Rack system.

On the other hand, the general solution methods, like Finite Element Method
cannot explicitly provide the contribution of the pure deformation modes in a
general buckling mode. The shell Finite Element Analysis (FEA) is the most
commonly used instrument to study the behaviour of thin-walled members.
Recently, an original method based on Generalised Beam Theory (GBT) was
developed by the author [10] in order to decompose the elastic buckling modes
provided by the shell FEA into pure buckling modes of Global, Distortional and
Local nature. The main feature of this method lies in using only the GBT cross-
sectional deformation modes instead of member mode shapes.
3 Analysis of the effect of holes patterns on coupled instability 107

One goal of this paper is to present the latest developments which enable the
modal identification method to analyse isotropic thin-walled members with
arbitrary holes in terms of shape (rectangular and circular). The second goal is to
study the effect of holes on the critical elastic resistance of perforated thin-walled
members, which could lead in the near future to an optimisation procedure with
respect the perforations distribution and shape for a given member.

2. GENERALISED BEAM THEORY SHORT PRESENTATION

Figure 2 presents the terminology of the local coordinate system and the
corresponding displacement field used for the study of thin-walled members with
arbitrary cross-section.

Fig. 2 Thin-walled member: local axes/displacements.

In classical GBT, one assumes the classical Kirchhoff-Love hypotheses


( zz = xz = yz = 0 ) and also Vlasovs simplifying hypotheses (null membrane
transverse extensions ssM , L = v = 0 and membrane shear strains xs
M ,L
= u + v = 0 ,
i
where ( ) = ( ) x , ( ) = ( ) s . According to GBT, any displacement is
considered as a linear combination of n orthogonal pure deformation modes, each
one expressed as a product of two functions
n
u( s, x ) = uk ( s )k ( x ),
k =1
n
(1)
v( s, x ) = vk ( s )k ( x ),
k =1
n
w( s, x ) = wk ( s )k ( x ),
k =1

where uk ( s ), vk ( s ), wk ( s ) are the components of the cross-sectional deformation


modes and k ( x ) are the amplitude functions describing their longitudinal
108 Mihai Nedelcu 4

variation. The number of pure deformation modes n depends on the cross-section


type (unbranched, branched, open, closed, etc.), the number of fold-lines and
intermediate nodes. The warping displacements u(s) are considered to have linear
variation along the entire cross-section, a consequence of the Vlasovs simplifying
assumptions.
The 1st GBT step (the cross-sectional analysis) provides the cross-sectional
deformation modes. Figure 3 presents the in-plane deformations of the first 22 pure
deformation modes for a typical rack section. Notice that there are 4 Global
deformation modes (axial, bending and torsion deformations), 6 Distortional
deformation modes and the rest are all Local deformation modes.

Fig. 3 Rack section: the cross-sectional pure deformation modes.

Next, the member equilibrium equations are written based on the principle of
virtual work applied in its variational form. The GBT system of differential
equations in modal formulation has the expression [11]:

( )
Cikk IV Dikk + Bikk = X jik W j0k (2)

where C, D, B and X are the cross-section linear and geometrical stiffness matrices.
The vector W 0 (having 4 components) contains the resultants of the applied pre-
buckling stresses, namely (i) axial force ( W10 = N ), (ii) bending moments
( W20 = M y , W30 = M z ), and (iii) bimoment ( W40 = B ).
The GBT 2nd step (the member stability analysis) consists in solving the
above differential equations system, yielding the critical load factors ( ) and the
corresponding modal amplitude functions ( k ( x ) ). This step is not used by the
method described in this paper.
5 Analysis of the effect of holes patterns on coupled instability 109

3. BUCKLING MODE DECOMPOSITION FROM SHELL FEA

The described method extracts from a buckling shell FEA the amplitude
functions k ( x ) of the pure deformation modes. It is based on the special orthogo-
nality properties of the cross-section deformation modes uk ( s ), vk ( s), wk ( s ) and
their derivatives. A special case of clear orthogonality can be seen if one analyses
the transverse bending stiffness matrix B (a diagonal matrix):
1
Bkk = Kwk wk ds = mk mk ds (3)
s s K
where mk ( s ) = Kwk ( s ) are the cross-sectional transverse bending moments,
K = Et 3 (12(1 2 )) is the plate bending stiffness. The diagonal shape of matrix B
proves the orthogonality properties of the 2nd derivatives of the cross-section
displacements wk ( s ) or of the moments mk ( s ) . In [10] the diagonal matrix B is
used to extract from shell FEA the amplitude functions k ( x) of the Local and
Distortional deformation modes. The procedure is now briefly presented for one
simple example using a C-section member.
The modal identification method starts with a buckling shell FEA of a thin-
walled member. Next, the cross-sectional displacement field is extracted on a mesh
of points along the members axis (0 xP L, with L being the bars length).
Suppose that the general buckling FE deformation in one point P along the
longitudinal axis of the member is given as a combination of three pure modes
(3 Global, 5 and 6 - Distortional) as presented in Fig. 4.

Fig. 4 Combination of three pure modes.

From Eq. (1) the transversal displacements given by the shell FEA have the
expression:
wFE ( s, x P ) = w3 ( s )3 ( x P ) + w5 ( s )5 ( x P ) + w6 ( s )6 ( x P ) (4)
and the transversal curvatures can be easily calculated
wFE ( s, x P ) = w3 ( s )3 ( x P ) + w5 ( s )5 ( x P ) + w6 ( s )6 ( x P ) (5)
110 Mihai Nedelcu 6

Suppose we know the curvature functions wk ( s ) of all the cross-sectional


deformation modes from the GBT 1st step and we can also extract the same
functions from the shell FEA ( wFE ( s, x P ) ). In order to find the value of the
amplitude function 6 ( xP ) , we simply make the following integration

s Kw6wFE ds = s Kw6 ( s)w3 ( s)tds 3 ( xP ) + (6)


+ Kw6 ( s ) w5 ( s )tds 5 ( xP ) + Kw6 ( s ) w6 ( s )tds 6 ( x P )
s s

and because B is a diagonal matrix, we know that

s Kwi wk ds = 0 for any i k (7)

so, the first two terms of Eq. (6) will be eliminated, thus giving

Kw w
s
6 FE ds = Kw6 ( s ) w6 ( s )tds 6 ( xP ) = B666 ( x P )
s (8)
Finally, the value of the amplitude function in point P is found

6 ( xP ) =
s Kw6wFE ds (9)
B66
In [10] the computation of the integral from the above equation is explained
in detail. The above described procedure can be used to identify all the pure
buckling modes which involve cross-sectional distortions (D and L modes), using
the general formula

i ( xP ) =
s Kwi wFE ds for i 5 (10)
Bii
This is not the case of the Global deformation modes (i = 14) involving
axial extension, major/minor axis bending and torsion. For this reason, two
different stiffness matrices introduced by Eq. (2) are currently used, the warping
stiffness matrix C (a diagonal matrix) and the geometrical stiffness matrix X1 (not
diagonal) standing for the stiffness degradation due to pre-buckling axial com-
pression. Their expressions are given below (for matrix C we have Membrane and
Bending components):

s
Cik = CikM + CikB = E tui uk ds + s Kwi wk ds
(11)
1
X 1ik =
A s ( vi vk + wi wk )tds
7 Analysis of the effect of holes patterns on coupled instability 111

where A is the cross-sectional area. Similarly with the procedure described above,
matrix X1 is used for the extraction of the amplitude functions k ( x) for all pure
deformation modes, and matrix C is used for the derivatives of the amplitude
functions k ( x ) which are needed for the description of the warping displacements
(see Eq. (1a)). This process was recently described in detail in [12].
Having the amplitude functions and their derivatives, the entire displacement
field (dGBT) is recreated and compared with the initial one (dFE), extracted from
shell FEA. An error vector is constructed derr = dFE dGBT and the approximation
error of the proposed method is measured as the norm of the error vector relative to
the norm of the FE displacement vector (an estimation criterion also used by
Adany et al. [13]):

d errT d err
error = (12)
d FE T d FE
Finally, the method provides the modal identification and participation, the
goal of the entire procedure, using the same formula proposed by Silvestre et al.
[14], an easy and intuitive expression based only on the amplitude functions. The
modal participation factor (Pi) is introduced as follows:

Pi =
L i ( x) dx (13)
k =1 L k ( x) d x
n

Fig. 5 Thin-walled member: continuous and perforated regions.

One goal of this paper is to present the extension of the modal identification
method for the special case of perforated thin-walled members with arbitrary holes.
First, the member is divided in distinct regions along its length as shown in Fig. 5:
continuous regions given by the intervals d, and regions with holes given by the
intervals dh.
These intervals dont have to be equal, there are no theoretical limitations
concerning the shape, size and distribution of the holes. The buckling shell FEA is
performed and the displacement field is extracted for each buckling mode
112 Mihai Nedelcu 8

considered in the analysis. For any cross-section P (0 xP L) inside the


continuous regions, the method described in the previous section can be directly
applied. For the perforated regions, the method can be applied if a minor
modification is performed, taking into account the missing segments along the
cross-section. The cross-sectional deformation modes remain the same, so the GBT
1st step is still applied for the continuous cross-section. The modification is
introduced with respect the stiffness matrices. Basically, the components of these
matrices are given by the virtual work produced by unit values of the amplitude
functions, or their derivatives. So, for each missing segment, the virtual work is
eliminated, meaning the integrals of Eq. (11) are calculated only on the real
cross-sections of the perforated regions.
Concerning the shape and pattern of holes, the simplest case to apply the
modal identification method described in this paper, is for members with
rectangular holes regularly distributed along the members length as presented in
Fig. 6 (left side).

Fig. 6 Thin-walled member with rectangular and circular holes.

In this case there are only two cross-sections that need to be analysed, one for
the continuous (S0) and one for the perforated regions (S1), yielding two sets of
stiffness matrices. However, the method has no theoretical limitations for non-
regular patters and arbitrary shapes of holes. The complications are just of
technical order, a circular hole will force the analysis of more than one cross-
section along the perforation (Sc1..5). This technical detail is currently solved as it
can be see in the illustrative example.
This is not the only technical complication; the circular holes (or rectangular
holes with round corners) will produce an irregular FE mesh. The shell FEA
displacement field is extracted in the FE nodes and in order to apply the mode
decomposition method, the cross-sectional discretisations in shell FEA and GBT 1st
step, have to be identical (an easy task for a regular FE mesh). For an irregular FE
discretisation (Fig. 7), the GBT and FEA discretisations are different and the shell
FEA displacement field has to be adapted by interpolation on the GBT mesh of
points, keeping the approximation errors to a minimum, a task that is currently
under work.
9 Analysis of the effect of holes patterns on coupled instability 113

Fig. 7 Irregular FE mesh for circular holes.

4. ILLUSTRATIVE EXAMPLE

The parametric study is completed on an axially compressed typical storage


rack steel upright with and without holes. Assuming that this lipped channel is a
cold-formed member, the thickness t = 2 mm is constant for all walls. The bars
length is L = 1500 mm, the material properties are: Youngs modulus E = 210GPa
and Poisson's ratio = 0.3 and cross-sectional dimensions are given in Fig. 8. The
member is locally pinned at one end and simple-supported (meaning pinned locally
and globally) and free to warp at the other end. The simple-supported end is axially
compressed by an edge uniform load which gives a resultant force of 1kN. Seven
perforation patterns were analysed as presented in Fig. 8. The first one (R0h) is
without holes, the next 5 display different configurations of rectangular holes, and
the last one displays circular holes. The name of the perforated members with
rectangular holes is constructed as follows: R(from rack) + number of holes
along the cross-section + rh (from rectangular hole) + hole transversal
dimension. The 5th configuration presents 3 rectangular holes, 1 on web, 2 on
flanges, and the 6th configuration presents 5 rectangular holes, 1 on web, 4 on
flanges. All the rectangular holes have the longitudinal dimension of 50mm and the
longitudinal center-to-center hole spacing of D = 90 mm. For circular holes, the
radius is R = 30 mm and the longitudinal center-to-center hole spacing is the same
D = 90 mm. The holes position and dimensions are chosen in such a way that the
same material quantity is removed for each configuration. In this way, the
influence of perforation pattern can be analysed from the point of view of critical
load and also of the pure modes participation provided by the modal identification
method presented in this paper. Local buckling is always preferable knowing that it
has the best post-buckling strength reserve. From this point of view the Global
buckling is the most unfavourable case, and the Distortional buckling is the
intermediate case.
114 Mihai Nedelcu 10

Fig. 8 The analysed perforation patterns.

Buckling shell FEA of the thin-walled members were performed using


ABAQUS [15] for the first 50 eigenvalues. Rectangular S4 shell finite elements
were considered in a highly regular mesh, size LFE = 10 mm in longitudinal
direction, with the exception of the member with circular holes. For the last one, a
special application was developed in MATLAB [16] which creates the Abaqus
input file (inp extension) and in which the FE discretisation is introduced as regular
as possible (see the last buckling shape of Fig. 13) otherwise it is quite difficult not
to obtain complicated FE meshes as given in Fig. 7. For the continuous regions the
cross-sectional discretisation is constant along the member length and identical
with the one used by the GBT cross-sectional analysis (Fig. 8): 2, 5-2-5 and 11
intermediate nodes between the corners, respectively in the flange lips, flanges and
web. For perforated regions, the rectangular holes require the analysis of an extra
cross-section the stiffness matrices are modified as explained in the previous
section. The circular holes of radius R = 30 mm, required the analysis of R/ LFE = 3
extra cross-sections (due to symmetry with respect the holes centre).
The GBT cross-sectional analysis performed on the continuous cross-section
provides a number of n = 51 pure deformation modes: 4 Global, 6 Distortional and
41 Local (the first 22 modes were presented in Fig. 3). In order to implement the
11 Analysis of the effect of holes patterns on coupled instability 115

described method of buckling identification and to calculate the modal participation,


another MATLAB application was written. It uses the FE displacement field of 151
cross-sections given by the shell FE sides normal to the member longitudinal axis.
Method validation. The analysis provides the pure modes participation based
on Eq. (13) for the first 50 general buckling modes provided by shell FEA and also
the approximation error based on Eq. (12). Fig. 911 present these results for
member configurations R0h (max. error = 1.32%), R1rh60 (max. error = 2.54%)
and R1ch30 (max. error = 5.93%).
For all seven configurations, the maximum approximation error was the last
one (5.93% for circular holes), buckling mode no. 38 caused by the local buckling
of a very small region of the member. The higher value of the approximation error
is due mainly to the simplifying assumptions introduced by the conventional GBT,
especially the Vlasovs hypotheses and the linear variation of the warping
displacements u(x,s) according the s axis. Refining the longitudinal mesh and the
GBT cross-sectional discretisation improves the accuracy of the presented method
but not to drastic values (detailed explanations of this phenomenon are given in
[17]). Nowadays GBT was extended in order to handle non-linear variation of the
warping displacements along the cross-section wall midline, and also can take into
account the cross-section deformations due to the wall transverse extensions. New
pure modes are introduced, namely shear modes and transverse extension modes.
The implementation of these two features will further improve the precision of the
presented method, and is currently under work.

Fig. 9 Member R0h: pure modes participation and approximation errors.


116 Mihai Nedelcu 12

Fig. 10 Member R1rh60: pure modes participation and approximation errors.

Fig. 11 Member R1ch30: pure modes participation and approximation errors.


13 Analysis of the effect of holes patterns on coupled instability 117

The effect of the perforation pattern. For all seven configurations, the 1st
general buckling mode is analysed. Figure 12 presents the effect of the perforation
pattern based on two factors: the pure modes participation and the critical load.

Fig. 12 1st general buckling mode: modal identification, critical load


and approximation errors.

Figure 13 presents the deformed shape of the 1st shell FEA buckling mode for
all seven configurations, together with the corresponding normalised amplitude
functions ( x ) of the first most significant pure modes based on their participation
provided by the presented method. The loaded and simply-supported end is at the
left side.
It can be concluded from the numerical and visual results given in the last
5 figures that for this member, by placing only one large hole in the middle of the
cross-section will highly favour the Distortional buckling and the critical load will
actually increase in comparison with the unperforated member, a phenomenon
already reported in literature (e.g. [18]). The Local deformations decrease, because
the longitudinal normal stresses are directed through the top and bottom regions of
the web, which are much stiffer due to their interaction with the flanges. In the
same time the Distortional buckling decreases the post-buckling strength reserve.
By placing 2 or 3 holes on the web will produce almost only Local buckling of the
members, due to the unstiffened web regions between the holes, but the critical
load is drastically decreasing. Placing a small-medium hole in the web leads to a
Local-Distortional coupled instability.
118 Mihai Nedelcu 14

Fig. 13 1st FEM buckling mode and the amplitudes


of the most relevant pure deformation modes.

One application of the presented method could be, in future, as follows:


having the coupling quantified, the user can try other perforation patterns to find
the optimal configuration which will provide the greatest participation of the Local
buckling for an acceptable value of the critical load. Of course this procedure must
be validated by geometric and material nonlinear collapse analyses (capable to
provide the real response and the post-buckling behaviour of the member) also
using the modal identification method described in this paper, a study that is
currently under work.
15 Analysis of the effect of holes patterns on coupled instability 119

4. CONCLUSIONS

This paper presents the latest developments of a GBT-based method


(originally developed in [10]) capable of identifying in a general buckling mode
provided by shell FEA, the modal participation of the pure deformation modes of
Global, Distortional and Local type. This modal identification is a valuable
quantitative tool for assessing coupled instabilities and the effect of perforation
patterns. Its high speed must also be mentioned, and depends almost entirely of the
time required by the shell FEA. For all these reasons, the proposed method is
elegant, extremely fast and a promising candidate to be combined with a general
purpose finite element code, providing in this way the buckling mode identification
for arbitrary thin-walled members with and without holes.

Receveid on July 16, 2014

REFERENCES

1. SCHARDT, R., Verallgemeinerte Technische Biegetheorie (in German), Springer-Verlag, Berlin,


1989.
2. SCHARDT, R., Generalized beam theory an adequate method for coupled stability problems,
Journal of Thin-Walled Structures, 19, pp. 161-180, 1994.
3. DNY, S., SCHAFER, W. B., A full modal decomposition of thin-walled, single branched open
cross-section members via the constrained finite strip method, Journal of Constructional Steel
Research, 64, pp. 1229, 2008.
4. BASAGLIA, C., CAMOTIM, D., SILVESTRE, N., Non-linear GBT formulation for open-section
thin-walled members with arbitrary support conditions. Journal of Computers and Structures,
89, 2122, pp. 19061919, 2011.
5. LI, Z., SCHAFER, B.W., Buckling analysis of cold-formed steel members with general boundary
conditions using CUFSM: Conventional and constrained finite strip methods, 20th International
Specialty Conference on Cold-Formed Steel Structures Recent Research and Developments
in Cold-Formed Steel Design and Construction, 2010.
6. BEBIANO, R., PINA, P., SILVESTRE N. and CAMOTIM, D. , GBTUL Buckling and Vibration
Analysis of Thin-Walled Members, DECivil/IST, Technical University of Lisbon (http://www.
civil.ist.utl.pt/gbt), 2008.
7. *** Elastic Buckling Analysis of Thin-Walled Members by Finite Strip Analysis, CUFSM
(http://www.ce.jhu.edu/bschafer/cufsm/).
8. DAVIES, J. M., LEACH, P., TAYLOR, A., The design of perforated cold-formed steel sections
subject to axial load and bending, Journal of Thin-Walled Structures, 29, pp. 141157, 1997.
9. MOEN, C., SCHAFER, W. B., Direct strength method for design of cold-formed steel columns
with holes, Journal of Structural Engineering, 137, 5, pp. 559570, 2011.
10. NEDELCU, M., GBT-based buckling mode decomposition from finite element analysis of thin-
walled members, Journal of Thin-Walled Structures, 54, pp. 156163, 2012.
11. BEBIANO, R., SILVESTRE, N., CAMOTIM, D., GBT formulation to analyze the buckling
behaviour of thin-walled members subjected to non-uniform bending, Journal of Structural
Stability and Dynamics, 7, 1, pp. 2354, 2007.
120 Mihai Nedelcu 16

12. NEDELCU, M., CHIRA N., CUCU, H.L., POPA A.G., Buckling mode decomposition of thin-
walled members with holes, 5th International Conference on Structural Engineering, Mechanics and
Computation, Cape Town, South Africa, 2-4 September 2013.
13. DNY, S., JO, A., SCHAFER, W. B., Buckling modes identification of thin-walled members
using cFSM base functions, Journal of Thin-Walled Structures, 48, pp. 806817, 2010.
14. SILVESTRE, N., CAMOTIM, D., Second-order generalised beam theory for arbitrary
orthotropic materials, Journal of Thin-Walled Structures, 40, pp. 791820, 2002.
15. *** HIBBIT, KARLSSON & SORENSEN INC. ABAQUS Standard (Version 6.3), 2002.
16. *** MATLAB Version 7.1.0246 Documentation, The Mathwork Inc., 2005.
17. NEDELCU, M., CUCU H. L., Buckling modes identification from fea of thin-walled members
using only GBT cross-sectional deformation modes, Journal of Thin-Walled Structures,
Article in Press, 2013.
18. MOEN C., SCHAFER, W. B., Elastic buckling of cold-formed steel columns and beams with
holes, Journal of Engineering Structures, 31, 12, pp. 28122824, 2009.
IMPERFECTION SENSITIVITY
OF THIN PLATES LOADED IN SHEAR

FRANC SINUR1, PRIMO MOE1, KLEMEN REJEC2,


GAPER LUAR3, DARKO BEG1

Abstract. The paper deals with imperfection sensitivity analysis of longitudinally


stiffened thin plates loaded in shear. The aim of this paper is to demonstrate the use of
optimization method for direct determination of the worst imperfection shape. New
linear constraints are considered in optimization method in order to find not only the
worst but also realistic imperfection shape. Further on, different solutions of arbitrary
shapes that are used to estimate the worst imperfection are introduced and compared.
Finally, a parametric analysis seeking the worst imperfection shape is performed and
discussed.

Key words: plate buckling, worst initial imperfections, longitudinal stiffeners, sensitivity
analysis, GMNIA.

1. INTRODUCTION

The increased capacity of computers and the possibilities of numerical tools


available today facilitate the use of advanced calculation models in everyday
design practice. Thus, in design of structures the use of material and geometrical
nonlinear analysis considering imperfections to design structures is growing. Such
analysis allows a detailed insight in the behaviour of structures or structural
elements. The design approach, however, requires a lot of experience in order to
build a convenient and proper numerical model and to interpret the results
correctly. The initial boundary conditions, constitutive material laws and initial
imperfections have an important influence on the behaviour and resistance of the
structure or structural element under consideration. At the University of Ljubljana,
Chair for Metal Structures, an advanced optimization method for direct determination
of the most unfavourable imperfection of the structural element was developed [1, 2].
This optimization method, based on direct and sensitivity analysis and optimization
algorithms, was implemented on thin plates subjected to shear stresses. The

1
University of Ljubljana, Faculty of Civil and Geodetic Engineering, Slovenia
2
University of Ljubljana, Faculty of Civil and Geodetic Engineering, Slovenia
3
Brogemeinschaft Kuhlmann Gerold Gnther Eisele, Germany
122 Franc Sinur, Primo Moe, Klemen Rejec, Gaper Luar, Darko Beg 2

optimization function is determined as the minimum value of the load proportional


factor (LPF), which is a function of the imperfection shapes (buckling modes,
Fourier terms, etc.), and it presents the worst imperfection shape for the structure.
The paper presents an extended study, which deals with the search for the
most unfavourable imperfection shape for representative slender plates (stiffened
plates) subjected to pure shear. The constraints of the optimization process were
determined by the maximum allowable imperfection amplitude obtained from the
production tolerances and design rules for plated structural elements [35] and by
the maximum curvature of the imperfect geometry. Special attention was devoted
to the constraint of the maximum curvature.

2. METHODOLOGY FOR THE DETERMINATION


OF THE MOST UNFAVOURABLE IMPERFECTION

Initial imperfections are the consequence of the production process and


cannot be avoided. When dealing with slender elements, i.e. thin plates subjected to
compressive stresses, the results of geometrical material nonlinear analysis considering
imperfections strongly depend on the imperfection shape and its amplitude. Thus, it
is important to consider the most unfavourable actually possible imperfection
shape, which will lead to safe and reliable assessment of the resistance of the
analysed structural element.
A computationally less expensive optimization method that retains the
generality of the optimization based definitely worst imperfection approach was
developed by Kristani and Korelc [1, 2, 6]. The basic idea of the approach is to
replace the nonlinear optimization problem with an iterative procedure that involves
only linear optimization problems. Within the iteration the objective function for
the minimum ultimate load is constructed by means of a fully nonlinear direct and
first order sensitivity analysis. Constraints on the shape and the amplitude of the
imperfections are taken into account. When carefully constructed, they remain
linear, therefore enabling the use of efficient and readily available linear program-
ming algorithms for obtaining the solution of the corresponding optimization
problem.

2.1. REPRESENTATION OF IMPERFECTIONS

The applied initial imperfection shape with specified amplitudes has to


represent a change in the geometry of a structure in the most unfavourable way so
that the ultimate load of the imperfect structure is as small as possible. The
imperfections are represented as a linear combination of the chosen basic shapes
within maximum out-of-plane amplitude e0 prescribed by the principle of equivalent
geometrical imperfections. Equivalent geometrical imperfections include geometrical
3 Imperfection sensitivity of thin plates loaded in shear 123

and structural imperfections. Geometrical imperfections can be augmented to include


structural imperfections that are not included directly into the finite element model.
Structural imperfections arise from the manufacturing method; e.g. residual stresses
produced by welding. The geometry of an imperfect structure X is defined by:

j=1 j j ,
N
X= Xp + (1)

where X p is the initial perfect geometry, j are the unknown shape parameters
and j are the base shapes. The unknown shape parameters j are obtained as the
solution of the optimization problem. The base shapes can be chosen arbitrarily,
but they have to be linearly independent in order to have a well-defined minimum
of the corresponding optimization problem. The overall numerical efficiency of the
procedure strongly depends on the number of the base shapes (N).

2.2. DESCRIPTION OF THE ALGORITHM

In the presented approach a fully geometrically and materially nonlinear


analysis is used. When dealing with thin-walled structures with moderate thickness,
it is necessary to take into account geometrical and material nonlinearity. Since the
algorithm starts from the beginning with imperfect structure, bifurcation points
usually do not occur prior reaching the limit point in the load-deformation curve.
Within this method the most unfavourable initial imperfection shape is
sought, defined by shape base and shape parameters at which the ultimate load
will be the lowest [1]. Unknown shape parameters are evaluated iteratively by an
optimization process. The iterative procedure for the kth step can be written as:
X k = X k 1 + X k

i =1 ik i ,
N
X k =
(2)
ik = ik 1 + ik

i =1 ik i
N
Xk =

where Xk is the imperfect geometry, ik the increment of the imperfection


parameters, Xk the increment of the imperfection and Xk the total imperfection.
The increment of the imperfection parameters in the kth iteration ik is obtained as
a solution of the corresponding optimization problem [1, 2]. The flowchart of the
method is illustrated in Fig. 1. The algorithm starts with the first base shape 1,
normalized by amplitude e0, as the initial guess X0 for the geometry of the
imperfect structure:
124 Franc Sinur, Primo Moe, Klemen Rejec, Gaper Luar, Darko Beg 4

e0
i =1
i0 = 0; i0 = max i
0 (3)
i 1
X 0 = X p + 10 1

and then improves the solution by solving a sequence of optimization problems


until the convergence condition ik < tolerance is reached. Within each step of
the iterative procedure a fully nonlinear direct and sensitivity analysis of the
structure with imperfect geometry Xk is performed, followed by the formulation
and solution of the optimization problem.

