COMITE EURO-INTERNATIONAL DU BETON

RC ELEMENTS
UNDER CYCLIC
LOADING
STATE OF THE ART REPORT

I Thomas Telford
Published by Thomas Telford Publications, Thomas Telford Services Ltd, 1 Heron
Quay, London E14 4JD

First published 1991 as CEB Bulletin d'Information No. 210 Behaviour and
analysis of reinforced concrete structures under alternate actions inducing
inelastic response - volume 1: general models
Thomas Telford edition published 1996

Distributors for Thomas Telford books are

USA: American Society of Civil Engineers, Publications Sales Department, 345
East 47th Street, New York, NY 10017-2398
Japan: Maruzen Co. Ltd, Book Department, 3-10 Nihonbashi 2-chome, Chuo-ku,
Tokyo 103
Australia: DA Books and Journals, 648 Whitehorse Road, Mitcham 3132, Victoria

A catalogue record for this book is available from the British Library
Classification
Availability: Unrestricted
Content: Subject area review
Status: Committee guided
Users: Designers, civil engineers

ISBN: 978-0-7277-3548-5

©CEB, 1996

All rights, including translation reserved. Except for fair copying, no part of this
publication may be reproduced, stored in a retrieval system of transmitted in any
form or by any means, electronic, mechanical, photocopying or otherwise, without
the prior written permission of the Publisher, Books, Publications Division, Thomas
Telford Services Ltd, Thomas Telford House, 1 Heron Quay, London El4 4JD.

This book is published on the understanding that the author is solely responsible for
the statements made and opinions expressed in it and that its publication does not
necessarily imply that such statements and/or opinions are or reflect the views or
opinions of the publishers.
Preface
During the second half of the 1980s, the CEB, with its new future Model
Code already under development, felt that the time had arrived to adjourn
and make known its ideas on the broad and rapidly evolving field of non-
linear analysis of structures. In fact, while the number of applications of
non-linear analyses for important and unconventional structures was
increasing fast, and recourse to such analyses was explicitly allowed by
the Model Code, no guidance was available to the designer for the selection
of the models more suited to each specific case. On the other hand, the
existence and the accessibility of a number of sophisticated finite element
codes for the non-linear analysis of reinforced concrete structures, was a
cause more for concern than for relief, because of their lack of validation
over a sufficient spectrum of different applications.
The terms of reference given by the CEB to the new group to be formed
stated that it should concern itself with those models and methods of
analysis adequate for dealing with general load and stress histories. This
decision was obviously of great consequence, in that it involved the
exclusion of a number of classical and well proven models and methods, but
it was certainly a necessary one: the assumption of a proportional increase
of all loads acting on a structure is not only unrealistic but is also not
conservative in assessing the likelihood of attaining a given limit-state.
Furthermore, since a fairly large number of real loads fluctuate with time,
and some of them may also reverse their action, it was finally decided to
characterize the scope of the group's work with the widest of its attributions,
by titling it 'Behaviour and analysis of reinforced concrete structures under
alternate actions inducing inelastic response'.
From the early discussions on the plan of activity, the members concurred
that the first step had to be the preparation of a document of fundamental
nature. The document would contain a review of the pertinent existing
knowledge in the area of constitutive modelling of concrete, steel, bond and
of their interaction. It should also discuss the problems encountered in
assembling the various elements with the purpose of constructing the model
of an element made of reinforced concrete. In selecting the material for
inclusion, the main criterion would be its ability to reproduce, adequately if
crudely, non-proportional stress-strain paths, with reversed loading.
Whether physically or empirically based, very simple or sophisticated,
long-established or brand new, the models would be presented in framework
as rational as possible, certainly accompanied by comments on their
advantages and limitations, but without attempts to rank them on the basis
of their relative merits.
This programme of work, with the addition of a final part dealing with the
solution strategies for the numerical integration of non-linear equations, has
materialized in the present document. Admittedly distant from the
immediate needs of professional practice, it aims at being a basic, com-
prehensive and relatively easy-to-read document, hopefully useful as a
reference for non-specialists wishing a more conscious approach to the use
of the available sophisticated computational tools.
The material in this book has been diffused initially as CEB Bulletin 210.
It is now being published integrally in an editorially revised form.

Paolo E. Pinto
Rome, March 1996
 Acknowledgements

General Task Group Chairman Paolo E. Pinto, Italy

22 of the CEB
Members E. C. Carvalho, Portugal
V. Cervenka, Czechoslovakia
V. Ciampi, Italy
R. Eligehausen, Germany
M. N. Fardis, Greece
F. C. Filippou, USA
P. Gambarova, Italy
G. Konig, Germany
G. Via, Italy
Chapters were drafted by the following.
1. Concrete in compression, V. Ciampi, P. E. Pinto, G. Via
Appendix to chapter 1, J. Ozbolt
2. Concrete in tension, G. Konig, H. Duda
3. Reinforcing steel, F. C. Filippou
4. Bond between concrete and steel, G. L. Balazs, R. Eligehausen
5. Interface behaviour, P. Gambarova, M. Di Prisco
6. Finite element modelling of reinforced concrete, M. N. Fardis
7. Solution strategies for the non-linear structural equations, V. Ciampi
Grateful acknowledgements are due to the following, who kindly reviewed
the text.
Professor O. Buyukozturk (USA) for chapter 1
Professor J. C. Walraven (Netherlands) for chapters 2 and 5
Professor E. Ramm (Germany) for chapter 7
Drs Fehling and Rothe of the Technische Hochschule of Darmstadt also
provided helpful contributions to various parts of the project.
Contents

1. Concrete in compression 1
1.1. Introduction, 2
1.2. Experimental behaviour of concrete under general
multiaxial stress histories, 5
1.3. Models derived from the theory of elasticity, 9
1.4. Models based on the theory of plasticity, 16
1.5. Models based on the mechanism of microcracking
or elastic damage, 26
1.6. Models based on the association of plasticity and
microcracking mechanics, 31

Appendix: General non-local microplane model for concrete 33

A.I. Introduction, 33
A.2. Review of the microplane model, 33
A.3. Unloading, reloading and cycling loading, 35
A.4. Incremental macroscopic stress-strain relations, 36
A.5. Rate effect, 36
A.6. Numerical algorithm in each load step, 36
A.7. Numerical examples, 37
A.8. Conclusions, 41

2. Concrete in tension 42
2.1. Introduction, 42
2.2. The tensile behaviour of concrete, 42
2.3. The fictitious crack model, 43
2.4. Models for the bulk behaviour, 43
2.5. Models for the stress-crack width relation, 43
2.6. Rheological material model for the cr-w relation, 47
2.7. Continuous function model, 54

3. Reinforcing steel 58
3.1. Introduction, 58
3.2. Giuffre, Menegotto and Pinto model, 60
3.3. Simple non-linear stress-strain model, 63

4. Bond between concrete and steel 70

4.1. Introduction, 70
4.2. Bond behaviour under monotonically increasing slip, 70
4.3. Bond behaviour under unidirectional cyclic loading, 72
4.4. Bond behaviour under alternating loading, 79
4.5. Conclusions, 97

5. Interface behaviour 98
5.1. Introduction, 98
5.2. Aggregate interlock, 102
5.3. Dowel action, 122
5.4. Concluding remarks, 132
6. Finite element modelling of reinforced concrete 134
6.1. Introduction, 134
6.2. Finite element modelling of the reinforcement and
of its interaction with concrete, 135
6.3. Modelling of the concrete component, 136
6.4. Composite steel/concrete modelling, for two-dimensional
homogeneously reinforced elements, 161

7. Solution strategies for non-linear structural equations 168

7.1. Non-linear equilibrium equations, 168
7.2. Static analysis (special problems), 172
7.3. Dynamic analysis, 173

References 179
1. Concrete in compression
In the stress space, the locus of all the stress combinations for which a
proportionally loaded concrete specimen reaches its maximum load-
carrying capacity is usually called the 'failure surface' or, a better term,
the 'ultimate strength surface'.
It is a convenient simplification to assume that an ultimate strength
surface constructed for proportional loading may still be used for defining
failure conditions for more general load cases, provided the individual
components of the stress tensor do not deviate too much from monotonic
path.
In dealing with the modelling of the behaviour of concrete over its
complete range of response, the knowledge of the ultimate strength surface
is important, since it allows identification of regions of stress states beyond
which complete failure is reached according to different mechanisms.
The different failure modes are shown schematically in the following
figure (adapted from Pramono and Willam1), with the corresponding
representative stress points in a ultimate strength surface plotted in the
Rendulic plane. For example, experiments indicate the existence of a
transition point (TP) which separates brittle softening behaviour from a
ductile regime, this latter occurring when the lateral confinement ratio
increases. The hatched area in Fig. 1 corresponds to softening post-peak
behaviour which, however, exhibits markedly different characteristics
depending on whether the specimen is under one principal compressive or
tensile stress.
For the case of predominant tensile stresses, in which failure occurs along
a well-defined direction and in the form of a localized crack, fracture
mechanics approaches have long since been introduced to derive stress-
displacement relationships of the cracked zone. This approach is still the
most commonly used although it has to be complemented with some criteria
for 'smearing' the discrete phenomenon over an equivalent continuum. All
other cases, from the compression failure modes to the mixed ones, are
treated in the classical context of the constitutive laws for continuous media.
This abrupt change from a discrete to a continuous type of modelling is
obviously not satisfactory from a physical standpoint.
There are in fact some recent schools of thought2 supporting the view that

Fig. I. Failure modes of plain •

concrete under different \r\ H
stress combinations ' T
 RC ELEMENTS UNDER CYCLIC LOADING

microcracking under increasing stresses is, at a microscopic level, a non-

directional phenomenon which can be described through the use of scalar
variables, also called damage variables. Cracks at a macroscopic level are
merely the trajectories of the damaged points. By using different damage
variables related to the inelastic parts of the deformation in compression and
in tension respectively, a unified treatment of crushing, cracking and any
'intermediate' damage state can be achieved.
The same unified treatment is achieved by a different class of models,
also based on a simplified physical description of the material behaviour at
the microscopic scale.3'4 These models, known as 'microplane models',
have been extensively studied and developed in very recent times, with
satisfactory results. Although presently their use is not widespread outside
scientific circles, their validity is sufficiently documented to warrant a
rather detailed presentation, which may be found in the Appendix to the
present chapter.
It must be conceded, however, that looking at the advanced state of the
practice the two classes of models just mentioned are represented in a very
modest percentage of the applications. By and large, present practice is still
based on separate models to describe 'mainly compressive' and 'mainly
tensile' behaviour types: this established conventional categorization of the
models will be maintained in this text, under the simplified headings of
'Concrete in compression' and 'Concrete in tension'.

1.1. Introduction The modelling of the three-dimensional stress-strain behaviour of concrete

under general loading conditions has been the subject of intensive research
in the last 20 years, motivated, at least partly, by the practical need to
provide adequate constitutive relationships to the powerful computer codes
being developed during the same period. Exhaustive reviews of the work
done on this subject have been carried out.5'6
Not all the material, however, is relevant to the purpose of the present
document: the models that will be discussed have been selected on the basis
of their capacity to reproduce the aspects thought to be essential for a
realistic description of concrete behaviour under repeated, strongly inelastic
actions. These aspects are as follows:
(a) The capacity of accounting for inelastic non-proportional unloadings
and reloadings.
(b) An adequate non-holonomic relationship between the state of stress
and the stiffness of the material.
(c) The capacity of accounting for the stress degradation as a function of
the load history, including post-peak behaviour.
It may be stated that strict compliance with the above requirements is far
from common, even for the more advanced recent proposals. Acknow-
ledging this fact, the point arises whether it is feasible to set a basis for a
comparative rating of the different models. At present this appears to be
debatable.
In principle, models endowed with plausible physical bases should
obviously be preferred to those which are little more than mathematical
algorithms capable of reproducing a necessarily incomplete set of
experimental findings. On the other hand, one should consider that all of
Conversion factor the so-called physically-based models treat the concrete material as a
continuum, which is less and less the case as the stress-strain state
1 inch = 254 mm increases. Therefore, they may also be seen as mathematical abstractions,
1 foot - 305 mm with a closeness to actual behaviour only achieved because of the use of
1 kip = 445 kN
appropriate empirically-derived functional dependencies.
 CONCRETE IN COMPRESSION

From a different standpoint, models could be compared in terms of sheer

accuracy, irrespective of their theoretical bases. This criterion, although
satisfying the requirement of objectivity, does not necessarily provide the
best guidance for practical use. Experience actually shows that, depending
on the particular application, models of varying degrees of accuracy may
give similar results in the analysis of a structure.
Even a complete lack of certain features, which characterize very
markedly the behaviour of concrete as observed in the laboratory (for
example cyclic strength and stiffness degradation), may prove uninfluential
on the final results in cases where the overall behaviour is dominated by the
phenomena of extensive cracking and of the yielding of steel. How much
generality and accuracy is really needed is therefore a problem-dependent
question.
Almost all the categories of models that can and have been used for the
non-linear analysis of RC structures under general load histories will find a
place in the presentation to follow, with one notable exception: the
endochronic model, which was specialized for concrete in the 1970s by
Bazant and Bhat,7 on the basis of a previous theory developed by Valanis.
This model, although it has proved to be quite accurate in simulating the
observed behaviour of concrete, is hindered in the applications by the large
number of parameters required and, possibly for this reason, it appears not
to have undergone any further development in the last decade.
Without intending to create a hierarchy among the various models,
remarks will be offered on their limitations within each category, and on the
limitations of the categories themselves. The resulting overall picture will
hopefully serve as a reference for a reasoned choice of the model to be
employed in each particular case.

1.1.1. Classification of models

Several rather arbitrary alternatives are possible for categorizing the variety
of constitutive laws for compressed concrete elaborated so far: the one
chosen for this presentation is illustrated in Table 1.
A first separation is drawn at a high conceptual level, by distinguishing
between models that are formulated directly at a macroscopic level, and
models that are constructed starting from a (simplified) treatment of the
micromechanics of the matter.
Within the 'macroscopic' models, three broad categories are distin-
guished, with the understanding that the boundaries between them are
sometimes blurred, and that 'coupled' models have also been tried
successfully.
1.1.1.1. Elasticity-based models. These are so called because they all
share a Hookean formulation, obviously in incremental form. Historically,
the models of this class were the first to be utilized for multi-dimensional
analyses of RC structures, initially limited to cases of proportional loading
but, after suitable developments, also for repeated and cyclic loading.
From a physical point of view, the elastic model bears only a faint
resemblance to the internal mechanisms proper to concrete materials in the
stress ranges of interest; the reason for its adoption is clearly a matter of
practical convenience, not of a physical belief. In fact, the basic
mathematical structure valid for elastic solids has been (sometimes quite
successfully) forced to follow the complicated non-holonomic behaviour of
concrete by deducing the tangent constitutive matrix entirely from the
observed behaviour: this origin justifies the frequently used attribute of
empirical models.
1.1.1.2. Plasticity-based models. These models owe their name to the
links they retain with classical and/or advanced plasticity theories. The
RC ELEMENTS UNDER CYCLIC LOADING

Macroscropic models
Models derived from Models based on Models based on
theory of elasticity theory of plasticity progressive damage
Equivalent Invariant- Classical Bounding surface Fracturing Continuum
uniaxial based damage variables damage
Elwi- Stankowski- Vermeer- Fardis- Dougill17 Krajcinovic18
Murray8 Gerstle10 De Borst12 Chen14
Lemaitre19
Buyukozturk- Shafer- Han- Chen- Resende-
Shareef9 Ottosen" Chen13 Buy ukozturk15 Martin20
Yang et al. Krajcinovic-
Fonseka21
Mazars22
Coupled models
Micromechanics models
Microplanes Ba2ant-Prat

Table 1. Concrete under

compression

physical model behind elastic-plastic theories involves the occurrence of

irreversible slips within the crystalline structure of the material. As such, the
model is certainly a more appropriate startintg point for describing the
macroscopic behaviour of metals, rather than that of concrete. This well
recognized fact has not prevented plasticity theory being used with some
success in a number of analyses involving RC structures.
Plasticity models will be subdivided into two groups, the first related to
'classical' plasticity theory, the second embracing some substantial recent
developments.
Classical plasticity is meant here to include the last stage of evolution of
this theory, up to the formulation of hardening mechanisms and loading
functions tailored specifically for concrete materials. An exhaustive review
of these classical models has been carried out;5 a recent proposal within this
framework is represented by the work of Han and Chen,13 which will be
described in a later section of this Chapter.
The introduction of internal state variables, also called 'plastic damage'
variables due to the fact that they depend on the plastic part of deformation
only, has allowed plasticity theories to incorporate some important features
that were previously beyond reach, for example, the description of falling
branches, the coupling between the elastic and the plastic behaviours, and
the inelastic unloadings and reloadings.
The plastic damage variables group of models, which will be given a
separate place in the present framework, is often, although not necessarily,
associated with a further new concept, i.e. the one making use of 'bounding
surfaces'. As will be discussed subsequently in detail, the bounding surface
idea can be looked upon as a mathematical tool having a clear physical
counterpart, which happens to be very effective for describing the strain
hardening and softening mechanisms, as well as those of unloading and
reloading. The intrinsic qualities are enhanced when used in conjunction
with the damage variables. One model belonging to this latter type,
proposed in 1985 by Yang et al.,16 will be used in the following to illustrate
the capabilities of the approach.
 CONCRETE IN COMPRESSION

The physical concept of a gradual growth of microcracks within the

material under increasing applied stress seems a priori more suited than that
of irreversible dislocations for explaining the observed behaviour of
concrete. Attempts to produce stress-strain laws based on this internal
mechanism have followed two main lines, known respectively as fracturing
models and continuum damage models; these have differences which are
more of a formal than of a substantial nature.
As an introductory outline, the first approach can be said to move in
duality with the theory of plasticity, by associating an elastic component of
stress to a fracturing stress decrement. This latter is governed by a potential
function which is defined in the strain space much in the same way as the
loading function of the plasticity theory is defined and operates in the stress
space. A brief account of the first model using this approach, i.e. that
proposed in 1976 by Dougill,23 is given below.
The second approach is based on the use of one, or of a set, of state
variables which are meant to quantify the internal damage corresponding to
a given stress-strain history.
The damage and microcracking models referred to above account for the
progressive degradation of the material occurring at the microscale level
through the introduction of appropriate variables directly into the macro-
scale stress-strain relationship. A more physical, and conceptually more
promising approach would consist in trying to develop the macroscale
stress-strain relationship starting from the mechanics of the microstructure.
A few attempts along this line have been made, with only one, the so-called
'microplane model', having reached a stage of practical implementation. It
is a highly simplified model of the microstructure in which the contiguous
grains forming the material exchange forces on planes passing through their
contact points: the number of force components assumed to act on these
planes and the stress-strain laws attributed to each of them determine the
behaviour of the model at the macroscale level. The analytical description
of this model, accompanied by comparisons between predicted and
observed behaviours, at both the material and the structural scale, is given
in the Appendix at the end of this chapter.

1.2. Experimental 1.2.1. Introduction

behaviour of concrete Until the late 1970s the information available on the behaviour of concrete
under general subjected to three-dimensional states of stress was essentially confined to
muitiaxial stress cases of monotonically increasing and proportional, or nearly proportional,
histories loading histories. Further, the differences in the test apparatus and methods
used to measure the response of concrete specimens had the effect of
scattering the results rather widely, so that even the partial knowledge
gained was not totally reliable.
Notable progress was achieved during the late 1970s as the result of a
large cooperative experimental research project24 conducted by seven
laboratories working under unified testing procedures. The triaxial tests
programme consisted of a hydrostatic pre-loading at levels ranging from
75-200% of the unixial strength, followed by different types of deviatoric
paths up to the specimen failure. The main conclusions drawn from this
research can be summarized as follows.
• Considerable scatter of measured strains under applied muitiaxial
stresses are to be expected for concrete, no matter what test method is
used.
• The behaviour of concrete can be represented conveniently in terms
RC ELEMENTS UNDER CYCLIC LOADING

of the two first stress and strain invariants (octahedral values). The
underlying assumption of isotropy is not substantially faulted, at a
macroscopic level, up to stress states close to failure.
• The direct relationships between volumetric and deviatoric stresses
and strains are governed by the evolution of two independent moduli
K and G.
• A coupling effect between octahedral shear stress and volumetric
strain is systematically observed, and can be described by
introducing a coupling modulus H.

According to the findings above, three stress and strain dependent moduli
appear to be sufficient for describing the behaviour of concrete under
monotonically increasing loads. Within the same restricted range of
response, a discussion on the features of concrete behaviour which are
universally agreed, together with some guidance on the possibility of
analytical modelling, are given in the Synthesis Report prepared by CEB in
1983.17
Concurrently with the described efforts aiming at screening and
systematizing the available experimental knowledge, computer codes for
the analysis of structures subject to general load types were being developed
at a fast pace, thus creating the need for constitutive laws able to follow
non-monotonic, non-proportional loadings and unloadings. This necessity
was readily recognized by the research field, which was prompt to respond
with a spectrum of advanced analytical models overcoming the previous
limitations, sometimes even beyond the boundaries of previously
established knowledge. These models have been the subject of a review,
last updated in the early 1980s.6
The broadening of the experimental knowledge to cover at least the most
basic non-proportional loading paths is a very recent process, and much
work remains to be done. Only a few of the laboratories active in concrete
research have made such types of tests, the largest amount of data coming
from the University of Colorado,25 Imperial College in London,26 the
Bundesanstalt fur Material Priifung in Berlin,28 and the Air Force Weapons
Laboratory in the New Mexico State University.29 The data obtained by
these institutions constitute the basis for validating all the models that will
be presented below.

1.2.2. Main experimental results

For the sake of conciseness, reference shall be made here to the results
obtained at the University of Colorado and Imperial College only; however,
these results are representative of those from the other sources. All the tests
have been performed by applying triaxial axisymmetric principal stresses to
prismatic specimens. Being stress-controlled, they stop at peak stress; hence
they do not give information on the softening branches. To explore the
response of concrete in terms of the two first invariants, two fundamental
stress paths have been used by most of the laboratories: these paths are
illustrated in Fig. 2.
In the test shown in Fig. 2(a) the specimen is subjected to hydrostatic
loadings and unloadings applied at different levels of the deviatoric stress
To. This type of test is used to obtain in the first place the direct relationship
between oo and eo (which might be a function of To), and also the 'coupling'
relationship between a0 and the deviatoric strain 70.
In the test of Fig. 2(b), the specimen is subjected to deviatoric loadings
and unloadings, applied at different levels of hydrostatic stress &o- From this
test, the direct relationship between deviatoric stress and strain (TO — 70) is
 CONCRETE IN COMPRESSION

Fig. 2. Basic stress paths

utilized

= O3 o2 =
(a) (b)

obtained, as well as the coupling between the deviatoric stress TQ and the
volumetric strain eo-
From the first type of test one can observe that for continuously
increasing a0 the curve <r0 — eo starts with an initial maximum value of the
tangent bulk modulus K = Ko; the slope of the curve subsequently decreases
in an almost linear fashion with increasing cr0, up to a certain threshold
beyond which K remains approximately constant with a residual value of
about 15-20% of Ko. Unloadings and reloadings up to the same stress point
show an almost linear elastic behaviour with a modulus close to Ko. No
significant dependency of K from the amount of deviatoric stress is
observed.
The coupling between a0 and 70 is shown in Fig. 3. For a given increase
of the hydrostatic stress <T0, the amount of deviatoric strain 70 taking place is
an increasing function of TQ. This deformation appears to be irreversible
upon the unloading of CTQ, and it is also insensitive to a subsequent reloading
of cr0, at least up to its previously attained maximum value.
The direct relationship between TQ and 70 as obtained from the test type of
Fig. 2(b) is shown in Fig. 4. The three curves refer to different <TO values:
their common feature is a softening behaviour. The increase of cr0 also
increases the maximum shear deformation attainable prior to failure. As for
the volumetric quantities, in this case also unloadings and reloadings occur
almost elastically and with a modulus close to the initial one, at least for
unloading points sufficiently far from the ultimate stress.
Finally, in Fig. 5 the curves To — eo for different values of the hydrostatic
pressure are represented. There are two main notable features: the coupling

-225
r = 0-69 r = ;>07

O
-1-50

-0-75
I \
CU

Fig. 3. Coupling between

volumetric stress ao and 4
deviatoric strain 70 J'o
 RC ELEMENTS UNDER CYCLIC LOADING

Fig. 4. Relationship between

To and 70 as function of o§

i
10
/ •

= 2-28
— -
/

.
-

Jo = 3 0 8
o0 = 5-36

1
0-5

ICL

12 16

effect increases for higher confining stresses, as indicated by the different

slopes of the curves; furthermore, beyond a level of r 0 which is a function of
the applied confining stress <7o, the incremental deformation changes from
compaction to dilatancy. The total deformation prior to failure can give
either a volume increase or a decrease, depending on the value of <r0. For
low values of cxo one gets the well-known phenomenon of shear dilatancy.
The unloading-reloading behaviour is similar to that previously described
for the coupling between <TQ and 70: no deformation is recovered upon
unloading, no additional deformation occurs until TQ reaches the value it had
at unloading.
A further important objective of the research programme carried out at
the University of Colorado has been that of finding out unloading and
reloading criteria valid for less obvious cases than those described in the
above. This is an aspect of fundamental importance in view of the analytical
modelling of the behaviour of concrete: it involves the validation, or the
proposal of alternative more realistic criteria, of the only available rules
which essentially derive from classical plasticity theory.
The stress paths chosen for the purpose are described in Fig. 6(a) in terms
of the deviatoric stresses, while Fig. 6(b) contains the curves a, — e,
(/ = x, y, z) for each different path.

1-5

5-36

10

7/
o = 2 28

0-5

Fig. 5. Coupling between

deviatoric stress TO and
volumetric strain £Q
/ / -2 -4
fo
-6
ICL

-8
 CONCRETE IN COMPRESSION

14 - 1

12 -

a0 = 8 ksi 10

6- 8
/ jf 3/
V
6 /2
4 / 1
1/ fr
J
2
cu
0 ' . 1 . 1 . 1 , 1 . 1 . 1 . 1 1 1 . 1 . 1 . 1 . 1 . 1 . 1 . 1

10 12 14

Fig. 6. Non-proportional loading tests: (a) stress paths; (b) curves <7j — e* for each path

The state of stress is first increased from zero to a hydrostatic value of o-0
= 560 kg/cm2, which is henceforth held constant. It is noted that the initial
state of stress is such as to induce inelastic response. Path 1 consists of
increasing az, while ax and ay are decreased so as to keep 00 constant.
It can be seen in Fig. 6(b) that the curve az — ez proceeds inelastically
(softening behaviour), while the two unloadings are approximately elastic.
When az is subsequently decreased to its original value, all the three cr, — e,
branches are linear elastic. In path 2 the stresses az and ax exchange their
role, but the increase of ax in this case is such as to remain below the value
previously attained by az.
It can be seen that the curve ax — ex proceeds inelastically, while the two
unloadings of o~z and ay are approximately elastic. This observation is
already in contrast with the |To|max criterion, according to which path 2
should have been entirely within the elastic range.
The further paths shown in Fig. 6(a) and 6(b) support the statement that
an adequate criterion for distinguishing loading can be based on the prin-
cipal stresses: loading occurs (and inelastic behaviour with it) whenever one
(or more) principal stress exceeds its previously attained maximum value.
In concluding this brief overview of the available experimental
information on the behaviour of concrete under multiaxial stress states, it
has to be stressed that the tests carried out up to now have not explored
some of the phenomena whose existence and importance are well
recognized in the uniaxial case, and which must be presumed to play a
no less vital role in the three-dimensional case. These phenomena include,
among other aspects, the post-peak behaviour and the stiffness degradation
due to repeated loadings and unloadings. The more advanced mathematical
models that will be presented in the following paragraphs are in fact able to
describe these phenomena: their calibration, however, cannot be based at
present other than on mono- or bi-dimensional results, so that they must be
cautiously looked upon as reasonable extrapolations from known data.

1.3. Models derived The distinctive aspect of this voluminous and still widely used category of
from the theory of models is that the incremental relationship between strains and stresses
elasticity maintains the same form as for the elastic solids, i.e.
day — Djju deu (1)
RC ELEMENTS UNDER CYCLIC LOADING

where D,/W is the usual stiffness tensor, but its elastic parameters are now
made functions of the current stress state (tangent values).
Since the physical behaviour of concrete subjected to non-proportional
stress paths in the range approaching failure is substantially different from
that of an elastic body, the formal structure of equation (1) is a priori
inadequate for the general case. Despite this intrinsic limitation, the
assiduous work completed over the past decade to improve the realism and
enlarge the scope of these models, coupled with the firm experimental basis
from which they have emerged and against which they have been carefully
calibrated, have earned these models a stable place within the class of the
so-called 'empirical' models.
The discriminating feature for these models to be considered in the
present context is the inclusion of a criterion for distinguishing loading from
unloading and reloading, with different behaviours for each of these cases:
otherwise, the observed strong non-holonomy of the stress-strain relation-
ship could not be reproduced. Other aspects, such as, for example, an
accurate description of the coupling between deviatoric and volumetric
components of stresses and strains, although certainly important in some
cases, are not deemed equally essential.
The number of parameters entering the matrix D,//W is potentially very
large so that some initial hypotheses have to be introduced to get an
acceptable balance between accuracy, ease of correlation with experimental
results, and simplicity. Two main ideas have been pursued so far.
(a) To consider inelastic concrete as an orthotropic material, with
behaviour which, along each of the principal directions, is governed
by an 'equivalent' uniaxial stress-strain law: Elwi and Murray,8
Buyukozturk and Shareef.9 This model requires the definition of a
minimum of four state parameters.
(b) To describe the behaviour of concrete through a relationship between
the stress and strain invariants (Stankowski and Gerstle,10 Shafer and
Ottosen"). Depending on the number of invariants used, the two
mentioned approaches require the definition of four and eight
parameters, respectively.
A brief account of the two categories of model is given below, with closer
reference to the papers of Elwi and Murray and Stankowski and Gerstle,
respectively.

1.3.1. Orthotropic, equivalent uniaxial models

The matrix of the tangent stiffness tensor is given the form8

0 0 0
(A* 12^3 I + 0 0 0
symmetrical 0 0 0
(2)
<f>G\2 0 0
<j>G2i 0
G31

with <f> = 1/(1 - fi22 -M23 ~ M31 -2/xi2M23Al3i). a n d M?2 =

fj%3 = vn,vyi, /X3, = 1/31 ^i 3 . £,-,/ = 1-3 are the tangent Young's moduli in
the three (principal) axes of orthotropy, and G, 2 , G 23 , G 3 | the tangent shear
moduli in the planes parallel to the coordinate planes 1-2, 2 - 3 , and
3-1.