Fig. 1 Flowchart of the method for the determination


of the most unfavourable imperfection [1].
5 Imperfection sensitivity of thin plates loaded in shear 125

2.3. SHAPE BASE

The shape base is used to find the most unfavourable initial imperfection.
The obvious choice are the buckling modes of the structure obtained by the initial
buckling analysis. Alternative and cheaper to evaluate are the eigenvectors of
initial elastic tangent matrix. Other possible shapes are deformation shapes of the
structure in elastic and plastic range and arbitrarily defined shapes. In this work
only the arbitrarily defined shapes were used.

2.4. ARBITRARILY DEFINED SHAPES

Shape base defined in terms of fourier series


The base shapes were defined as two-dimensional sine functions in order to
keep all edges straight, without any imperfection. The only parameter that defines
the considered shapes is the number of terms of Fourier series that was considered.
The shapes were defined as:
S1 S 2 = {s1 s 2 ; s1 S1 , s 2 S 2 } , (4)

ix
S1 = sin ; i = 1, 2, ... n , (5)
A
iy
S 2 = sin ; i = 1, 2, ... m , (6)
B
where A and B represent the length and the width of the analysed plate, n defines
the number of half sine waves in x and m in y direction, respectively. The number
of shape base is equal to product nm. The base shapes defined according to
expression (4) for n = m = 3 are illustrated in Fig. 2.

Specially defined base shapes


Another type of base shapes called specially defined shapes were considered
in the analysis. The shapes were defined with sine function, where one half wave is
assumed in transverse direction and n waves in longitudinal direction. The number
of shapes in transverse direction depends on the number of longitudinal stiffeners,
e.g. for two longitudinal stiffeners the possible number of shapes is 6 n, where
6 corresponds to 6 possible shapes in transverse direction, as presented in Fig. 3.
For shape no. 1 the length of the half sine wave is equal to the width of plate B.
Shapes no. 2 and 3 correspond to wave length 2B/3 (as the stiffeners are equally
spaced); it means that the wave is stretched over two subpanels with the maximum
amplitude located at the position of the longitudinal stiffener. For the last 3 shapes,
shapes no. 4, 5 and 6, the shape is defined with wave length B/3 and is positioned
over one subpanel, while the other subpanels remain straight.
126 Franc Sinur, Primo Moe, Klemen Rejec, Gaper Luar, Darko Beg 6

Fig. 2 Base shapes for m = n = 3.

Shape 1 Shape 2 Shape 3

Shape 4 Shape 5 Shape 6

Fig. 3 Specially defined shapes for n = 1.


7 Imperfection sensitivity of thin plates loaded in shear 127

2.5. NUMERICAL MODEL

The numerical model consists of simply supported plate, longitudinal


stiffeners and three stiff bands (Fig. 4). The boundary conditions were as follows:
all four edges were simply supported; the left edge was restrained also in the
longitudinal direction. The vertical supports in the other three edges were defined
in nodes between base plate and stiff band. Additional stiff band was modeled in
edges in order to simulate the same conditions as in the left edge, to keep the edge
undeformed in its longitudinal direction. The load was applied as shear line load
along edges, as presented in Fig. 4. The amplitude of the load was set to plastic
shear resistance of the plate.

Fig. 4 Numerical model boundary conditions and load application.

3. LINEAR CONSTRAINTS

By using optimization approach to find the most unfavourable imperfection


shape, the linear constraints in terms of maximum imperfection amplitude have to
be defined. Another parameter which significantly influences the solution for the
most unfavourable imperfection is, as already mentioned, the maximum curvature.
These two parameters were the key terms that influenced the solution for the most
unfavourable imperfection shape defined by using optimization approach.
When numerical tools are used to design a structure, in most cases equivalent
geometric imperfections are used. They take into account both geometric as well as
structural imperfection (residual stresses) by increased amplitude of geometric
imperfection. Annex C of EN 1993-1-5 provides recommendations on possible
types of initial imperfection shapes and amplitudes that should be accounted for in
FEM design of thin plated structural elements.
The amplitudes of the imperfections were limited according to the EN 1993-1-5
specifications, while the influence of the curvature limit was studied within the
study presented in this paper, and based on it the maximum allowed curvature was
defined.
It has already been demonstrated [1, 7] that out-of-plane displacement constraint
only might lead to initial imperfections that are unrealistic and consequently very
128 Franc Sinur, Primo Moe, Klemen Rejec, Gaper Luar, Darko Beg 8

conservative due to high reduction of the ultimate resistance. Thus, an additional


constraint, such as maximum curvature, should be considered in the optimization
process. Sensitivity analysis was performed in order to establish the influence of
the curvature amplitude on the behaviour and the resistance of the plate subjected
to in-plane stresses and to define the constraints for parametric study.
In Table 1 the maximum amplitudes for global and local imperfections taken
from EN 1993-1-5 (Annex C) are given, where nst corresponds to the number of
longitudinal stiffeners. The maximum amplitude of the imperfection is defined as:
e0 = eG + 0.7e L , (7)
with global imperfection eG as the leading and local eL as the accompanying
imperfection. The simplest function which was used to generate initial imperfection
was sine function with the maximum out-of-plane amplitude and consequently also
the maximum curvature at the mid span. For this imperfection shape the curvature
is very small and could be too strict for real cases. A series of extra functions was
defined; they are similar to sine function with maximum amplitude shifted to xmax =
2/5A, 1/3A and 1/4A, where A is the panel length. These functions were defined to
get different values of the curvature, while the maximum out-of-plane amplitude
remained the same. The functions were defined as a 4th order polynomial:
f ( x) = ax 4 + bx 3 + cx 2 + dx + e . (8)
Table 1
Maximum imperfection amplitudes defined in EN 1993-1-5

A B
Global imperfection amplitude eG min ,
400 400
A B
Local imperfection amplitude eL min ,
200 ( nst + 1) 200

Coefficients a, b, c, d and e were defined to fulfil the following conditions:


f(0) = 0, f(x = a) = 0, f(x = xmax) = emax, f(x = xmax) = 0 and f(x = a)= /a. The
last condition arises from the first derivation if sine function is evaluated at x = A.
5

200 400 600 800 1000

th
Fig. 5 4 order polynomials with maximum at 1/4, 1/3, 2/5 and 1/2 of the span.
9 Imperfection sensitivity of thin plates loaded in shear 129

It is obvious that the maximum curvature will be obtained for the curve with
the largest shift of the maximum amplitude position, i.e. at x = A/4 (Fig. 5). The
upper bound for the maximum curvature is defined with the curvature obtained for
elastic bending moment, when the first fibre starts to yield. This value is constant
for all lengths of the plate and depends solely on the plate thickness t:
2 y
yield = . (9)
t
In Fig. 6 the maximum curvature for the selected 4th order functions and the
upper bound curvature for three different plate thicknesses are plotted (t = 10 mm,
20 mm and 30 mm). The maximum curvature among the given functions is found
for the one with maximum amplitude at x = A/4, which is also evident from Fig. 5.
The upper bound yield depends only on the plates thickness; therefore, the
curvature does not change due to the change of the panel size. For a plate with
thickness t = 20 mm, yield is smaller than the curvature obtained with function with
maximum at x = A/4, when panel length is less than A = 600 mm. Therefore, up to
this point the maximum curvature is limited with the upper bound limit yield and
from this point forward the limit is defined from maximum curvature that is
defined with function with maximum at x = a/4.

Fig. 6 Curvature as a function of panel length a [mm].

The influence of the curvature amplitude on the most unfavourable


imperfection shape solution and consequently on LPF for longitudinally stiffened
plate (two cases: one and two longitudinal stiffeners) subjected to pure shear is
gathered in Table 2. By increasing the maximum allowed curvature the resistance
of the plate drops, especially when the maximum curvature is increased from the
130 Franc Sinur, Primo Moe, Klemen Rejec, Gaper Luar, Darko Beg 10

curvature of the function with maximum at x = A/4 to yield and 2yield. Further on,
the imperfection shape is very smooth for stricter curvature limit and vice versa,
the imperfection shape becomes very rough when the curvature limit is set higher.
This is a good reason why the maximum curvature should also be applied; it
influences the imperfection shape in a way for it to become more realistic. For the
studied cases a curvature limit defined with a function with maximum at x = A/4 is
found to be a reasonable limit for the constraint because:
The obtained imperfection shape is smooth.
The resistance drop induced by higher curvature limits is too large and
results in excessively conservative values of LPF.
The most unfavourable imperfection shape given in Table 2 proves that free
edges of the numerical model are not straight. This is due to the fact, that the last
line of the elements represents stiff band. The plate is straight where the stiff band
is connected to the base plate which can also be identified from the figures bellow.

Table 2
Comparison of the most unfavourable imperfection shapes for different curvature constraints
f(x=a/4)=emax 1Yield 2Yield
max. curvature
0.000155 0.000338 0.000676
[1/mm]
LPF 0.844 0.790 0.766
1 stiffener

Most unfavourable
imperfection

max. curvature
0.000161 0.000338 0.000676
[1/mm]
LPF 0.877 0.837 0.828
2 stiffeners

Most unfavourable
imperfection

4. NUMERICAL ANALYSIS REQUIRED NUMBER OF BASE SHAPES

The most unfavourable imperfection was sought for plates stiffened with one,
two and four longitudinal stiffeners (Fig. 7). The material was modelled as bilinear
without any strain hardening. The yield strength was set to fy = 355 MPa, elastic
modulus to E = 210 000 MPa and Poisons ratio to = 0.3.
11 Imperfection sensitivity of thin plates loaded in shear 131

Fig. 7 Geometry and dimension notations of studied stiffened plates.

First, the sensitivity analysis to determine the required number of base shapes
and to define the mesh density was performed. This analysis was done for all three
configurations of the plate (1, 2 and 4 stiffeners) with panel aspect ratio =1, global
panel slenderness B/t = 200 and normalized flexural stiffness of stiffener = Isl / Ip = 16.4,
where Isl is the flexural stiffness of the stiffener taking into account the effective
width of the plate and Ip is the flexural stiffness of the plate itself.

4.1. MESH DENSITY

Sensitivity analysis was performed for four different mesh densities: M10,
M20, M40 and M80. The number corresponds to the number of elements used
along the plates length A. In Fig. 8 the influence of mesh density on the LPF factor
is shown. Each curve corresponds to one of the plates configurations. On y-axis
the LPF factor obtained is normalized to the LPF factor obtained for mesh density
M10, while on x-axis the mesh density is plotted. The mesh density M40 was found
to be appropriate for further parametric study as no significant drop of LPF was
obtained with a more dense mesh.

1.00
0.95
LPFMx / LPFM10

0.90
0.85
0.80
0.75
one stiffener
0.70
two stiffeners
0.65 four stiffeners
0.60
0 20 40 60 80
Mesh density

Fig. 8 The influence of mesh density on LPF.

4.2. MINIMUM NUMBER OF BASE SHAPES

As the computational time significantly increases with the growing number


of base shapes, the minimum number of base shapes that gives a sufficiently
132 Franc Sinur, Primo Moe, Klemen Rejec, Gaper Luar, Darko Beg 12

reliable and accurate solution needs to be defined. The analysis was performed for
base shapes defined in terms of Fourier series and for specially defined base shapes
presented in Chapter 2.
The influence of the number of base shapes considered in the analysis is for
plate with one and four longitudinal stiffeners shown in Fig. 9. The minimum
required number of base shapes depends on the number of longitudinal stiffeners.
Fewer shapes are needed for a plate with one stiffener than for a plate with four
stiffeners. The results are presented for both forms of initial shapes; defined as
Fourier terms and specially defined shapes. In Fig. 9 the legend notations NFouRed = x
describe how many terms of Fourier series are considered. For instance, NFouRed = 4
means that the first 4 terms of Fourier series are used to define base shapes as
described in Chapter 2. This results in 42 = 16 shapes. Similarly, label NGlobObl = x
provides information on the number of specially defined shapes in the longitudinal
direction. The shapes in transverse direction are defined in Chapter 2 and depend
on the number of longitudinal stiffeners (3 for one stiffener, 6 for two stiffeners
and 15 for four stiffeners). NGlobObl = 4 means that in longitudinal direction
functions with 1, 2, 3 and 4 half waves were considered and therefore for a plate
with four longitudinal stiffeners 415 = 60 base shapes are obtained.
For a plate stiffened with one longitudinal stiffener at least 16 base shapes
(NFouRed = 4) are required to get a reliable and accurate solution, which does not
change significantly if higher terms of Fourier series are used. For specially
defined shapes the minimum required number of base shapes is 9 (NGlobObl = 3).
As already mentioned, a higher number of base shapes is required for plates with
larger number of longitudinal stiffeners. This is especially evident when special
types of base shapes are used. For 4 stiffeners the minimum number of shapes that
has to be considered is 45 (NGlobObl = 3), whereas only 16 (NFouRed = 4) shapes
are needed if the shapes are defined by using combinations of Fourier terms. If,
however, more than 50 base shapes were considered, problems with convergence
occurred in the analysis.
Based on the presented sensitivity analysis, the minimum number of base
shapes was defined for the following parametric study. The most unfavourable
imperfection shape was sought by using base shapes defined as Fourier terms (16
shapes considered, NFouRed = 4), and specially defined shapes ( 3 i =1 i considered
n _ st

shapes, where n_st is the number of longitudinal stiffeners, NGlobObl = 3).


The most unfavourable imperfection shape for plate with 4 longitudinal
stiffeners is shown in Fig. 10. Increased number of shapes considered (NFouRed = 8
and NGlobObl = 4) does not produce any different imperfection shape than that
obtained with NFouRed = 4 and NGlobObl = 3.
13 Imperfection sensitivity of thin plates loaded in shear 133

TERMS OF FOURIER SERIES SPECIALLY DEFINED SHAPES


1 stiffener 1 stiffener

4 stiffeners 4 stiffeners

Fig. 9 LPF factor obtained with different number of base shapes included in analysis.

Imperfection shapes defined with Fourier Terms

NFouRed=2 NFouRed=4 NFouRed=8

Imperfection shapes defined with specially defined shapes

NGlobObl=2 NGlobObl=3 NGlobObl=4

Fig. 10 The most unfavourable imperfection shapes


obtained with different numbers of base shapes considered.
134 Franc Sinur, Primo Moe, Klemen Rejec, Gaper Luar, Darko Beg 14

4.3. PARAMETRIC STUDY


A parametric study was performed to determine the influence of the plates
slenderness and the stiffness of longitudinal stiffeners on the most unfavourable
imperfection shape. The varied parameters within were:
Number of longitudinal stiffeners n = 1, 2, 4.
Slenderness of subpanel b/t = 35, 50, 75, 100.
Stiffness of longitudinal stiffeners = 18, 35, 110.
Three different stiffnesses of longitudinal stiffener were studied. The stiffener
with stiffness = 35 ensures that elastic global buckling of the subpanel is equal to
elastic global buckling of the entire stiffened panel. In all cases the initial
imperfection shape 1 was defined as a combination of global G and local L
imperfection shape as recommended in EN 1993-1-5.
In Fig. 11 the results obtained for plate stiffened with one stiffener and
subpanel slenderness b/t = 75 are presented. The main results that were obtained
and will be discussed are the initial imperfection and the final deformation shape as
well as the LPF. The initial imperfections are compared to the deformed shape
obtained at maximum LPF.
For all defined most unfavourable imperfection shapes that were calculated
using shapes based on Fourier terms and also on specially defined shapes the LPF
was always smaller compared to the LPF calculated with imperfections determined
according to the EN 1993-1-5 recommendations (combination of global and local
imperfections G + 0.7 L). Smaller LPF was found for imperfection shapes
defined with Fourier terms. The imperfection shape defined with Fourier terms is
smoother and very similar to the final deformation shape. Three buckling waves
with orientation in the direction of tension field formation characterize the worst
imperfection shape found with base shapes based on Fourier terms.
The difference of this LPF compared to the LPF obtained with the EN 1993-1-5
imperfection shape is much smaller if the worst imperfection shape is calculated
with specially defined shapes. The imperfection shapes obtained are much more
diverse with several waves. Tendency to global deformation shape was observed
meaning that the imperfection shape approaches to the deformation shape if pool of
specially defined shapes is increased. However, the number of the base shapes
considered in this case was not high enough to get similar worst imperfection shape
as with Fourier terms.
The results for plate stiffened with two longitudinal stiffeners and subpanel
slenderness b/t = 50 are shown in Fig. 12. Also in this case the worst imperfection
shape found is similar to the deformation shape. In contrast with the previous case,
the similarity between the imperfection shape and the deformation shape is found
also for the solution based on specially defined shapes. The LPF factor is, in
comparison to the LPF obtained with the EN 1993-1-5 imperfections (values in
brackets) in all cases smaller, where the difference varies from 3.6% to 6.9%. The
results again prove that the worst imperfection shape is similar to the deformation
shape. Similar conclusions were observed also for other parameters that were
analysed within the parametric study.
15 Imperfection sensitivity of thin plates loaded in shear 135

Shapes based on Fourier terms Specially defined shapes


Imperfection shape Deformation shape Imperfection shape Deformation shape
18

LPF=0. 797 (-11.3%) LPF=0.871 (-3.1%)


35

LPF=0.838 (-9.5%) LPF=0. 894 (-3.5%)


110

LPF=0.859 (-8.9%) LPF=0.913 (-3.2%)


Fig. 11 Comparison of the most unfavourable imperfection and deformed shape
for plate stiffened with one stiffener, b/t = 75.
Shapes based on Fourier terms Specially defined shapes
Imperfection shape Deformation shape Imperfection shape Deformation shape
28

LPF=0.928 (-6.9%) LPF=0. 931 (-6.6%)


55

LPF=0. 939 (-5.9%) LPF=0.948 (-5.0 %)


165

LPF=0.955 (-4.4%) LPF=0.963 (-3.6%)


Fig. 12 Comparison of the most unfavourable imperfection
and deformed shape for plate stiffened with two stiffeners, b/t = 50.
136 Franc Sinur, Primo Moe, Klemen Rejec, Gaper Luar, Darko Beg 16

5. CONCLUDING REMARKS

The worst imperfection shape to obtain minimum resistance can be defined as


superposition of different shapes. With the performed study we tried to answer the
following questions:
how the arbitrary shapes should be defined in order to get consistent and
reliable results,
how to find, next to maximum out-of-plane displacement, an additional
linear constraint that will provide more realistic imperfection shapes and
is there a simple way to define the worst imperfection shape?
It has been shown that the imperfection shape defined as a superposition of
Fourier terms results in imperfection shapes with the lowest LPF and that the worst
imperfection shape is smooth and similar to the deformed shape. The imperfection
shape estimated with specially defined shapes gives a higher value of LPF and the
corresponding shape is not as smooth as the one gained with terms of Fourier
series. Therefore, the shapes defined with Fourier series are found to be more
appropriate in the process of seeking the worst imperfection shape.
In the presented work an additional parameter, i.e. the maximum curvature of
the plate, was introduced and analysed in order to influence the solution of the
optimization method when seeking the worst imperfection shape. The main
objective of the maximum curvature constraint is to influence the imperfection
shape in a way that the final result is a realistic imperfection shape. In this step of
the analysis engineering judgment is required to assess whether the imperfection
shape is realistic or not. The maximum curvature limit was defined with a shift of
the maximum amplitude from the mid span of the plate to of the plate span with
maximum out-of-plane amplitude defined according to EN 1993-1-5. For very
short plates the maximum curvature was limited to yield.
One of the main observations, which should be further verified, is that the
worst imperfection shape could be defined as deformed shape evaluated at the
maximum LPF. Of course, this deformed shape should fulfil the following limits:
out-of-plane imperfection amplitude and maximum curvature amplitude, otherwise
the imperfection shape will significantly reduce the resistance of the element.
In many cases the 1st buckling mode is found as the worst imperfection, but
this is not always the case. When searching for ultimate resistance of the structure
it is a good practice to perform initial imperfection analysis with several buckling
modes. But this might be time consuming; therefore, it would be enough to
estimate the resistance with the 1st buckling mode and with deformation shape as it
was found to be the worst initial imperfection.

Received on July 16, 2014


17 Imperfection sensitivity of thin plates loaded in shear 137

REFERENCES

1. KRISTANI, N., Limit State Design Using Exact Sensitivity Analysis and Shape Optimization,
PhD Thesis, University of Ljubljana, Faculty of civil and geodetic engineering, Ljubljana, 2008.
2. KRISTANI, Niko, KORELC, Joe, Optimization method for the determination of the most
unfavorable imperfection of structures, Comput. Mech., 42, 6, pp. 859872, 2008.
3. *** CEN, Eurocode 3: Design of steel structures. Part 15: Plated structural elements (EN 1993-1-5),
European Committee for Standardisation, Brussels, 2006.
4. BEG, D., KUHLMANN, U., DAVAINE, L., BRAUN, B., Design of Plated Structures. Eurocode
3: Design of Steel Structures. Part 15: Design of Plated Structures, 1st edition, European
Convention for Constructional Steelwork, Berlin, Ernst & Sohn, Brussels, 2010.
5. JOHANSSON, B., MAQUOI, R., SEDLACEK, G., MLLER, C., BEG, D., Commentary and
worked examples to EN 1993-1-5 Plated structural elements, 1st edition, Office for Official
Publications of the European Communities, Joint Research Centre European Commission,
Luxembourg, 2007.
6. KORELC, J. AceFEM, Mathematica finite element environment, University of Ljubljana, Faculty
of Civil and Geodetic Engineering, Ljubljana, 2010.
7. KUHLMANN, U., BEG, D., ZIZZA, A., SINUR, F., REJEC, K., Tragverhalten von Blechen mit
Lngssteifen unter Interaktion von Biegemoment und Querkraft (in German), Research Report,
Universitt, Institut fr Konstruktion und Entwurf, Stuttgart, 2012.
PART II: DESIGN CODIFICATION ORIENTED STUDIES

LATERAL TORSIONAL BUCKLING RESISTANCE OF


CASTELLATED BEAMS

LSZL DUNAI, BALZS KVESDI, DVID WISCHY, BARNABS BZA*

Abstract. Castellated beams are likely used structural elements in building structures
especially in deck systems and in steel frames. The castellated beams are usually
produced by cutting a traditional I-girder and welded by creating large openings in the
web. The web openings can reach 7080% of the whole web depth, therefore the
effect of the holes on the structural behavior and in the design of castellated beams is
not negligible. The openings have influence on the shear and on the lateral torsional
buckling resistances. Most of the previous investigations in this topic were focusing
on the shear buckling resistance determination and there is only a few which dealing
with the lateral torsional buckling resistance. Due to the reduced out-of-plane stiffness
of the web, however, different failure modes can be relevant for castellated beams
comparing to traditional I-girders. In the paper the lateral torsional buckling resistance
of castellated beams is studied by advanced numerical model and analysis. The lateral
torsional buckling resistances are determined for different specimen geometries
covering a wide parameter range. Based on the numerical investigations enhanced
design methods are proposed to determine the lateral torsional buckling resistances.

Key words: castellated beam, lateral torsional buckling resistance, web distortion.

1. INTRODUCTION

Castellated beams are likely used in floor and roof systems of building
structures and as main girders of steel frames, as illustrated by the examples of Fig. 1.
The holes in the web are usually produced by cutting of a hot-rolled I-girder and
welded again by increasing the web depth of the original girder. The bending
stiffness can be significantly increased by this manufacturing process without any
increase in the self-weight of the girder. This fact makes the application of
castellated beams an efficient and economical solution compared to traditional
welded or hot-rolled I-girders. The commonly used web openings in the practice
can reach 7080% of the whole web depth, what should be considered in the
design. It is known, that the lateral stiffness of the web is smaller for the castellated
beams than for traditional I-girders due to the web openings, therefore the
importance of the distortion in the structural behavior has a larger importance,
which can results in decrease in the lateral torsional buckling (LTB) resistance.
*
BME, Megyetem rkp. 3, Department of Structural Engineering, 1111 Budapest, Hungary

Ro. J. Techn. Sci. Appl. Mechanics, Volume 59, N 12, P. 139227, Bucharest, 2014
140 Lszl Dunai, Balzs Kvesdi, Dvid Wischy, Barnabs Bza 2

There are numerous previous investigations dealing with the shear buckling and
bending resistance determination of the castellated beams, but there are only a
smaller number of previous investigations available in the international literature
dealing with the LTB resistance. The majority of the previous investigations in this
topic are focusing on the critical bending moment for castellated beams and the
LTB ultimate behavior is rarely studied in details. The current study has the aim to
analyze the complex nonlinear phenomena and to work out proposal for the lateral
torsional buckling resistance of castellated beams.

Fig. 1 Application examples for castellated beams [1, 2].

To investigate the special structural behavior of castellated beams under


bending and to study the effect of the web distortion on the LTB resistance a
numerical research program is executed. The numerical model is developed and
verified based on the results of previous investigations and it is used to determine
the critical bending moment and the ultimate resistance for various girder
geometries with different openings. The applicability of the general LTB design
method of the EN1993-1-1 [3] is studied in details. The previously developed
design equations for the determination of the critical bending moment are
evaluated and modifications are proposed. The cross-section modulus and the
relevant buckling curves to be applied are also determined to ensure safe and
economic design. The applicability of the simplified LTB design method of the
EN1993-1-1 [3] is also studied and modification factors are developed to consider
the special structural behavior.

2. LITERATURE REVIEW

The structural behavior of the castellated beams is a widely investigated


research field by different researchers in the last decades. In the first generation of
the castellated beams the openings were mainly polygonal, and nowadays most of
the openings are circular. Therefore the main part of the current investigations are
focusing on castellated beams with circular openings and smaller number of studies
3 Lateral torsional buckling resistance of castellated beams 141

are focusing on girders with polygonal openings. The previous research investigations
can be separated on three different topics: (i) shear and bending cross-sectional
resistance, (ii) shear buckling and (iii) lateral torsional buckling resistance. Large
research program was executed on the shear and bending resistance by Chung et al.
[4] in 2000 and by Liu et al. [5] in 2003. Based on their investigations design
models were developed for the determination of the bending and shear resistance of
the castellated beams, which were based on a Vierendeel mechanism between the
openings. Hagen et al. [6], [7] investigated the bending and shear capacity in 2009
and design methods were proposed to determine the bending and shear resistances
and for the M+V interaction.
Shear buckling behavior was investigated by Shanmugan et al. [8] in 2002
and by Soltani et al. [9] in 2012. Numerical research program was performed to
investigate the shear buckling resistance by various opening geometries and sizes.
Tsavdaridis [10] highlighted the importance of the opening geometry on the shear
buckling behavior in 2012 and demonstrated different Vierendeel mechanisms
using various opening geometries.
Investigations dealing with the LTB behavior of castellated beams was started
by Nethercot and Kerdal [11] in 1982. The structural behavior of the castellated beams
were compared to the conventional I-girders and it was concluded, that the LTB
behavior of the castellated beams are similar to the conventional I-girders. If the
cross-section properties are calculated based on the smallest net section geometry,
the application of the standard design methods leads to safe design. Mohebkhan in
2004 [12] made numerical investigations dealing with the effect of the bending
moment diagram on the critical bending moment. The conclusion was that the
modification factor due to the bending moment diagram is not constant, it depends
on the slenderness of the analyzed beam. Zhirakian et al. [1315] made extensive
research on this research field between 2006 and 2012. The structural behavior of
the castellated beams was studied and the effect of web distortion on the load
carrying capacity was described. The lateral displacements and the deformed shape
of the web was studied and characterized based on the web distortion. The
conclusion was that the distortion of the web has influence on the LTB failure
mode of the castellated beams, but the resistance can be approximated by the
design method used for conventional I-girders by several modifications. Sweeden
[16] executed a large numerical research program in 2010 to determine the critical
bending moment for castellated beams under different loading conditions and
girder geometries. Based on the study a modification factor (LB) is proposed to
consider the effect of the loading condition, the bending moment diagram, the
opening sizes and the distance between the openings. Ellobody [17] performed a
numerical investigation in 2011 and determined the LTB resistance for more than
one hundred castellated beams with different geometries and opening sizes. Ellobody
also studied the application possibilities of high strength steel for castellated
beams, and he found that it is an efficient solution. Furthermore the calculated LTB
resistance was also compared to standard design models and it was found that if the
142 Lszl Dunai, Balzs Kvesdi, Dvid Wischy, Barnabs Bza 4

observed failure mode is the combination of LTB and web distortion the standard
design method developed for conventional I-girders may lead to unsafe design. If
the web distortion has negligible effect on the failure mode, the LTB resistance can
be determined on the same way as for conventional I-girders. Lakui et al. [18]
executed an experimental research program on 5 test specimens in 2008 and the
test results were extended by numerical simulations. Based on these studies it was
concluded that the buckling curves according to the EN1993-1-1 [3] can be applied
also for castellated beams with minor changes. Since the castellated beams are
produced mainly form hot-rolled sections by cutting and welding, the buckling
behavior can be classified in between the traditional welded and hot-rolled girders.
Based on the literature review two major conclusions can be drawn, as
follows. Most of the previous investigations were executed in the last 10 years,
what shows the actuality of the problem. The majority of the previous studies
describe the special structural behavior of the castellated beams, classify the
observed failure modes. But there are only a limited number of previous studies
dealing with design method development. Beside that all the previous studies
dealing with the design of castellated beams are focusing on the determination of
the critical bending moment and on the applicability of the general LTB design
method of the EN1993-1-1 [3]. No investigations are found by the authors dealing
with the simplified LTB method for the application of castellated beams.
The aims of the current investigations are (i) to study the applicability of the
available research results in a larger parameter range, (ii) to improve the previously
developed general LTB resistance models for castellated beams and (iii) to study
the applicability of the simplified LTB design method of the EN1993-1-1 [3] for
castellated beams.