10
 CONCRETE IN COMPRESSION

To get the values of the tangent moduli £, recourse is made to the

'equivalent uniaxial strain' concept of Darwin and Pecknold.27 The concept
is based on the experimental finding that in a specimen subjected to a three-
dimensional state of principal stresses the curve cr, — e, in each direction
maintains approximately the same form irrespective of the stress states in
the other directions (including the state of zero stress). This fact implies that
a single normalized curve can be used for any composite state of stress,
provided that the normalizing parameters (peak stress and corresponding
strain, ultimate strain and corresponding stress) are adjusted for the
particular stress ratios. With reference to such a normalized curve, a
variation of the strain in the abscissas produces by definition the same
variation of stress which the specimen would experience in its actual three-
dimensional stress state: whence the appellative of 'equivalent uniaxial'
strain: e,u.
Consistently with the concept above, any suitable expression can be used
to describe the uniaxial response, for both compression and tension. The
same function may cover the entire range, or different functions may be
taken to describe portions of it: a requisite is that for continuous loading in
one direction the derivative of the adopted curve be continuous everywhere.
30
As an example, the generalized Saenz equation proposed by Elwi and
Murray is reported here.

0-, = (3)

where
=
RE EQI
e (c cr,f e,c
?ff- 1 1
R =RE
(Re -
and the physical meaning of the parameters aic, e/c, cr,f, e,? is illustrated in
Fig. 7.
The current (tangent) value of the modulus Et is now obtained upon
differentiation of equation (3): E, = d<r,/de,u, which yields an expression of
the type
Ei — Ei(eiu/aic, e,c, an, e,f) (4)
and the current value of e m is the integral along the loading path

£. ^3 (5)

Fig. 7. Stress-equivalent
strain curve
RC ELEMENTS UNDER CYCLIC LOADING

For the Poisson's ratio v two choices have been made. One is to consider
v as a constant,9 which is a reasonable approximation only to 70-80% of
peak stress, beyond which v must increase in order to reproduce the well-
known dilatancy effect in the range approaching failure. For v = constant,
the matrix D of equation (2) becomes
h) v{\ + 0 0 0
(1 - v2)E2 i/(l + 0 0 0
symmetrical (1 — 0 0 0
<j)G\2 0 0 (6)
SGiT. 0

with </> = 1 - 3v2 - 2i/\

Elwi and Murray, on the other hand, assumed v to vary as a function of
the ratio e m /e, c , that is, they considered a different Poisson's ratio for each
principal direction. Based on the uniaxial test data of Kupfer et a/.,31 the
proposed expression for v, is:
v; — VQ{\ + a e,u + be2u + c e3u)
with a, b, c = constants, and e,u = e,u/e,c.
1.3.1.1. Loading and unloading criterion. The loading-unloading
criterion is implicit in the formulation, due to the separate consideration
of the behaviour along the three axes: along each principal direction there
can be either loading or unloading independently of each other. This
'uniaxial' treatment lends itself easily to be extended to any specified
unloading-reloading behaviour, for example of the type indicated with
dotted lines in Fig. 7: this generalization has been implemented by
Buyukozturk and Shareef.
1.3.1.2. Parameters of uniaxial a — e(U curves. The values of the
stresses at failure aic for any combination of them are obtained by definition
through an appropriate failure surface: the well-known five parameters
failure surface of Willam and Warnke 32 is used, for example, by
Buyukozturk and Shareef. For the strain values corresponding to aic, two
alternatives have been proposed. One is to use a surface in the 'equivalent
strains' space having the same form of the failure surface in the stress space,
that is, a surface obtained from the latter by simply replacing a-,c with e,c.
The second alternative9 is to use an empirical expression relating aic with
e,c, of the type

where / c ' and ecu are the uniaxial compressive strength and the uniaxial
ultimate compressive strain, respectively.
The remaining parameters in equation (3) are the coordinates (o/f, e,f of
a point sufficiently far away on the softening branch of the curve. It is
known that experimental determination of the softening branch, especially
at large strains, is test-dependent and often even questioned on theoretical
grounds. The empirical proposal made by Elwi and Murray consists of
taking e,f = 4e,c, aic = 4ai{.

1.3.2. Invariants-based models

In the simpler approach by Stankowski and Gerstle,110" a deformability
matrix relating the increments of the first two stress and strain invariants is
given in the form

12
 CONCRETE IN COMPRESSION

\
[l/Y
where Aao, Ae0 are the hydrostatic stress-strain increments, and Aro, A70
the deviatoric stress-strain increments.
Notice that in assigning a generic stress increment: Acr,(/ = 1,2,3), the
corresponding Acr0 and ATO are uniquely defined, and so are the increments
Aeo, A70 from equation (7). These two latter quantities, however, are not in
general sufficient for the determination of the principal strain tensor
Ae,(/— 1,2,3), unless the stress-strain state is axisymmetric. For the
general case, it is assumed that the increments of the deviatoric strain
increments are parallel to the corresponding deviatoric stress increments,
although it is recognized that this is only a rough approximation of the real
behaviour.
1.3.2.1. Loading criterion and loading surface. Loading is defined as
the case in which one or more of the principal stresses increases beyond its
previously attained highest value. For stress states other than loading (i.e.
unloading and reloading) deformations are assumed to vary elastically
according to the tangent modulus at the origin for the 'direct' components of
strains (i.e. Aeo =/(A(7o) and A70 = / ( A r o ) ) , while the 'coupled'
components of strains (i.e. Aeo —/(Aro) and A70 =/(Aoo)) are assumed
to be completely unrecoverable.
Although the model being described is of a purely empirical origin and
could therefore be presented and used as such, it is possible to find certain
links with classical flow theories which are useful in providing a better
insight into its physical bases.
To this purpose, the concept of a loading surface well known in strain
hardening plasticity is recalled, and the model under consideration is
attributed a loading surface which changes its shape according to the
evolution of the stress state. More precisely, for stress states below 60-70%
of the peak stress, the loading surface is assumed as normal to the direction
of the increasing 07 (principal stress loading criterion), while for higher
stresses it tends to coincide with the failure surface. The mechanism of this
evolution will be clarified below.
1.3.2.2. Expressions of the moduli. The expressions of the bulk
modulus K and of the shear modulus G are given directly based on a large
number of experimental results.
K = Ko{\ - Ck <T0)
G = Go( 1 - ro/Tou) [GO = Gin( 1 - Cga0)} (8)
Equation (8) shows that both K and G display a softening behaviour with
increasing cr0 and To, respectively, and also that Go is a decreasing function
of the hydrostatic stress a. (rou is the current ultimate deviatoric strength for
the angle of similarity defined by the current stress state.)
As already mentioned, unloading and reloading proceed elastically with
moduli KQ and G,n, respectively. Expressions for the two coupling moduli

H= YY= —
Ae0 A70
have been derived for general load histories by Stankowski and Gerstle,
based on the results of ad hoc tests. If recourse is made to the loading
surface, however, the use of these empirical expressions is no longer
necessary.
Consider the stress state represented by point A in Fig. 8. Since there is
still distance between point A and the failure state, the loading surface is

13
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 8. 'Principal stress rule' Ai0 (Expansion)

(point A) and 'failure surface
rule' (point B)

assumed to be a rectangular box with faces parallel to the coordinate planes,

and the plastic component of the strain increment has the same direction as
the corresponding principal stress increment. It follows that the ratio between
the volumetric and deviatoric plastic strain increments is fixed, and given by
1
"V2
Using equation (7) it is easy to show that this flow rule is tantamount to
specifying the values of the coupling moduli H and Y as functions of K and
G. One finds
1 1
H = 2y/(2)G- Y = K-
1-G/Go ' ,/(!)" \-K/K0
It is seen from Fig. 8, that the volumetric strain increment
corresponds to a compaction which is in a fixed ratio with the distortion
A7Q, while it is well known that for To approaching its ultimate value TQU,
the total strain increment changes from positive to negative, i.e. to volume
increase.
This fact is taken into account by introducing a corrective function
F(TO/T 0U )

Aeg = F(TO/TQU)
P
A7 0 V2
obeying the following conditions

• F(TO/TOU = 0) = 1, that is, normality to the principal stress loading

function for To = 0
• F(T O /TO U = a) = 0, that is, zero plastic volume variation for
T 0 = QTo u
• = 1) = — (dTOu/d<To)-1, that is, normality to the failure
F(TO/T 0U
surface for TQ = Tou.

14
 CONCRETE IN COMPRESSION

The function F(TO/TOU) may be seen as the parameter governing the

evolution of the loading surface. It can be used directly, however, to modify
the expressions of the moduli Hand Y of equation (9) by replacing the factor
y/2 with i/(2)/F.
As already noted, the model of Stankowski and Gerstle is applicable
without further arbitrary hypotheses to the case in which two of the stress
and strain quantities are equal, i.e. to the axisymmetric case. A second
limitation is the inability to follow the behaviour of the material beyond the
peak stress state.
Both of these limitations are overcome by a more recent proposal by
Shafer and Ottosen," which may be regarded as a generalization of the
work by Stankowski and Gerstle; in fact, by elimination of some additional
terms and by consideration of the axisymmetric case, the latter model
degenerates into the former one. Only a few highlights of the newer model
will be given here, taking into account the fact that its calibration for
concrete materials is still incomplete.
The basic hypoelastic stress-strain relationship is initially presented in
the form

• * * l fiTl no,
{ B2 B 3\ \
L
AwJ
with the following definitions for the quantities Ae and Aw
Ae = SjjAejj/3T0 Aw = skmsIC&Aeik/'iTl
which show Ae and Aw to bear similarities with the second and third
incremental strain invariants, respectively.
After splitting the two 'diagonal' moduli A\ and B2 into an elastic and a
'corrective' part

=
B2 B\ + B2 = 2Go + B
and with the aid of some analytical manipulations it is possible to transform
equation (10) into a complete tensorial relationship of the form
ACT,,- = DijU Aekt (11)
containing eight parameters to be defined.
1.3.2.3. Loading criterion and loading surface. Characteristic of the
model is the presence of two separate loading functions, one for deviatoric
and one for volumetric loading. Bearing in mind the form of the constitutive
relationship equation (10), the loading functions are cast in the form
volumetric loading AL, = (A', Ae0 + A'2Ae + A'3Aw)
deviatoric loading ALd = (B\ Ae0 + B'2Ae + B'3Aw) (12)
where the primed quantities are suitable combinations of the unprimed ones.
Unloading occurs when AL, and/or ALd < 0, and the separate effects are
directly accounted for in the stress-strain relationship, since this latter can
be expressed as a function of ALy and ALd a s given by equation (12).
It is important to note that since the loading criteria are formulated in
terms of strain rate quantities, they are applicable also for the softening
range.
1.3.2.4 Determination of material parameters. A detailed explanation
of the procedure for calibrating the eight parameters to match the behaviour

15
RC ELEMENTS UNDER CYCLIC LOADING

proper to concrete is beyond the scope of the present review. However,

33
according to Eberhardsteiner et ai, calibration is achieved through the
following five conditions.
(a) Affinity between the deviatoric loading function and the failure
surface. This condition (which does not represent necessarily a close
approximation to the physical reality) yields two relationships
between the unknown parameters.
(b) Application of a 'cap' to the loading and failure surfaces in the range
of prevailing compressive volumetric stresses. This condition gives
two additional relationships between the unknown parameters.
(c) Application of equation (11) to represent proportional loading. By
equating equation (11) to the Hooke's law in incremental form the
equivalent tangent expressions of the modulus of elasticity: Et and of
Poisson's ratio ut are obtained as functions of the basic parameters.
This in turn allows to provide directly the functions Et and ut valid
for the uniaxial case (Et is obtained by differentiating the a — e
formula of Sargin,34 which includes a descending branch, ut is given
an expression based on tests by Van Mier,35 and then to generalize
the uniaxial expressions by introducing into them suitable equivalent
parameters of stress and strain.
(d) Application of equation (11) to represent pure hydrostatic loading.
(e) Imposing the condition of vanishing determinant of the material
stiffness matrix upon reach of the failure surface.

1.4. Models based 1.4.1. Classical plasticity

on the theory of Plasticity theory is based on the separation of the total strain increment into
plasticity an elastic and a plastic component

The elastic component is by definition fully recoverable, and is governed by

an elastic constitutive law which is generally assumed as linear and
isotropic. The plastic component is totally unrecoverable and starts to exist
when the point representative of the stress state reaches for the first time a
surface defined in the same space which is usually called the 'yield surface'.
Within the theory of perfect plasticity, the yield surface is a fixed surface
and the stress point can never lie outside of it. For plastic strain increments
to occur, the stress point must lie on the yield surface and the 'loading
index' must take a zero value. This index is defined as the internal product
between the stress increments vector and the unit vector defining the normal
to the surface (pointing outward). Thus
. P n -f df(amn)
de
U>0 lf
-^T-< = 0

= 0 = yield surface).
Geometrically, the condition above states that during plastic loading the
stress increments can only move on a plane tangent to the yield surface. A
negative value of the loading index implies that the stress point is moving
toward the interior of the yield surface, and no plastic strains occur in this
case.
A fundamental postulate of perfect plasticity is that plastic flow takes
place along the direction of the normal to the yield surface (normality rule).
The magnitude of the plastic flow is undetermined.
Perfect plasticity is unsuited in many respects for describing the
behaviour of concrete. One of the reasons for this is the gradual

16
 CONCRETE IN COMPRESSION

development of non-linearity actually exhibited by concrete materials,

which starts from stress levels far below those defining the yield surface.
An extension of the theory which allows incorporation of this kind of
behaviour involves the introduction of the concept of hardening. According
to this extended theory the onset of plastic deformation is governed by a
'first loading' surface. This surface defines the upper limit of elastic
behaviour and it is obviously located well within the outer surface
describing the fully plastic stage of the material.
The first loading surface evolves in the course of the deformation process
following a particular hardening mechanism, but keeps its function of
controlling the development of the plastic strains. For the latter to occur, it
is necessary that the stress point belongs to the current loading surface; if
the loading continues (i.e. no unloading takes place) then the stress point
remains attached to the loading surface during its evolution.
Three different hardening mechanisms have been mainly used so far.

(a) Isotroplc hardening refers to a mechanism according to which the

loading surface undergoes an isomorphic expansion depending on a
parameter usually related to the total (integrated) plastic
deformation.
(b) Kinematic hardening defines a mechanism in which the loading
surface does not change its shape but translates rigidly depending
either on the plastic strain increment (Prager's rule) or on the current
stress level (Ziegler's rule).
(c) The mixed mechanism is a combination of the former two
(Hodge).
Hardening plasticity also assumes, as perfect plasticity does, that plastic
flow is normal to the current loading surface. The normality rule, coupled
with the condition that the stress point must remain on the loading surface
('consistency condition'), completely determines the plastic strain incre-
ment. This is shown, by way of an example, with reference to the isotropic
hardening mechanism, for which the loading surface takes the expression

/ f a ) = *(ep) [ep
The consistency condition now writes

Introducing the plastic modulus tF defined as

one obtains

whence for the single components deL taking into account the normality
assumption, one has
df
. P= J day Qf_ ,

damn damn)

17
RC ELEMENTS UNDER CYCLIC LOADING

Exhaustive and detailed information on the various models which have

been implemented in the past within the above framework is available.5
Although research effort is still being expended to try to improve the
descriptive capacity of elastic-plastic models in relation to the actual
behaviour of concrete as observed in the tests, it is possible to discuss in
general terms some limitations which are inherent to the category as such,
and seem therefore very difficult to overcome.
In the first instance, contrary to widely recognized experimental
evidence, elastic-plastic models cannot but give unloadings and reloadings
which follow the initial elastic moduli, disregarding the phenomenon of
stiffness degradation and of energy dissipation due to hysteretic loops.
Secondly, the normality rule as applied to the loading surface does not
reproduce with sufficient accuracy the actual process of plastic deforma-
tions, in particular the ratio between the volumetric and deviatoric plastic
strains. It is known, for example, that a concrete specimen subjected to a
deviatoric stress increment exhibits a volumetric strain variation which is
contraction or dilation, depending on whether the state of stress is low or
approaching the ultimate, respectively. Additionally, the theoretical argu-
ments supporting the 'normality rule' (i.e. Prager's stability postulate) cease
to be a necessary condition for materials having deformation mechanisms
other than, or additional to, plastic crystalline dislocations, like frictional
slips or internal microcracking. This being the case for materials such as
concrete and many soils, the theoretical grounds for the normality rule do
not hold any more, while the experiments clearly show that this rule has to
be relaxed. A number of recent plasticity models incorporate this
modification, by using different types of 'non-associated' flow rules.
As a final point regarding the classic elastic-plastic theory, one must note
that this theory cannot account for a feature of the behaviour of concrete
which is clearly demonstrated in the tests: in the course of inelastic loading
histories a 'state of damage' develops and accumulates within the material.
Notwithstanding their severe theoretical and factual limitations, elastic-
plastic models in their numerous variants are still perhaps the models most
widely used in the advanced computer programs for example, ABAQUS,
ADINA, ASKA, DIANA. This is due in part to historical reasons since
plasticity theory, being relatively aged, has been investigated in depth and it
is thus thoroughly consolidated, a quality which the newer models,
appealing though they may be, obviously cannot possess. A second reason
is related to the present uncertainty about the importance of the role of
compressed concrete on the overall behaviour of the structures and, con-
sequently, about the real need of pushing the 'realism' of the models closer
than it already is.
At the closure of this section devoted to plasticity models it seems
appropriate to give an outline of a particular recent proposal (Han and Chen,
1987),l3 the main distinctive aspect of which is in the treatment of the
hardening mechanism and in the flow rule adopted. The reason behind the
choice of the model is to show that, without rejecting some fundamentals of
plasticity theory, one can depart enough from it to be able to move towards
a different category of models, denominated 'bounding surface and damage
models', which will be the subject of the next paragraph.
The model of Han and Chen defines a rather sophisticated evolution of
the loading function, from the initial yield surface to the final failure
surface, as illustrated in Fig. 9. The failure surface is expressed in the
general form
r-n{am, 0) = O
where

18
 CONCRETE IN COMPRESSION

Uniaxial compressive
Fig. 9. Non-uniform " loading path
hardening plasticity model Failure surface
(from Han and Chen13)

Loading surface

r = (2J 2 ) 1/2 , am = i / , and 6 = ^ = L ° d e angle

In particular, use of the five parameter Willam and Warnke32 failure

surface is suggested for r{.
The loading surfaces are defined as

where K is a shape factor.

The starting point for the construction of a loading surface is the base
surface (Fig. 10) described by
fb = r - Kort = 0
which represents an affine contraction of the failure surface. The shape
factor
K = K[o~m, Ko)
defining the corresponding loading surface, is such as to satisfy the
following conditions (see Fig. 10): for triaxial tension the yield surface
coincides with the failure surface, in the mixed tension-compression zone
the loading surface departs from the failure surface giving origin to a
hardening-plastic region and, finally, in the triaxial compression zone it first
follows closely the base surface but for higher hydrostatic compression
closes up at the hydrostatic axis for a value — om — p.
A suggested form of the function K is given in Fig. 11.
Each loading surface is then characterized by the hardening parameter Ko,
which varies between the value Ko = Ky it attains at the initial yield and the
value A"o = 1 at ultimate strength.
The hardening parameter also regulates the value of p, which is expressed
as
p = A/{l-K0)
with A = constant, so that it becomes infinite for Ko = 1.
Since all the loading surfaces intersect the uniaxial compressive loading
path, the base plastic modulus //£ for each loading surface may be defined
by its corresponding modulus value in an experimental uniaxial
compressive stress-plastic strain curve.
The plastic modulus to be used for a general stress state in the constitutive

19
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 10. Construction of yield

Tension-
surface (from Han and
Chen13)

Tension-tension
,= 0

relation is then obtained by multiplying the value of the base plastic

modulus by an empirical modification factor which is a function of the
hydrostatic pressure and Lode angle

In order to realistically describe volume contraction and dilatancy, the

direction of the vector of plastic strain is defined by a non-associated flow
rule and made proportional to the gradient dg/daij of a plastic potential
function g^f. This latter is chosen to be of the Drucker and Prager type,
given by

where a represents the plastic dilatancy factor and 0 is a constant with no

influence on the flow rule: a is a linear function of the hardening parameter
Ko and takes on negative values at the beginning of yielding and positive
values when the failure surface is approached.
The outlined model leads to an incrementally linear stress-strain law with
a non-symmetrical matrix due to the non-associated flow rule. The model
has been checked to describe correctly the following important features of
concrete behaviour

(a) work-hardening and ductile behaviour in compression, and linear

brittle behaviour in tension
(b) hydrostatic pressure sensitivity with change from compaction to
dilation near failure
(c) non-linear volume dilatancy.

Still, other no less important features remain beyond reach of this model,

K(om)

Fig. 11. Shape factor as

function of a,n (from Han
and Chen )

20
 CONCRETE IN COMPRESSION

i.e. inelastic unloading/reloading, post-peak behaviour and strength/stiffness

degradation. To cope with these phenomena, additional concepts have to be
introduced.

1.4.2. Bounding surface and damage models

An idea, originated by Dafalias and Popov36 and commonly referred to
under the name of 'bounding surface', can be said to represent a major
breakthrough in widening the scope and enhancing the flexibility of
traditional plasticity theory.
The bounding surface is a surface in the stress space defined through the
following properties.

(a) It encloses all the admissible stress states of the material.

(b) It evolves as a function of one or more variables which account for
the past loading history.
(c) Plastic strains can occur for stress states well within the surface.
(d) To each point representative of the stress state a corresponding
'image' point is defined on the bounding surface, according to a
mapping rule which is part of the constitutive modelling.
(e) The plastic strains rate is a function of the distance between the
current stress point and its image on the surface, and increases when
the distance decreases.

The concept of a bounding surface has been used for modelling the
behaviour of such diverse materials as steel,36 soils,37 and concrete.1 l6 ' 38
Depending on the specific application, the concept may require adaptations
and additions with respect to those basic features which have been
mentioned. For example, metal plasticity requires an enclosed surface to
define yielding of the material, while in some other models a second
internal surface has been introduced with a meaning and a purpose
analogous to that of the loading surface in hardening plasticity. The general
formulation of bounding surface plasticity is briefly summarized in the
following. The bounding surface is given by the expression

where the bar over the a-^ indicates that they belong to F = 0. The state
variables qn accounts for the effects of the load history and are functions of
the accumulated plastic strains. Following Dafalias and Herrmann,37 the
plastic strains rate is given by

where L is the loading index (positive if plastic strains are to occur) and Ry
is the gradient of an appropriate potential function. The loading index L is
taken as proportional to the internal product of the gradient of a loading
surface by the vector of the stress increments

L N d

The proportionality factor is the inverse of the plastic modulus: hP. The
definition of L given above applies to perfect as well as to hardening
plasticity and has the advantage of correctly indicating 'loading' when the
material is in the softening range. (In this case Ny d<7,j becomes negative,
but since ffl is also negative their product is positive.)
Note that L is not only a loading index, but gives also the modulus of the
strain increment vector.

21
RC ELEMENTS UNDER CYCLIC LOADING

As an alternative to the previous definition, the index L can be defined

with reference to the bounding surface, i.e.
1
r, j-
ij d(7ij
7w
where Ny is the gradient of the F = 0 at the 'image ay, and Hp is a ficticious
plastic modulus associated to the bounding surface.
The consistency condition as applied to the 'image' point belonging to F
= 0 yields
dF A_ dF
•^—dcrij + —dqn =0
day dqn
whereupon one obtains for the plastic modulus Hp

In those cases where a second internal surface defining loading is not

introduced, then
Ny=N{i
is adopted.
Finally, the actual plastic modulus fP is defined through a function
relating it to Hp and to the distance 8 between ay and ay. This function is
often assumed of the form
Hp = Rp +f(6)
and it is such that for 8 > 0 one has Hp > Hp while for 8 = 0 the two
moduli become equal.
From the presentation above it appears that the classical elastic-plastic
theory is but a limiting case of the bounding surface formulation, which can
be obtained by choosing the function f{8) to be infinite for 8 > 0 and zero
for 8 = 0. Conversely, one can observe that classical elastic-plastic theory
can be easily transformed into a bounding surface formulation, simply by
assuming the failure surface to coincide with the bounding surface, and by
introducing a proper inverse mapping from the points of the bounding
surface to the actual stress points.
1.4.2.1. Damage variables. The use of variables related to the total
(integrated) plastic strain is a feature common to all plastic-hardening
models, including the one under consideration. A new useful perspective
consists of looking at these variables as a measure of a state of damage
progressing within the material. In most of the models proposed so far, only
a reduced number of such variables is used, and these have a scalar
character. This latter assumption is clearly motivated by reasons of sim-
plicity, since in principle the damage should be dependent on the 'direction'
of the stress-strain path, and therefore vectorial or tensorial measures
should be more appropriate for its description.
The existence of a state of damage is reflected in the behaviour of the
material as a general deterioration of the mechanical characteristics, namely
strength and stiffness. The reduction of strength is dealt with, in the bounding
surface approach, through the contraction of the bounding surface itself.
One possibility for modelling stiffness degradation is obtained with the
introduction of the plastic damage variables into the potential functions
from which the elastic strain tensor is obtained. This procedure is called
elastic-plastic coupling.

22
 CONCRETE IN COMPRESSION

A number of constitutive laws for concrete based on the bounding surface

concept and using plastic variables to describe damage have been proposed
in the recent past.1 16'38 The one proposed by Yang, Dafalias and Herrmann
in 1985 offers a good example of the flexibility of the concept in
reproducing most of the experimentally known behavioural properties of
concrete.
The constitutive model of Yang, Dafalias and Herrmann is based on a
non-classical plasticity formulation characterized by

(a) two simultaneous but independent plastic mechanisms governed by

two loading surfaces
(b) scalar internal variables which represent a progressive damage
process
(c) change of elastic properties in the course of plastic deformation
(elastic-plastic coupling, stiffness degradation)
(d) concept of bounding surface used for describing the hardening
process in one of the two plastic mechanisms.

The components of the plastic strain rates, for each one of the two loading
surfaces, are expressed, according to the general formulation previously
given, as
A$ = LNl (13)
where L, loading index, (>0 for plastic loading), is given by

L = ±N»d*u (14)
with Njj = Q£-, with / = 0 = loading surface and H = generalized plastic
modulus.
L also gives the intensity of the plastic strain increment, the direction
being given by Njj which may be different from the normal to the loading
surface Nj- (non-associated flow rule).
Damage variables are included in the formulation to represent the effect
of progressive damage, mainly due to microcracking. These internal
variables may be directly identified or are connected to a scalar ep, which is
a measure of an effective accumulated deviatoric plastic strain
4^ '/2
The elastic part of the total strain is expressed as

(15)
«-<*;
where ^(a/y, &), the complementary free energy density, depends on the
damage variable k, which, in turn, is related to ep.
The total strain rate may be written as
de;y = d 4 + de? = de)/ + d6^ + de? (16)
where the elastic part has been divided into a reversible contribution
d9
= Cijlm dalm, Cijlm{k) = (17)
ooij oai

which is present at both loading and unloading, and in the elastic-plastic

coupling contribution

23
RC ELEMENTS UNDER CYCLIC LOADING

which is present only at loading.

By using equations (13), (14), (17) and (18), equation (16) takes the form
of an incremental strain-stress relationship, which can be explicitly inverted
to produce a relationship between stress rates and total strain rates, this
latter being more directly useful in computation. Also the loading index
(equation (14)) may be given a more convenient expression in terms of total
strain rate; as will be noted later, this may be used to avoid ambiguity in the
definition of loading and unloading at post-failure states.
The bounding surface, F{a,j, ep) = 0, is a surface in stress space which
encloses all current states and evolves in connection with ep. The loading
surface is always enclosed in it; contact between the two is possible, but not
intersection (same normal vector in the contact point). When the two
surfaces are in contact, and as far as plastic loading continues, the current
stress point belongs to both surfaces and a consistency condition, AF = 0, is
satisfied for the bounding surface, which implies

(19)
L
When plastic loading is active but the two surfaces are not in contact the
stress point belongs only to the loading surface. An image stress point on the
bounding surface, atj, is defined through a proper mapping rule, together
with a distance S between 07, and its image <7,y.
The plastic modulus in equation (14) is then defined as
H = H +f{8) (20)
H is still given by equation (19), computed at al}, and/(5) is a positive
decreasing function of 8 such that/(O) = 0 (which implies H = H when the
two surfaces are in contact).
Of the two plastic mechanisms used to formulate the concrete model,
only the first one uses the concept of bounding surface.
The loading surface/! = 0 and the bounding surface F = 0 are in this case
two similar cones in the space of the principal stresses, with the same origin
and same axis, that is the hydrostatic axis. Equivalently, the three stress
invariants / = an, J = (1/2 sy sp) and the Lode angle 6 in the n plane, may
be used to define them (Figs 12 and 13).
The loading surface always passes through the current stress point. When
plastic loading associated to /1 = 0 occurs, the computation of the plastic
strain rate requires the definition of the mapping rule between the current
stress point and its image on the bounding surface; this simply consists of a

F= 0

It 0

Fig. 12. Loading surface and bounding surface in Fig. 13. Loading surface and bounding surface
the I-J plane'6 in the it plane is

24
 CONCRETE IN COMPRESSION

Fig. 14. Non-associated Compaction

plastic flow mechanism16

f, = 0

radial projection on the it plane which also defines an adimensional distance

8 (see Fig. 13).
The plastic modulus H can be computed according to equation (20). Since
the bounding surface, F = 0, is supposed to contract with increasing ep, H
(equation (19)) is always negative.
The expression used iox f(8) is:

f(8) = Eoh 8in

with Eo = initial Young's modulus, h a shape parameter and 8m the value

that 8 assumes at the beginning of each plastic loading stage. f(6) is always
positive and its value varies between +oo for 8 — £,n and zero for 8 = 0.
The plastic strain rate, in this first plastic mechanism, is not associated,
that is, is not normal to f\ = O(Njj ^ N?). This is required in order to
correctly model the volumetric compaction/dilatancy caused by a pure
deviatoric stress increment. If an associated flow rule were used the effect
would always be dilatancy. As shown in Fig. 14, the vector N9 is
constructed by keeping the component of the vector N" in the deviatoric
plane, and then by adding to it a new component parallel to the hydrostatic
axis in such a way that the desired compaction/dilatancy effect is obtained.
The scalar value of this component is specified as a function of the distance
8 between the loading and the bounding surfaces, in accordance with the
experimental fact that dilatancy occurs near failure while compaction occurs
when the stress state is sufficiently distant from it.
The second plastic mechanism regulates the plastic deformation for
increasing compression. No bounding surface concept is adopted in this
case. The loading surface f2 = 0, which is a plane perpendicular to the
hydrostatic axis (Fig. 12) undergoes a simple isotropic hardening, that is,
may only translate towards increasing / values if 'pushed' by a stress
increment when the stress point lies on it. The flow rule is associated, that is
only volumetric deformation occurs for increments of hydrostatic com-
pression. The plastic modulus H2 is directly given as a function of /.
The elastic part of the strain increment is computed using equations (17),
(18), starting from an expression for \& in the form
2
\&- L / L / 2
2G{k)~
This is the usual relationship for linear isotropic elasticity (with K bulk
modulus and G shear modulus) with the only difference that G is now a
function of the damage parameter k. This is related to ep but increases only
if the current stress point is sufficiently close to the bounding surface, that

25
RC ELEMENTS UNDER CYCLIC LOADING

is, to failure. The specified relationship between G and k implies stiffness

degradation with increasing damage.
Compared to models based on classical plasticity the present constitutive
model is superior because it includes stiffness degradation during pro-
gressive damage processes, shear compaction-dilatancy phenomena, and
post-failure strain softening behaviour. All the analytical expressions
involved are completely defined, and only one parameter, the compressive
strength fc, has to be specified.