3. NUMERICAL MODELING

In the first phase of the research a numerical model is developed and verified
on the basis of previous results. Using the models the critical bending moment is
determined by bifurcation analysis (GNB) and the LTB resistance is calculated by
nonlinear FE simulation (GMNI). The numerical model is developed in Ansys [19]
environment by 4-node-thin shell elements. Linear elastic material model is used in
the GNB and linear elastic-hardening plastic material model using isotropic
hardening rule with von-Mises yield criterion is applied in the GMNI analysis with
the characteristics as follows: Youngs modulus 210000 MPa; the yield plateau is
modeled to 1% strains and from the yield stress it follows linear hardening with a
reduced modulus until reaching the ultimate stress by 15% strain level. At the ends
of the analyzed girders simply supported conditions are used, allowing warping and
rotation at both ends; no internal lateral supports are applied.
5 Lateral torsional buckling resistance of castellated beams 143

In the study three loading conditions are analyzed: (i) uniformly distributed
load along the whole length, (ii) concentrated force at the mid-span, and (iii)
concentrated bending moments in the ends. Three load positions within the cross-
section are investigated, namely the load is placed in the (i) upper flange, (ii) center
of gravity and (iii) lower flange. The numerical model is verified based on the
investigations of Sweeden [16] by the comparison of the calculated critical bending
moments. The results of the numerical simulations are also compared to the test
results of Showkati et al. [15]. Both comparisons showed good agreement with the
published results. The typical failure mode of the analyzed girders can be seen in
Fig. 2. It can be observed that the deflected shape of the cross-section shows
significant distortion of the web, what should be considered in the LTB resistance
calculation.

Fig. 2 Observed failure mode.

4. NUMERICAL RESEARCH PROGRAM

The numerical research program is completed with two aims. The first one is
to determine the critical bending moment of the analyzed girders and the second
one is to determine the LTB resistance of the castellated beams with different geometries.
The analyzed parameters are the followings: (i) diameter of the holes (d), (ii)
distance between the openings (s), (iii) flange width and thickness (bf; tf); web
depth and thickness (hw; tw), (iv) span of the girder (L), (v) shape of the holes
(diagonal or polygonal), (vi) load type (uniformly distributed, concentrated, end-
moment), (vii) load position (upper flange, center of gravity, lower flange). The
notations are shown in Fig. 3 and the investigated parameter range can be seen in Table 1.

Fig. 3 Notations.
144 Lszl Dunai, Balzs Kvesdi, Dvid Wischy, Barnabs Bza 6

The parameters are determined on the bases of typical girder geometries used
in the practice and by the product description of manufacturers producing
castellated beams. In the cases of the diameter of the holes (d) and the distance
between the openings (s) the parameter range is extended in order to analyze the
distortional phenomenon in details. The total number of studied girder geometries
is 240. For each girders the critical bending moments and the LTB resistances are
determined with and without openings, in order to investigate the reduction due to
the openings.

Table 1
Parameter range
parameter investigated parameter range
d / hw 0,3 0,4 0,5 0,6 0,7 0,8
s / hw 1,2 1,4 1,6 1,8 2,0 2,5 3,0 4,0
bf / tf 10 12 14 16 20 30
hw / tw 40 50 60 70 80 90
L / hw 7,5 10 12,5 15 20 25
opening shape circular, polygonal
load position upper flange, center of gravity, lower flange
load type uniformly distributed, concentrated, end-moment

In the study three typical failure modes are observed, as shown in Fig. 4. In
the absence of holes the failure mode is LTB with quasi-straight deflected webs
(Fig. 4a). By increasing the diameter of the openings the web distortion becomes
more critical and it results in decrease in the LTB resistance, as shown in Fig. 4b, c.

a, no holes b, d/hw=0,2 d, d/hw=0,8


Fig. 4 Typical failure modes.

Based on the parametric study the following conclusions can be drawn:


1. The opening size has significant influence on the failure mode of the
castellated beams since the web distortion becomes more critical by increasing the
opening size. Figure 5 shows the relationship between the opening size and the
resistance reduction factor (factor 1,0 refers to the girder without openings). It is
7 Lateral torsional buckling resistance of castellated beams 145

concluded that the critical bending moment and the LTB resistance decreases
quasi-linearly by increasing the opening size. The resistance decrease for girders
with larger web depth is larger, as shown in Fig. 5, what can be explained by the
effect of web distortion. Webs can give more efficient lateral support in the case of
girders with relatively smaller web depth.

Fig. 5 Effect of d/hw on the resistance reduction due to openings.

2. It is observed that the distance between the holes has also a quasi-linear
effect on the critical bending moment and on the LTB resistance, as shown in Fig.
6. The resistance reduction for three girders with two opening sizes are presented in
Fig. 6. The dashed lines show the resistance decreases of the girders with smaller
openings and the continuous lines represent the results of larger openings. The
resistance reduction tendencies are the same, the values depend on the d/hw; hw/tw;
bf/tf ratios.

Fig. 6 Effect of s/d on the resistance reduction due to openings.


146 Lszl Dunai, Balzs Kvesdi, Dvid Wischy, Barnabs Bza 8

3. The analysis of the cross-section geometry (bf/tf; hw/tw) showed that the
tendencies are the same as for the conventional I-girders without web openings but
on a smaller resistance level. This observation proves that the general design
method of the Eurocode standard can be applied also for castellated beams with
changes taken the effect of the holes into account.

5. LTB RESISTANCE DETERMINATION METHOD

In the LTB design method of the castellated beams the calculation of the
cross-section resistance and the critical bending moment have the main interest.
The basic principle of the slenderness calculation suggest to use the cross-section
modulus which results in the largest normal stress in the flanges. If the holes are
equally spaced along the girder length and the slope of the bending moment
diagram between the holes is not significant, it can be assumed, that the maximum
stress can be approximated by using the net section properties, as proposed by
Sweeden [16]. In this proposal the modification factors taking the effect of the
holes in the critical bending moment into account. The results of the Sweedens
design method are compared to the current numerical results. It is observed, that
the proposed modification factors give good approximation in a specific parameter
range. But the calculated critical bending moments according to Sweedens [16]
design method does not follow all the tendencies observed in the numerical
calculations (e.g. for the d/hw ratio). Therefore the calculation method of the critical
bending moment (Mcr) is investigated in details and refined according to the current
results. The basis of the method is the equation of the Mcr developed by Sweeden
and the proposed enhanced design method is the following: the LTB design
resistance of the castellated beam can be determined by Eq. (1).
Wy f y , (1)
M b , Rd = LT
M1

where: Wy is the cross-section modulus of the net section according to the


classification made on the original cross-section without openings, and LT is the
reduction factor using the buckling curve c according to EN1993-1-1 [3].
The slenderness of the castellated beam can be calculated by Eqs. (2)(9).
Wy f y , (2)
LT =
M cr

M cr = CB , p CB , CB , L LB M 0,cr , (3)

where: KCB,p depends on the load position: upper flange: 0,7; center of gravity:
1,00; lower flange: 1,4;
9 Lateral torsional buckling resistance of castellated beams 147

KCB,p depends on the load type: uniformly distributed load: 1,0; concentrated
force: 1,21; end-moments: 0,95.
0.25
L , (4)
CB , L =
15 h w

0,03 hw bf (5)
LB = + 1,21 0,002 10
b t t
f w f
tf

s s
1, 5

= CB , Hb + 0,25 0,07 CB , wD , (6)
hw hw

hw
CB , Hb = 0,172 + 0,248 , (7)
bf
d
0 , 08 0 , 24
h hw
CB , wD = 3 w , (8)
tw

2 E Iz I w L2 G I t
. (9)
M 0,cr = +
L2 Iz 2 E Iz

The notations in Eqs. (48) are shown in Fig. 3; the notations used in Eq. (9)
are the same as given in the EN1993-1-1 [3] with the only difference, that the value
of It and Iz should be calculated based on the net cross-section geometry. Using the
results of Eq. (3) and the numerical simulations the reduction factors for the
analyzed girders are presented in Fig. 7.

Fig. 7 Calculated reduction factors.


148 Lszl Dunai, Balzs Kvesdi, Dvid Wischy, Barnabs Bza 10

The results show that the buckling curve c of the EN1993-1-1 [3] gives a
good approximation of the numerical results for castellated beams in the analyzed
parameter range. These results prove the conclusions of Lakui et al. [18], that for
castellated beams with h/bf > 2 the buckling curve c and with h/bf < 2 the
buckling curve b could be used. Note that in the case of the current investigations
the h/bf ratio is larger than 2 for all the analyzed specimens. Accordingly they
belongs to the buckling curve c as proposed by Lakui et al. [18]. It can be
concluded that the current numerical study proves the applicability of this proposal.
Figure 8 shows the comparison of the critical bending moment and the LTB
resistance between the enhanced design method and the numerical results. It can be
seen on the diagrams, that the calculated values with the proposed enhanced design
method gives a good approximation to the numerical results. The average ratio
between the results of the developed design method and the numerical calculations
is 0,927 for the critical bending moment and 0,909 for the lateral torsional buckling
resistance.

a b
Fig. 8 Comparison of the developed design method and numerical results: a) critical bending
moment; b) LTB resistance.

6. SIMPLIFIED METHOD FOR LTB RESISTANCE CALCULATION

In the simplified LTB design method the equivalent compression flange is


created from the whole flange and from a certain part of the compressed web. The
simplified LTB design method is an efficient and accurate tool in the case of lateral
distortional buckling failure modes, too, and therefore the applicability for
castellated beams is studied. The holes slightly reduce the area of the equivalent
compression flange, therefore the consideration of the holes is implemented in the
cross-section modulus. For this purpose an equivalent cross-section is introduced
having the same web depth as the original castellated beam with an average hole
size calculated by Eq. (10).
11 Lateral torsional buckling resistance of castellated beams 149

d 2
n
4 hw (10)
heq = 1
L hw 2


where: n total number of the holes along the girder length (L); d hole diameter.
The geometrical presentation of the equivalent girder geometry is shown in
Fig. 9.

Fig. 9 Original and equivalent beam geometries.

The cross-section modulus (Weq) of the equivalent beam should be


determined and the LTB resistance can be calculated using this value in Eq. (11).
On the basis of the numerical results the LTB resistance of the castellated beams
can be determined by the same way as for the conventional I-girders with the
modifications described in the Eqs. (11) (14).
fy
M b. Rd = CB , p Weq . (11)
M1
The original LTB calculation method is extended by the factor CB,p,
considering the effect of the load position, determined by:
0,8 if the upper flange,
1,1 if the center of gravity and
1,4 if the lower flange is loaded.
The slenderness of the equivalent compressed T-bar can be calculated by
Eqs. (1214). The modification factors of CB,L and CB,w takes the effect of the
span and the cross-section geometries into account. The value of kc can be
determined according to the EN1993-1-1 [3].
k c Lc
f = CB , w CB , L , (12)
i fz 1
0 , 5 0,1
L h
CB , L = CB, w = w . (1314)
15 hw 60 t w
150 Lszl Dunai, Balzs Kvesdi, Dvid Wischy, Barnabs Bza 12

The calculated resistances according to the enhanced simplified LTB design


method and the numerical simulations are compared and good agreement is
presented in Fig. 10. The average ratio between the lateral torsional buckling
resistances determined by the developed design method and the numerical analyses
is 0,902. Note that the differences are slightly larger than for the general LTB
design method, but it is considered acceptable from the practically required
accuracy point of view.

Fig. 10 Comparison of the enhanced simplified LTB design method


and the numerical results.

7. SUMMARY AND CONCLUSIONS

The lateral torsional buckling behavior of the castellated beams are investigated
in the current paper. More than 240 girder geometries are studied: girders with and
without openings, geometrical and loading conditions are investigated to determine
the effect on the LTB resistance. For each analyzed girders the critical bending
moment and the LTB resistance are determined. The calculated critical bending
moments are compared to the design method of Sweeden [16] and a refinement is
proposed to predict Mcr. On this basis the applicability of the LTB design method
of the EN1993-1-1 [3] for castellated beams is investigated. The investigations
proved the applicability of the proposal of Lakui et al. [18] regarding to the
applicable buckling curves. The numerical results proved the applicability of the
LTB resistance model by the following minor changes for castellated beams.
1. The cross-section properties of the net section should be used in the cross-
section modulus determination.
2. The critical bending moment should be calculated according to Eqs. (2)(9).
3. The buckling curve c according to the EN1993-1-1 [3] can be applied in
the analyzed parameter range (h/bf > 2).
13 Lateral torsional buckling resistance of castellated beams 151

The applicability of the simplified LTB design method for castellated beams
is also studied. The results of the numerical simulations proved its applicability by
the following changes:
1. The cross-section properties (Weq) of the proposed equivalent section
should be used in the LTB resistance calculation.
2. The slenderness of the equivalent compression flange should be determined
according to Eqs. (1214) with minor changes of the slenderness calculation
method developed for traditional I-girders.
3. An additional modification factor (CB,p) considering the load position is
proposed for castellated beams.

Received on March 30, 2014

REFERENCES

1. http://www.bouwenmetstaal.nl
2. http://www.newsteelconstruction.com
3. Eurocode 3: Design of steel structures Part 1.1: General rules and rules for buildings, EN
1993-1-1, 2005.
4. CHUNG, K.F., LIU, T.C.H., KO, A.C.H., Investigation of Vierendeel mechanism in steel beams
with circular web openings, Journal of Constructional Steel Research, 57, pp. 467490, 2001.
5. LIU, T.C.H., CHUNG, K.F., Steel beams with large web openings of various shapes and sizes:
finite element investigation, Journal of Constructional Steel Research, 59, pp. 11591176,
2003.
6. HAGEN, N.C., LARSEN, P.K., AALBERG, A., Shear capacity of steel plate girders with large
web openings, Part I: Modeling and simulations, Journal of Constructional Steel Research, 65,
pp. 142150, 2009.
7. HAGEN, N.C., LARSEN, P.K., Shear capacity of steel plate girders with large web openings, Part
I: Design guidelines, Journal of Constructional Steel Research, 65, pp. 151158, 2009.
8. SHANMUGAM, N.E., LIAN, V.T., THEVENDRAN, V., Finite element modelling of plate girders
with web openings, Thin-Walled Structures, 40, pp. 443464, 2002.
9. SOLTANI, M.R., BOUCHAIR, A., MIMOUNE, M., Nonlinear FE analysis of the ultimate
behavior of steel castellated beams, Journal of Constructional Steel Research, 70, pp. 101
114, 2012.
10. TSAVDARIDIS, K.D., DMELLO, C., Optimization of novel elliptically-based web opening
shapes of perforated steel beams, Journal of Constructional Steel Research, 76, pp. 3953,
2012.
11. NETHERCOT, D.A., KERDAL, D., Lateral-torsional buckling of castellated beams, The Structural
Engineer, 60, 1982.
12. MOHEBKHAH, A., The moment-gradient factor in lateral-torsional buckling on inelastic
castellated beams, Journal of Constructional Steel Research, 60, pp. 14811494, 2004.
13. ZIRAKIAN, T., SHOWKATI, H., Distortional buckling of castellated beams, Journal of
Constructional Steel Research, 62, pp. 863871, 2006.
14. ZIRAKIAN, T., Lateral-distortional buckling of I-beams and the extrapolation techniques,
Journal of Constructional Steel Research, 64, pp. 111, 2008.
152 Lszl Dunai, Balzs Kvesdi, Dvid Wischy, Barnabs Bza 14

15. SHOWKATI, H., GHAZIJAHANI, T.G., NOORI, A., ZIRAKIAN, T., Experiments of elastically
braced castellated beams, Journal of Constructional Steel Research, 77, pp. 136172, 2012.
16. SWEEDEN, A., Elastic lateral stability of I-shapes cellular steel beams, Journal of Constructional
Steel Research, 67, pp. 151163, 2011.
17. ELLOBODY, E., Nonlinear analysis of cellular steel beams under combined buckling modes,
Thin-Walled Structures, 52, pp. 6679, 2012.
18. LAKUI, V.T., DZEBA, I., ANDROIC, B., The buckling curve for lateral torsional buckling
resistance of castellated beams, 5th European Conference on Steel Structures (Eurosteel 2008),
Graz, Austria, September 3-5, 2008.
19. ANSYS v14.5, Canonsburg, Pennsylvania, USA.
UNIFIED SLENDERNESS LIMITS FOR STRUCTURAL STEEL
CIRCULAR HOLLOW SECTIONS

LEROY GARDNER1, KWAN HO LAW2, CRAIG BUCHANAN3,

Abstract. Circular hollow sections (CHS) are widely used in a range of structural
engineering applications. The sections may be hot-finished or cold-formed from a
variety of metallic materials with a range of yield strengths. The design of these
sections is covered by all major design codes, yet there are significant differences in
the treatment of local buckling, as considered through cross-section classification.
Cross-section classification criteria relate to rotation capacity and strength requirements
(attainment of the plastic or elastic moment in bending and the yield load in
compression), while the relative performance of structural CHS is governed by
susceptibility to local buckling and is influenced by cross-section slenderness,
material stiffness and yield strength, forming process (affecting geometry, material
homogeneity and residual stresses), material strain hardening characteristics and
ovalization. Furthermore, the classification criteria and reliability requirements vary
among the different structural design codes. This paper presents a review of 153 test
results on CHS in bending, covering structural steel, aluminium, stainless steel and
very high strength steel. Based on the available test data, current codified provisions
in the European, North American and Australian Standards are reassessed, and
following reliability analyses new unified slenderness limits are proposed for
structural steel CHS.

Key words: circular hollow sections, reliability analysis, section classification, slenderness
limits, tubular construction.

1. INTRODUCTION

Circular hollow sections (CHS) have been manufactured and used since the
early 1800s for structural members such as columns, beams, tension members and
trusses [1]. They are thin-walled structural elements, and therefore a primary
consideration in their design is local buckling. Current design codes adopt the
concept of cross-section classification for the treatment of local buckling in thin-walled
tubular members, but there is significant variability between the slenderness limits
employed to separate the individual classes.
1
Imperial College London, Department of Civil and Environmental Engineering, South Kensington
Campus, London, SW7 2AZ, UK
2
The University of Hong Kong, Department of Civil Engineering, Pokfulam Road, Hong Kong
3
Imperial College London, Department of Civil and Environmental Engineering, South Kensington
Campus, London, SW7 2AZ, UK
154 Leroy Gardner, Kwan Ho Law, Craig Buchanan 2

EN 1993-1-1 (2005) [2] and BS 5950 (2000) [3] for structural steelwork,
together with EN 1993-1-4 (2006) [4] for stainless steel and EN 1999-1-1 (2007)
[5] for aluminium, define four behavioural classes of cross-section, based upon
their susceptibility to local buckling. Class 1 cross-sections, called plastic sections
in BS 5950, are capable of reaching and maintaining their full plastic moment Mpl
in bending by forming plastic hinges with sufficient rotation capacity for plastic
design. Class 2 cross-sections, referred to as compact sections in BS 5950, are also
capable of reaching their full plastic moment in bending but have a lower deformation
capacity. In Class 3 cross-sections, called semi-compact sections in BS 5950, local
buckling prevents attainment of the full plastic moment and the bending moment
resistance is limited to the yield moment Mel. Class 4 cross-sections, commonly
referred to as slender sections, exhibit local buckling before the yield stress is
reached. The moment-rotation characteristics of the four behavioural classes are
illustrated in Fig. 1. AISC 360 (2005) [6] and AS 4100 (1998) [7] effectively
define three classes of cross-section: Class 1 cross-sections are referred to as
compact, there is no equivalent to Class 2 sections, Class 3 sections are termed
non-compact, while Class 4 cross-sections are referred to as slender sections. The
Class 3 limit also separates cross-sections that are fully effective in compression,
where the section capacity is taken as the yield load, from those that fail by local
buckling prior to the attainment of the yield load. For both compression and
bending the cross-section level resistance is determined based on an effective
cross-section defined by the CHS diameter-to-thickness (D/t) ratio.
Applied moment M

Mpl
Class 2
Mel Class 1

Class 3

Class 4

Rotation

Fig. 1 Four behavioural classes of cross-sections.

In this paper, the factors influencing local buckling and the structural
response of circular hollow sections are discussed; test data on structural steel,
stainless steel and aluminium tubes are collated and analysed; and slenderness
limits prescribed in a series of international design standards, which exhibit
3 Unified slenderness limits for structural steel circular hollow sections 155

significant variability, are evaluated. This study focuses on structural hollow


sections rather than cylindrical shells with very high diameter-to-thickness ratios.
Finally unified slenderness limits for structural steel CHS are proposed following
reliability analyses.

2. FACTORS AFFECTING LOCAL BUCKLING


AND STRUCTURAL RESPONSE OF CHS

Local buckling occurs in thin-walled sections when the applied compressive


stress exceeds a critical value, and is characterised by local ripples in the cross-
section wall. Local buckling and the structural response of CHS are influenced by a
number of factors, which are discussed in this section.
The elastic buckling stress of a cylindrical shell in the axis-symmetric mode
(cr) is given by Eq. 1:
2E t
cr = , (1)
3(1 2 ) D
where E is the Youngs modulus, is Poissons ratio and D/t is the diameter-to-
thickness ratio of the section and the geometric parameter that controls local
buckling. The susceptibility to elastic buckling in preference to yielding depends
on the yield strength of the material fy, and this therefore also appears in
slenderness parameters, such as those adopted in design codes. With higher yield
strengths, the slenderness of the section effectively increases; i.e. the section is
more susceptible to local buckling prior to yielding. Previous studies have however
shown that the slenderness limits derived for normal strength steel tubes become
conservative when applied to very high strength steel [8]. This is linked to the
presence and influence of initial geometric imperfections and residual stresses.
Residual stresses are typically induced in structural components through plastic
deformation and differential cooling during manufacture. Their influence on
structural members is to cause premature yielding and loss of stiffness, often
leading to a reduction in load carrying capacity. In high strength steel sections,
residual stresses are a smaller fraction of the yield strength and therefore their
detrimental effect is reduced compared with normal strength steel [9].
Initial geometric imperfections can also have a significant influence on the
strength of thin-walled sections [10] by amplifying buckling deformations and
hence expediting the initiation of yield. The effect of imperfections is less
detrimental to the response of high strength structural components. A modified
imperfection factor for columns which reduces with increased yield strength has
been proposed to reflect this behaviour [9]; this issue has also been highlighted in
the context of local buckling [11].
156 Leroy Gardner, Kwan Ho Law, Craig Buchanan 4

Local buckling and the structural response of CHS are also influenced by the
stress-strain behaviour of the constituent material, which is largely controlled by its
chemical composition and physical properties, but is also affected by the section
forming process. Generally, there are two different types of stress-strain curves
yield point and round house. Hot-finished sections typically have a yield point
stress-strain curve, where stress is linearly proportional to strain up to the yield
point, after which a yield plateau and strain hardening may be observed. A round
house stress-strain curve deviates from linearity at low stresses and displays a
gradually yielding behaviour and no sharply defined yield point. Stainless steel and
aluminium exhibit this type of behaviour as the basic material response; cold-
formed steel sections also display a rounded stress-strain curve. This is due to the
Bauschinger effect, whereby residual stresses resulting from plastic deformations
induced during production cause deviation of the stress-strain response from
linearity upon load reversal. Resistance to local buckling depends on the stiffness
of the material, and hence local buckling is promoted by any loss of stiffness due to
yielding or nonlinearity. Gradual loss of stiffness as opposed to a sharp yield point
is usually regarded as being beneficial in terms of structural performance [12, 13],
with a greater degree of strain hardening enabling higher compressive and moment
capacities in stocky sections of low D/t ratios. Research is currently being
undertaken to utilise strain hardening for enhanced section capacity in low D/t ratio
circular hollow sections.
A further factor to be considered in the response of CHS in bending is
ovalization. This refers to the gradual flattening of a tube under bending resulting
from the inclined nature of the forces in the tube wall that arise in the deformed
configuration [14, 15]. The material and geometric properties of structural metallic
tubes preclude failure by ovalization wholly in the elastic range, with yielding or
local buckling being the key factors limiting structural resistance. However,
ovalization may contribute to failure since hoop stresses are induced in the wall of
the tube that will influence the onset of plasticity, and there is a reduction in local
curvature of the most heavily compressed region of the tube, which facilitates the
onset of local buckling.

3. EXISTING SLENDERNESS PARAMETERS AND LIMITS

Slenderness parameters for CHS in all structural design codes include the
geometric diameter-to-thickness ratio D/t and the material yield strength fy.
However there is a range of slenderness values, due to the yield strength being
normalized by different values in the codes, which are summarized in Table 1
along with the treatment of Class 4 (slender) sections.
5 Unified slenderness limits for structural steel circular hollow sections 157

Table 1
CHS slenderness parameters adopted in different structural design codes

Guidance on effective properties for Class 4


Cross-section
Design code (slender) cross-sections
slenderness parameter
Compression Bending
EN 1993-1-1 (2005) D fy
1 1
Structural steel t 235
BS 5950 (2000) D fy
Clause 3.6.6 Clause 3.5.6.4
Structural steel t 275
AISC 360 (2005) D fy
Section E7 Section F8
Structural steel t E
AS 4100 (1998) D fy
Section 6.2 Section 5.2
Structural steel t 250
EN 1993-1-4 (2006) D f y 210 000
1 1
Stainless steel t 235 E
EN 1999-1-1 (2007) D y f
Clause 6.1.5 Clause 6.1.5
Aluminium t 250
1
Note: No effective section properties are provided but designer is directed to EN 1993-1-6 (2007)
[16] for shells.