1.5. Models based These models aim at reproducing, at a phenomenological macrolevel, the
on the mechanism of internal mechanism of microcracking and growth of defects and micro-
microcracking or cavities which arise in brittle materials, such as concrete, as deformation
elastic damage progresses. The weakening of material in these models is revealed by a
decrease of stiffness without permanent deformation when the material is
unloaded. This mechanism may also be termed elastic damage.
This idea has been implemented in two formally distinct ways, which are
conceptually very much related. The first, proposed by Dougill,23 and
known as the theory of progressively fracturing solids, has been developed
in strong analogy with the theory of plasticity. The second, based on the
explicit introduction of damage variables in the constitutive relation,
belongs to the vast field of investigation known as continuous damage
mechanics.

1.5.1. Progressive fracturing material models

Following Dougill's proposal,23 in an elastic progressively fracturing
material the response at a point is given by the generalized Hooke's law

in which the elasticity tensor C,,/,* is affected by the history of deformation.

Two kinds of behaviour are considered
(a) 'loading' during which degradation occurs
(b) 'unloading' or 'reloading' which are characterized by a purely elastic
behaviour.
To distinguish between these two modes of behaviour and to describe the
conditions which develop during degradation a loading function is
introduced F(eij,Ha) which defines a 'fracture surface' in strain space
F(e,7, Ha) = 0 (22)
in terms of the quantities Ha, which are dependent on the history of
progressive fracture. The fracture surface plays the same role in the theory
of progressively fracturing material as does the yield surface in the theory of
hardening plasticity. Thus any deformation path that lies entirely within a
particular fracture surface is accompanied by a linear elastic behaviour with
unchanging values of Cjjhk and Ha. However, during loading, F = 0 and the
state of strain is described by a point lying on the fracture surface. This
surface expands to adapt to the current state of deformation and in
accordance with the manner in which Ha change with degradation. The
resulting behaviour is that of a material which
(a) loses stiffness due to stable progressive fracturing during loading
(b) unloads in a linear elastic manner, with stiffness depending on the
extent of progressive fracture prior to unloading
(c) has the property that the material may always be returned to a state of
zero stress and strain by linear elastic unloading.

26
 CONCRETE IN COMPRESSION

Fig. 15. Elastic and

fracturing components of
stress

In order to obtain the flow rule for the material, consider the incremental
version of equation (21)
dehk k thk (23)

This expression suggests that the stress increment day is the sum of an
elastic component do? given by the first term in equation (23) and a
component — da1- = dCyhk ehk, caused by the change in stiffness and termed
the fracture stress decrement. With reference to a unixial case the two terms
are sketched in Fig. 15.
Application of Il'iushin's postulate of plasticity and of the restriction that
the total work done in a small cycle of deformation should be non-negative
leads to the requirement that
AW{ = ]j de,:/ > 0 (24)
The flow rule then follows by analogy with the theory of hardening
plasticity

da ij = d A — (25)
oe-y
and the consistency condition of the loading surface may be used to
determine the scalar function dA. Assuming, for example, the simple form
of F, F = F(e,y) — H(Wl), which depends on a single scalar H, expressed
as a function of the fracturing work W1, we get

(26)
dey dWl
where dWf = 1/2 da\: ey is the shaded area indicated in Fig. 15.
Using all the previous relations finally we get

—2 dehk (27)
dH (dF/demn)e
which is the general constitutive equation for a progressively fracturing
solid.
An analytical expression of the function dVfldH has been deduced by
Spooner and Dougill on the basis of data obtained by uniaxial tests. From
the previous equations, and the additional postulate that the change in
stiffness caused by an increment of deformation is independent of the
deformation path, Dougill has obtained an expression for the stiffness

27
RC ELEMENTS UNDER CYCLIC LOADING

degradation rate this is necessary for the updating of the current

stiffness tensor.

1.5.2. Elastic damage models

The second approach is based on the so-called continuous damage
mechanics which studies degrading material behaviour in terms of the
evolution of a continuous defect field. In the simplest case a scalar measure
of damage D is utilized to describe the continuous distribution of
microdefects in the form of voids and microcracks.
Considering the elementary example of a uniaxial test and the decrease of
the effective load bearing section Aeff in Fig. 16. the nominal value of stress
is simply

or = - - = creff(l -d) (28)

A
where aeS = P/AeS.
In this case Aes/A0 = 1 — d represents the intact load bearing fraction of
Fig. 16. Damage as reduction the cross-sectional area Ao. As a result the damage variable d represents the
of the bearing area damaged fraction of the nominal cross section area
Ap - Ae{{
d = (29)
Ao
This simple example can constitute the basis for a generalization to more
complex relationships. In fact it is reasonable to postulate that, given a
general constitutive relationship valid for the intact material in the symbolic
general form

e
U =fij(a'hk) (30)

the same constitutive law may hold for the damaged material by the simple
substitution of an effective stress tensor aeg in place of a.
Assuming that linear elastic behaviour governs the effective stress-strain
relationship
= Ee (31)
then the nominal stress-strain relationship reduces to the simple secant
relationship
a = E(l -d)e = E'e (32)
where 0<d< 1.
This shows a very simple and direct relationship between damage and
stiffness reduction; such a relationship may be useful for identifying the
damage level through experimental measurements of elastic moduli at
unloading. These simple considerations introduce the simplest form of an
elastic damage constitutive relationship. This needs to be completed with
the definition of an evolutionary law for the damage parameter as a function
of the stress and strain history.
The outlined version of an elastic damage model, with a single damage
parameter, is the simplest constitutive relationship of this category. In spite
of its simplicity it has been implemented by various authors, Resende and
Martin,20 Krajcinovic and Fonseka,21 Mazars,22 with some success in
describing the behaviour of concrete.
Also evident is the close relationship of this elastic damage formulation
with the progressively fracturing material introduced by Dougill;23 the only
difference being that here no fracture surface or flow rule, reminiscent of

28
 CONCRETE IN COMPRESSION

classical plasticity, are required. Use is made instead of an internal variable

representing the extent of internal damage together with an evolution law
defining the rate of damage.
In order to give a better understanding of the structure of simple damage
constitutive laws for the description of concrete-like behaviour, the proposal
by Resende and Martin,20 will be briefly presented. This work, although
partly superseded by further developments, ° remains a good example of a
formulation which is built completely within the framework of elasticity
and damage.
The constitutive law is given in incremental form and relates invariant
quantities, such as hydrostatic stress rate &m and corresponding volumetric
strain rate ev, deviatoric stress rate and conjugate deviatoric strain rate e;
this latter is defined as

e=- ^ (33)

The damage variable d is assumed to be a non-decreasing function of the

total deviatoric strain, in the form
d=A(crm, e)e e>0 (34)
where the dependence on am is used to model the observed fact that, for the
same cumulated deviatoric strain e, damage is less at larger hydrostatic
compression.
The relation connecting s and e in total terms is defined by
s = G0(l — d)e (35)
which is formally identical to equation (32). In incremental terms it
becomes

GQ(\-d)e-G0Aee e>0
[ }
G0{l-d) e<0

As for the volumetric strain rate, ev, it is assumed to depend both on &m
and on the deviatoric strain, to make it possible to reproduce the known
behaviour of concrete dilation under shear deformation; it is expressed as
the sum of an elastic component and of an inelastic damage component
e v - eev + ^ (37)
The elastic component e® is a non-linear function of hydrostatic stress in
order to obtain the effect of progressive incompressibility of the material as
am increases; in incremental terms it is written as
am =f(eev, ev)eev (38)
The damage component is in turn expressed as

^ = eP + e^ = (c, + c2e)d + c3de (39)

that is, as the sum of a permanent term and a recoverable one.

The first term produces a behaviour of the type shown in Fig. 17, where,
after an initial compaction, dilation follows, due to uplift in sliding at
microfracture asperities; the different curves in the figure show dependence
on the level of hydrostatic compression.
The second term is specially used to account for the unloading behaviour.
At unloading it gives a negative contribution to stiffness contribution. This

29
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 17. Volumetric strain as

function of the deviatoric
strain Compaction

Dilation Increasing
hydrostatic
pressure

is larger the larger is the value of the damage variable d prior to unloading.
At large values of d the unloading stiffness appears negative, as also
experimentally observed (see Fig. 18).
The final volumetric relation is obtained upon substitution in equation
(38) of the difference ev - e^, in place of e®
om = f (
and hence

f (eev, e v )[e v - (ci + c2e)A e -

(40)
f(eev, ev)(ev-c3de) e <0
The two equations (36) and (40) define the structure of the constitutive
relationship in invariant form; from these a generalization follows to a full
tridimensional stress-strain relationship.
Continuous damage mechanics has a broader scope than the limited
version outlined here with particular reference to elastic damage form-
ulations. A spontaneous generalization is, for example, to define more than
one damage variable or tensorial damage variables.
A rigorous formulation of such general constitutive relationships, which
is directly connected to the thermodynamics of irreversible processes and to
the internal variables formulation of constitutive relationships is presented
for example by Krajcinovic.4' Models of such generality are not available at
the moment for concrete.

Fig. 18. Volumetric strain as

function of volumetric stress

30
 CONCRETE IN COMPRESSION

1.6. Models based With reference to the actual behaviour exhibited by real materials such as
on the association of concrete, a positive feature of models based on progressive fracturing or
plasticity and elastic damage lies in their natural ability to reproduce both the hardening
microcracking and the softening range, together with stiffness degradation. In spite of these
mechanics qualities, the complete reversibility proper to these models makes them
unsuited for describing the behaviour of concrete, which is known to present
unrecoverable deformation. Since this phenomenon is well treated by
plasticity theory, the idea has arisen of combining the plastic and the
fracturing mechanisms into a single composite or mixed model. This has the
additional appeal of reproducing at macrolevel the underlying mechanical
processes of plastic slippage and microcracking both present in the real
behaviour. The coupled mechanism is illustrated qualitatively in Fig. 19.
Figure 19 represents in (a) a typical elastoplastic-hardening relationship
with unloading stiffness equal to initial stiffness; in (b) an elastic damage
relationship with stiffness degradation as deformation progresses, but no
residual strains; and in (c) a more realistic relationship where both stiffness
degradation and residual strains are present.
The first authors to exploit this possibility were Ba2ant and Kim.42 Many
more proposals have followed. A very brief account of these is given here
with reference to Fig. 20 which attempts a unitary frame in which such
proposals can be accommodated. The boxes relative to elastoplastic and
continuous damage models have already been commented on in specific
parts of this Chapter. Here a short comment is given on the most recent
proposals of the mixed or composite type.
In the pioneering work by BaZant and Kim a classical elastoplastic
mechanism with a Drucker and Prager type surface and mixed type
hardening is coupled with a progressively fracturing mechanism described
according to the Dougill23 theory. Two loading surfaces are introduced, one
of which controls the plastic deformation and the other the fracturing stress
relaxation.
A similar idea has been used by Han and Chen.43 They associate to a
Dougill type fracturing mechanism an elastoplastic model which has the
peculiarity of being defined in strain space as an alternative to the more
usual formulation in stress space. This allows the adoption of a single
loading function for the components of the stress increments of the two
types.
In recent times classical plasticity has found some effective association
with the continuous damage theory. Among the papers devoted to the
formalization of the different categories which can be generated by this
approach a special mention has to be given to papers by Krajcinovic ' and
Lemaitre.44
A specific proposal which refers to concrete has been presented by

Fig. 19. (a) elastoplastic

model with hardening;
(b) elastic damage model;
(c) combined model (c)

31
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 20. Combinations of -jPlastic damage vari<lblos

constitutive models
Elastoplasticity -

(Bazant-Bhat [7],
Han-Chen) [43]
(Resende) (Lubliner ef a/.)
Progressive [40] [2]
fracturing

Continuum (Krajcinovic,
damage theory - Lemaitre,...)
[41.44]

Continuum damage
mechanics

Resende.40 In this proposal the continuous damage model is more elaborate

than the one already mentioned by Resende and Martin.20 In fact two
damage mechanisms are present: one used for representing tensile cracking,
the other to describe the shear-induced microcracking observed mainly in
compression. The plastic strain component intervenes when there is a
variation of the volumetric component of stress and is used to allow for
residual deformation for purely hydrostatic loading histories.
Finally a model which puts together elastoplasticity with damage
variables and continuum damage mechanics is the one proposed by
Lubliner et al.2 In such a model plastic damage is associated to a single
variable which is used to define the contraction of the yielding surface.
Stiffness degradation is, on the other hand, associated to two different
groups of internal variables. The variables of the first group find their
collocation in the elastic damage mechanics since they depend on total
deformation. The other variables are associated only to plastic deformation.
With such variables the stiffness matrix of the material changes and
degradates as deformation progresses.

32
 CONCRETE IN COMPRESSION

Appendix: General non-local microplane

model for concrete
A.1. Introduction A new general non-local microplane model for concrete is presented, which
is capable of predicting concrete behaviour, not only in the case of full
three-dimensional monotonic loading, but also in the case of general three-
dimensional cyclic conditions, including strain-softening and rate effect.
This model represents a constitutive model in which the material is
characterized by a relation between the stress and strain components on
planes of various orientations. These planes may be imagined to represent
the damage planes or weak planes in the microstructure, such as contact
layers between aggregate pieces in concrete. The latest version of the
microplane model, developed by Bazant and Prat45 was shown to be capable
of predicting the behaviour of concrete in monotonic loading for a broad
range of stress-strain histories using only a few material parameters. This
model, together with the non-local concept, was implemented in a finite
element code.46 In a number of numerical examples, its capacity to
realistically predict the structural response has been demonstrated.
However, to predict the concrete behaviour not only in the case of
monotonic loading but also more generally in the case of the cycling loading
and rate effect, a further generalization of this model is necessary and
reported below.
It should be noted that the basic concept behind this microplane model
was advanced in 1938 by G. I. Taylor and developed in detail for plasticity
by Batdorf and Budianski in 1949, under the name 'slip theory of plasticity'.
That name became unacceptable when the idea was extended in 1984 to
include damage and friction. Extension to strain-softening damage required
various modifications, such as replacing the static constraint with the
kinematic constraint (the static one is unstable for strain softening) and
combining the elastic and inelastic strains.45

A.2. Review of the A.2.1. Basic hypotheses and strain components

microplane model The basic hypotheses used in the present cyclic microplane model are as
follows.
• Hypothesis I. The strains on any microplane are the resolved
components of the total macroscopic strain tensor e(/. This represents
a kinematic constraint and yields the relations (Fig. 21)

k),- = nie,7; eN = n;n7e,y-; CD = n/n/e,;,- — ev (41)

CT = ^Miri + ekk; CM = meM = m,-n/eI7;
eK = ke T = k/nye/,- (42)

Hypothesis II. Each microplane resists normal and shear strain

components. As a consequence of Hypothesis I, the shear stress or,
does not have in general the same direction as the shear strain ex-
Hypothesis III. In order to correctly model the concrete behaviour in
the case of very high confining pressure, the response on each
microplane is assumed to depend on volumetric strain ey = tkk/3.
Hypothesis IV. The stress-strain curves of each microplane are path-
independent as long as there is no unloading on the microplane.
During each unloading and reloading, which is defined separately on
Fig. 21. Microplane strain each microplane, the curves of the stress and strain differences from
components the state at the start of unloading are also path-independent. Thus, all

33
RC ELEMENTS UNDER CYCLIC LOADING

the macroscopic path-dependence is produced by various combin-

ations of loading and unloading on all the microplanes. It may be
noted that some microplanes may get unloaded even for
macroscopically monotonic or virgin loading, thus making it path-
dependent. The number of possible macroscopic path directions is
enormous (for 21 microplanes there are 2 2 possible tangential
stiffness matrices in each loading step, due to all possible
combinations of loading and unloading).
Hypothesis V. The volumetric and deviatoric responses on each
microplane are assumed to be mutually independent, except that in
the case of high volumetric compression shear response depends on
the volumetric strain (otherwise it would not be possible to model the
influence of high lateral compression stresses on the axial
compression strength).