In order to make a direct comparison between the various design codes, the
slenderness limits have been converted to a common basis, using the slenderness
parameter adopted for stainless steel in EN 1993-1-4. This is appropriate since the
EN 1993-1-4 slenderness parameter includes the Youngs modulus E and the
material yield strength, which can therefore reflect the different material behaviours.
Aluminium in particular has a significantly lower Youngs modulus than both structural
steel and stainless steel. The values adopted for the material Youngs modulus are:
210 000 N/mm2 for structural steel, 200 000 N/mm2 for stainless steel and 70 000
N/mm2 for aluminium. The modified slenderness limits are presented in Table 2.
From Table 2, it can be observed that the Class 3 slenderness limits in
compression are fairly consistent between the structural steel and stainless steel
design codes, but a more relaxed limit is applied to aluminium. The Class 1 and 2
slenderness limits in bending are also fairly consistent across the range of design
codes and materials. However, the Class 3 slenderness limits in bending show
significant variation. It should be noted that EN 1993-1-1 and EN 1999-1-1 adopt
the same Class 3 slenderness limit for both compression and bending, 90.0 and
171.6 respectively.
158 Leroy Gardner, Kwan Ho Law, Craig Buchanan 6

Table 2
Summary of CHS slenderness limits in different structural design codes
Material Structural steel Stainless steel Aluminium
EN AISC EN EN
Design code BS 5950 AS 4100
1993-1-1 360 1993-1-4 1999-1-1
Class 1 limit in
50.0 46.8 62.6 53.2 50.0 42.9
bending
Class 2 limit in
70.0 58.5 - - 70.0 90.8
bending
Class 3 limit in
90.0 163.8 277.0 127.7 280.0 171.6
bending
Class 3 limit in
90.0 93.6 98.3 87.2 90.0 171.6
compression

The Class 3 limit is of particular practical significance because it represents


the borderline between fully effective and slender sections, with the latter requiring
additional calculation effort for designers. There are two principal reasons for the
variation in this slenderness limit between the different design codes. The first
relates to the pool of available structural performance data, noting that
classification limits are often sensitive to the slenderness range of test data upon
which they are based [17]. The Class 3 limit for CHS in bending in EN 1993-1-1
was derived on the basis of tests on stocky sections [18]; whereas the same limit in
AISC 360, which is significantly more relaxed, was based on a far wider range of
slenderness values [1922]. The second reason relates to the different regional
practices in terms of structural reliability. The partial safety factors adopted in the
different design codes are summarised in Table 3.

Table 3
Partial safety factors for cross-section resistance adopted in different design codes
Material Structural steel Stainless steel Aluminium
EN BS AISC AS EN EN
Code
1993-1-1 5950 360 4100 1993-1-4 1999-1-1
Partial safety
1.00 1.00 0.901 0.901 1.10 1.10
factor
Note: 1 Partial factor appears in the numerator, while others appear in the denominator; 1/0.9=1.11.

Reliability of the design provisions for cross-section resistance depends upon


both the adopted slenderness limit and the partial safety factor. The target reliability
index and material over-strength are also influential, as are any possible regional
differences in manufacturing standards and tolerances. EN 1993-1-1 employs a
partial safety factor of unity, while AISC 360 and AS 4100 adopt a value of 1.11
7 Unified slenderness limits for structural steel circular hollow sections 159

(0.9 in the numerator). The EN 1993-1-1 limits would therefore be expected to be


stricter, since the limit itself has to effectively compensate for the disparity in
safety factors. Reliability analyses have been performed and unified slenderness
limits are proposed for structural steel CHS in the following section.

4. EVALUATION OF TEST DATA


AND PROPOSED SLENDERNESS LIMITS

A total of 153 test results on circular hollow section beams of different materials
and configurations under bending have been collated in this study. The following
tests were undertaken: 52 tests on hot-finished structural steel sections [15, 18, 20, 21],
33 tests on cold-formed structural steel sections [21, 2325], 21 tests on fabricated
structural steel sections [19, 22, 26], 12 tests on very high strength structural steel
sections [27], 20 tests on stainless steel sections [28, 29] and 15 tests on aluminium
sections [30]. The tests were conducted in three different configurations: 25 in pure
bending, 119 in four-point bending and 9 in three-point bending. The cross-section
slenderness of the beams varied from 20.4 to 294.5 (using the EN 1993-1-4 measure
of slenderness from Table 1). A graph of the ultimate test moment normalised by the
elastic moment capacity plotted against the cross-section slenderness is shown in
Fig. 2. The Class 3 slenderness limits in bending from the design codes are also
shown. The collated test results display the anticipated trend of decreasing normalised
moment capacity with increasing slenderness, though there is significant scatter in
the data, which is believed to relate to the factors discussed previously. The superior
performance of the very high strength structural steel sections is particularly evident.
In order to obtain a unified slenderness limit achieving a consistent level of
safety and incorporating the uncertainty in the test results and the variability of the
basic variables (material and geometric properties) in the design expression, a
reliability analysis in accordance with EN 1990 (2002) [31] was performed, as
outlined in [32]. The analysis was performed on the 106 tests on hot-finished, cold-
formed and fabricated structural steel sections. Since no formula for deriving
effective section properties for Class 4 CHS is provided in EN 1993-1-1, a
modified expression based on the BS 5950 provisions was adopted in calculating
the design moment capacity for these sections, as given by Eq. 2:
90 235
Weff = Wel , (2)
D /t fy
where Weff and Wel are the effective and elastic section moduli, respectively.
160 Leroy Gardner, Kwan Ho Law, Craig Buchanan 8

Fig. 2 Normalised test moment capacity versus cross-section slenderness,


with Class 3 slenderness limits from codes.

The parameters assumed in the statistical analyses were based on previous


findings on the mechanical properties of structural steel: the ratio of mean to nominal
yield strengths (i.e. the material over-strength) was taken as 1.16 and the
coefficients of variation of yield strength and geometric properties were taken as
0.05 and 0.02 respectively [33, 34]. These values originate from industrial data
obtained from European steel producers. The results of the analysis and a summary
of the key statistical parameters are presented in Table 4. The following symbols
are used: kd,n = design (ultimate limit states) fractile factor for n tests, where n is the
population of test data under consideration; b = average ratio of experimental to
model resistance based on a least squares fit to the test data; V = coefficient of
variation of the tests relative to the resistance model; Vr = combined coefficient of
variation incorporating both model and basic variable uncertainties; and M0 = factor
by which the mean curve should be reduced to provide a reliable design curve.

Table 4
Summary of statistical analysis parameters for EN 1990
n kd,n b V Vr M0
106 3.18 1.10 0.131 0.142 1.24

A least squares regression fit to the test data set is plotted in Fig. 3, which is
then scaled down by the required safety factor of 1.24 obtained from the reliability
analysis to produce the design curve. The unified Class 3 slenderness limit (where
the design curve passes through Mu/Mel = 1.0) for steel sections was found to be
9 Unified slenderness limits for structural steel circular hollow sections 161

100 with partial factor of 1.00 adopted in EN 1993-1-1 and 135 for the AISC 360
and AS 4100 partial factor of 0.9 (in the numerator), with the latter design curve
being scaled down by a factor of 1.12 (=1.240.9) from the mean.

Fig. 3 Normalised test moment capacity versus cross-section slenderness, with statistical analysis.

5. CONCLUSIONS

The factors affecting local buckling in CHS and the treatment of this
instability in various structural design codes have been discussed. A large disparity
in the Class 3 slenderness limits in bending was observed between the different
design codes. Towards the establishment of unified slenderness limits, the results
of 153 bending tests on CHS were examined, and reliability analyses were
performed in accordance with EN 1990. Revised structural steel Class 3
slenderness limits of 100 for EN 1993-1-1 and 135 for AISC 360 and AS 4100
were proposed. These slenderness limits provide a unified treatment across the
design codes since the more relaxed slenderness limit proposed for AISC 360 and
AS 4100 is offset by the inclusion of the partial safety factor of 1.11 (0.9 in the
numerator) adopted in these codes. Further investigation is underway in this area,
and into utilising strain hardening for enhanced section capacity in low D/t ratio
circular hollow sections.

Received on March 14, 2014


162 Leroy Gardner, Kwan Ho Law, Craig Buchanan 10

REFERENCES

1. DUTTA, D., Structures with Hollow Sections, Wiley VCH, Weinheim, 2002.
2. EN 1993-1-1 (2005) Eurocode 3: Design of steel structures Part 11: General rules and rules for
buildings, CEN, 2005.
3. BS 5950-1 (2000) Structural use of steelwork in building Part 1: Code of practice for design
rolled and welded sections, BSI, 2000.
4. EN 1993-1-4 (2006) Eurocode 3: Design of steel structures Part 14: General rules
supplementary rules for stainless steels, CEN, 2006.
5. EN 1999-1-1 (2007) Eurocode 9: Design of aluminium structures Part 11: General structural
rules, CEN, 2007.
6. AISC 360 (2005) Load and Resistance Factor Design Specification for Structural Steel Buildings,
American Institute of Steel Construction, Chicago, 2005.
7. AS 4100 (1998) Steel Structures, Standards Australia, Homebush, New South Wales, Australia,
1998.
8. ZHAO, X. L., Section capacity of very high strength (VHS) circular tubes under compression,
Department of Civil Engineering, Monash University, Australia, 2000.
9. IABSE, Use and application of high-performance steels for steel structures, Structural Engineering
Documents, 2005.
10. ZHAO, X. L. and JIAO, H., Imperfection, residual stress and yield slenderness limit of very high
strength (VHS) circular steel tubes, Journal of Constructional Steel Research, 59, 2,
pp. 233249, 2003.
11. ZHAO, X. L. and JASPART, J. P., Width-to-thickness ratios for classification of tubular sections,
Eurosteel 2005, June 8-10, 2005, pp.1.4183190, 2005.
12. SCHILLING, C. G., Buckling strength of circular tubes, Journal of the Structural Division,
ASCE, 91, 5, 1965.
13. KATO, B., Local buckling of steel circular tubes in plastic range, Stability of Structures under
Static and Dynamic Loads, ASCE, pp.375391, 1977.
14. BRAZIER, L. G., On the Flexure of Thin Cylindrical Shells and other Thin Sections,
Proceedings of the Royal Society A, 116, 773, pp.104114, 1927.
15. ELCHALAKANI, M., ZHAO, X. L. and GRZEBIETA, R. H., Plastic mechanism analysis of
circular tubes under pure bending tests to determine slenderness limits for cold-formed
circular hollow sections, International Journal of Mechanical Sciences, 44, 6, pp.11171143,
2002.
16. EN 1993-1-6 (2007) Eurocode 3: Design of steel structures Part 1-6: Strength and Stability of
Shell Structures, CEN, 2007.
17. GARDNER, L. and CHAN, T. M., Cross-section classification of elliptical hollow sections, Steel
and Composite Structures, 7, 3, pp.185200, 2007.
18. SEDLACEK, G., STRANHNER, N., LANGENBERG, P., RONDAL, J. and BOREAEVE, P.,
Rotation Capacity of Hollow Beam Sections, Final Report-CIDECT Research Project, No 2P,
RWTH, Aachen, 1995.
19. SHERMAN, D. R., Supplemental tests for bending capacity of fabricated pipes, Department of
Civil Engineering, University of Wisconsin Milwaukee, 1984.
20. JIRSA, J. O., FOOK-HOY, L., WILHOIT, J. C. and MERWIN, J. E., Ovaling of pipelines under
pure bending, Paper OTC 1569, 4th Annual Offshore Technology Conf., Houston, Texas, 1972.
21. SHERMAN, D. R., Tests of circular steel tubes in bending, Journal of the Structural Division,
ASCE, 102, 11, 1976.
22. SHERMAN, D. R., Bending Capacity of Fabricated Tubes, Department of Civil Engineering.
University of Wisconsin Milwaukee, 1983.
23. KOROL, R. M., Critical buckling strains of round tubes in flexure, Department of Civil
Engineering, Mc Master University, Hamilton, Ontario, Canada L8S 4L8, 1978.
11 Unified slenderness limits for structural steel circular hollow sections 163

24. ELCHALAKANI, M., ZHAO, X. L. and GRZEBIETA, R. H., Plastic slenderness limits for cold-
formed circular hollow sections, Australian Journal of Structural Engineering, Institutions of
Engineers, 3, 3, pp.116, 2001.
25. ELCHALAKANI, M., ZHAO, X. L. and GRZEBIETA, R. H., Bending tests to determine
slenderness limits for cold-formed circular hollow sections, Journal of Constructional Steel
Research, 58, 11, pp.14071430, 2002.
26. STEPHENS, M. J., KULAK, G. L. and MONTGOMERY, C. J., Local buckling of thin walled
tubular steel members, Structural Engineering Report No. 103, University of Alberta, Canada,
1982.
27. ZHAO, X. L. and JIAO, H., Section slenderness limits of very high strength circular steel tubes in
bending, Department of Civil Engineering, Monash University, Australia, 2004.
28. RASMUSSEN, K. J. R. and HANCOCK, G. J., Design of cold-formed stainless steel tubular
members II: Beams, Journal of Structural Engineering, 119, 8, pp.23672386, 1993.
29. STEEL CONSTRUCTION INSTITUTE, Development of the use of stainless steel in
construction, Final Report to European Coal and Steel Community-WP3 (Contract 7210-
SA/842), Steel Construction Institute, 2000.
30. MOORE, R. L. and MARSHALL, H., Beam and torsion tests of aluminium-alloy 61S-T tubing,
Technical notes No. 867, National advisory committee for aeronautics, Washington, 1942.
31. EN 1990 (2002) Eurocode: Basis of Structural Design, CEN, 2002.
32. LAW, K. H., Instabilities in Structural Steel Elliptical Hollow Section Members, Department of
Civil and Environmental Engineering, Imperial College London, London, 2010.
33. BYFIELD, M. P. and NETHERCOT, D. A., Material and geometric properties of structural steel
for use in design, The Structural Engineer, 75, 21, pp.363367, 1997.
34. BYFIELD, M. P. and NETHERCOT, D. A., Correspondence on material and geometric properties of
structural steel for use in design, The Structural Engineer, 76, 20, pp.402, 1998.
In memoriam of Darko Beg, Ph.D., Professor of Steel
Structures, University of Ljubljana, Slovenia, one of
the most active members of ECCS/TWG8.3, passed
away on 11th February 2014.

EVOLUTION OF EUROCODE 3 AMENDMENTS


TO EN 1993-1-5 FOR PLATE BUCKLING

ULRIKE KUHLMANN1, BENJAMIN BRAUN2

Abstract. Starting in 2010 the European Commission initiated the process of evolution
of the first generation of Eurocodes. Based on Mandates M/466 EN and M/515 EN,
CEN/TC250 created several expert groups which deal with this evolution work. This
paper reports about the common ongoing work of Working Group TC250/SC3/EN
1993-1-5 and ECCS Technical Working Group 8.3 (Plate Buckling). Amendments
which have been already prepared in order to improve the ease-of-use and to cover
technical development are presented. An outlook which tasks are to be addressed in
further work is given.

Key words: Eurocode, EN 1993-1-5, evolution, amendments, plate buckling.

1. INTRODUCTION

The launch of the first generation of Structural Eurocodes marked the


outcome of more than 30 years of collaborative work which started in 1975 with
the objective to create common European standards for the design of building and
civil engineering structures. Today 10 Structural Eurocodes provide rules for basis
of design, actions on structures and structural design rules for the use of all major
construction materials such as concrete, steel, timber, masonry and aluminum. By
December 2010 national standards which were in conflict with any of the 58
Eurocode parts had to be withdrawn.
Starting in March 2010 the European Commission sent Mandate M/466 EN
Programming Mandate addressed to CEN in the field of structural Eurocodes to
the European Committee for Standardization (CEN) in order to initiate the process
of evolution of the first generation of Eurocodes. Such an evolution has been
considered necessary to sustain the users confidence in the standards and to take
market developments, innovation and research into consideration both through the
revision of existing standards and the development of new standards.

1
University of Stuttgart, Institute of Structural Design, Pfaffenwaldring 7, 70569 Stuttgart, Germany
2
Space Structures GmbH, Fanny-Zobel-Strasse 9, 12435 Berlin, Germany
2 Evolution of Eurocode 3 165

CEN responded to this mandate in June 2011 by submitting a detailed


standardisation work programme (consisting of several project proposals) to the
European Commission. In December 2012 the European Commission forwarded
the follow-up Mandate M/515 EN Mandate for amending existing Eurocodes and
extending the scope of structural Eurocodes to CEN. In general, the mandate
foresees that an additional Eurocode on structural glass and substantial additions to
the existing standards are developed.

2. EVOLUTION OF EUROCODE 3 PART 1.5

CEN Technical Committee 250 Structural Eurocodes (CEN/TC250)


prepared the above mentioned standardisation work programme and has now the
mandate to prepare the second generation of Eurocodes. For existing Eurocodes
several subcommittees (SC) already exist within CEN/TC250 and it is SC3 which
deals with Eurocode 3 on Structural Steel. Following Mandate M/515 EN,
CEN/TC250/SC3 created several Working Groups (formerly called Evolution
Groups) for the evolution of the different parts of Eurocode 3. This paper reports
about the work of Working Group EN 1993-1-5 which deals with the design of
plated structural elements (plate buckling).
Today, aside of other tasks, the technical development of rules for plate
buckling is mainly driven forward by the European Convention for Constructional
Steelwork (ECCS) and its Technical Working Group 8.3 (ECCS/TWG83). Therefore it
became obvious that both groups are brought closely together in order to minimize
work and time effort. For that reason Working Group EN 1993-1-5 and the
ECCS/TWG83 group have common meetings and work jointly together on the
further development of EN 1993-1-5. Within the task SC3.T4 of the Mandate
M/515 EN the following sub-tasks have been formulated:
reduction in number of Nationally Determined Parameters (NDPs),
enhanced ease of use,
imperfections for flat plate elements,
improved interaction rules for plates,
improved patch loading rules for plates,
stiffener design,
harmonization of design rules for stiffened plated elements,
guidance for use of FEM in design,
development of advanced design rules for extended girder applications such
as corrugated webs.
The mandate will have four overlapping phases with EN 1993-1-5 being in
the second phase which is planned starting in 2015. Since there will be a large
number of amendments dealing with all Eurocode 3 parts which are presented to
CEN/TC250/SC3 in the end, it has been decided within ECCS/TWG83 group and
166 Ulrike Kuhlmann, Benjamin Braun 3

Working Group EN 1993-1-5 to have as many amendments ready as possible and


as early as possible. This paper summarizes the amendments which have been
presented to CEN/TC250/SC3 for decision. Symbols and abbreviations are given at
the end of this paper. Nevertheless, it is recommended to the reader to have the
standard EN 1993-1-5 at hand for the following sections.
In the end, an outlook is given on the proposals which are currently under
preparation. Besides that, discussions are ongoing in order to harmonize EN 1993-
1-5 with other Eurocode parts such as EN 1993-1-1 e.g. on Class-4 cross-sections
and EN 1993-1-3, e.g. on Annex D dealing with the use of the effective thickness
method.

3. AMENDMENTS COVERING ENHANCED USE

3.1. RESISTANCE OF LONGITUDINALLY STIFFENED PLATES


SUBJECTED TO DIRECT STRESSES; INTERACTION
BETWEEN PLATE AND COLUMN BUCKLING
The determination of the interpolation coefficient in Clause 4.5.4(1) which
identifies whether the plate tends to plate buckling or column-like buckling is an
extensive calculation procedure as it requires the use of several different sections of
EN 1993-1-5 i.e. Section 4.5.3, Annex A.1, Annex A.2.1 and Annex 2.2.
A simplification of this procedure has been already discussed in the COMBRI
project [1] but was elaborated in detail by Darko Beg [24]. The amendment aims
to simplify the existing design rules by giving a direct calculation method for the
interpolation coefficient , which will thus lead to a strong consolidation of the whole
calculation procedure of Clause 4.5.4(1). For longitudinally stiffened plates the
interpolation coefficient may be obtained directly from one of the following equations.
For orthotropic plates with at least three stiffeners the interpolation
coefficient may be obtained from:
bsl ,1 1 +
= k , p 2 1 for 1 ; (1)
bc

(1 + )
2
2
1
= for = 1, 0,5, 0,5 < 4 , (2)

but 0 1 .
Parameters k,p, , bsl,1, bc, , and are specified according to Annex A.1
and Figure A.1, EN 1993-1-5.
For orthotropic plates with one or two stiffeners the interpolation coefficient
may be obtained from:
4 Evolution of Eurocode 3 167

bsl ,1 Asl ,1t 2


= k , p 2
1, (3)
bc I sl ,112(1 2 )
where k,p is known from relevant computer simulations, or according to Annex
A.2, EN 1993-1-5.
For the most common case a ac and by using 4.5.3(3), EN 19931-5, for cr,c
the interpolation coefficient can be expressed as:
a 4bt 3
= 4 for a ac . (4)
4 (1 2 )b12b22 I sl ,1
Parameters k,p, , a, ac, b, b1, b2, t, , Asl,1 and Isl,1, are specified according to
Annex A.2, EN 19931-5. The geometrical values bsl,1 and bc are specified in Figure
A.1, EN 1993-1-5.
For the case with two stiffeners in the compression zone should be
calculated for the three cases given in Annex A.2.1(7), EN 1993-1-5, and the
lowest value is taken as relevant.

3.2. RESISTANCE OF LONGITUDINALLY STIFFENED PLATES


SUBJECTED TO DIRECT STRESSES; EFFECTIVE AREA

Numerical studies in [1] have shown that Equation (4.5) in Clause 4.5.1(4)
and the resulting effectivep areas due to plate buckling may lead to unsafe results
for plates with weak stiffeners.
As a result stiffened plates having weak longitudinal stiffeners should be
considered as unstiffened plates regarding their resistance to direct stresses and
their effectivep area should be calculated as unstiffened plates according to Section
4.4, EN 1993-1-5. Longitudinal stiffeners should be considered as weak stiffeners
if their relative bending stiffness is less than 25, where is defined by:
E Is E t3
= , with D =
bD 12 1 2
.
( ) (5)

3.3. SHEAR RESISTANCE OF LONGITUDINALLY STIFFENED GIRDERS

Studies in [1, 5] have shown that for unstiffened web panels or panels stiffened
by open cross-section stiffeners the assumption of hinged boundary conditions is a
requirement for the use of the shear buckling curves in Clause 5.3(1). Thus the
factor w according to Table 5.1, EN 1993-1-5, is only valid for slendernesses w
which are determined for plates with hinged boundary conditions. However, it can
be shown that closed-section longitudinal stiffeners have a beneficial effect on the
168 Ulrike Kuhlmann, Benjamin Braun 5

overall plate buckling resistance compared to open section stiffeners. This is


particularly the case for closed-section longitudinal stiffeners connected to the end-
posts. In such a situation, additional rigidity is provided to the end-posts by the
longitudinal stiffener. In Clause 5.3(2) and Clause 5.3(4) this effect is not
considered so far.
Thus in Clause 5.3(2) it should be added that for webs stiffened by closed-
section longitudinal stiffeners connected to the end posts and vertical stiffeners, the
end posts may always be considered as rigid. Even in the worst situation of a small
closed-section longitudinal stiffener connected to a large transverse stiffener, the
torsional restraint remains significant (even higher than in the case of very large
open stiffeners connected to small transverse stiffeners) and that the average
stiffness ratio in the case of closed sections is of an order of magnitude 100 times
higher than for open sections.
Thus Clause 5.3(4) should be modified such that the second moment of area
of an open-section longitudinal stiffener should only be reduced to 1/3 of its actual
value when calculating the shear buckling coefficient k.

4. AMENDMENTS COVERING TECHNICAL DEVELOPMENT

4.1. RESISTANCE OF STEEL PLATE GIRDERS


SUBJECTED TO PATCH LOADING

The contributions of several recent doctoral studies [610] underline the need
of a further modification of the plastic resistance Fy against patch loading which
appears in the current version of EN 1993-1-5. According to these studies, the
current definition of the plastic resistance overestimates patch loading capacity in
certain cases (hybrid girders) whereas this capacity is slightly underestimated for
others (very slender girders).
Thus Chapter 6 should be modified as follows. In Section 6.4 the reduction
factor F for effective length for resistance should be obtained from:
1 .0
F = 1.0 , (6)
F + F2 F
where

F =
1
2
( (
1 + F 0 F F 0 + F , ) ) (7)

l y t w f yw
F =
Fcr (8)
6 Evolution of Eurocode 3 169

F 0 = 0.75, F 0 = 0.50. (9)

It should be noted that the values according to Equation (9) are based on a
value of M1 = 1.1.
In Section 6.5 the effective loaded length ly should be calculated as follows:
bf
m1 = , (10)
tw
2
h
m2 = 0.02 w if F > 0.5 (11)
t

f

m2 = 0 if F 0.5 ,
For box girders, bf in Equation (10) should be limited to 15tf on each side
of the web.
For types a) and b) in Fig. 6.1, EN 1993-1-5, ly should be obtained using:

(
l y = s s + 2 t f 1 + m1 . ) (12)

4.2. INTERACTION BETWEEN PATCH LOADING,


BENDING MOMENT AND SHEAR FORCE

If steel structures are subjected to the combination of bending, shear and


patch loading, the interacting stability behaviour should be taken into consideration
in design. The combined loading situation can often occur in case of bridge girders
during launching. Therefore the determination of the load carrying capacity under
the combined loading situation is an important aspect of the bridge design. In the
current version of the EN 1993-1-5 there is no standard design method to take the
interaction of these three effects into account and there has been a very small
number of previous investigations in the literature about this topic. Consequently,
no formulation for the interaction between transverse force, bending moment and
shear force is given in Chapter 7. Based on a number of recent studies [1113],
Section 7.2 should be replaced such that if the girder is subjected to a concentrated
transverse force acting on the compression flange in conjunction with bending
moment and shear force, the resistance should be verified using Sections 4.6, 5.5,
6.6, EN 1993-1-5, and the following interaction expression:
1.6
3.6 F
1 + 3 1 Ed +2 1.0 , (13)
2 VEd
170 Ulrike Kuhlmann, Benjamin Braun 7

where
M Ed
1 = , (14)
M pl , Rd

VEd
3 = . (15)
Vbw, Rd

4.3. RESISTANCE OF GIRDERS


WITH CORRUGATED WEBS SUBJECTED TO PATCH LOADING

The current version of the Annex D gives no proposal to determine the


resistance against transverse force in case of girders with trapezoidally corrugated
webs. It should be noted that Annex D, EN 1993-1-5, is the only part of Eurocode
3 which deals with girders with corrugated webs. The objective of this amendment
is to improve the standard with a design method to determine the patch loading
resistance of girders with trapezoidally corrugated webs based on [1418].
The design resistance of trapezoidally corrugated webs can be determined
according to Equation (16) provided that the compression flange is adequately
restrained in lateral direction.
The design method can be used if the load is applied through the flange and
restrained by shear forces in the web (Fig. 1).

FS

ss
SS
hw

V1,S
V2,S

Fig. 1 Loading type.

In case of girders with inclined webs, the internal forces to be taken into
account are the components of the external load in the plane of the web.
The design resistance of the trapezoidally corrugated web to local buckling
under transverse force should be taken as:
FRd = t w f yw s s k . (16)
8 Evolution of Eurocode 3 171

k is the modification factor due to corrugation angle, which should be calculated


from, Fig. 2.
a1 + a2
k = . (17)
a1 + a4

Fig. 2 Corrugated web geometry.


is the reduction factor due to local buckling which should be calculated from:
1.9 0.798
= 1.00 , (18)
p p
2

where

f yw
p = (19)
cr
2
k 2 t
cr = E w (20)
12 (1 )
2
ai
k = 1.11 , ai = max(a1;a2), (21)
i.e. the maximum fold length to which the load is applied (Fig. 2).
The design method is applicable for girders with a fold length larger than
h t
ai w + 260 w . (22)
tw 11.5

5. OUTLOOK

5.1. M-V INTERACTION OF LONGITUDINALLY


STIFFENED PLATE GIRDERS

The M-V interaction formula of Section 7.1, EN 1993-1-5, is an extension of


the plastic interaction to slender cross-sections. It covers the load bearing
behaviour general but has only been checked for longitudinally unstiffened panels
172 Ulrike Kuhlmann, Benjamin Braun 9

[19]. The current M-V interaction formula is based on an empirical model which
was developed on the basis of only few experimental results and that assigns a pure
shear loading to the web-core of a symmetric cross-section. Recent experimental
work regarding the interaction of bending moment and shear force for longitude-
nally stiffened girders was conducted in [20] opening some additional questions
concerning M-V interaction for stiffened girders.
In the frame of [21] six full scale girders were tested and served as a database
for subsequent numerical analyses. It was shown that the gross cross-section check
at the edge of the panel can cover also the stability resistance of the panel subjected
to M-V interaction in the case of symmetric cross-sections. For unsymmetric cross-
sections a new proposal has been derived which, in combination with the gross
cross-section check, gives very reliable results. However, the partial safety factor
determination is a crucial part in this work. Thus, the amendment will cover not only
M-V interaction but also basic issues with respect to partial safety factor determination.

5.2. DESIGN OF INTERMEDIATE TRANSVERSE STIFFENERS

In general, transverse stiffeners at girder ends and at intermediate supports


are designed as strong double-sided stiffeners. In contrast to this, intermediate
transverse stiffeners are designed as single-sided open-section stiffeners which are
intended to increase the strength and the stiffness of the web. The latter kind of
intermediate stiffener is usually not subjected to external loads. Instead effects
from tension field action and deviation forces from longitudinal stresses dominate.
If such a design is based on EN 1993-1-5, the sizing leads to significantly larger
cross-sections in comparison to what is necessary according to numerical analysis.
This can be mainly attributed to the overestimation of the axial force in the
stiffener due to tension field action [22, 23].
Further research has been carried out [24] whether a stiffness-only approach
is justifiable in contrast to a combined approach which checks stiffness and force.
It can be shown that the design requirements of a rigid intermediate transverse
stiffener may be obtained by fulfilling simple stiffness criteria which simplifies the
design procedure while considering all relevant effects. The amendment will define
a minimum required second moment of area for rigid transverse stiffeners which
covers the design requirements imposed by the different loadings.

5.3. TRANSVERSE BENDING MOMENTS


IN CORRUGATED WEB GIRDERS

In corrugated web girders which are subjected to combinations of shear force


and bending moment, transversal bending moments Mz occur. Current research
studies the effect of transverse bending moments in both bridge and building
structures. In particular the influence of support conditions and of Mz on the
10 Evolution of Eurocode 3 173

bending resistance is investigated [25]. Only few reports have dealt with the effect
of Mz on the bending resistance so far [26].
Therefore, a large number of different girder geometries with trapezoidal and
sinusoidal web shape have been analysed numerically. The results show that the
influence of Mz is more dominant for bridges than for building structures. In
general, support conditions play an important role. In comparison, Eurocode rules
give a reduction in bending resistance that is up to seven times higher than the
maximum reduction identified in the numerical analyses. It can be concluded that
the reduction factor for bending resistance is negligible with respect to ultimate
behavior. However, when first yielding is considered as the limit, the reduction can
be significant. Thus, a formula which determines the additional stresses in an
elastic analysis is under preparation. The amendment will be fully addressed in EN
1993-1-5 since it is the only part in Eurocode 3 where corrugated web girders are
dealt with.

5.4. APPLICATION OF THE REDUCED STRESS METHOD

EN 1993-1-5 provides two methods for the design of plated structures:


effective width method and reduced stress method. The reduced stress method has
been introduced at a very late stage of the development of current EN 1993-1-5 as
an alternative verification method to the effective width method. In [27] the
background to this method has been outlined and shortcomings with respect to the
biaxial compression state have been discussed. It is shown that the reduced stress
method offers advantages over todays numerical procedures which allow the
elastic critical load factor of the full stress field to be determined in a single step,
thus taking interaction into account in a very efficient way. However, a
modification has been proposed for biaxially compressed plates in [11] since the
interaction verification in its pure format based on the Von Mises yield criterion is
not able to represent the actual mechanical behaviour. The proposal leads to appropriate
and plausible results without over-complicating the interaction equation.
In general, the reduced stress method offers the possibility to consider the
effect of tensile stresses in plates with multiaxial stress state. However, tensile stresses
have not only a stabilizing effect on buckling, but also a plastic destabilizing
influence. In [28] first results how tensile stresses influence the buckling behavior
are presented. This work will be further developed and combined with the research
on biaxially compressed plates into the common amendment which will improve
the reduced stress method in its actual format.

6. CONCLUSIONS

The decision within ECCS/TWG83 group and Working Group EN 1993-1-5


to prepare as many amendments as early as possible has already generated an
174 Ulrike Kuhlmann, Benjamin Braun 11

amount of considerable amendments for the further development of EN 1993-1-5.


These amendments have been presented in this paper in a similar way as presented
to CEN/TC250/SC3 for decision and gained the preliminary acceptance. Thus they
can be easily implemented in the future code when it will be developed in the
frame of CEN/TC250 Mandate M/515 EN. A large number of proposals is still to
come including the topics summarized in the outlook and the tasks formulated in
Mandate M/515 EN.

Acknowledgements. The authors gratefully acknowledge the contribution of the different


amendments authors and the collaboration among the members of Working Group TC250/SC3/EN
1993-1-5 and ECCS Technical Working Group 8.3.

Received on July 16, 2014

SYMBOLS AND ABBREVIATIONS

a
= ; aspect ratio of length a and width b
b
F0 reduction curve parameter

=
A sl
; ratio of gross area sum of longitudinal stiffeners Asl
bt
and plate gross area
235
=
f y N/mm 2

E Is
= ; relative bending stiffness of a stiffener
bD
M1 partial safety factor

F ; p ; p slenderness parameters

F0 reduction curve parameter


Poissons coefficient
cr elastic critical buckling stress

2 ; stress ratio of edge stresses 1 (larger) and 2 (smaller)


=
1
12 Evolution of Eurocode 3 175

; F; w reduction factors
a plate length

I sl ,1 b12 b22
ac = 4.33 4 ; plate parameter
t3 b
Asl,1 gross area of a longitudinal stiffener
b plate width
b 1; b 2 subpanel widths
bc plate width under compression
bf flange width
bsl,1 distance between stress neutral axis and longitudinal stiffener

D E t 3 ; plate bending stiffness


=
(
12 1 2)
E Youngs modulus
fyw web yield strength
hw web height
Isl,1 second moment of area of a longitudinal stiffener
Is second moment of area of longitudinal stiffener for out-of-plane
bending, its cross-section including a participating width of web
of 10t each side of each stiffener-to-web junction
ly effective patch loaded length
k; k,p; k buckling coefficients
m1; m2 patch loading parameters
MEd design bending moment
Mpl,Rd design plastic moment resistance
(irrespective of cross-section class)
ss patch loading length
t plate thickness
tf flange thickness
176 Ulrike Kuhlmann, Benjamin Braun 13

tw web thickness
VEd design shear force
Vbw,Rd design shear resistance of the web contribution

REFERENCES

1. KUHLMANN, U., et al., Valorisation of Knowledge for Competitive Steel and Composite Bridge,
RFCS, European Commission, Final Report, 2009.
2. JOHANSSON, B., MAQUOI, R., SEDLACEK, G., MLLER, C., BEG, D., Commentary and
worked examples to EN 1993-1-5 Plated Structural Elements, European Commission,
2007.
3. BEG, D., KUHLMANN, U., DAVAINE, L., BRAUN, B., Eurocode Design Manual Design of
Plated Structures, ECCS (Ed.), Ernst & Sohn, 2010.
4. BEG, D.: Background document to Amendment AM-1-5-2011-2, SC3 document N1863, 2012.
5. DEGE, H., BOISSONNADE, N., MAQUOI, R., Numerical investigations on web panels stiffened
by one or two trapezoidal stiffeners, Proceedings of the 5th International Conference on
Coupled Instabilities in Metal Structures, CIMS 2008, Sydney, Australia, 2008, pp. 324332.
6. DAVAINE, L., Formulations de la rsistance au lancement dune me mtallique de pont raide
longitudinalement, PhD Thesis D05-05, INSA de Rennes, France, 2005.
7. GOZZI, J., Patch loading resistance of plated girders, PhD Thesis 2007:30, Lule University of
Technology, Sweden, 2007.
8. CLARIN, M., Plate Buckling Resistance Patch Loading of Longitudinally Stiffened Webs and
Local Buckling, PhD Thesis 2007:31, Lule University of Technology, Sweden, 2007.
9. CHACON, R., Resistance of Transversally Stiffened Hybrid Steel Plate Girders to Concentrated
Loads, PhD Thesis, Universitat Politcnica de Catalunya, Barcelona, Spain, 2008.
10. CHACON, R., MIRAMBELL, E., REAL, E., Hybrid steel plate girders subjected to patch
loading. Part 2: Design proposal, Journal of Constructional Steel Research, 66, pp. 709715,
2010.
11. BRAUN, B., Stability of steel plates under combined loading, PhD Thesis, No. 20103, Institute
of Structural Design, University of Stuttgart, 2010.
12. KVESDI, B., ALCAINE, J., DUNAI, L., MIRAMBELL, E., BRAUN, B., KUHLMANN, U.,
Interaction behaviour of steel I-girders under bending, shear and transverse force Part I:
Longitudinally unstiffened girders, Journal of Constructional Steel Research, 2014 (under
review).
13. KVESDI, B., ALCAINE, J., DUNAI, L., MIRAMBELL, E., BRAUN, B., KUHLMANN, U.,
Interaction behaviour of steel I-girders under bending, shear and transverse force Part II:
Longitudinally stiffened girders, Journal of Constructional Steel Research, 2014 (under
review).
14. KVESDI, B., Patch loading resistance of girders with corrugated webs, PhD Dissertation,
Budapest University of Technology and Economics, Hungary, 2010.
15. KVESDI, B., BRAUN, B., KUHLMANN, U., DUNAI, L., Patch loading resistance of girders
with corrugated webs, Journal of Constructional Steel Research, 66, pp. 14451454, 2010.
16. KVESDI, B., BRAUN, B., DUNAI, L., KUHLMANN, U., Design of girders with trapezoidally
corrugated webs under patch loading, shear and bending interaction, Steel Construction
Design and Research, 5, pp. 1622, 2012.
17. KVESDI, B., BRAUN, B., KUHLMANN, U., DUNAI, L., Enhanced design method for the
patch loading resistance of girders with corrugated webs, Proceedings of the 5th European
14 Evolution of Eurocode 3 177

Conference on Steel and Composite Structures, EUROSTEEL 2008, Volume B, September 3-


5, 2008, Graz, Austria, pp.11551160, 2008.
18. KUHLMANN, U., BRAUN, B., KVESDI, B., The patch loading resistance of girders with
corrugated webs, Proceedings of the 5th International Conference on Coupled Instabilities in
Metal Structures, CIMS 2008, Sydney, Australia, pp. 309316, 2008.
19. VELJKOVIC, M., JOHANSSON, B., Design for buckling of plates due to direct stress, Nordic
Steel Construction Conference, Helsinki, Finland, June 20-23, 2001.
20. SINUR, F., Behaviour of longitudinally stiffened plated girders subjected to bending-shear
interaction, PhD Thesis, Faculty of Civil and Geodetic Engineering, University of Ljubljana,
2011.
21. KUHLMANN, U., BEG, D., ZIZZA, A., SINUR, F., REJEC, K., Bending-shear interaction of
longitudinally stiffened steel plates, DFG Report No. KU 1130/14-1, 2013.
22. HENDY, C. R., PRESTA, F., Transverse web stiffeners and shear moment interaction for steel
plate girder bridges, Proceedings of the International Conference on Steel Bridges, ECCS,
June 46, 2008, Guimares, Portugal, 2008, p. 8.
23. PRESTA, F., Post-buckling behaviour of transversely stiffened plate girders, PhD Thesis,
Cosenza, Universita degli studi della Calabria, 2007.
24. SINUR, F., BEG, D., Intermediate transverse stiffeners in plate girders, Steel Construction
Design and Research, 5, pp. 2332, 2012.
25. KVESDI, B., JAGER, B., DUNAI, L., Stress distribution in the flanges of girders with
corrugated webs, Journal of Constructional Steel Research, 79, pp. 204215, 2012.
26. BALAZ, I., KOLEKOVA, Y., Influence of Transverse Bending Moment in the Flange of
Corrugated I-Girders, Procedia Engineering, 40, pp. 2631, 2012.
27. BRAUN, B., KUHLMANN, U., Reduced stress design of plates under biaxial compression, Steel
Construction Design and Research, 5, pp. 3340, 2012.
28. ZIZZA, A., KUHLMANN, U., Tensile Stresses on the Buckling Behaviour of Slender Plates,
IABSE Young Engineers Colloquium 2014, Dresden, Germany, March 10, 2014, pp. 1011.
BUCKLING CURVES
FOR HEAVY WIDE FLANGE STEEL COLUMNS

H. H. SNIJDER1, L.-G. CAJOT2, N. POPA2, R. C. SPOORENBERG3

Abstract. This paper proposes existing European buckling curves to be used for
checking the resistance of heavy wide flange columns made from mild and high-
strength steel, failing by flexural buckling. Buckling curves are not available in the
current Eurocode3 EN 1993-1-1, for height-to-width ratios h/b > 1.2 and flange
thicknesses tf > 100 mm. The buckling curves are evaluated according to the statistical
procedure given in Annex D of EN 1990 using finite element analyses. Residual stress
models as described in literature were used to define the initial stress state of the
column in the finite element model. A large database was created containing the ratio
between the elastic-plastic buckling resistance obtained from finite element analysis
and the buckling resistance obtained from the proposed buckling curve for a wide set
of column configurations from which a partial factor Rd was deduced. Different
section types with flange thicknesses tf >100 mm were investigated: the stocky HD
and more slender HL type, featuring h/b = 1.23 and h/b = 2.35 respectively. The
materials investigated were:
Quenched and Self-Tempered (QST) steel available under the proprietary name
HISTAR 460 (High Strength ArcelorMittal) with a yield stress of 460 N/mm2;
steel grade S460;
steel grade S355;
steel grade S235.
For as far as available, statistical information on these materials was used to estimate
the partial factor for material properties m. Then the partial (safety) factor M1 can be
calculated as M1 = Rd m. Based on the criterion that M1 should not exceed 1.05,
buckling curves are suggested which can be used together with M1 = 1.0. Buckling
curves to be included in Eurocode3 EN 1993-1-1 are finally proposed for heavy wide
flange columns in S235 to S500, with cross-sections with height-to-width ratios
h/b>1.2 and flange thicknesses tf > 100 mm.
This paper is an extended and more complete version of an earlier paper [1].

Key words: buckling curves, finite element analyses, heavy wide flange sections, high-
strength steel, mild steel, partial factor, statistical evaluation.

1. INTRODUCTION

The advent of Quenched and Self-Tempered (QST) steel sections which


combine high strength (i.e. nominal yield stress greater than 430 N/mm2) with good
ductility and weldability has led to a broadening of the possibilities in steel
1
Eindhoven University of Technology, Department of the Built Environment, P.O. Box 513,
5600 MB Eindhoven, The Netherlands
2
ArcelorMittal, Long Products, 66, rue de Luxembourg, L-4009 Esch/Alzette, Luxembourg
3
Iv-Consult b.v., P.O. Box 1155, 3350 CD Papendrecht, The Netherlands
2 Buckling curves for heavy wide flange steel columns 179

construction. This manufacturing method can also be applied to produce heavy


wide flange sections, i.e. wide flange sections with flanges thicker than 40 mm. At
the moment, heavy wide flange QST sections are manufactured by ArcelorMittal
under the proprietary name of HISTAR (HIgh-STrength ARcelorMittal). Two
grades are currently produced: HISTAR 355 and (high-strength) HISTAR 460,
which possess a yield stress of 355 N/mm2 and 460 N/mm2 respectively, not
considering reduction of yield stress with increasing material thickness. Heavy
wide flange HISTAR 460 sections have already been applied worldwide, with the
majority in the United States where the US equivalent of HISTAR 460, Grade 65,
is covered by ASTM A913 [2, 3].
Besides the high yield stress, HISTAR 460 sections have improved material
properties for wide flange sections possessing thick flanges. For HISTAR 460 a
smaller reduction in yield stress needs to be incorporated for greater material
thicknesses according to ETA-10/0156 [4] when compared to other grades (e.g.
S460M and S500M according to EN 10025-4 [5]) as illustrated in Fig. 1. For
HISTAR 460 and S460 sections with flange thicknesses exceeding 100 mm, the
yield stress is 450 and 385 N/mm2 respectively.
minimum yield
stress [N/mm2]

Fig. 1 Decrease of yield stress of HISTAR 460, S460 and S500


with increasing material thickness.

For S355 and S235, depending on whether the material is classified as a non-
alloy structural steel, normalized fine grain structural steel or thermomechanical
fine grain steel according to EN 10025-2 [6], EN 10025-3 [7] and EN 10025-4 [5],
respectively, a substantially reduced yield stress must be used to account for a
reduction in material properties for thick plated parts. For the present study with
sections possessing flange thicknesses between 100 and 150 mm, the yield stress
for S235 steel is 195 N/mm2. For S355 the lowest yield stress value according to
the three different standards is selected: 295 N/mm2.
180 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 3

As such, heavy wide flange HISTAR 460 sections are used to their best
advantage when the ultimate limit state is the governing design criterion. This is
the case when applied as gravity columns in multi-story buildings, beams in short-
or medium-span bridges or chord- and brace members as part of truss-like
structures. In these situations the design is most often controlled by the flexural
buckling resistance of the member for which due allowance has to be made
according to the relevant design codes.

Table 1
Buckling curve selection table according to Eurocode3, EN 1993-1-1
Buckling curve
S 235 S460
Buckling
Cross-section Limits S 275
about axis
S 355
S 420
Rolled I-sections y-y a a0
tf 40 mm
z-z b a0
h/b > 1.2
y-y b a
40 < tf 100 mm
z-z c a

y-y b a
tf 100 mm
h/b 1.2 z-z c a

y-y d c
tf > 100 mm
z-z d c

HISTAR 460 falls in the category S460 in Table 1. Small and medium-sized
HISTAR 460 sections with flange thickness tf smaller than or equal to 40 mm are
assigned to buckling curve a or a0 depending on the value of the height-to-width
ratio h/b. Heavy HISTAR 460 sections which have a flange thickness smaller than
or equal to 100 mm can be designed according to buckling curve a. Heavy sections
possessing a flange thickness in excess of 100 mm and an h/b-ratio smaller than 1.2
are assigned to buckling curve c. For heavy HISTAR 460 sections having h/b-
ratios greater than 1.2 and flanges thicker than 100 mm no buckling curves are
available. The same is true for heavy sections in S460, S355 and S235 for h/b > 1.2
and tf > 100 mm.
In order to arrive at buckling curves reflecting the buckling response for
heavy sections in HISTAR 460, S460 , S355 and S235 with flange thickness larger
than 100 mm and h/b-ratios greater than 1.2, a combined experimental and
numerical study was initiated by ArcelorMittal in Luxemburg and set up and
executed by Eindhoven University of Technology in the Netherlands. The experiments
consisted of residual stress measurements performed on two different heavy wide
flange section types made in steel grade HISTAR 460. A residual stress model was
4 Buckling curves for heavy wide flange steel columns 181

proposed which can be used for heavy wide flange QST sections. The testing
procedure and the derivation of this so called QST residual stress model are
detailed in [9]. Since the manufacturing process of heavy S460 sections is identical
to that of HISTAR 460 sections, the residual stress model of the latter was applied
to the S460 analyses. As no experimental data is available to model the residual
stresses state for heavy HL sections made from S235 or S355 an assumption was
made on their distribution. For the S235 members the so called ECCS residual
stress model [10] featuring a bilinear stress pattern over the web height and flange
width was used to define the initial stress state. For the sections made from S355
steel the QST residual stress model was adopted in addition to the ECCS residual
stress model.
In the present paper, existing ECCS buckling curves are proposed to check
the flexural buckling resistance of heavy HISTAR 460 sections. The reliability of
the suggested buckling curves is evaluated according to annex D of EN 1990 [11].
The buckling resistance for a wide set of columns in HISTAR 460, S460, S355 and
S235 is evaluated with the finite element method using the residual stress models
mentioned to define the initial stress state of the column and with widely accepted
geometric imperfections.

1.1. EARLIER APPROACHES FOR DERIVATION OF BUCKLING CURVES

From the earliest experiments on pin-ended columns failing by flexural


buckling it was observed that the slenderness (ratio between length and radius of
gyration) of the member has profound influence on the buckling response. This led
to the development of the buckling curve concept, relating the load a column can
withstand before instability occurs to its non-dimensional or relative slenderness
(slenderness normalized against the steel properties). Important references include
[12, 13]. The studies underlying the buckling curve concept were often based on a
two-fold approach: to obtain the elastic-plastic buckling resistance through full-
scale column testing and to conduct theoretical (and later numerical) analyses to
replicate and supplement the experimental results. The theoretical analyses were
expanded to include a wide set of columns not part of the experimental plan from
which design rules (buckling curves) were proposed. The accuracy of the buckling
curve was often evaluated through comparison with characteristic values from full-
scale tests performed, where good agreement between the buckling curve and test
justified the selected buckling curve.

1.2. STATISTICAL EVALUATION OF RESISTANCE MODELS

The earlier approaches to arrive at buckling curve formulations have become


obsolete as with the appearance of EN 1990 Annex D [11] Design assisted by
testing it is now possible to make a statistical evaluation for new design rules and
182 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 5

existing ones and to quantify the variability of salient parameters. In brief, the EN
1990 states that the design resistance (Rd) may be obtained directly from the
quotient of the characteristic (Rk) strength and the partial factor M:
Rk
Rd = , (1)
M
where M can be subdivided as follows:
M = Rd m , (2)
where: M is the partial factor for a material property, also accounting for model
uncertainties and dimensional variations according to EN 1990 [11] or general
partial factor according to EN 1993-1-1 [8]; Rd is the partial factor associated with the
uncertainty of the resistance model; m is the partial factor for a material property.
A distinction for the general partial factor M is made depending on the failure
mode of the member under investigation. In the present study, columns are investigated
for which loss of stability is the governing failure mode. Therefore the general
partial factor is in line with EN 1993-1-1 denoted 1 throughout this paper.
The general partial factor serves as a reduction for the capacity: high
M1-values impose a larger reduction on the buckling capacity compared to lower
M1-values.
One of the earliest studies concerning the statistical evaluation of resistance
models was carried out by Sedlacek et al. [14] at RWTH Aachen, Germany.
Although the investigation was performed prior to the final appearance of EN
1990, it adopted the same methodology. The study was aimed at finding new
imperfection factors for the resistance model of Eurocode3 (EN 1993-1-1) to check
the lateral-torsional buckling resistance of rolled and welded beams. The reliability
of the old resistance model, originally from the DIN, in addition to the new
resistance model was re-evaluated. The statistical evaluation was based on 144
lateral-torsional buckling tests.
A probabilistic assessment of the existing design rules to check the lateral
torsional buckling resistance of beams was performed by the University of Coimbra,
Portugal for wide flange beams. The partial factor associated with the uncertainty
of the resistance model Rd was computed for different load cases and section types
using the three different design models for lateral torsional buckling available in
EN 1993-1-1, Rebelo et al. [15]. The evaluation of the partial factors was based on
the solution results from finite element analyses conducted on a wide set of beam
configurations.
In the accompanying paper by Simoes da Silva et al. [16], the partial factor
for the material properties m was determined based on tensile tests conducted on
sections made from different steel grades. Using equation (2), the factors from [15, 16]
were used to arrive at M1-values for different load cases and steel grades.
6 Buckling curves for heavy wide flange steel columns 183

1.3. PRESENT STUDY


In the present study the methodology adopted by Rebelo et al. [15] and in
line with Annex D of EN 1990 will be used to check the reliability of proposed
buckling curves for heavy wide flange sections.
A large database is created containing the ratio between the elastic-plastic
buckling resistance obtained from non-linear finite element analyses and buckling
resistances obtained from the suggested buckling curve for a wide set of column
configurations. It is mentioned that the numerically obtained elastic-plastic buckling
resistance serves as a replacement of the ultimate resistance found in a column
buckling test. The ratio between both buckling resistances for a specific set of
columns is used to compute the partial factor Rd associated with the uncertainty of
the resistance model.
For S460, statistical literature information concerning the relevant material
property, being the yield stress fy, is used to estimate m < 1.0. Then the general
partial factor M1 is computed according to equation (2). Since for HISTAR 460
such information is not yet available m = 1.0 can be conservatively used to
compute the general partial factor M1 according to equation (2). This general
partial factor can have a lower value in the nearby future pending the availability of
a lower m-value representing the variability in mechanical properties of HISTAR
460 steels. For S355 and S235 m = 1.0 was also conservatively adopted to compute
the general partial factor M1 according to equation (2).
The study is limited to heavy wide flange sections which possess a flange
thickness (tf) greater than 100 mm and for which the height-to-width ratio (h/b) is
greater than 1.2. The selected sections for the present study are listed in Table 2. In
the present study a distinction is made between HD en HL sections, having an
h/b-value of approximately 1.23 and 2.35, respectively.
Table 2
Heavy wide flange sections offered by ArcelorMittal with h/b > 1.2 and tf > 100 mm
Section name: Weight
h b tw tf h/b
European per m
[mm] [mm] [mm] [mm] []
American Imperial [kg]
HD 400 900
900 531 442 65.9 106 1.20
W1416605
HD 400 990
990 550 448 71.9 115 1.23
W1416665
HD 400 1086
1086 569 454 78 125 1.25
W1416730
HD 400 1202
1202 580 471 95 130 1.23
W1416808
HD 400 1299
1299 600 476 100 140 1.26
W1416873
HL 920 1194
1194 1081 457 60.5 109 2.37
W3616.5802
HL 920 1269
1269 1093 461 64 115.1 2.37
W3616.5853
HL 920 1377
1377 1093 473 76.7 115.1 2.31
W3616.5925
184 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 7

2. BUCKLING CURVE FORMULATION

The theoretical resistance defining the maximum compressive force a column


can withstand before failing in a flexural buckling mode is determined using EN
1993-1-1 [8]. The buckling resistance for a column can be verified as follows:
N Ed
1.0 , (3)
N b,Rd
where: NE,d is the design value of the compression force; Nb,Rd is the design
buckling resistance.
The design buckling resistance is given by:
Af f
N b,Rd = , (4)
M1
where A is the cross-sectional area, fy is the yield stress, M1 is the general partial
factor for instability and is the buckling reduction factor. This check is only valid
for sections belonging to cross-sectional class 1, 2 or 3. The product of the cross-
sectional area and the yield stress is known as the squash load of the cross-section
or Npl. Using equations (3, 4), the verification of the buckling resistance can be
rewritten as follows:
N Ed
1.0 . (5)
N pl /M1
The buckling reduction factor can be computed according to:
1
= but 1.0 , (6)
+ 2 2
where

(
= 0.5 1 + ( 0.2 ) + 2 . ) (7)

The relative slenderness can be determined as follows:

= N pl N cr , (8)

where Ncr is the elastic critical force of the column. The imperfection factor
attains one of the values as listed in Table 3, depending on the cross-section, steel
grade and buckling case (weak-axis or strong-axis buckling) under consideration.
8 Buckling curves for heavy wide flange steel columns 185

Table 3
Imperfection factor for buckling curves
Buckling curve a0 a b c d
Imperfection factor 0.13 0.21 0.34 0.49 0.76

A graphical representation of the buckling curves is shown in Figure 2. Based


on the selected buckling curve and corresponding imperfection factor a theoretical
resistance can be computed for a heavy section if the relative slenderness is
known. This value will be compared to the elastic-plastic buckling resistance
obtained from non-linear finite element analysis (Section 3).
1

a0
0.8 a
b
Relative resistance [-]

c
0.6 d

0.4

0.2

0
0 1 2 3
Relative slenderness [-]

Fig. 2 Buckling curves from Eurocode3.

3. FINITE ELEMENT MODEL

3.1. ELEMENTS

The geometrical and material non-linear analyses on the columns containing


imperfections (GMNIA) were performed in the ANSYS v.11.0 implicit
environment. The columns were modelled with beam elements. The 3D three node
finite strain element (BEAM189) was selected for the analyses as it can describe
plasticity, large deformations and large strains. A user-defined cross-section was
modelled based on nominal dimensions (Table 2). The cross-section is subdivided
into different cells to capture growth of plastic zones across the cross-section. Each
186 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 9

cell contains four integration points where the stresses are evaluated (Fig. 4a). Two
integration point locations in longitudinal direction of each element describe
progressive yielding along the length of the column. A total of 20 elements along
the length of the member were considered sufficient. Earlier research studies on
column buckling have shown that this element type is able to replicate
experimental elastic-plastic buckling tests with good accuracy thereby taking into
account the effects of residual stresses [17, 18].