A.2.2. Microplane stress-strain relations

In the case of virgin loading the behaviour of each microplane strain
component is described by path-independent total stress-strain relations of
the form
cry = Fy(ey); crD = F D ( € D ) ; crM = (43)
To represent unloading and application of the non-local damage concept, it
is convenient to define the stress-strain relation in the form of continuum
damage mechanics

cry = Cyey; ot> = CDCD; crM = (44)

in which, except for volumetric compression,

CM = CK = CK(1 - (45)
where Cy, CD, CM, CK represent the secant moduli Cy = Fy(ev)/ev,
CD = F D (e D )/e D , CM = FM(eM)/eM, CK = FK(eK)/eK; C v , C&, CM, C£
are the initial values of Cy, CD, CM, CK; and cjy, CJD, WM, UK are the
volumetric, deviatoric and shear damage on the level of the microplane. The
shear microplane moduli are equal for both directions (CM = CK). The
damage microplane functions for the virgin loading are approximated as

ev > 0 : uiy= 1 — exp

•)

£D >0: u'D = 1 — exp

(-iffD
6D <0: ato = 1 — exp
(-110
-exp - — (46)
1
«
in which e$ = e^ if ey > 0, and e$ = e^ — ^ e y if cy < 0, where
e\, e2, «3, ^4, m, n and k are empirical material constants. The dependence
of e5 on the volumetric strain ey reflects internal friction and represents the
additional kinematic constraint of a scalar type. In volumetric compression

34
 CONCRETE IN COMPRESSION

behaviour there is no damage (u;y = 0) and the response for virgin loading
is described by

(47)

where a, b, p and q are empirical constants.

A.3. Unloading, In the previous version of microplane model45 it was possible to represent
reloading and cycling only the first unloading, and even that with considerable errors compared to
loading experiments. To model unloading, reloading and cycling loading and do so
even for general triaxial stress states, more complex rules on the microplane
level are needed. After a number of trials, the following unloading-
reloading rules, which are different for each microplane strain component,
have been chosen and verified.
In contrast to virgin loading, the stress-strain relation must be in the
incremental form and is as follows
day = Cydey; dot) = (48)
where Cy, Co, CM, CK represent unloading-reloading tangent moduli
which are defined as follows (see Fig. 22)
\ -a)a/(e-ei)
e > eP; = e P - aP/C0 + {3(e - e P ) (49)
e < ep; = 0

where, <7p and ep denote the positive or negative peak stress and the corres-
ponding strain for each microplane component, using values <rp~, ep" and dp
ep for the positive and negative peaks; a and /? are empirically chosen
constants with values between 1 and 0. The general shape of the unloading-
reloading rules for the microplane components is shown in Fig. 22.

"D

op 7

(b)

Op

CM.K
Fig. 22. Stress-strain
relations on microplane:
(a) volumetric tension-
compression; (b) deviatoric;
(c) volumetric compression-
tension; (d) shear (C)

35
 RC ELEMENTS UNDER CYCLIC LOADING

A.4. Incremental Since on each microplane the kinematic constraint is used, the microplane
macroscopic stress- stresses cannot be exactly equal to the resolved components of the
strain relations macrostresses (static constraint). The micro-macro equivalence of the
stresses can be enforced only approximately, which is done by means of the
principle of virtual work. This yields an incremental macroscopic stress-
strain relation of the form
= CijrsAers - (50)
in which

- -z- -
1 / kn/i r« - CD)
7s L ->

- ( msnr)C u

- ( kjiii)(krns KJ F(n)dS (51)

kjnT) A o d F(n)dS (52)

Here, Cyrs is the macroscopic incremental material stiffness tensor, and ^

are the corresponding macroscopic inelastic stress increments. These values
are calculated using an efficient numerical integration scheme over the
sphere that involves 21 plane directions.47'48
The values of the material parameters used in the calculations are the
same as described earlier,45 except for two additional constants a and 0 for
each microplane component. The optimal values of a and (3 can be obtained
by fitting a set of cyclic test data. Present experience indicates that a and (3
may be taken as the same for all concrete types.

A.5. Rate effect Experimental evidence indicates that concrete stiffness, strength and
ductility are rather sensitive to the deformation rate. This is known as the
rate effect. This effect is no doubt caused by creep in the bulk of test
specimen as well as time-dependent rupture of bonds in the fracture process
zone, which both cause stress relaxation. A simple rheological model,
UMIUII
- I1IIUIII
IIUUUI
consisting of one elastic spring coupled in series with a parallel coupling of
a damage unit and viscous dashpot (Fig. 23), is introduced for each
microplane component — volumetric, deviatoric and shear.49 It can be
shown that the viscosity of the dashpot must be highly stress-dependent. It
41 should be also noted that a similar rheological model for the rate effect in
damage (but without the stress-dependence of viscosity) has recently been
Fig. 23. Rheological model
introduced by de Borst.50 Dropping the subscripts V, D, M and K, the stress-
for each microplane
component strain relation for each microplane component may be written as

(53)

where e is the total microplane strain including the flow (creep) strain, the
upper dots denote the time derivatives, Ct and Co are material parameters.

A.6. Numerical In every iteration of the load or time step r, based on the kinematic
algorithm in each load constraint (equation (41)), the known macrostrains e//r and their known
step increments Ae/,r can be used to calculate the strains and strain increments
on each microplane. Furthermore, the known values of CN = ey + CD, CM,
CK, ACN = Aey + ACD, ACM and ACK, are used to calculate the stresses on

36
 CONCRETE IN COMPRESSION

each microplane by solving equation (53). Each of these equations can be

solved by using a central difference approximation. However, such an
approximation often appears unstable in the case of strain softening, and,
even if the computations remain stable, a large error is usually accumulated,
with the result that the stress is not reduced exactly to zero at very large
strain.
These drawbacks can be eliminated by the use of the so-called
exponential algorithm, initially developed for aging creep of concrete51
and later extended to creep with strain softening and applied by Bazant and
Ozbolt,46 and BaZant and Chern.52 Another possibility is to take advantage
of the secant formulation of the microplane model and from the known total
microplane strains (eei + e^a), using equation (44) directly calculate the
microplane stresses from equation (44) or more accurately using the
exponential algorithm. The creep deformations can be in that case directly
calculated from equation (53). In both cases the basic stress-strain curve for
each microplane must be followed.
Implementation of the foregoing numerical procedures as well as a
numerical algorithm employed in each load step of the finite element
analysis is described in more detail.46 Furthermore, the non-local concept
proposed by Pijaudier-Cabot and Bazant53 is also implemented in the
present cyclic response model. A more detailed description of the non-local
implementation is given in.46

A.7. Numerical The behaviour of the general microplane model is demonstrated on the
examples material level, using only one uniformly strained finite element, loaded in
three different ways — tensions, compression and shear. Cyclic behaviour
of the three-point bend and compression specimens in plane stress is also
simulated. It should be noted that the plane stress state is a relatively
complex state to simulate with the microplane model. The microplane
model is a full three-dimensional model and in plane stress finite elements
the lateral strains (out-of-plane strains) need to be calculated from the
condition that the lateral stresses are zero. The basic material parameters
used are: initial Young's modulus E = 20000MPa, Poisson's ratio // = 0-18.
The microplane material parameters are chosen as follows a = 0005, b =
0-043, p = 0-75, q = 200, «, = 000007, e2 = 0-0020, e3 = 00020, e4 = 0,m =
0-85, n = 2-25, k = 2-25. Most of these values are the same as in Bazant and
Prat,45 except m, n and k (which have been adjusted so that descending

2-5- 40-
Displacement control: Displacement control:
cte/df = infinite dt/df = infinite
dt/df = 005 s" 1 de/df = 0-1E-4/0-1E-4 (1/s)
2-0-
de/df = 0025 s"' 30-

20-

10-
0-5-

00002 00004 0002 0004 0-006

Strain Strain
(a) (b)
Fig. 24. Calculated concrete response under monotonic uniaxial tension: (a) and compression (b) with different
rates

37
RC ELEMENTS UNDER CYCLIC LOADING

Displacement control: Displacement control:

No rale effect 25- df/df = 0 0 5 s " '

1-5-

i
0-5-

-0-5-

Monotonic loading

-1-5
00002 00004 00002 00004
Strain Strain

Displacement control:
2-5- df/df = 0025 s '

Monotonic loading

00002 00004
Fig. 25. Calculated concrete behaviour in uniaxial
Strain cyclic tension with three different rates

stress-strain curves were more steep). The load was introduced by

prescribing the displacement increases with different rates.
Figure 24 shows the stress-strain curves in the cases of uniaxial mono-
tonic tension and compression, with different loading rates. Similar calcula-
tions, again using different rates of loading, are carried out in the case of
cyclic tension (Fig. 25), cyclic compression (Fig. 26), and cyclic shear
(Fig. 27). Fig. 28 shows comparison between the uniaxial cyclic test results
of Sinha et al.54 and the present calculations.

40
Displacement control: Displacement control:
no rate effect df/df = 1 0 s" 1

Monotonic loading

0002 0004 0006

strain
Strain
Fig. 26. Compressed concrete behaviour in uniaxial cyclic compression with two different rates

38
 CONCRETE IN COMPRESSION

16
Displacement control: Displacement control
no rate effect df/df = 1 0 s '

8-

w 0

55

-8"

-16
-0010 -0005 0 0005 0010 -0010 - 0 005 o 0005 0010
Strain Strain
Fig. 27. Calculated concrete behaviour in cyclic shear with two different rates

The present results indicate that the cyclic microplane model can
realistically predict the cyclic behaviour of concrete in different stress-
strain conditions using the same material parameters. The results shown are
only one among many possible material responses that can be modelled.
The microplane model parameters, the parameters that control loading and
unloading on each microplane in each direction as well as the rheological
model parameters are here kept constant. Varying these parameters as
material functions of the strain state and history it would be possible to
obtain a number of different kinds of responses.
Furthermore, the results of the analysis indicate that the rate effect has
significant influence on the stress-strain response. If the rate of loading
decreases, the peak stress also decreases while the post-peak descending
stress-strain curve becomes less steep. The present model can predict the
drop of stresses after repeated unloading-reloading cycles in post-peak
strain softening. As will be demonstrated later, this effect is significant
when simulating the structural behaviour.
Finally, to demonstrate the response of the present model in finite
element applications, the behaviour of the three-point bend (Fig. 29(a)) and
compression specimens (Fig. 29(b)) is analysed. Four-point isoparametric
quadrilateral plane stress finite elements with four integration points are
used. In both cases, symmetric response is assumed (Fig. 29). The material

Displacement control: dt/df = 12-5 s

x = Elem
(b)

Fig. 28. Comparison of cyclic uniaxial compression between: (a) test data (Sinha et al: ) and (b) calculated results

39
 RC ELEMENTS UNDER CYCLIC LOADING

F/2
0 Q
3| -

1 E

g
II

Thickness = 100 mm
1L
1/2 ,
1 ?i (i r. r
ft/2 = 150 mm j

Thickness = 1 0 mm
(a) (b)

Fig. 29. Geometry and finite element mesh for (a) three-point bend specimen, (b) compression specimen

parameters are the same as in previous examples except ei = 000004and m

= 0-5.
Figure 30(a) shows the load-displacement curve obtained for the three-
point bend specimen. The specimen is loaded using displacement control at
the upper nodes (see Fig. 29). The displacement rate u = 62-5 mm/s. One
cycle is performed prior to the peak load, and the next one deep in the
softening range. It is noted that after the second cycle the material is
severely damaged, so that in a further loading a drastic decrease of strength
is observed.
In the subsequent calculations of the same specimen, one finite element at
the bottom of the specimen was assumed to be weaker, having a tensile
strength limit 10% lower than the other elements (the shaded element in Fig.
29). The specimen was loaded up to 94% of failure load (obtained by
monotonic loading). This was followed by cyclic loading between 0 and
94% of peak load. The loading consisted of a force at the specimen top with
the rate of increase dF/dt = 5000 kN/s. The results indicate (Fig. 30(b)) a
significant increase of the displacements due to material damage. As a
consequence of the material damage due to cycling, failure is obtained after
the third cycle, slightly before reaching the monotonic descending branch
(see Fig. 30(b)).
Figure 31 shows the load-displacement cyclic response of a cubic
specimen, subjected to controlled displacement at the top of the specimen

40 32
Displacement control:
do/df = 62-5 mm/s
Fu = 25-46 kN
0-94 Fu
30- 24

16-

Monotonic loading, displacement control:

do/dt = 31-25mm/s
Cyclic loading, load control:
dF/df = 50000 kN/s

0-1 0-2 0-3 0-4 0 0-1 0-2 0-3 0-4

Displacement: mm Displacement: mm
(a) (b)
Fig. 30. Load-displacement curve of three-point bend specimen with cycling: (a) in post-peak strain-softening
range; (b) before reaching peak load

40
 CONCRETE IN COMPRESSION

Fig. 31. Load-displacement

curve for specimen with fixed Displacement control: du/df = 250 mm/s
loading platens loaded in
compression

0-125 0-250 0-375 0-500 0-625

Displacement: mm

with the rate u = 25 mm/s. To simulate bonded (non-sliding) rigid platens at

the top and bottom of the specimen, the horizontal displacements of the top
nodes in the finite element mesh are fixed. The calculated load-displace-
ment curve indicates large residual strains, and a significant decrease of the
concrete strength after reloading. These effects are here larger than those
obtained in the analysis using only one finite element.
The calculated hysteretic loops are wider when the rate effect is taken
into account and have approximately the correct area. The unloading begins
by a steep slope, in agreement with laboratory observations. For shear
cycling, the calculated hysteretic loops exhibit the characteristic pinched
form with a nearly zero stiffness at loading reversal, a fact which is also
known from experiments.35'55
Several other two- and three-dimensional finite element calculations not
shown here, 5657 confirm the model capabilities.

A.8. Conclusions The conclusions reached in this Appendix are listed below.
(a) The present general microplane model is capable of realistically
predicting the behaviour of plain concrete under a broad range of
strain states and histories using the same material parameters. The
model is fully three-dimensional. Together with the non-local strain
concept, the model can be effectively used in finite element
simulation of the failure process in the concrete structures under
rather general loading types and histories.
(b) The rate effect can be implemented in the microplane model by
combining damage with Maxwell's rheological model. The model is
able to give an approximately correct hysteretic loop area and a steep
initial unloading slope. For shear, it exhibits the pinched form of
hysteretic loops. Generally, increasing the rate increases the concrete
strength and the post-peak descending branch of the stress-strain
curve becomes steeper.
(c) The examples indicate that the cyclic microplane model, together
with rate effect, predicts material damage due to repeated unloading-
reloading cycles quite realistically.