3.2. BOUNDARY CONDITIONS

All selected column configurations for the present investigations were simply
supported. The column was pin-supported and torsionally restrained at the bottom.
The same boundary condition was applied at the top with the exception that
vertical translation was permitted. For the evaluation of strong-axis buckling, the
column was restrained against weak-axis deflections by translational supports
along the length (Fig. 3).

(a) weak-axis buckling (b) strong-axis buckling


uy=uz=x=0 F uy=uz=x=0 F

uy=0
L

x y
z z

y y
ux=uy=uz=x=0 ux=uy=uz=x=0

Fig. 3 Boundary conditions.

3.3. RESIDUAL STRESSES

An individual residual stress value was set for each integration point in the
cross-section. The stress value specified for each integration point is assigned to
the tributary area belonging to that integration point, rendering a step-wise initial
stress pattern over the cross-section (Fig. 4b). Here the pattern for the residual
stress model of [9] for HISTAR 460 is shown. The residual stresses are constant
across the flange thickness and web thickness. After inserting the residual stresses
into the element, a first solution step was issued to verify internal equilibrium of
the residual stress model. Insignificant differences were observed between the residual
10 Buckling curves for heavy wide flange steel columns 187

stress model and the stresses after solving, indicating correct implementation of the
residual stress model (Fig. 4c).

Fig. 4 Finite element discretization of cross-section and implemented residual stresses (for HD 4001202).

For HISTAR 460 and for S460, the QST residual stress model of [9] was
adopted as depicted in (Fig. 5a). This model has a parabolic shape along the web
height and flange width. The magnitude of the residual stresses in the web is
related to the section dimensions.

a) QST [9] b) ECCS [10]


Fig. 5 Residual stress models.

Currently no experimental data on the residual stress distribution in mild steel


HL and HD sections is known to the authors. In order to include the effect of
residual stress on the resistance of these sections when failing by strong- and weak-
axis buckling an assumption is made about their distribution. Two different
residual models are selected to make an educated guess concerning the residual
stresses: the ECCS residual stress model [10] commonly used for wide flange hot-
rolled sections (Fig. 5b) and the earlier derived QST residual stress model (Fig. 5a).
188 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 11

As the latter model has residual stress values exceeding the (reduced) yield stress
of S235 steels, which is physically not possible, the QST residual stress model is
not used for the S235 columns. The ECCS residual stress model is featured by a bi-
linear stress distribution along the web height and flange widths. The extreme
values are at the flange tips and web center (compression) and web-to-flange
junction (tension) and set at 30 % of the (unreduced) yield stress. The unreduced
yield stresses, fy = 235 N/mm2 and fy = 355 N/mm2 for S235 and S355 respectively,
were used which is conservative in view of defining the residual stress values.
Using fy = 355 N/mm2 may be very conservative since normally fy = 235 N/mm2 is
used with the ECCS residual stress model regardless the steel grade.
The residual stress values at the most critical locations for the QST and
ECCS residual stress models are listed in Table 4 and Table 5, respectively.

Table 4
Residual stress values for QST residual stress model [9] [N/mm2]
frt frc wrt wrc
Steel grade Section
(tension) (compression) (tension) (compression)
HD 400 900 81 135 81 101
HISTAR460 HD 400 990 81 135 81 101
HD 400 1086 81 135 81 102
S460 HD 400 1202 81 135 81 95
HD 400 1299 81 135 81 97
S355 HL 920 1194 180 225 180 219
HL 920 1269 180 225 180 220
HL 920 1377 180 225 180 201

Table 5
Residual stress values for ECCS residual stress model [10] [N/mm2]
Flange tip/web center Web-to-flange junction
Steel grade Section
(compression) (tension)
S235 all 0.3 235 = 71 0.3 235 = 71
S355 all 0.3 355 = 107 0.3 355 = 107

3.4. MATERIAL MODEL

A bi-linear material model was applied to describe the materials response to


loading (Fig. 6). A fixed yield stress value fy was used to define the onset of
yielding. This value is based on the steel properties, thereby taking into account a
reduction in yield stress due to the thickness of the flanges. A more generally
accepted value for the Youngs modulus of 200 000 N/mm2 was adopted to define
the elastic stage of the material. Strain hardening increases the ultimate load of
stocky columns but hardly affects columns in the intermediate and high slenderness
ranges. Neglecting strain hardening is a conservative approach and is in line with
what has previously been done by other researchers, e.g. [15, 16]. Therefore, no
strain hardening effects were included.
12 Buckling curves for heavy wide flange steel columns 189

Steel grade Yield stress


fy
[N/mm2]
HISTAR 450
460
S460 385
S355 295
S235 195
E = 200 000 N/mm2

Fig. 6 Material model.

3.5. GEOMETRIC IMPERFECTIONS

The shape of the geometric imperfection was based on the buckling mode
belonging to the lowest eigenvalue from a linear buckling analysis. This resulted in
a sinusoidal bow imperfection. The amplitude defining the maximum deviation
from the ideal geometry was L/1000, where L is the height of the column. This
approach is generally accepted for the determination of buckling curves. A similar
approach, but then for lateral torsional buckling of beams, was used in [15]. The
value L/1000 for the imperfection amplitude is recommended in [10]. This value is
expected to be conservative since it is very likely that the real imperfections of the
heavy sections considered here are smaller than L/1000. The value L/1000 is a
design imperfection amplitude for use in numerical analyses.

3.6. SOLUTION

All elastic-plastic buckling GMNIA are load-controlled. A force with


specified magnitude was applied at the top of the column. The Arc-Length method
was selected to solve the non-linear equilibrium iterations. The Arc-Length method
was selected in preference to the conventional Newton-Raphson method as the
former is able to describe the decreasing load-deflection curve beyond the maximum
load whereas the latter will abort the solution when the maximum resistance has
been reached. The load was divided into four load steps which in turn were further
divided into substeps or load increments. For each load-increment a number of
equilibrium iterations were performed to arrive at a converged solution. The
solution was considered solved when the out-of-balance load vector is smaller than
0.05 % of the load increment. The ultimate strength or flexural buckling resistance
of the column (Nult;FEM) was identified as the maximum load on the load-deflection
curve. The elastic buckling load (Ncr;FEM) is obtained from a linear buckling
analysis (LBA) using the Block-Lanczos extraction method of eigenvalues.
190 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 13

3.7. PLOTTING RESULTS IN BUCKLING CURVE

For each column configuration for which the ultimate resistance is evaluated
through non-linear finite element analyses, the reduction factor is obtained by
normalizing the ultimate load against the squash load of the cross-section (Npl;FEM).
N ult;FEM
FEM = , (9)
N pl;FEM
where the squash load of the cross-section is computed according to:
N pl;FEM = Af y , (10)
where A is the cross-sectional area of the element and fy is the yield stress.
The factor FEM is labeled as the experimental resistance for comparison
with the theoretical resistance . The relative slenderness of the column can be
computed by taking the square root of the ratio between the squash load of the cross-
section and the elastic buckling load evaluated from a linear buckling analysis:
FEM = N pl;FEM N cr;FEM . (11)
Note that equation (11) is similar to equation (8) but the squash load is now based
on that of the FEM model and the elastic buckling load is calculated with a LBA.

3.8. FINITE ELEMENT RESULTS


3.8.1. Steel HISTAR 460
Typical load-deflection curves as obtained from the finite element analyses
are shown in Fig. 7a.

1 1.2
a b HD 920x1377
L=14400 mm strong-axis
1
0.8
HD 400x1202
L=8000 mm weak-axis HD 400x1202
0.8
Reduction factor [-]

L=8000 mm weak-axis
N ult;FEM/Ncr;FEM [-]

0.6

0.6

0.4
0.4
HD 920x1377
0.2 L=14400 mm strong-axis
0.2
Column curve a

0 0
0 10 20 30 40 50 0 1 2 3
Lateral displacement [mm] Relative slenderness [-]

Fig. 7 Finite element output: load-deflection curves for HISTAR 460 (a)
and corresponding data in buckling curve diagram (b).
14 Buckling curves for heavy wide flange steel columns 191

In Fig. 7b the ultimate loads from Fig. 7a are plotted in the buckling curve
diagram using the equations (911) in addition to buckling curve a. Plotting the
ultimate load for a specific group of columns in the buckling curve diagram in
addition to a buckling curve allows a first estimate to be made as to whether that
specific buckling curve is on the conservative or unconservative side.

3.8.2. Steel grade S460


Typical load-deflection curves as obtained from finite element analyses are
shown in Fig. 8 for four HD 400x1299 columns in S460 failing by weak-axis buckling.
In Fig. 9 the ultimate loads from Fig. 8 are plotted in the buckling curve
diagram using the equations (911) in addition to buckling curve b to show that
there is a reasonable fit with buckling curve b slightly on the conservative side.

160000 40000

elastic buckling
elastic buckling
120000 30000
Load [kN]

Load [kN]

80000 20000 elastic-plastic buckling

elastic-plastic buckling

40000 10000
HD 400x1299 HD 400x1299
S460 S460
L=6 m L=12.4 m
0 0
0 40 80 120 160 0 50 100 150 200
Lateral displacement [mm] Lateral displacement [mm]
16000 8000

elastic buckling elastic buckling

12000 6000 elastic-plastic buckling


elastic-plastic buckling
Load [kN]
Load [kN]

8000 4000

4000 2000
HD 400x1299 HD 400x1299
S460 S460
L=18.8 m L=25.2 m
0 0
0 100 200 300 400 0 250 500 750 1000
Lateral displacement [mm] Lateral displacement [mm]

Fig. 8 Finite element output: load-deflection curves for elastic and elastic-plastic
buckling analyses of HD 4001299 columns in S460.
192 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 15

1
HD 400x1299
S460
weak-axis
0.8

Reduction factor [-]


0.6

0.4

0.2 Buckling curve b


FEM

0
0 1 2 3
Relative slenderness [-]

Fig. 9 Finite element results plotted in buckling curve diagram


for buckling of HD 4001299 columns in S460.

3.8.3. Steel grades S355 and S235


Typical load-deflection curves for strong axis buckling as obtained from
finite element analyses are shown in Fig. 10 for columns with different lengths
with cross-section HL 9201377 in S355 and S235 using the QST and ECCS
residual stress models.

50000 50000

S355
40000 QST model 40000

30000 30000
Load [kN]

Load [kN]

S235 S355
ECCS model QST model
20000 20000

S235
ECCS model
10000 10000
HL 920 x 1377 HL 920 x 1377
L = 30 m L = 50 m

0 0
0 50 100 150 200 0 250 500 750 1000
Lateral displacement [mm] Lateral displacement [mm]

Fig. 10 Finite element output: load-deflection curves


for buckling analyses of HL 9201377 columns.
16 Buckling curves for heavy wide flange steel columns 193

In Fig. 11 the ultimate loads from Fig. 10 are plotted in the buckling curve
diagram using the equations (911) in addition to buckling curve a to show that
there is a reasonable fit.
1

S235 ECCS
0.8 Model
L=30 m
Reduction factor [-]

0.6 S355 QST


Model

L=50 m
0.4

buckling curve a
0.2

0
0 1 2 3
Relative slenderness [-]

Fig. 11 Finite element results plotted in buckling curve diagram


for buckling of HL 9201377 columns.

4. STATISTICAL EVALUATION
AND SUGGESTED BUCKLING CURVES

4.1. PARTIAL FACTOR EVALUATION PROCEDURE

The partial factor evaluation procedure follows Annex D of EN 1990 and is


applied here in a similar way as in [15]. For any heavy QST column i a comparison
can be made between its experimental resistance (re,i) and its theoretical resistance (rt,i):
re,i
Ri = . (12)
rt,i
In the present study the experimental resistance refers to FEM from equation
(9) for a column i failing by flexural buckling as obtained from non-linear finite
element analysis, so re,i = FEM. The theoretical resistance of column i refers to the
buckling reduction factor according to the buckling curve formulation from EN
1993-1-1 (equation (6)), so rt,i = . It is noted that in order to arrive at a theoretical
resistance a selection for a buckling curve (imperfection factor from Table 3) must
already be made. A value of Ri smaller than 1.0 or larger than 1.0 reflects an
unconservative or a conservative theoretical resistance model, respectively. For any
194 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 17

group of column configurations belonging to a set with sample size n, the mean
value correction factor Rm and corresponding variance can be determined:
n
1 1 n
Rm =
n
Ri , R2 =
n 1 i =1
( Ri Rm )2 . (13)
i =1

When plotting the experimental resistance on the y-axis and corresponding


theoretical resistance on the x-axis for all column configurations belonging to
subset n, the points will be distributed around the so-called estimator line:
re = Rmrt.
For each column configuration belonging to subset n an error term i is
introduced:
re, i
i = . (14)
rt , i RM

A logarithmic transformation is performed:


i = 1n ( i ) . (15)

For the logarithmic error terms belonging to sample size n, the mean value
and corresponding variance are determined as follows:
2
1 n 1 n
=
n i =1
i, 2 =
n 1 i =1
(
i ) . (16)

The variance can be used to compute the coefficient of variation as follows:

V = exp 2 1 .( ) (17)

When using a subset with a sample size n > 100 the partial factor associated
with the uncertainty of the resistance model can be determined as follows:
1
Rd = 1.0 ,
Rm exp ( kd,n Q 0.5Q 2 ) (18)

for which:

(
Q = 1n V2 + 1 ) (19)

and where kd,n is the characteristic fractile factor: 0.8 3.8 = 3.04.
18 Buckling curves for heavy wide flange steel columns 195

So, finally equation (18) gives the partial factor Rd belonging to a suggested
buckling curve based on a set of column configurations.

4.2. PARTIAL FACTOR FOR SUGGESTED BUCKLING CURVES

Non-linear finite element analyses were carried out for the heavy wide flange
cross-sections of Table 2. Per steel grade, cross-section, buckling axis and where
appropriate residual stress model, at least 100 analyses were performed indicated
by the sample size n.

4.2.1. Steel HISTAR 460


For HISTAR 460 all eight cross-sections were considered, (Table 2). For
each cross-section the weak-axis and strong-axis buckling response was evaluated.
The relative slenderness of the investigated columns was in the range between 0.31
and 3.3.
Plotting the finite element results in a buckling curve diagram permits a first
judgment on the suitability of the buckling curve to represent the column strength
for heavy HISTAR 460 sections. In case the chosen buckling curve is positioned
below the finite element results, it will provide conservative column strength
values. The buckling curve can be regarded as unconservative when the finite
element data is below the buckling curve. Figure 12 (left) shows the finite elements
results for a HL 4001202 section in HISTAR 460 buckling about its weak axis in
a buckling curve diagram in addition to buckling curve b.
Similar trends are found when plotting the theoretical column strength rt,i =
against its numerical counterpart re,i = FEM such as shown in Fig. 12 (right). In
case the buckling curve produces column strengths similar to the finite element
results, the data is positioned on the line re = rt. Data distributed above the line re =
rt indicates that the buckling curve provides conservative values for the column
strength. Unconservative columns strengths are found when the data points are
below re = rt. When the buckling curve formulation represents column strengths
different from those obtained with finite element analyses, the data points will be
distributed around the line re = Rmrt, where Rm is mean value correction factor
according to equation (13), [11, 19]. This line will give a better description of the
correlation between the theoretical and numerical values in comparison to re = rt.
The partial factor associated with the uncertainty of the resistance model is
evaluated for each buckling curve. The corresponding Rd-values for each section
type and buckling axis are presented in Table 6. For unfavorable buckling curves
the Rd-value is lower in comparison to more favorable buckling curves for a
majority of the investigated cases. Hence, relating the elastic-plastic buckling
response of a heavy HISTAR 460 section to a more favorable buckling curve is at
196 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 19

the expense of a higher partial factor Rd. The most favorable buckling curve
selected for heavy HISTAR 460 sections failing by flexural buckling is based on
the criterion Rd < 1.05, as denoted in bold in Table 6. This is associated with a
target value Rd = 1.0 so that values greater than 1.05 cannot be accepted. The 1.05
boundary is arbitrary but it is believed to be reasonable.

1 1
HD 400x1202 HD 400x1202
HISTAR 460 HISTAR 460
0.8 weak-axis 0.8 weak-axis
Reduction factor [-]

0.6 0.6

0.4 re,i [-] 0.4

FEM
0.2 Buckling curve b 0.2 re=Rmxrt
FEM re=rt

0 0
0 1 2 3 0 0.2 0.4 0.6 0.8 1
Relative slenderness [-] rt,i [-]

Fig. 12 Finite element data in buckling curve (left) and compared against theoretical solutions for
buckling curve b (right) weak-axis buckling of HD 400x1202 in HISTAR 460.

Table 6
Partial factors Rd per buckling curve for HISTAR 460
Buckling curve
Buckling Sample
Heavy section a0 a b c d
axis size n
Weak 104 1.161 1.054 0.994 0.978 0.947
HD 400900
Strong 119 1.015 1.000 0.998 0.984 0.950
Weak 104 1.160 1.054 1.003 0.991 0.968
HD 400990
Strong 111 1.016 1.010 1.017 1.015 0.995
Weak 109 1.159 1.053 1.000 0.989 0.964
HD 4001086
Strong 109 1.015 1.012 1.022 1.021 1.004
Weak 110 1.160 1.053 0.997 0.983 0.955
HD 4001202
Strong 134 1.016 1.009 1.017 1.013 0.992
Weak 110 1.160 1.053 0.999 0.985 0.957
HD 4001299
Strong 137 1.017 1.009 1.011 1.003 0.978
Weak 100 1.275 1.146 1.017 0.982 0.953
HL 9201377
Strong 166 1.053 0.992 0.983 0.964 0.922
Weak 101 1.298 1.167 1.031 0.974 0.933
HL 9201194
Strong 106 1.073 0.991 0.967 0.940 0.887
Weak 103 1.287 1.158 1.030 0.985 0.953
HL 9201269
Strong 101 1.056 0.993 0.985 0.969 0.931
20 Buckling curves for heavy wide flange steel columns 197

The differences between sections belonging to the same type (HD or HL) and
buckling around the same axis (weak or strong) are relatively small, indicating that
section geometry for the same section type has little influence on the partial factor.
In general the partial factors for an identical buckling curve are greater for the
weak-axis buckling case than those for the strong-axis buckling case. This reflects
the more detrimental influence of residual stresses for columns failing by weak-
axis buckling. Assuming that a Rd -value smaller than 1.05 allows Rd = 1.0 to be
used, HD and HL sections failing by weak-axis buckling should be assigned to
buckling curve b. Buckling curve a0 is assigned to HD sections failing by strong
axis buckling. HL sections buckling about the strong axis should be checked by
buckling curve a. These results are summarized in Table 7.

Table 7
Proposed buckling curves for HISTAR 460 sections based on Rd

Cross-section Limits Buckling about axis Buckling curve


Rolled I-sections

HD section: y-y a0
h/b 1.23 z-z b

HL section: y-y a
h/b 2.35 z-z b

4.2.2. Steel grade S460

For S460 four different cross-sections were considered, namely HD 400900,


HD 4001299, HL 9201194 and HL 9201377 (Table 2). For each cross-section
the weak-axis and strong-axis buckling response was evaluated. The relative
slenderness of the investigated columns was in the range between 0.23 and 3.33.
Figure 13 (left) shows the finite elements results for a HL 9201377 section
in S460 buckling about its strong axis in a buckling curve diagram in addition to
buckling curve a. In Fig. 13 (right) the same results are plotted with the theoretical
column strength rt,i = on the horizontal axis against its numerical counterpart
re,i = FEM on the vertical axis.
198 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 21

1 1
HL 920x1377 HL 920x1377
S460 S460
strong-axis strong-axis
0.8 0.8
Reduction factor [-]

0.6 0.6

re,i [-]
0.4 0.4

FEM
0.2 Buckling curve a 0.2 re=Rmxrt

FEM re=rt

0 0
0 1 2 3 0 0.2 0.4 0.6 0.8 1
Relative slenderness [-] rt,i [-]

Fig. 13 Finite element data in buckling curve (left) and compared against theoretical solutions
for buckling curve b (right) strong-axis buckling of HL 9201377 in S460.

The partial factor associated with the uncertainty of the resistance model is
evaluated for each buckling curve. The corresponding Rd-values for each section
type and buckling axis are presented in Table 8. Again the partial factor should be
smaller than 1.05 to allow Rd = 1.0 to be used. The associated values are denoted in
bold in Table 8.
Again the differences between sections belonging to the same type (HD or
HL) and buckling around the same axis (weak or strong) are relatively small and
the partial factors for an identical buckling curve are greater for the weak-axis
buckling case than those for the strong-axis buckling case.
The proposed buckling curves for S460 sections are the same as for HISTAR
460 (Table 7) except for HL sections buckling around the weak axis where
buckling curve c seems to be more appropriate. However, it should be noted that
for that case the partial factors belonging to buckling curve b (Rd = 1.066 and Rd =
1.058 for the sections HL 9201194 and HL 9201377, respectively) just slightly
exceed Rd = 1.05.
Table 8
Partial factors Rd per buckling curve for S460
Buckling curve
Heavy section Buckling axis Sample a a b c d
size n 0

HD 400900 Weak 116 1.209 1.092 1.004 0.982 0.952


Strong 106 1.033 1.002 1.009 1.005 0.983
HD 4001299 Weak 131 1.208 1.091 1.005 0.980 0.945
Strong 105 1.035 1.003 1.008 1.002 0.978
HL 9201194 Weak 100 1.356 1.218 1.066 0.978 0.935
Strong 115 1.101 0.999 0.963 0.934 0.880
HL 9201377 Weak 108 1.355 1.214 1.058 0.990 0.972
Strong 102 1.084 0.995 0.988 0.975 0.939
22 Buckling curves for heavy wide flange steel columns 199

4.2.3. Steel grades S355 and S235


Since HL cross-sections apparently result in less favorable buckling curves
than HD cross-sections, only HL cross-sections were considered for 355 and S235.
For two cross-sections in S355 and S235 strong-axis buckling was considered: HL
9201377 and HL 9201194 (Table 2). For cross-section HL 9201377 made from
steel grade S355 buckling about the weak-axis was considered. The relative
slenderness of the investigated columns was in the range between 0.21 and 3.58.

1 1
HL 920x1377 HL 920x1377
S355 ECCS S355 ECCS
strong-axis strong-axis
0.8 0.8
Reduction factor [-]

0.6 0.6
re,i [-]

0.4 0.4

FEM

0.2 Buckling curve a 0.2 re=Rmxrt

FEM re=rt

0 0
0 1 2 3 0 0.2 0.4 0.6 0.8 1
Relative slenderness [-] rt,i [-]

Fig. 14 Finite element data in buckling curve (left) and compared against theoretical solutions for
buckling curve b (right) strong-axis buckling of HL 9201377 in S355 with ECCS residual stress model.

Figure 14 (left) shows the finite elements results for a HL 9201377 section
in S355 with ECCS residual stress model, buckling about its strong axis in a
buckling curve diagram in addition to buckling curve a. In Fig. 14 (right) the same
results are plotted with the theoretical column strength rt,i = on the horizontal axis
against its numerical counterpart re,i = FEM on the vertical axis.
The partial factors associated with the uncertainty of the resistance model are
evaluated for each buckling curve and shown in Table 9. Also the residual stress
model used is indicated. Again the partial factor should be smaller than 1.05 to
allow Rd = 1.0 to be used. The associated values are denoted in bold in Table 9.
For S355 and S235 Table 9 suggests buckling curve a for strong-axis
buckling and buckling curve c for weak-axis buckling. It should be noted that the
strong-axis partial factors associated with buckling curve a obtained with residual
stress model QST are quite close to the 1.05 boundary value. It should also be
noted that no knowledge is available with respect to the real residual stress models
and levels for these heavy sections in S355 and S235.
200 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 23

Table 9
Partial factors Rd per buckling curve for S355 and S235
Buckling curve
Heavy Steel Res. str. Buckl. Sample
a0 a b c d
section gr. mod. axis size n
S235 ECCS Strong 115 1.122 1.026 1.016 1.018 1.008
HL ECCS Strong 120 1.093 1.002 0.985 0.972 0.937
9201377 S355 Strong 115 1.151 1.037 0.995 0.983 0.949
QST
Weak 120 1.502 1.347 1.170 1.029 0.964
S235 ECCS Strong 142 1.129 1.029 1.021 1.008 0.990
HL
ECCS Strong 123 1.095 1.014 1.005 1.001 0.983
9201194 S355
QST Strong 122 1.160 1.044 0.994 0.986 0.958

4.3. STATISTICAL INFORMATION ON MATERIALS

The presented analyses so far were limited to the computation of Rd-values.


Using available statistical information on the partial factor for the material m and
using equation (2) may produce lower M1-values than the Rd-values, since
m-values are generally smaller than 1.0. The relevant material property is the yield
stress. So, if statistical information on the yield stress is available such that a
m-value can be evaluated, then the general partial factor M1 can be calculated with
equation (2). As the yield stress of a single coupon can never be lower than the
nominal yield stress, as this would lead to the member being rejected, it can be
reasonably assumed that a m-value of 1.0 is a conservative value to account for the
variability of the material properties. So, if statistical information on the yield
stress is not available, the partial factor for the material m can be safely set equal to
m = 1.0 and the general partial factor based on equation (2) is equal to M1 = Rd.

4.3.1. Steel HISTAR 460


Since for HISTAR 460 published statistical information on the yield stress is
not yet available m = 1.0 is used to compute the general partial factor M1
according to equation (2): i.e. M1 = Rd. Assuming that M1 = 1.0 is the target value
for the general partial factor, then the (arbitrary but reasonable) criterion for
choosing a buckling curve is that M1 < 1.05. So, in fact for HISTAR 460 with M1 = Rd,
also the partial factor should fulfil the requirement Rd < 1.05. This means that for
HISTAR 460 the buckling curves of Table 7 can be used with M1 = 1.0. As soon as
a database becomes available containing the yield stress from a wide set of coupon
tests on HISTAR 460 sections, a value of m lower than 1.0 can be obtained
resulting in either a lower general partial factor or a more favorable buckling curve.

4.3.2. Steel grade S460


For S460, statistical literature information concerning the yield stress fy is
used to estimate m < 1.0. Two RFCS (Research Fund for Coal and Steel) projects
provide statistical data: OPUS [20] and PROQUA [21].
24 Buckling curves for heavy wide flange steel columns 201

OPUS [20] provides statistical data for the yield stress of S460M in the
flange thickness range of 16 < tf < 40mm: the mean value is fy,m = 521.1 N/mm2,
the standard deviation is fy, = 26.75 N/mm2 and the coefficient of variation then is
Vfy = fy,m/ fy, = 0.051. The ratio between mean and nominal yield stress can be
calculated as R = fy,m/ fy,nom = 521.1/460 = 1.13. If it is assumed that this ratio also
applies to S460 cross-sections with flange thicknesses tf > 100mm, then with a
reduced nominal yield stress for thickness fy,nom = 385 N/mm2, the mean value
becomes: fy,m = fy,nom R = 385 1.13 = 435 N/mm2. Keeping the coefficient of
variation the same, the material partial factor can be calculated using:
f y,nom
m =
(
f y,m 1 1.64V fy ). (20)

This results in m = 385/(435(11.64 0.051)) = 0.966. If this material partial


factor is used with the partial factors Rd of Table 8 in equation (2), more favorable
general partial factors M1 are obtained. This does not affect the buckling curves for
HD sections nor does it affect the buckling curve for strong-axis buckling of HL
columns but it does affect the buckling curve for weak-axis buckling of HL
columns. For HL columns in weak-axis buckling and for buckling curve b the Rd-
values are 1.066 and 1.058 for HL 9201194 and HL 9201377 respectively.
Multiplied (equation (2)) by m = 0.966 the M1-values become 1.030 and 1.022
respectively. So M1 < 1.05 meaning that M1 = 1.0 may be used in this case together
with buckling curve b. This means that also for S460 the buckling curves of Table
7 can be used with M1 = 1.0.
Statistical data given in PROQUA [21] together with plant measurements
support the value m = 0.966 used.

4.3.3. Steel grades S355 and S235


Though statistical information on the yield stress of S355 and S235 is readily
available, there are other reasons (to be mentioned hereafter) not to use this
information.

5. BUCKLING CURVES FOR EUROCODE3

Though for HISTAR 460 and S460 the buckling curves of Table 7 may be
used for heavy sections with h/b > 1.2 and tf > 100 mm in combination with the
general partial factor M1 = 1.0, these buckling curves cannot be incorporated in
Eurocode3 EN 1993-1-1 [8] if the format of the buckling curve selection table
(Table 1) is to be kept, since no distinction is made between HD and HL cross-
sections. For that reason, the most unfavorable buckling curves for HD and HL
202 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 25

cross-sections of Table 7 are used in the proposed new Eurocode3 buckling curve
selection table (Table 10). Also the proprietary name HISTAR can obviously not
be mentioned in the proposed Eurocode3 buckling curve selection table.
For S355 and S235 it was shown in section 4.2.3 that for heavy sections with
h/b > 1.2 and tf > 100 mm the buckling curves a and c may be used for strong and
weak-axis buckling respectively, together with Rd = 1.0. This result may even be
improved using statistical data on the yield stress resulting in m < 1.0 and thus in
either a lower value of the general partial factor M1 or in more favorable buckling
curves than mentioned. However, this does not make sense as long as no further
and better information is available on residual stress distributions in heavy sections
with h/b > 1.2 and tf > 100 mm in S355 and S235. In fact, information on residual
stresses is not available at this moment at all. For that reason it is proposed to
conservatively use the buckling curves b and c, which fit nicely in the table when
comparing with buckling curves for other cases. Often, moving from S460 to lower
steel grades means a shift of one buckling curve and the current proposal is in line
with that.
The QST residual stress model [9] as used in the present analyses is
representative for any heavy wide flange section having similar cross-sectional
dimensions and made with the Quenched and Self-Tempered process. As such, the
residual stress model can be used to define the initial stress state in heavy sections
made from grade S460 but also from S500 as these grades are manufactured with
identical methods by ArcelorMittal as HISTAR 460 steel. Grade S500 also has the
same nominal yield stress after reduction to account for material thickness effects
(i.e. 450 N/mm2, see Fig. 1). For these reasons S500 can be added in the last
column of Table 10.

Table 10
Proposed buckling curve selection table for Eurocode3, EN 1993-1-1
Buckling Buckling curve
about S235 S460
Cross-section Limits axis S275 S500
S355
S420
Rolled I-sections y-y a a0
tf 40 mm
z-z b a0

y-y b a
h/b > 1.2 40 < tf 100 mm
z-z c a

y-y b a
tf > 100 mm
z-z c b
y-y b a
tf 100 mm
z-z c a
h/b 1.2
y-y d c
tf > 100 mm
z-z d c
26 Buckling curves for heavy wide flange steel columns 203

6. CONCLUSIONS

In this paper buckling curves are proposed to check the flexural buckling
resistance of heavy wide flange columns, which have a flange thickness tf >100
mm and a height-to-width ratio h/b > 1.2. These sections are currently not covered
by Eurocode3 (EN 1993-1-1).
A database was created containing the elastic-plastic buckling resistance for a
wide set of heavy HISTAR 460, S460, S355 and S235 columns (both the stocky
HD-type and slender HL-type) failing by weak-axis and strong-axis buckling. The
buckling resistance was evaluated using non-linear finite element analyses using an
earlier proposed residual stress model [9] and the ECCS residual stress model [10]
to define the initial stress state.
The numerical buckling loads were compared against theoretical values,
where the latter correspond to the buckling resistances for a selected buckling
curve according to EN 1993-1-1. Based on the ratio between both values, a partial
factor Rd associated with the uncertainty of the resistance model was evaluated
according to Annex D of EN 1990 for each of the five buckling curves.
Aiming at a target value for the general partial factor of M1 = 1.0, meaning
that the resulting M1-values should not exceed 1.05, buckling curves are proposed.
For cross-sections with h/b > 1.2 and tf > 100 mm in steel grades S460 and S500
the buckling curves a and b are proposed for strong and weak-axis buckling
respectively, while for these cross-sections in steel grades S235 up to and including
S420 the buckling curves b and c are proposed for strong and weak-axis buckling
respectively.
Quenched and Self-Tempered (QST) steel cross-sections are currently
manufactured under the proprietary name HISTAR (HIgh STrength ARcelorMittal)
by ArcelorMittal. For stocky HD cross-sections in HISTAR 460 with h/b 1.23
and tf > 100 mm it was shown that the buckling curves a0 and b can be used for
strong and weak-axis buckling respectively. For slender HL cross-sections in
HISTAR 460 with h/b 2.35 and tf > 100 mm it was shown that the buckling
curves a and b can be used for strong and weak-axis buckling respectively.

Received on July 16, 2014

REFERENCES

1. SPOORENBERG, R.C., SNIJDER, H.H., CAJOT, L.-G., POPA, N., Buckling curves for heavy
wide flange QST columns based on statistical evaluation, Journal of Constructional Steel
Research, 101, pp. 280289, 2014.
2. AXMANN, G., Steel going strong, Modern Steel Construction, 43, pp. 5661, 2003.
3. POLLAK, B.S., Designing with Grade 65, Modern Steel Construction, 44, 2004.
4. European Organisation for Technical Approvals, European Technical Approval ETA-10/0156,
Long Products made of HISTAR 355/355L and HISTAR 460/460L, Deutsches Institut fr
Bautechnik, 2010.
204 H. H. Snijder, L.-G. Cajot, N. Popa, R. C. Spoorenberg 27

5. European Committee for Standardization, Hot rolled products of structural steels Part 4:
Technical delivery conditions for themo-mechanical rolled weldable fine grain structural
steels, EN-10025-4, 2004.
6. European Committee for Standardization, Hot rolled products of structural steels Part 2:
Technical delivery conditions for non-alloy structural steel, EN 10025-2, 2004.
7. European Committee for Standardization, Hot rolled products of structural steels Part 3:
Technical delivery conditions for normalized/ normalized rolled weldable fine grain
structural steels, EN 10025-3, 2004.
8. European Committee for Standardization, Eurocode 3. Design of steel structures, general rules and
rules for buildings, Brussels: EN 1993-1-1, 2004.
9. SPOORENBERG, R.C., SNIJDER, H.H., CAJOT, L.-G., MAY, M.S., Experimental investigation
on residual stresses in heavy wide flange QST steel sections, Journal of Constructional Steel
Research, 89, pp. 6374, 2013.
10. VOGEL, U., et al., Ultimate Limit State Calculation of Sway Frames with Rigid Joints, ECCS
Publication No 33, 1984.
11. European Committee for Standardization, Eurocode Basis for structural design, Brussels, EN
1990, 2001.
12. BEER, H., SCHULZ, G., Bases thoriques des courbes europennes de flambement, Construction
Mtallique, 3, 1970.
13. RONDAL, J., MAQUOI, R., Single Equation for SSRC Colomn-Strength Curves, Journal of the
Structural Division, Proc. of the ASCE, 105, ST1, pp. 247250, 1979.
14. SEDLACEK, G., UNGERMANN, D., KUCK, J., MAQUOI, R., JANSS, J., Background
documentation Document 5.03 (partim) Evaluation of test results on beams with cross-
sectional class 1-3 in order to obtain strength functions and suitable model factors, Eurocode
3, Editorial Group, 1989.
15. REBELO, C., LOPES, N., SIMOES DA SILVA, L., NETHERCOT, D., VILA REAL, P.M.M.,
Statistical evaluation of the lateral-torsional buckling resistance of steel I-beams. Part 1:
Variability of the Eurocode 3 resistance model, Journal of Constructional Steel Research, 65,
pp. 818831, 2009.
16. SIMOES DA SILVA, L., REBELO, C., NETHERCOT, D., MARQUES, L., SIMOES, R., VILA
REAL, P.M.M., Statistical evaluation of the lateral-torsional buckling resistance of steel I-
beams, Part 2: Variability of steel properties, Journal of Constructional Steel Research, 65,
pp. 832849, 2009.
17. BAN, H., SHI, G., SHI, Y., WANG Y., Overall buckling behavior of 460 MPa high strength steel
columns: Experimental investigation and design method, Journal of Constructional Steel
Research, 74, pp 140150, 2012.
18. SHI G., BAN, H., BIJLAARD, F.S.K., Tests and numerical study of ultra-high strength steel
columns with end restraints, Journal of Constructional Steel Research, 70, pp. 236247, 2012.
19. BIJLAARD, F.S.K., Eurocode 3 Design of steel structures Present status and further
developments, Steel Construction Design and Research, 1, 1, pp. 1623, 2008.
20. BRACONI, A., et al., OPUS, Optimizing the seismic performance of steel and steel-concrete
structures by standardizing material quality control, European Commission, Technical Steel
Research, Steel products and applications for building construction and industry, Final report,
Directorate-General for Research, EUR 25893, 2013.
21. CAJOT, L.-G., et al., PROQUA, Probabilistic quantification of safety of a steel structure
highlighting the potential of steel versus other materials, European Commission, Technical
Steel Research, Steel products and applications for building construction and industry, Final
report, Directorate-General for Research, EUR 21695 EN, Luxembourg, Office for Official
Publications of the European Communities, 2005.
PROBABILISTIC ASSESSMENT OF THE IMPACT
OF STRAIGHTNESS TOLERANCES IN EN 1090-2
ON THE STABILITY DESIGN OF STEEL COLUMNS

ANDREAS TARAS*

Abstract. This paper addresses a current, practical topic which has interesting
implications for design practices and research in the steel construction industry: the
impact of new, relaxed fabrication tolerances on the strength of compression
members. The current debate on this topic held at a code committee level was
triggered by the introduction of increased tolerances for the out-of-straightness of
steel members in the new European standard for the execution of steel structures for
constructional steelwork, EN 1090-2 [1]. The extent to which these changes are still
covered by the present design rules is currently unknown. To date, a consistent,
logically coherent justification for the acceptance of these changes is still missing.
The study presented in this paper is therefore intended as a contribution to this debate.
In the broader scientific context, it is intended to serve as an example for the
possibility of answering questions regarding the impact of changes to production
tolerances and manufacturing habits on design rules by probabilistic approaches.
Key words: buckling, imperfections, tolerances, structural reliability, eurocode 3,
probabilistic assessment, steel columns.

1. INTRODUCTION

The recent introduction of the harmonized European standard EN 1090-2 [1]


for the fabrication and erection of steel structures brought about a number of
innovations to the allowable shape deviations (or tolerances) of members, plates
etc. In the case of compression members, the most significant and controversial
change, when compared to most previous national standards, was the introduction
of an out-of-straightness limit of = L/750. This value is noticeably larger than the
value of L/1000 known to form the basis of the stability design rules for column
and beam-column buckling in Eurocode 3 EN 1993-1-1 [2] (EC3). Therefore, the
pertinent ECCS and CEN committees are currently confronted with the question of
what impact must be expected to the safety level of slender steel structures
designed in accordance with EC3 and fabricated following EN 1090-2.
This paper presents the results of a study that addressed this question for the
most relevant case of steel compression members (columns). Thereby, the case
of weak-axis buckling of one exemplary section was considered; this allowed for
*
Graz University of Technology, Institute for Steel Structures, Austria
206 Andreas Taras 2

an exemplary representation of the possible changes to the implicit safety level, in


terms of relative differences between different scenarios and tolerance prescriptions.
The study makes use of plausible scenarios for future column production habits and
of an advanced probabilistic and mechanical-deterministic methodology. This
methodology includes Monte Carlo simulations based on statistical distributions
of the basic variables of the stability problem, coupled with geometrically and
materially non-linear FEM analyses (GMNIA, see section 4.2) of imperfect members
with random distributions of these same variables. A large number (exceeding
2000) of single simulation runs was carried out for this study. The results of these
FEM calculation runs were then evaluated, as results of numerical tests, in
accordance with Annex D of EN 1990 [3].

2. TOLERANCES FOR COMPRESSION MEMBERS


IN INTERNATIONAL CODES

Most codes regulating the execution of steel structures distinguish between


manufacturing and erection tolerances. The former, also called shop fabrication
tolerances, are concerned with deviations from the nominal dimensions and are
measured in the workshop. They mainly serve as control quantity for the production
quality in the shop or factory. The latter are measured on site, after erection. They
are a control quantity for the quality of the erection works on site.
Table 1 summarizes the regulations of the out-of-straightness tolerances for
compression members according to different international codes. The following
comments can be made:
i. The manufacturing (shop) tolerances for out-of-straightness of compression
members were fixed at a value of L/1000 in all codes of practice preceding the
publication of the new European standard EN 1090-2. As a side note, it should be
mentioned that hollow sections produced according to the standard EN 10210-1
(2006) had and still have a larger tolerance limit of L/500 [4].
ii. The value of L/1000 is identical to the imperfection assumptions on
which the European column buckling curves (now found in the Eurocode) are
based, e0 = L/1000 [5, 6]. Historically, this tolerance value was actually the initial
motivation for the adoption of e0 = L/1000 in the non-linear numerical calculations
that were carried out during the development of the European column buckling
curves. With L/750, EN 1090-2 increases the tolerance limit by 1/3.
iii. In previous national European and current international codes, the
sections devoted to erection tolerances did not generally include provisions
regarding the out-of-straightness of compression members. The AISC code only
recently added a clarifying drawing where columns are shown to have L/1000
also in the erected configuration, but no comment is given on this and on how
this should be measured. The same can be stated about the former Austrian
Standard N B 4300-7 (1994). Other codes only mention positioning and
inclination restrictions for the erection of columns.
3 The impact of straightness tolerances on the stability design of steel columns 207

Table 1
Out-of-straightness of compression members (except hollow sections)
according to different international standards
Measurement Manufacturing Erection
Standard Country
Definition Tolerance Tolerance
AISC Code of
None
Standard Practice USA L/1000
L/1000)
(2005)
BS 5920-2 (2001)
& UK max(3mm ; L/1000) None
NSSS (2007)
Product standards
DIN 18800-7
DE hot-rolled: EN 10034 None
(2002)
welded: ISO 13920
N B4300-7 (1994) AUT L/1000 L/1000
(from EN 1090- ENV 1993-1-
Some EU L/1000
2) 1:1992 / None
countries max (3 mm; L/1000)
ENV 1090-1:1996
L/667 &
ECCS Rec. 1978 - L/1000
statistics
EN 1090-2:2008 CEN members L/750 L/750

iv. The lack of a specification of erection tolerances is explicable by the


difficulty and cumbersomeness associated with measuring a columns curvature
once it is built in. Furthermore, if the curvature is specified as an erection tolerance
it must also be specified how, and with what frequency, this quantity ought to be
measured on site this is not straight-forward and potentially controversial.
v. The only provision known to specify clear and uncontroversial acceptance
criteria for column out-of-straightness on site is contained in the ECCS European
Recommendations for Steel Construction (1978), see [7]. While this code also
uses the value of L/1000 for workshop fabrication, it loosens this value to
L/667 = 0.0015 L on site. According to this recommendation, if individual
measured values exceed this value, the construction contracts parties are expected
to agree upon the subsequent procedure by judging the fitness-for-purpose of the
compression element.
vi. Additionally to this criterion, the emphasis is placed on the statistical
distribution of the measured curvature. This is schematically illustrated in Fig. 1.
A series of measurements is accepted as conforming to the requirements when the
mean value plus times the estimated standard deviation of measurements
lies within the tolerance limit. The value of decreases with a rising number of
measurements n, spanning a range of 3.7 for n = 6 to 2.1 for n = 100. They are
meant to lead to a 90% confidence level regarding the non-exceedance of the
tolerance value.
208 Andreas Taras 4

1.2
OK accep tance criteria for out-of-straightness

tolerance limit
m not OK 1) single value i L / 667
1.0
m = i /n ; sn = ( m) / (n 1)
scaled probability density [-]

L/667
i

0.8 2) m + sn L / 667

.sn
0.6

0.4 .s n

0.2

0.0
/L
00

05

10

15

20

25

30
00

00

00

00

00

00

00
0.

0.

0.

0.

0.

0.

0.

Fig. 1 Visualization of the acceptance criteria for column curvature according to [7].

vii. EN 1090-2 does specify erection tolerances for the out-of-straightness of


compression members. However, no indication is given about how, and how often,
these ought to be measured. One must assume that, in theory, all columns must be
measured and checked against the tolerance criterion if EN 1090-2 applies as part
of a construction contract.

3. COLUMN OUT-OF-STRAIGHTNESS COVERED


BY THE EUROCODE 3 BUCKLING RULES

3.1. BUCKLING TESTS AND NUMERICAL CALCULATIONS

The code survey in the preceding section confirmed that the tolerance
definition for column straightness in EN 1090-2 differs from and is more
generous than all other important international codes, including those that were
valid (with earlier editions) in Europe when the current EC3 rules were created. In
this section, a closer look is taken at the imperfections that can be considered to be
covered by these rules. As is discussed in more detail in [5] and [6] and was
more recently summarized in [8], the current EC3 column buckling rules are based
on an extremely thorough, international experimental campaign, combined with an
equally thorough theoretical/numerical study, both carried out under the auspices
of ECCS. The experimental campaign consisted of both large scale buckling tests
and small-scale auxiliary tests that included the measurement of yield stress,
cross-sectional geometry and column out-of-straightness. While the buckling tests
5 The impact of straightness tolerances on the stability design of steel columns 209

formed the direct experimental basis of the current EC3 rules, the auxiliary tests
were originally mainly used for the purpose of documentation and test quality
control. Figure 2 shows the measurements of the out-of-straightness amplitude
for some of these tests; as can be seen, the limit of /L = 1/1000 = 0.001, valid at
the time of the tests, was exceeded by a non-negligible amount of tested columns,
especially for smaller, slender cross-sections like the IPE 160 shown in the figure.
However, the statistical distribution of the out-of-straightness had a fairly narrow
scatter-band, with a mean value of /L of m = 0.00085 = 1/1176 and a standard
deviation of s = 0.0002 = 1/5000. This would have fulfilled the above-mentioned
procedure in the 1978 ECCS recommendation for field measurements, which also
required L/1000 in the shop. The probabilistic/statistical evaluation of the full-
scale tests corresponding to the above measurements for IPE 160 sections is
illustrated in Fig. 3, leading to a maximum value of the necessary safety factor M*
of approximately 1,13. In the main diagram, the test results are plotted over the
nominal value of the slenderness , evaluated statistically and shown as vertical
lines with markers at the values m and m 2s. This evaluation was found to be
reproducible by numerical simulations of a model beam with nominal member
geometry, bow-imperfections of amplitude e0 = L/1000 and standardized patterns
of residual stresses [5]. This is also shown in Fig. 3, where the line marked with fy,k
represents the characteristic curve (evaluated with the measured value m 2s of
the yield stress), while fy,nom identifies the curve evaluated with nominal values of
yield stress. As can be seen, the characteristic GMNIA line describes the m 2s
points of the test result evaluation very well. This knowledge was then used to
numerically determine the shape and position of the 5 European column buckling
curves in [5].

n=150 measurements
30 6
m=0.00085 m=0.00032
IPE 160 s=0.0002 HEM-sections s=0.0003
25 5
[-] [-]
20 4
frequency [-]

frequency [-]

15 3

10 2

5 1

0 0
00

02

04

06

08

10

12

14

0. 0 0
0. 0 1
0. 02
0. 03
0. 04
0. 05
0. 06
0. 07
0. 08
0. 09
10
00

00

00

00

00

00

00

00

00
00
00
00
00
00
00
00
00
00
00
0.

0.

0.

0.

0.

0.

0.

0.

0.

/L [-] a) /L [-] b)

Fig. 2 Measured initial curvature as fraction of the column length


for some of the ECCS column tests.
210 Andreas Taras 6

Fig. 3 Original ECCS test results [6] for IPE 160 sections compared with GMNIA analyses
and safety-factor evaluation of the current, calibrated EC3 design formula.

3.2. IMPLICIT EQUIVALENT IMPERFECTIONS FOR SECOND-ORDER


ANALYSIS AND THEIR RELATIONSHIP
TO STRAIGHTNESS TOLERANCES

In order to obtain mechanically consistent and meaningful design equations


for column buckling, the ECCS column bucking curves were transformed into
Ayrton-Perry type design formulae by Rondal and Maquoi [9]. These design
expressions, which represent a combination of second-order internal forces with a
linear cross-sectional failure criterion and equivalent, generalized and calibrated
imperfections, are the familiar expressions currently found in EC3, Eqs. 1 and 2.
1
= 1.0, (1)
2 2
+
1
(
= 1 + + ; = 0.2 . )
2
with (2)
2
The starting point for the derivation of these equations is represented by a
first-yield buckling condition using second-order internal forces resulting from a
sinusoidal pre-deformation.
N N e0 1
+ 1.0, (3)
A f y W f y 1 (N/N cr )
with: N axial force ; Ncr elastic (Eulers) critical buckling load;
A cross-sectional area; W section modulus (plastic or elastic);
e0 amplitude of a sinusoidal imperfection shape (out-of-straightness);
fy yield strength.
7 The impact of straightness tolerances on the stability design of steel columns 211

Equation 3 can be rewritten in a normalized form by introducing the EC3


variables = N/(A.fy), = (A f y ) / Ncr , and = A e0 / W , leading to Eq. 4:


+ 2
= 1.0. (4)
1
Equation 4 is a quadratic equation, the lowest solution of which is given by
Eq. 1 and 2. Thus, in the Eurocode column buckling curves, the original elastic
term for (= A. e0 / W) was substituted by a calibration term containing the
buckling curve coefficient . This substitution can be reversed to obtain the
implicit equivalent imperfection amplitudes e0 contained in the Eurocode column
buckling curves. In the following Eq. 5, these are represented as a fraction of the
column length L:
1 W
( e0 /L ) =
W
= 0.2 .
A L A L
( ) (5)

In some reviews (e.g. [10]) of the current Eurocode design formulae for
column buckling, it is argued that the loosening of the curvature tolerances /L
may be justifiable on the basis of the above equivalent imperfections for second-
order calculations. Acknowledging the fact that the values of the generalized
imperfection coefficient also have to cover the influence of residual stresses in
the Ayrton-Perry formulation, it is argued (assumption 1) that, at = , the
influence of the residual stresses is practically zero. Furthermore, it is implicitly
assumed (assumption 2) that the buckling curve a0 ( = a0 = 0.13) is representative for
a column with practically no locked-in stresses. The following formula can then be
derived for strong-axis flexural column buckling FBy-y:

2 I y a0 2 i y a0 fy
( e0 / L )lim,GEOM,FBy y = = . (6)
h A L h E
In a third assumption (assumption 3), the ratio iy/h is set to 0.5. With this
value, the following theoretical limit value of the purely geometrical imperfection
that is already covered by the current buckling rules can be calculated, for the most
inconvenient case of steel grade S235:
0.13 235
( e0 / L )lim,GEOM,FBy y = 2 0.5 = 1/ 722. (7)
210000
Since e0 /L = 1/722 is larger than /L = 1/750, one may believe that the new
tolerance limit is covered by the current design rules. In fact, this line of
argumentation appears to represent the main background for the loosening of the
straightness tolerances for columns.
212 Andreas Taras 8

While at first glance this derivation is appealing, a number of problems


become evident when a more in-depth verification of its basic assumptions is
performed. The following points can be brought forward:
i. First of all, the assumption regarding iy/h is not realistic for most commercial
I & H-sections, where this value moves in the range of 0.350.42. This simple
corrective remark, regarding only one of the three assumptions made above, would
disprove the above argumentation, since e.g. with iy/h = 0.4, e0 /Llim,GEOM,FBy-y =
= 1/903, which is smaller than /L = 1/750.
ii. If assumption 2 were true, applying Equ. 7 in the case of weak-axis
flexural buckling FBz-z of typical I- and H-sections (which have ratios of iz/b~0.25)
would correspondingly lead to the conclusion that a tolerance of only /L = 1/1444
is covered by the design rules an even smaller value than the one that was
previously valid at /L = 1/1000.
iii. If assumption 2 is dropped, or said not to apply for weak-axis buckling of
a member made of S235 steel, the next most plausible buckling curve for this
assumption is line b, the highest curve for weak-axis buckling of S235 I-sections.
In this case, the covered imperfection is calculated as follows:

0.34 235
( e0 / L )lim,GEOM,FBz z 2 0.25 = 1/ 552. (8)
210 000

This is larger than both /L = 1/1000 and /L = 1/750, and (importantly)


larger than any value ever recorded in the ECCS tests (Fig. 2), it is hard to imagine
that this could be covered by the present rules.
These two points confirm that it is not purposeful to attempt to draw any
conclusions regarding the permissibility of loosening geometrical shape tolerances
by inference from considerations stemming from the second-order beam theory and
the generalized imperfection amplitude factor . The current column buckling
formulae are essentially based on curve-fitting of a second-order equation onto
previously established buckling curves, derived from tests and numerical analyses
(geometrically and materially non-linear analyses, including imperfections: GMNIA,
see section 4.2). Thereby, these manipulated second-order equivalent imperfections
lost much of their physical meaning, particularly in the quantitative sense that is of
interest here. Finally, it must be remembered that, from a safety point of view, the
values of must cover more than just residual stresses and column curvature:
since the structural designer eventually calculates a columns strength with nominal
values of cross-sectional geometry and material strength, but in reality these values
scatter, the values of must also account for the scatter of these quantities. This is
not acknowledged in the above considerations.
9 The impact of straightness tolerances on the stability design of steel columns 213

4. PROBABILISTIC STUDY METHODOLOGY

4.1. GENERAL PRINCIPLES

In this paper, a probabilistic approach was used in order to assess the impact
of changes to the straightness tolerances on the safety of the stability design rules
for steel compression members. Thereby, First Order Reliability Methods were
considered, following the general philosophy employed in the Eurocodes (EN 1990
[3]). Possible scenarios for the future impact of loosened straightness tolerances
were considered in a numerical parametric study with randomly generated input
parameters, whereby physical tests were simulated/replaced by non-linear numerical
tests with scattering input parameters. In the steel research community, this type of
test data generation is generally referred to as a type of Monte Carlo
assessment procedure, in reference to related, fully probabilistic approaches using
random input data generation. These test results were then analyzed statistically,
using the FORM procedures of EN 1990, in order to obtain representative values
for the necessary safety factors of the design rules.
The study was performed exemplarily for flexural buckling about the
weak z-z axis (FBz-z-) and for one section. This apparent limitation is not
particularly relevant for the aims of this paper, which are to highlight potential
relative differences in the safety level prompted by the chance in tolerances, rather
than absolute values. Thus, a representative I-section geometry had to be chosen
for the Monte Carlo simulations of this paper. The IPE 160 section made of S235
steel, already shown in Fig. 3, was selected because it was very thoroughly studied
during the development of the European column buckling curves and thus
comparisons with physical tests are possible.

4.2. NUMERICAL METHODOLOGY

The modelling techniques sketched in Fig. 4 were used for the numerical
analyses with scattering input parameters, using the software package ABAQUS
(Dassault Systems). The boundary conditions of the member, as well as the chosen
shapes of imperfection and the used finite element mesh, are shown in the figure.
Single-span compression members with in-plane, out-of-plane and torsional restraints
at the supports were considered (end fork conditions). An end load eccentricity
in the same direction as the assumed column out-of-straightness was considered as
one of the possible scattering quantities.
Four-node linear shell elements (S4) with six degrees of freedom per node
and finite strain formulation were chosen to model the flange and web plates of the
studied sections. The contribution of shear stresses in plasticity is accounted for by
this modelling technique. The mesh density was left constant, with 16 elements per
flange and web plate generally found to be a sufficient number. The number of
nodes in longitudinal direction was varied between 100 and 200, depending on
214 Andreas Taras 10

total member length. Rigid coupling beams were used to connect the single plates
with each other. At the supports, the stiffness terms of these elements were
manipulated in order to obtain a stiff load introduction mechanism (allowing one to
define concentrated loads to simulate moments and axial loads) that nevertheless
allowed the cross-section to warp and rotate freely.