41
 2. Concrete in tension
2.1. Introduction There is a substantial difference in the tensile stress-strain behaviour of
concrete and, for example, steel. To understand the behaviour of concrete,
this difference has to be examined. Steel, having passed the yield point!
shows a strain-hardening behaviour. Referred to the real cross-section of the
material, the strain-hardening behaviour lasts up to the ultimate load.
Concrete shows no — or almost no — strain hardening. Strain softening
starts immediately after the elastic limit is reached. The stress-strain
behaviour is governed by micro- and macrocracking within a certain crack
band.

2.2. The tensile The tensile behaviour of concrete is illustrated in Fig. 32. A deformation-
behaviour of concrete controlled centric tension test may be performed on the specimen shown on
the left of Fig. 32. The specimen is loaded with the force P. The total
deformation is measured over the length /. At the left side of Fig. 32 the
specimen is plotted for the load steps A, B and C. The load steps are also
marked in the load-deformation curve at the right side. Load step A is
before peak load, load step B at peak load and load step C after peak load in
the descending branch of the load-displacement curve.
Already before the peak load is reached, some microcracking occurs (Fig.
32(a)). As the microcracking is uniformly distributed at the macrolevel, a
uniform strain over the length of the specimen may be assumed. The strain e
is plotted over the length of the specimen in Fig. 32 right next to the
specimen.
Immediately before the peak load, an accumulation of microcracks occurs
at the weakest part of the specimen. At the macrolevel this leads to an
additional strain over the length h of this weak part. A crack band of width h
develops (see Fig. 32(b)).
Having passed the peak load, the crack band localizes more and more.
The crack band width diminishes, and the deformation within the crack
band increases. The final failure occurs due to one single crack.
The total deformation of the specimen may be split up in the bulk
deformation — which is almost linearly elastic up to the peak — and the
deformation of the crack band.

W = Al-el
p
Fig. 32. Tensile behaviour of *
concrete (a) (b) (c)

42
 CONCRETE IN TENSION

2.3. The fictitious Hillerborg58 introduced the fictitious crack model, where he collected the
crack model deformation of the crack band into the crack width w of one single
'fictitious' crack. The relation between the crack width of the 'fictitious'
crack and the stress is the a-w relation. Two mechanisms contribute to the
stress transfer over the crack. As mentioned, it is actually a fictitious crack:
the crack width is the collected deformation of a band of microcracks.
Within this crack band material bridges transfer the load. After the
formation of a real single crack, the stress transfer is possible due to
aggregate interlock. In most cases a crack will run along the interface
between the aggregate grains and the cement matrix. The grains are pulled
out of the matrix and due to this, friction forces between grains and matrix
occur. The grains act like friction blocks and transfer friction forces over the
crack.

2.4. Models for the The bulk behaviour is described by the a — e relation of Fig. 32(c). This
bulk behaviour relation is linear almost up to the ultimate load. The modulus of elasticity is
equal to the initial tangential modulus of elasticity in compression.
Poisson's ratio is in the order of v - 015-0-25. Just before the ultimate
load is reached, the a — e relation bends off from the linear behaviour. For
the state of loading where the non-linear behaviour starts, the available
experimental results show great differences. In some experiments the non-
linearity starts at about half of the ultimate load, in other experiments the
curve is linear up to the peak load. These differences in the experimental
results are caused by different boundary conditions. Any source of non-
uniformity, like internal bending due to non-uniform cracking, eigenstresses
due to differential shrinkage and temperature, notch effects, etc., causes
non-linearities. Theoretical considerations59 show that even when the
material behaves linearly elastic up to the ultimate load, a specimen may
behave non-linearly before the ultimate load is reached. The non-linear
behaviour of test specimens is more a structural behaviour than a material
behaviour. Because of this it seems acceptable to assume linear-elastic
material behaviour up to the ultimate load, for both the loading and
unloading part.

2.5. Models for the A large number of experimental results are available for the a — w relation
stress - crack width in the case of monotonic loading. Fig. 33 shows some suggestions for the
relation formulation of a — w relations.5 ~63
Except for the linear one, which is often chosen because it is very simple, all
a — w relationships show the same behaviour, qualitatively. The first branch
of the relation is very steep, whereas the second branch is more flat. Since most
practical applications are not very sensitive to the exact shape of the a — w
relation, the choice of any of them is mainly a matter of practical convenience.
Contrary to the monotonic loading, only few models are available for the
case of cyclic loading. Fig. 34 shows the models of Rots et al.,64 Gylltoft,65
Reinhardt et al.66 and Yankelevsky and Reinhardt.67 These models can be
compared with the experimental results from Reinhardt et al. shown in Fig.
35. In spite of some differences in details, the models presented in Fig. 34
show some common features. For the description of the hysteresis loop they
use polygonal shapes, which are independent of the shape of the envelope.
Rots et al., Gylltoft, and Yankelevsky and Reinhardt use a a — e relation
(crack band model), Reinhardt et al. use a stress-crack width relation
(fictitious crack model).
The simple model used by Rots et al. (and others) connects the point on
the envelope where unloading starts straight with the origin of the
coordinate system. Unloading and reloading run on the same line, i.e.
there is no hysteresis loop. After running back and forth the envelope

43
 RC ELEMENTS UNDER CYCLIC LOADING

Fig. 33. Suggestions for the

cr-w relation: (a) linear;
(b) bilinear (Petersson)60;
(c) multilinear
(Gustafssonf;
(d) Gqpalaratnam and
Shah ; (e) Cornelissen et
alM; {f) Duda59

function is valid again. This and similar models approximate the real
behaviour very roughly. They should be used only if cyclic effects have
negligible relevance in the overall behaviour.
Gylltoft uses a set of lines parallel to the initial slope of the curve to
describe the cyclic behaviour. The path O A B C D in Fig. 34 describes
monotonic loading. The microcracking is considered to start when the
tensile strength is reached at point A. If unloading is to be performed after
this stage, e.g. from point E, the gradient of the unloading path is chosen to
be equal to the linear elastic gradient also used for the initial loading path.

la-

Fig. 34. Material models

for cyclic loading:
(a) Rots et al; (b) Gylltoft;
(c) Reinhardt et al.;
(d) Yankelevsky and
Reinhardt

44
 CONCRETE IN TENSION

Fig. 35. Experimental results

3
of cyclic tension test

When zero stress is reached the remaining strain is denoted ef. When
changing over to compressive stresses, partial closing of the existing
microcracks occurs, resulting in the strain-gap FG. The size of the gap is
chosen to be linearly proportional to ef(FG = gcet)- The parameter gc is
considered to be governed by the properties of the concrete. The microcrack
will not close completely. This constraint can comprise effects due to
imperfect matching of irregular crack surfaces, particles within the
microcracks which have come loose, and irreversible deformations. The
path GH in compression is parallel to path OA.
When reloading into tensile stress takes place, a partial opening of the
existing microcracks occurs, resulting in a strain-gap GF' = gogcei; go is
considered to be a material parameter. Being back in a state of tension
again, the path F'E' is parallel to path OA. When approaching the virgin
curve once more, the remaining part of the stress-strain curve is lowered in
order not to exceed the stress level at which the virgin curve was left earlier.
Point E' and E are situated on the same stress level. After this it is proposed
that path E'C should be followed until ultimate fracture occurs. Path E'C' is
parallel to path EC.
The model of Gylltoft is a better approximation to the real material
behaviour than the first simple model, but it must be pointed out here that
the shifting of the envelope line EC to the line E'C contradicts experimental
results. After running through the hysteresis loop, the load-displacement
curve of cyclic tension tests reaches the envelope curve of monotonic
loading again.
The model of Reinhardt et al.66 uses the a-6 envelope curve of equation
(54) suggested by Cornelissen et a/.63 which was already shown in Fig. 33

M") =fJ [l - (54)

C, = 3
C 2 = 6-93
wc = 160/zm
w =6
Loading and unloading steps are modelled as follows.
(a) After the peak value, the stress follows the envelope curve up to a\,
where unloading starts.
(b) Since the subsequent path is partly determined by the reloading
displacement £3, first 03 and £3 have to be calculated. The stress

45
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 36. Stress drop during 0-16

loading cycles

drop, ACT, can be obtained from Fig. 36. The diagram presents three
lines obtained from different experiments. Type b represents tensile/
tensile loading with the lower stress almost zero. Type c and d belong
to loadings where the lower stress is compressive —015/ct or —/ct,
respectively. The stress drop is dependent on <J\,8\ and the lower
stress oi- The stress 03 has to be calculated as 03 = o\ — ACT. The
crack opening 63 must be calculated fromCT3and equation (54) by an
approximate procedure.
(c) Unloading takes place at constant <5| until line A is reached. Further
unloading follows the lines A, B and C until equilibrium stress 01 and
crack opening 82 are reached. Line A is given by equation (55) and
line B by equation (56).

-=1f--0 (55)
/ct 6 \S3 )
CT
- 1 (56)
It 0-3&

Line C assumes an irreversible crack opening 8\ at increasing

compressive stress. For normal concrete a close fit to experimental
data is given by
8\ = 0-35.50'66 (57)
(d) Reloading starts from 02 at constant 8\ until the reloading stiffness
C 3 (line connecting the origin of the coordinate system with the point
(63,CT3)is reached at <rr). This stiffness is followed up to 03, after
which the envelope curve is followed.

Reinhardt et al.'s66 model, as described above, is a good description of

the hysteresis behaviour of concrete under cyclic tension. The only dis-
advantage is the necessity to calculate the stress drop ACT at the beginning of
unloading. ACT depends on the lower stress. In most normal applications and
in finite element analyses, in particular, it is impossible to know the lower

46
 CONCRETE IN TENSION

stress which will be reached in a hysteresis loop at the beginning of this

loop.
The model of Yankelevsky and Reinhardt67 assumes a unique envelope in
tension and focuses on unloading from and reloading to that envelope which
is assumed to be given. In the uniaxial stress-deformation plane several
focal points are defined. The focal points 0 and Z]-Z 6 are placed along the
tangent to the envelope curve at the coordinates' origin (see Fig. 34). The
auxiliary points A-M are found at the intersections of auxiliary lines con-
necting the focal point or the auxiliary points, respectively. The construction
of the polygonal paths over the auxiliary point has been detailed.67
The model of Yankelevsky and Reinhardt is a very detailed phenomeno-
logical description of the hysteresis behaviour of concrete. Due to the large
number of points and auxiliary lines the model is very complex. It will be
hard to fit this model in a finite element program. A weak point of the model
is that the stress drop between the auxiliary points A and M — the
beginning and the end of the hysteresis loop — do not depend on the lower
stress. This is not really true as shown in Fig. 36.
The four models mentioned can be classified as 'phenomenologicaF. The
material behaviour is determined by test and described by curve fitting; first
for the envelope and then for the hysteresis loops. Curve fitting describes
only the material behaviour but does not describe the physical background.
A new model, which can be classified as 'semi-physical', because it makes
use of mechanical elements such as blocks and springs, has been recently
proposed.59'68 Section 2.6 will be dedicated to a detailed presentation of this
model.
Another new model, the continuous function model, was recently
introduced by Hordijk and Reinhardt.69 This model will be introduced in
section 2.7. Whereas the basic idea and the numerical treatment of the
models of sections 2.6 and 2.7 differ very much, the outcome in terms of the
cyclic a — w relation is almost identical; once again, the choice between
these two models is a matter of practical convenience.

2.6. Rheological 2.6.1. Fundamentals

material model for the Transfer of stresses over a crack is possible due to the friction forces acting
a- w relation" between grains and matrix. Considering every single grain to be a friction
block, a concrete cross-section in which a crack is developing can be
described as a parallel arrangement of many friction blocks. To simplify
matters, it is assumed that each friction block / is working as long as the
crack width w is smaller than its ultimate crack width w,imax. With 3>(w) =
(complementary) distribution function of ultimate crack widths, this leads
to:
n = no<£(w)
where n = number of working friction blocks
n0 = number of working friction blocks for zero crack width
A given concrete area A is able to transfer the force P(w = 0) =/ ct A for
zero crack width. For a crack width w > 0 the transferable load P(w) is
proportional to the number of friction blocks which are still working for the
actual crack width. For the distribution function 4>(w) any monotonically
decreasing function with <&(w = 0) = 1 and $(w —• oo) = 0 can be chosen.
This leads to

where / ct = tensile strength of concrete

0ct(w) = tensile stress of concrete in the crack

47
 RC ELEMENTS UNDER CYCLIC LOADING

Fig. 37. Material model I I I-

°a.O —

2.6.2. Model for monotonically increasing crack width

In the first stage of the a — w relation, a steep decrease of the transferable
stress occurs. When the stress decreases to about one-third of / ct , the
following decrease is not as steep. The first branch may be interpreted as
successive failure of the primary bearing mechanism, that is, the adhesion
between grain and matrix. The second branch may be interpreted as
secondary bearing mechanism, that is, friction between grain and matrix
after cracking.
These two load transferring mechanisms can be described by two groups
of friction blocks, which are arranged in parallel as shown in Fig. 37.
Friction block A symbolizes the na friction blocks of the first group, friction
block B the nb friction blocks of the second group.
Group A describes the primary bearing mechanism, group B the
secondary one.
The transferred stresses for w = 0 will be designated with crao or
respectively.

The distribution function 3>a(w) must describe the rapid failure of the
primary bearing mechanism. The following formulation was chosen

The slower failure of the elements of group B is described by

$ b (w) = e~w/wb
One may therefore write

'w» (58)

The fitting of the function given in equation (58) to experimental results

is easily possible. In Fig. 38 the procedure is shown using experimental
results from Cornelissen et al.63 The sum resulting from <ra]o and crb0 is
equal to the tensile strength of concrete / ct ; the term with o^o controls the
first steep portion of the curve, the one with cr^o the flatter portion. Crack
widths wa and wb are defined as the values for which the contributions
^ ( w ) and <Tb(w) to (Ta.fi and <7b,o are reduced by a factor Me.

2.6.3. Model for cyclic loading

The material model introduced above can be extended to cyclic loading. A
material model is applicable for cyclic loading only if it is able to describe
all kinds of different deformation histories (lower stress = 0, > 0, < 0 or
variable) in a sensible way.

2.6.4. Description of hysteresis loops with rheological elements

Hysteretic material behaviour can be described by the combination of two

48
 CONCRETE IN TENSION

100
Fig. 38. cr-w relation fitted to '
experimental results Experimental
Material model

0-75-

0-50- <

0-25-

004 I 008 0-12 0-16

I w: mm
I

friction blocks and two springs as shown in Fig. 39. The symbols used are
defined as follows.
F\ = first spring
F2 = second spring
Ri = first friction block
R2 = second friction block
w = total displacement (= crack width)
M>\ = displacement of friction block Rx
w2 = displacement of friction block R2
P(w) = total load (= load in F{)
P\ = transferable friction force in Rx
P2 = transferable friction force in R2
K\ = spring stiffness of F\
K2 = spring stiffness of F2
If the mechanism shown in Fig. 39 is forced to a certain displacement, a
relation between load and deformation of the type shown in Fig. 40 is
obtained.