Fig. 4 Overview of the FEM modelling techniques and assumptions.

Since the studied section is a hot-rolled IPE 160, the fillets were included in
the calculations by adding equivalent beam elements that were placed in the
centroids of the flanges. By defining the cross-sections of these beam elements as
equivalent box or square hollow sections (SHS) of variable depth and wall thickness,
the total area and torsional stiffness of the modelled member could be calibrated to
precisely match the tabulated values given by the production standards for rolled
11 The impact of straightness tolerances on the stability design of steel columns 215

sections. The exact bending capacity, as well as the weak- and strong-axis sectional
inertia (moments of area) were thereby also approximated with negligible error (in
the range of less than 2% error).
In the calculation of all relative quantities (such as slenderness , the buckling
reduction factor , etc.), the properties of the actually modelled section (with the
equivalent SHS section) were considered, rather than the original rolled section,
thereby further reducing the error of the calculation results, when expressed again
in these relative variables.
In order to obtain realistic numerical values of the ultimate buckling strength,
Geometrically and Materially Non-linear Analyses with Imperfections (GMNIA)
were carried out these were already mentioned above and were also used for the
development of the column buckling curves themselves. As the name implies,
imperfections are included in these calculations, whereby both geometric and
structural imperfections (residual stresses) were considered:
i. The initial geometric imperfections were assumed to follow a sinusoidal
shape. Since a double-symmetric section failing in weak-axis flexural buckling was
studied, no twist of the section needed to be considered.
ii. The residual stresses were assumed to vary linearly over the single cross-
section components, following the provisions given by ECCS in [11].
In addition to the imperfections, the material non-linearity (stress-strain
curve), as well as geometric non-linearities (equilibrium in the deformed state),
were considered. Strain hardening was included in the calculations, again following
the long-established recommendations of ECCS in [11].

4.2. PROBABILISTIC EVALUATION ACCORDING TO EN 1990 [3]

EN 1990 Annex D contains a standardized procedure based on First


Order Reliability Methods that allows for the determination of appropriate values
(in a semi-probabilistic design concept) of partial safety factors M on the basis of
test results. A summary of the procedure, as applicable to the problem studied in
this paper, is given in the following.
EN 1990 recognizes the relevance of the accuracy of a design formulation
and of its ability to react to the variability of the input parameters that govern a
design problem. The design formulation therefore takes on an immediate, central
part in the statistical evaluation procedure of EN 1990. In the notation of the code,
the theoretical value of strength rt is expressed as a function of its individual input
parameters
rt = g rt ( X ) . (9)
216 Andreas Taras 12

As an example for a function grt, the buckling formula leading to a column


strength rt = A fy may be considered, with the vector of the basic variables X
containing such parameters as the cross-sectional geometry, the yield stress, the
column length, etc.
The theoretical strength rt is compared with the experimental strength re in
the methodology of Annex D. Thereby, the actual values of the basic parameters
should be determined for each individual test result re,i. This information about the
basic variables is used to calculate the specific strength prediction rt,i for a single
test result, which can then be plotted as shown in Fig. 5.
For a number of n pairs re,i/rt,i, plotted in the rt/re plane, a regression line
through the origin can then be calculated through least-square approximation, using
the following formula:
n n
( rt,i ) .
2
b = re,i rt,i (10)
i =1 i =1

re
design model with measured input data
compared with physica or numerical tests

V
b
rt
Fig. 5 Schematical representation of the interpretation of V as variance of the design model.

The coefficient of variation V of the error terms i (with I = re,i/(b.rt,i)) of the


design function is calculated as follows:

V = exp( s2 ) 1, (11)

with s2 =
1 n
(
i
n 1 i =1
) (12)

1 n
= i (13)
n i =1

r
i = ln e,i = ln i . (14)
b rt,i
Up to now, the methodology has only accounted for the differences between
a certain pool of test data results (usually in terms of strength) and the prediction of
13 The impact of straightness tolerances on the stability design of steel columns 217

the same strength according to the design function. In the next step, the sensitivity
of the design function itself to the variability of the basic input variables must be
accounted for, by calculating the error propagation term Vr,t. In the usual case of a
complex, multi-variable design function, Vr,t is calculated using the following
formulae:

VAR [ g r,t ( X ) ]
2
1 j
g
V r,t 2 = = r,t i , (15)
g r,t ( X m ) 2 rm,t 2 i =1 X i

g r,t
with i .. partial derivative for the variable Xi times its standard deviation.
X i
The log-normal variation coefficients can now be calculated as follows:

Q r,t = ln (V r,t 2 + 1) , (16)

Q = ln (V 2 + 1) , (17)

(
Q = ln Vr 2 + 1 , ) (18)
Vr
,
t

with Vr 2 = 2
+ V 2 . (19)

In the next step, the design value of the resistance rd is calculated:

Qrt 2 Q2
for n 100: rd = b g r,t ( X m ) exp( kd , kd ,n 0.5 Q 2 ), (20)
Q Q

for n > 100: rd = b g r,t ( X m ) exp( kd , Q 0.5 Q 2 ), (21)

with g rt ( X m )  rm,t representing the value of the design function evaluated with
the mean values of all basic input variables, and kd,n and kd, being the design
fractile factors for n and infinite single test results (see [3], Annex D, Table D.2).
Finally, the required partial safety factor M*, applicable for designs based on
nominal input data, can be calculated as follows
rtk rnom
M * = N  , (22)
rd n EC3 rd

with rnom = g rt ( X NOM ) , i.e. the design function evaluated with nominal values of
the input parameters. This is the quantity usually determined by designers in EC3.
218 Andreas Taras 14

4.3. RANDOM INPUT DATA GENERATION

As stated previously, simulated tests obtained from GMNIA calculations with


random input variables (Monte Carlo simulations) are nowadays increasingly
being used instead of expensive and time-consuming physical tests. This procedure
is fully legitimate, provided that the real scatter bands of the single input
parameters, as well as the way these parameters correlate, are known [12].
This paper makes use of so-called Monte Carlo simulations in order to
generate random sets of input parameters for these numerical tests. The numerical
values for the single calculations were generated using the standard inverse
transform method [13] on the basis of cumulative distribution functions (cdf) F for
the normal or log-normal distribution, depending on the modelled parameter. A
single, randomly generated value X is therefore obtained from F and a randomly
generated value U ~ Unif[0,1], i.e. a random, uniformly distributed real number in
the interval [0,1]:
X = F 1 (U ), U ~ Unif [ 0,1]. (23)

An exemplary illustration of the results of data generation of this kind are


shown in Fig. 6, where the frequency plot for randomly generated variable values
for column curvatures and load eccentricities is shown to follow the underlying
distribution (in form of its probability density function) very well.
An issue for the application of the FORM and Monte Carlo methods is the
question of the correlation of the single parameters. This is graphically illustrated
in Fig. 7 for the example of the initial curvature and the residual stresses in a
column, generated either without any correlation (a) or with strong correlation (b).
In the latter case, the maximum of one variable corresponds to the minimum of the
other variable, while no such relationship exists in Fig. 7a.
70 100
generated values
60 assumed pdf
number of MC tests [-]

80
number of MC tests [-]

50
L
e0 60 N
40
ecc.
30 40 L

20
20
10

0 0
00

50

00

50

00

50

00

50

00
0

8
0

0.

0.

1.

1.

2.

2.

3.

3.

4.
00

00

00

00

00

00

00

00

00

00

ecc.[mm]
0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

e0/L a) b)

Fig. 6 Exemplary representation of the generation of random variables


on the basis of distribution functions.
15 The impact of straightness tolerances on the stability design of steel columns 219

As a matter of fact, precisely the two parameters plotted in Fig. 7 are


sometimes suspected to be inversely correlated with each other, meaning that smaller
residual stresses are systematically present whenever high curvatures are present
(see e.g. [14, p. 387]). However, for the purposes of this paper, no systematic
correlation was considered. This is motivated by the following considerations:
i. No statistical evaluations exist that would definitively prove such
correlations between parameters that are relevant for buckling.
ii. More importantly, the statistical evaluation procedure integrated in the
Eurocode EN 1990 and discussed in section 4.2 is a First Order Reliability
Method, based on the assumption that all variables are independent. Since the
results of Monte Carlo simulations are intended to be evaluated using this procedure,
any information stemming from the inclusion of the correlation would be ignored
by the EN 1990 evaluation procedure itself.
100 100
no correlation strong correlation

80 80
res [N/mm]

res [N/mm]

60 60

40 40

20 20

0 0
00

02

04

06

08

10

12

14

00

02

04

06

08

10

12

14
00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00
0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.
e0/L a) e0/L b)

Fig. 7 Example: two variables that: (a) do not correlate at all; (b) strongly correlate.

5. CONSIDERED INPUT PARAMETERS AND


SCENARIOS REGARDING COLUMN STRAIGHTNESS

5.1. CONSIDERED INPUT PARAMETERS

In order to be truly representative of the real behavior of physical tests with


random input parameters, the numerical calculations in the Monte Carlo simulations
need to take into account the scatter of as many parameters as are necessary to
represent the scatter of the resulting buckling strengths with sufficient accuracy. In
this paper, the case of flexural buckling of a hot-rolled IPE-section about the weak
axis is documented; for this case, the main cross-sectional parameters (width, depth,
flange and web thickness) as well as the yield strength are of great significance. It
is thus important to make use of representative information regarding the scatter
band of these and other related quantities.
220 Andreas Taras 16

Some very valuable studies [1517] have already been published that deal
with a systematic statistical analysis of the properties of steel products for
structural steelwork. Nevertheless, no final consensus seems to have been reached
at the present stage as to what set of structural parameters can be regarded as truly
representative for the statistical properties of internationally manufactured steel
products that are placed onto the European market. Another early application of the
general approach used in this paper, which also contained representative input
parameters for the buckling of I-sections, is documented in [18].
As was stated in section 4.1, a representative I-section geometry had to be
chosen for the Monte Carlo simulations of this paper, and the IPE 160 section
made of S235 steel was selected for this purpose.
On the basis of the above-mentioned studies, scatter bands of the parameters
as shown in Table 2 were considered in this study, in addition to four scenarios
regarding the scatter of the member out-of-straightness discussed in section 5.2.
The normal or log-normal (ln) distribution was assumed in all cases. These values
represent a lower-bound consensus for the considered quantities. Sometimes
even higher mean values of the yield strength of S235 steel are mentioned in the
literature [17]. However, one must consider that the product standard for hot-rolled
steel products, EN 10025, only mentions minimum values of yield strength, with
no mention of a required scatter band or non-exceedance probability, meaning that
steel production could in fact comply with this standard without much over-
strength of the scatter band of the yield stress; thus, it is prudent to use scatter
band values for the yield strength that are not exceedingly optimistic.
Table 2
Parameter variation for the Monte Carlo simulations IPE 160 S235
Parameter mean value m standard dev. s
Initial curvature e0 See scenarios Fig. 8
Yield stress fy 285 N/mm 17.1 N/mm
Eccentricity 0.60 mm (ln) 0.45 mm (ln)
Residual stress res 0.20235 N/mm 0.05235 N/mm
Flange thickness tf 7.4 mm 0.37 mm
Web thickness tw 5.0 mm 0.25 mm
Depth h 160 mm 1.6 mm
Width b 82 mm 0.82 mm

5.2. SCENARIOS REGARDING OUT-OF-STRAIGHTNESS

In order to assess the possible impact of changes to the production habits for
steel compression members that may (plausibly) be caused by the loosening of
straightness tolerances in EN 1090-2, four different scenarios were considered in
this study. They represent four different assumptions regarding the scatter of the
amplitude of the column out-of-straightness. They are represented in Fig. 8 which
shows the histograms of the values of the initial column curvature (with index e0
instead of used to indicate that a sinusoidal equivalent imperfection shape
17 The impact of straightness tolerances on the stability design of steel columns 221

was assumed in the GMNIA calculations) which were randomly generated for the
Monte Carlo simulations. All scenarios are referred to the best-documented (by
tests and calculations) case of the weak-axis flexural buckling of IPE 160 sections
made of steel grade S235, see section 5.1. Scenario 1 (Fig. 8a) is used as reference
for the other 3 scenarios, thus the histogram of this scenario is plotted in light grey
in Fig. 8b to d. The red continuous lines in the plots shown the assumed,
underlying probability density functions, scaled to match the total frequency of
produced numerical tests, while the light-blue histograms show the randomly
produced input data. The vertical, dashed red lines show the locations of the
tolerance limits of L/1000 and L/750 in comparison with the produced data.
SCENARIO 1 SCENARIO 2
L/750

L/750
L/1000

L/1000
70 350

60 300
number of MC tests [-]

number of MC tests [-]

50 250

40 200

30 150

20 100

10 50

0 0
00

02

04

06

08

10

12

14

16

18

00

02

04

06

08

10

12

14

16

18
00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00
_ _
0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.
e0 /L a) e0 /L b)

SCENARIO 3 SCENARIO 4
L/750

L/750
L/1000

L/1000

80 80
number of MC tests [-]

number of MC tests [-]

60 60

40 40

20 20

0 0
00

02

04

06

08

10

12

14

16

18

00

02

04

06

08

10

12

14

16

18
00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

00

_ _
0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

0.

e0 /L c) e0/L d)

Fig. 8 Scenarios for the future development of the scatter band of column curvatures.

The scenarios can be described as follows:


SCENARIO 1. The production habits of steelwork fabricators are not
altered by the new manufacturing and erection tolerances. The deviations found
during the ECCS tests are thought to be representative of real production; thus, the
probability density function (pdf) for ~ e0 found in [18] on the basis of the test
results in [6] is assumed to be correct and still valid: m = 0.00085 L, s = 0.0002 L.
222 Andreas Taras 18

SCENARIO 2. All columns are produced with an initial curvature of


~ e0 = L/750. Although very unrealistic, this scenario is the most unfavourable
possibility still in full compliance with the new tolerance limits in EN 1090-2.
SCENARIO 3. Steelwork fabricators make use of the new, relaxed
tolerances and produce columns that are on average somewhat more curved, i.e. by
a shift of L/750L/1000 = L/3000 towards the right in Fig. 8c. The scatter of the
production is otherwise left unmodified. No measurements are undertaken, neither
in the shop nor on site, therefore columns with curvatures exceeding the erection
tolerance of = L/750 are not prevented from being used in the structure.
SCENARIO 4. This final scenario is almost identical to 3, with the
addition that all columns are thought to be measured on site. Columns that exceed
the erection tolerance value of = L/750 are not allowed; in practical terms, a new
random value was generated in this case until ~ e0 < L/750.

6. RESULTS

The results of the statistical analysis are shown in Fig. 9 to Fig. 12 for the
four different scenarios discussed in the previous section. In order to give a more
complete und understandable picture of the outcome of these simulations, the
form of representation with m 2s bars representing the statistical distribution
(mean value m +/ 2s standard deviations s) of the single numerical test results is
complemented by a plot showing the result of a reliability assessment in accordance
with EN 1990 Annex D, i.e. following section 4.2 of this paper up to Eq. 22. This
assessment was performed using the same input parameters of the statistical data as
contained in Table 2, and is shown in terms of the required values of the partial
safety factor M* (note: the * is used to differentiate this calculated value from the
normative value M1 found in Eurocode 3). This factor gives a clear indication of
the impact of the single scenarios on the safety level of the column buckling rules,
especially when compared to the current reliability level discussed in section 3.1
and shown in Fig. 3.
The figures can be commented upon as follows:
SCENARIO 1. Figure 9 shows the position of the m 2s points to lie
very close to the applicable ECCS column buckling curve b. This would not
represent any noticeable change with respect to the current reliability level, see Fig.
3. The maximum value of M* that was calculated by following the procedure of
EN 1990 is very similar to the one calculated by Mller [19] for the ECCS tests
(1.115 vs. 1.13). This is not surprising, since Scenario 1 simulates the case where
the initial curvatures are left untouched by the changes of tolerance.
SCENARIO 2. Figure 10 shows the position of the m 2s points to lie
noticeably lower than the applicable ECCS column buckling curve b. The
difference is most pronounced in the region of intermediate slenderness, around
19 The impact of straightness tolerances on the stability design of steel columns 223

= 1.0, where the m 2s points are very close to the line representing curve c.
The maximum value of M* has increased to 1.177 in this scenario.
SCENARIO 3. Figure 11 where the m 2s points start falling
significantly below the applicable buckling curve at a slenderness of = 0.8. In
terms of M*, a maximum value of 1.161 is calculated for this scenario.
SCENARIO 4. Figure 12 again, the m 2s points start falling significantly
below the applicable buckling curve at a slenderness of = 0.8. In terms of M*, a
maximum value of 1.155 is calculated, only slightly lower than for scenario 3.
SCENARIO 1
1.50
1

m+2s
2
1.25 z,nom

m
1.00
m-2s
1.20

0.75 1.15
z,nom

1.115

ENV
1.10
0.50 M*
IPE 160, S235 EC3: a
weak axis buckling 1.05
EC3: b
0.25 EC3: c EC3
1.00

0.00 0.95
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
0
2
4
6
8
0
2
4
6
8
0
0.
0.
0.
0.
0.
1.
1.
1.
1.
1.
2.
z,nom a) z,nom b)

Fig. 9 Monte Carlo simulation of FBz-z of an IPE 160 section S235: SCENARIO 1;
statistical distribution of the simulated tests in a - plot (a); reliability analysis acc. to EN 1990 (b).

SCENARIO 2
1.50
m+2s

1
2
1.25 z,nom
m

1.00
m-2s

1.20
1.177

0.75 1.15
z,nom

ENV
1.10
M*

0.50
IPE 160, S235 EC3: a
weak axis buckling 1.05
EC3: b
0.25 EC3: c EC3
1.00

0.00 0.95
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
0
2
4
6
8
0
2
4
6
8
0
0.
0.
0.
0.
0.
1.
1.
1.
1.
1.
2.

z,nom a) z,nom b)

Fig. 10 Monte Carlo simulation of FBz-z of an IPE 160 section S235: SCENARIO 2.
224 Andreas Taras 20

SCENARIO 3
1.50

m+2s
1
1.25 2
z,nom

m
1.00

m-2s
1.20
1.161
0.75 1.15
z,nom

ENV
1.10

M *
0.50
IPE 160, S235 EC3: a
weak axis buckling 1.05
EC3: b
0.25 EC3: c EC3
1.00

0.00 0.95
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

0
2
4
6
8
0
2
4
6
8
0
0.
0.
0.
0.
0.
1.
1.
1.
1.
1.
2.
z,nom a) z,nom b)

Fig. 11 Monte Carlo simulation of FBz-z of an IPE 160 section S235: SCENARIO 3.

SCENARIO 4
1.50
1
m+2s

2
z,nom
1.25
m

1.00 1.20
m-2s

1.155
0.75 1.15
z,nom

ENV

1.10
M*

0.50
IPE 160, S235 EC3: a 1.05
weak axis buckling EC3: b
0.25 EC3: c EC3
1.00

0.00 0.95
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
0
2
4
6
8
0
2
4
6
8
0
0.
0.
0.
0.
0.
1.
1.
1.
1.
1.
2.

z,nom a) z,nom b)

Fig. 12 Monte Carlo simulation of FBz-z of an IPE 160 section S235: SCENARIO 4.

7. DISCUSSION OF THE RESULTS

The results of the Monte Carlo simulations shown in Figs. 9 to 12 point out
that the reliability level of column buckling could be affected by the introduction of
new tolerance limits. The scenarios 2 to 4 all treated the possibility that columns
produced in the future have on average larger initial curvatures than was customary
up to now, while all other parameters were left (in terms of scatter band) unmodified.
21 The impact of straightness tolerances on the stability design of steel columns 225

Thus, it is not surprising that the calculations led to an (average) loss of column
strength, respectively of margin of safety. The magnitude of this loss is in the order
of 4 to 6%.
As far as the likelihood of the single scenarios is concerned, it must certainly
be admitted that scenario 2 is not plausible, since it cannot be expected that all
steelwork fabricators will produce columns that have a value of out-of-straightness
of exactly = L/750. However, this extreme scenario is still interesting when
compared to the (much more plausible) scenarios 3 and 4: these scenarios treat the
possibility that fabricators relax their fabrication habits with respect to column
curvatures, an outcome that, even if not necessary, must certainly be seen as
desired by the new tolerance specifications. The calculations in this section have
shown the differences between the extreme scenario 2 and the desired
scenarios 3 and 4 to be rather small. Interestingly, scenario 4 is plausible even if
every single column is verified for initial curvature on site and replaced if curvatures
above = L/750 are detected. Compared to the (similar) scenario 3, where no such
measurements are undertaken, the advantage stemming from these measurements is
negligible.
It is important to understand that it is often not the purpose of geometric
erection (i.e. on-site) tolerance limits to specify the extreme upper values of shape
deviations in the erected structure, which must never be exceeded in order to
design a structure safely with a given set of design rules. On the contrary, as the
presentation (in section 2, Fig. 1) of the acceptance criteria in the 1978 ECCS
recommendation has shown, the philosophy with respect to tolerances on site was
that single values above the specified tolerance (in that case, still /L = 1/1000) are
acceptable as long as the statistical distribution of these deviations is kept in
check, and the production in the shop is confirmed to produce columns within the
(shop) limits for . Other international codes implicitly followed a similar
philosophy, by assuming that site measurements were entirely unnecessary if /L =
L/1000 was checked in the shop, since the statistical distribution of values on site
could then be assumed to be acceptable.
Of course, this study only discussed some, possibly too pessimistic
scenarios. The correlation between tolerance limits and actual deviations is not
always quite as clear-cut as has been assumed here. To name one example, Chan &
Gardner [4] have found that the tolerance limits of = L/500 for the initial
curvature of cylindrical hollow sections may be unduly lax as evidenced by both
the observed structural performance and measured imperfections of real columns.
In other words, they observed a poor correlation between actual shape deviations
and tolerance limit. They then justified the use of the current buckling rules for
such sections, which are also based on GMNIA calculations with e0 /L = 1/1000
and fixed residual stresses, by the actual statistical distribution of the shape
226 Andreas Taras 22

deviations. In principle, this is not quite unlike Scenario 1 discussed above:


fabrication habits dont acknowledge actually laxer fabrication tolerances. The
conclusion in [4] is coherent: a reassessment of the tolerance limit towards more
realistic, lower levels, as opposed to adjusting buckling curves to match current
tolerance levels.
Other arguments could be brought up to declare the scenarios discussed
above as too pessimistic: for example, residual stress distributions could be lower
now than back in the 1960s and 70s, when the ECCS tests were carried out. Then
again, actual column imperfections might be smaller than the deviations measured
in the ECCS program. Finally, as stated before, a beneficial inverse correlation
between and res (i.e. res is systematically lower when is large) is sometimes
thought to exist. These are valid points, but purely speculative at this stage;
practically nothing is known about real column curvatures in buildings, and the
knowledge and data about residual stresses in mill- or shop-fabricated column sections
has also not yet reached a satisfying level in terms of statistical representativeness.

8. CONCLUSIONS

On the basis of the above findings, it can be concluded that the new manu-
facturing and erection tolerances for compression members, as contained in EN
1090-2, cannot be logically proven to be covered by current buckling rules. On
the contrary, the implicit intent of the new limits, i.e. allowing steelwork fabricators
to loosen their fabrication habits with regard to column straightness, has been
shown to (plausibly) lead to a drop of the reliability level of column buckling rules
by about 5%. Whether this is acceptable or not should be carefully considered by
the concerned code committees. In this sense, considering the already present
differences between (published, and confirmed in this paper) evaluations of the
appropriate safety factor for column buckling M* (i.e. values of around 1.1) and
the actual code recommendation for compression members in buildings (M1=1.0),
the implications of further potential loss of 5% of safety must be considered with
particular care.
As a more general conclusion, it was shown that the methodology presented
in this paper is a suitable tool for the assessment of the influence of parameter
scatter bands on the safety level of design rules for steel structures. It can be
applied to problems similar to the one studied in this paper, i.e. studies of the
impact of changes of production habits, as well as for product and design rule
development. Further applications of this method are discussed in [20].

Received on July 16, 2014


23 The impact of straightness tolerances on the stability design of steel columns 227

REFERENCES

1. CEN, Execution of Steel Structures and Aluminium Structures Part 2: Technical Requirements
for Steel Structures, EN 1090-2, Brussels, 2008.
2. CEN, Eurocode 3. Design of steel structures. General rules and rules for buildings, EN 1993-1-1,
Brussels, 2005.
3. CEN, Eurocode Basis of structural design, EN 1990, Brussels, 2002.
4. CHAN, T.-M., GARDNER, L., Flexural Buckling of Elliptical Hollow Section Columns, Journal of
Structural Engineering, 135, pp. 546-557, 2009.
5. BEER, H., SCHULZ, G., Bases Thoriques des Courbes Europennes de Flambement,
Construction Mtallique, pp.37-57, 1970.
6. SFINTESCO, D., Fondement Exprimental des Courbes Europennes de Flambement, Construction
Mtallique, 3, pp. 5-12, 1970.
7. ECCS, European Recommendations for Steel Construction, Brussels, 1978.
8. GREINER, R., TARAS, A., New design curves for LT and TF buckling with consistent derivation
and code-conform formulation, Steel Construction, 3, 3, pp.176-186, 2010.
9. RONDAL, J., MAQUOI, R., Formulations dAyrton-Perry pour le Flambement des Barres
Mtalliques, Construction Metallique, 4, pp.41-53, 1979.
10. SEDLACEK, G., NAUMES, J., BIJLAARD, F., MAQUOI, R., MUELLER, Ch., CEN Document:
N1639E Excerpt from the Background Document to EN 1993-1-1 Flexural buckling and
lateral buckling on a common basis: Stability assessments according to Eurocode 3, CEN
European Committee for Standardization Eurocode Sub-Committee SC3, Brussels, 2008.
11. European Convention for Constructional Steelwork, Ultimate Limit State Calculation of Sway
Frames with Rigid Joints, Editor: Vogel, U., ECCS, TC 8, Brussels, 1984.
12. SINGER, J., ARBOCZ, J., WELLER, T., Buckling Experiments Experimental Methods in Buckling
of Thin-Walled Structures, Volume 1 Basic Concepts, Columns, Beams and Plates, Wiley
& Sons, Chichester, 1998.
13. ROSS, S.M., Introduction to Probability and Statistics for Engineers and Scientists, Elsevier,
Amsterdam, 2009.
14. BALLIO, G., MAZZOLANI, F.M., Theory and Design of Steel Structures, Chapman and Hall,
London, 1983.
15. ALPSTEN, G., On the use of reliability assessment methods for structural design, Proceedings of
Euro-SiBRAM2002, Prague, Czech Republic, June 2426, 2002.
16. MELCHER, J., KALA, Z., HOLICKY, M., FAJKUS, M., ROZLYKA, L., Design characteristics of
structural steel based on statistical analysis of metallurgical products, J. of Const. Steel Res.,
60, pp. 795-808, 2004
17. SIMOES DA SILVA, L., REBELO, C., NETHERCOT, D., Statistical Evaluation of the Lateral-
Torsional Buckling Resistance of Steel I-beams, Part 2: Variability of steel properties, J. of Const.
Steel Res., 64, pp. 832-849, 2008.
18. STRATING, J, VOS, H, Computer Simulation of the E.C.C.S. Buckling Curve using a Monte-
Carlo Method, Proc. Intern. Coll. on Column Strength, Paris, 1972, pp. 334-355.
19. MLLER, C., Zum Nachweis ebener Tragwerke aus Stahl gegen seitliches Ausweichen,
Dissertation, RWTH Aacher, Shaker Verlag, 2003.
20. TARAS, A, Contribution to the Development of Consistent Stability Design Rules for Steel
Members, PhD Dissertation, Graz University of Technology, 2010.