1-0-1

I hwwvH h/vm P

— - w2 — - w, w

Fig. 39. Hysteresis model Fig. 40. Hysteresis model behaviour

49
RC ELEMENTS UNDER CYCLIC LOADING

The load P remains constant until point 2, when the direction of

deformation reverses. Between point 1 and point 2 the following equations
are valid
P =Pi+Pi

Wl = Wx--± (59)
K2
Since a crack does not open until the tensile strength of concrete is reached,
the mechanism must be able to transfer the maximum load for w = 0. This is
achieved by adjusting the displacements wx and w2 at point 1 in such a way
that both friction blocks R\ and R2 transfer exactly the maximum friction
force P\ or P2, respectively. At point 1 the following equations are valid

P =Pi+P2

w =0

L.

w2 = w , - ^ (60)
K2
After reversal of the deformation direction at point 2, spring Fi is
compressed until the friction force at /?i is equal to —Pi. When this occurs,
point 3 is reached. Between point 2 and point 3 both friction blocks are
motionless. Between point 2 and point 3 the following equations are valid
/.,, ... \ is-
— \W — W\)l\.\

W\ = W,

w2 = w\ (61)
where w\ - displacement W\ at point 2
w\ = displacement w2 at point 2
From point 3 on, R { slides in a negative direction. The stiffness of spring
F2 is now adjusted in a way that on achievement of a crack width w = 0, the
point 4* on the load axis will be reached. Point 4* has the ordinate — 2P\.
This leads to
p(w = 0) = - 2 P ,
w =0
from which

where wmax = maximum value of w ever reached in the deformation history.

Between point 3 and point 4 the displacement w{ results from equilibrium
conditions of friction block Ru block R2 is motionless. Between point 3 and
point 4 the following equations are valid

50
 CONCRETE IN TENSION

P =(w-wl)Kl
w2K2
Kl+K2
W =
2
At point 4 the deformation direction is changed again. Spring Fx is now
elongated until friction block R\ starts sliding in a positive direction at point
5. Both friction blocks Rx and R2 are motionless up to point 5. Between
point 4 and point 5 the following equations are valid
P ={w-w\)K\
W\ =w\

(64)
From point 5 on, R\ slides in a positive direction until point 6 is reached,
which is the starting point 2 of the hysteresis loop. The displacement wx
again results from equilibrium conditions of block R{. Between point 5 and
point 6 the following equations are valid

p = (w- wx)Kx
W2K2 + wKi - />,
K{ + K 2
1

w2 (65)
From point 6 on, both friction blocks slide in a positive direction. Now the
same relations as between point 1 and point 2 are valid again. If the
deformation direction will be changed once more, another hysteresis loop
will result. From the definition of the spring stiffness K2 in equation (62),
point 4* with the coordinates (w = 0, P = —2P\) will be reached in every
hysteresis loop, if the deformation w = 0 is achieved.

2.6.5. Extension of the material model for cyclic loading

The arrangement of the elements discussed in the previous paragraph and
illustrated in Fig. 39 has been shown to provide a satisfactory representation
of an hysteretic type of behaviour. In order to reproduce the full range of
behaviour, the model introduced in section 2.6.2 will be extended as shown
in Fig. 41. Each of the friction blocks of the element groups A and B will be
replaced by a combination of two friction blocks and two springs.
Additionally, element group C will be introduced.
2.6.5.1. Element groups A and B on cyclic loading. Equations (60) to
(65) describe the reaction of a mechanism consisting of the series
arrangement of two springs and two friction blocks. The mechanisms A
and B shown in Fig. 41 symbolize the na or n\, respectively, and consist of
mechanisms arranged parallel to each other of the same type.
According to the explanations in the earlier section 2.6.1, the behaviour
of an element group can be described by combining equations (60) to (65)
with the distribution functions. The behaviour of the element groups A and
B is shown in Fig. 42.
The same distribution functions $ a (w) and $\>(w) as used in Chapter 3
are adopted. The basic purpose of the distribution functions is to indicate
how many of the originally existing elements are working at the actual crack

51
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 41. Material model for

cyclic loading
\ I I-VW\AV| i-vww-—I

"b2,0 \ o b1.0 \

<\ KMMM I-VWW-

width. With increasing opening of the crack, the elements fail and will not
be reactivated when the crack is closed again. The distribution functions
must, therefore, refer to the maximum crack width (= wmax) ever reached in
the deformation history. The following equations are proposed for the two
functions <&a(w) and
_ , x /'"'max

$ a ( w ) = e v wa

= e (66)

- 1 for negative deformation direction

w
/w max x
( 1 ) C2 for positive deformation direction
\ w J
2.6.5.2. Element group C. A crack, once having opened, cannot be
closed again. During the opening of the crack, small particles separate from
the crack surface. These particles dislocate while the crack is open and
prevent the surfaces from fitting into each other again without constraint.
This behaviour of the material is modelled by element group C. Element
group C is only active within the hysteresis loop. The number of active
elements of element group C is defined by distribution function $c(w). For
<£c(w) a hyperbolic function is chosen

and therefore
max
—ac 0 ( — 1) for negative deformation direction
<Tc(w) = ' \ w )
max
—<*c 0 ( — 1) C2 for positive deformation direction
' V w /
(68)

Group A
0-4

0-2
Group B

30 10 20 30
w.ftm
Fig. 42. Hysteresis loop of -0-2J
element groups A and B

52
 CONCRETE IN TENSION

Fig. 43. Hysteresis loop of

element group C

± -0-4-

-0-8-

The behaviour of element group C is described by equation (68) and is

plotted in Fig. 43. The constant C2 influences the areea within the hysteresis
loop of element group C. Only values between 0 < C2 < 1 are reasonable.
In the case of C2 = 1, the curves for positive and negative deformation
directions are the same, hence, the area of the hysteresis loop is zero. For C2
= 0 the reloading branch of element group C is a vertical line from point 4
(Fig. 43) to the w-axis. In this case the area and, therefore, the energy
dissipation within the hystereris loop is at its maximum.
2.6.5.3. Total behaviour of the material model. The behaviour of the
total material model results from the addition of the stresses of the three
element groups A, B and C. In Fig. 44 this addition was carried out for the
hysteresis loops plotted in Figs 42 and 43.

2.6.6. Comparison between experimental results and model responses

The best way to compare the material model and experiments is through
axial tension tests. The values measured include the crack width and the
bulk deformation. The bulk deformation can be determined with sufficient
accuracy by assuming linear-elastic behaviour. Figs 45 and 46 show the
experimental results from Reinhardt et al.66 on the left side and the results
from the material model on the right side. In order to adapt the model to the
experimental results, the parameters o-a,o, 0"b,O) w& and wt,, which describe
the envelope of the curve, have been read from the experimental curve as
described in section 2.6.2, Fig. 38.

30

Fig. 44. Hysteresis loop of -1.2-1

material model

53
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 45. Comparison of

(a) experiment and
(b) material model,
minimum stress « —/ct

(b)

2.6.7. Final remarks

The 'rheological' model presented for the uniaxial behaviour of concrete
under tension appears to possess a number of favourable aspects, as
compared to the empirical models previously described. Some of these are
listed below.
• The material model is based on a mechanically clear and
reproducible conception.
• The envelope and the hysteresis loops are described by the same
model. It is not necessary,67 to work with a totally different definition
for the envelope and the hysteresis loops.
• The model only works with information which is available at the
actual state of loading.
• The model can easily be adapted for other concrete strengths or
mixtures. It is applicable to normal, light and high strength concrete.
• The relation between stress and crack width is described by simple
analytic relations. Because of this the model can easily be used
together with finite element programs. The model has been imple-
mented into the finite element program SNAP. 66 The analytical
simulation of available test results has provided satisfactory
conclusions.
• The a — w relation is integrable. The fracture energy GF is, therefore,
easy to calculate.

2.7. Continuous The model gives a description for the stress-crack-opening relation, while
function model crack opening is defined according to the fictitious crack model by
Hillerborg et al.5S It consists basically of continuous functions and is,
therefore, called the 'continuous function model'.

2.7.1. Loops starting from and returning to the envelope curve

The three basic equations are, respectively, empirical expressions for the

Fig. 46. Comparison of

(a) experiment and
(b) material model,
minimum stress « - / c t / 1 0

54
 CONCRETE IN TENSION

Fig. 47. Set-up for the

continuous function model
— (II)

Aa

eu: envelope unloading

er: envelope reloading
inc: increase

unloading curve (I), the gap in the envelope curve (II), and the reloading
curve (III) (see Fig. 47). It has been decided to use only characteristic points
in the a — w relation (/",, wc, weu, creu, 0L) as variables in the expressions.
The parameter wc is the crack opening where stress can no longer be
transferred and is defined as 5,14GF//t. The expressions chosen are based on
a close inspection of the experimental results. The softening relation is
described by the following equation

/ UA31 / W\ W 3
(69)
(ci —] expl-c 2 —I ( 1 + q ) exp(-c2)

with c\ =3,c2 = 6-93 and wc = 5,14GF//t.

Starting from point (weu, o~eu) at the envelope curve, the unloading curve
(I) is determined by

0-014 fa - - -0-57, 1 —

(70)
When reloading starts from a lower stress level CTL, the gap in the envelope
curve (II) is known by the expression for winc
Win
• = 0-l^i fa 1+3 (71)

The coordinates of the returning point at the envelope curve (wer, <rei) can
now be found with

= Weu + (72)
and equation (69). The reloading curve (III), starting from the point at the
lower stress level (H>L, <TL) up to point (wer, o~eT) at the envelope curve, is
determined by

1 / w - wh w-wL
1 - 1 -
w fir -

(73)
C 3 +l

55
RC ELEMENTS UNDER CYCLIC LOADING

with the coefficients c 3 and c4

0-71/ti

£3 = 3 3 1 - I—
ft
-1

c4 = 2 3, / t - ' + 0-5
/t

2.7.2. Inner loops

The above expressions describe the behaviour of a loop starting from and
returning to the envelope curve. Here, a proposal will also be given for the
behaviour when the deformation direction reverses within such a loop. For
the description of these so-called 'inner' loops, a counter i is used. This
counter is assumed to be zero for the envelope curve and is increased by 1
each time there is a reversal in the crack opening direction before the crack
opening at the previous reversal is reached (see Fig. 48).
The essential for inner loops in this model is that they return to the same
point from where they started. This means that a damaging effect due to
cycling within a loop, starting from and returning to the envelope curve, is
not taken into account. As far as notation is concerned, At is used for a point
where the direction changes from an opening crack into a closing crack and
L, is used for the opposite reversal in crack opening direction. The meaning
of the other variables used can be obtained from Fig. 48. The unloading
curves are determined by

(75)

= 0

Envelope
— . ^ curve

Fig. 48. Continuous function

model — procedure for inner
loops

56
 CONCRETE IN TENSION

Fig. 49. Model prediction for

arbitrary chosen loading
path

to
Q.

S
55

-1- I, = 3 MPa
G F = 100 J/m2
n-c = 171 10 6 m

-2-
10 15 20 25
6
Crack opening w: 10 m

and the reloading curves can be found with

= a r ,K,) - at, - Sl(^Li WA

'"i ) AaL,,_, (76)

(77)

After completion of an inner loop, the counter / decreases by 1. Special

attention has to be given to stress-crack-opening reversals at crack
openings corresponding to the gap in the envelope curve. For that case,
point A i (see insert in Fig. 48) must be determined by means of an iterative
procedure.

2.7.3. Model predictions and concluding remarks

Comparisons between experimental results and model predictions can be
found in Reference 71. There, the predictions have also been compared with
those by the focal point model by Yankelevsky and Reinhardt.67 Here, only
the result of a prediction for an arbitrarily chosen loading path is shown in
Fig. 49.
It can be concluded that the model gives an appropriate and complete
description for the cyclic behaviour of cracked plain concrete. It contributes
to improve the description of concrete in finite element codes and other
numerical programs.

57
3. Reinforcing steel

This Chapter presents a brief review of reinforcing steel models which can
be used in the finite element analysis of reinforced concrete structures
which are subjected to large alternate actions. Since a short summary of
models proposed before 1983 was presented in Bulletin d'Information
161,72 the present report mostly addresses models published since then and
gives only brief evaluations about earlier models.

3.1. Introduction Numerous models of the hysteretic stress-strain behaviour have been
published.73"77 Most of these are phenomenological in that they attempt to
describe the behaviour by parameters which cannot be directly derived from
the physical behaviour at the microscopic level. Even though such physical
models have been published, the large computational effort associated with
their use in the finite element analysis of large reinforced concrete structures
excludes them from consideration in the present report.
It is of interest to note that most analytical models proposed to date have
attempted to address the hysteretic behaviour of steel in general, rather than
reinforcing steel in particular. The range of cyclic strain history, to which
reinforcing bars are likely to be subjected, differs significantly from that of
structural steel members in that compressive strains are not as large as
tensile strains. This is caused by an interplay of bond deterioration and crack
closure, which prevents reinforcing bars from excessive yielding in com-
pression before spalling of concrete cover has occurred. As long as a section
is uncracked or the crack previously formed is closed and the concrete cover
has not spalled, compressive forces are largely carried by concrete, while, if
the crack is open, bond deterioration in the vicinity of the crack prevents
any large force build-up in reinforcing bars. This results in small negative
steel strain increments and consequently in small compressive steel stress
increments. In either case bars under compression hardly ever enter into the
inelastic compressive strain range before the concrete cover has spalled.
Thus cyclic strain histories of reinforcing bars tend to centre about an ever-
increasing tensile plastic strain. This fact can be used in the simplification of
the proposed models of the hysteretic behaviour of reinforcing bars.
Another important consideration in the search for an appropriate model is
its numerical efficiency. Many evaluations of the steel stress—strain relation
will be required in the non-linear finite element analysis of large reinforced
concrete structures. This suggests that the model should be as simple as
possible. Counteracting the need for simplicity is the requirement of
accuracy, which is based on the observation that cracks running through the
depth of a RC member can remain open during load reversals, causing the
hysteretic response of the member to be entirely controlled by the behaviour
of reinforcing steel. In particular, the Bauschinger effect has to be accounted
for. Even though multilinear models have been proposed in the past, it
seems that simple non-linear models have a clear advantage. In any case, the
opposing requirements of numerical efficiency, i.e. simplicity on the one
hand and accuracy on the other, call for a compromise between the two
extremes.
Some of the characteristic features of the hysteretic behaviour of
reinforcing steel are shown in Fig. 50. To retain accuracy the proposed steel
models should account for the non-linear monotonic envelope, in particular,
the onset of strain hardening, the Bauschinger effect and the isotropic strain
hardening under plastic strain reversals. Several non-linear analytical
models have been proposed to date. To a large extent they are devised to

58
 REINFORCING STEEL

Fig. 50. Features of Strain hardening branch

hysteretic steel stress-strain
behaviour of reinforcing steel

Increase in strength
beyond initial yield stress
after reversal:
isotropic strain hardening

Early departure from

linear elastic response
' after reversal:
/ Bauschinger effect

Degradation in steel
modulus after reversal:
cyclic strain softening

represent the cyclic behaviour of structural steel thus failing to take

advantage of the special strain history of reinforcing steel in reducing the
complexity of the model. The most successful among them fall in the group
of variable parameter models, with the parameters of the model varying
with strain history. Within this group one can distinguish four types of
models

(a) an explicit algebraic equation for stress in the form78"82

a = a(e) (78)
(b) an explicit algebraic equation for strain in the form77'83'84
e = e(a) (79)
(c) an implicit algebraic equation of the form73'76
f(e, a)=0 (80)
(d) a first order differential equation of the form74'85

— = Ep = f(a) (81)

where ep is the plastic strain, e denotes the steel strain, a denotes the steel
stress and B? denotes the plastic modulus. Models in categories (a) and (b)
are the most popular to date, because of their simplicity. The most
successful in category (a) is the model proposed by Giuffre, Menegotto and
Pinto, while in category (b) most models are based on modifications of the
model originally proposed by Ramberg and Osgood. There appears to be
some advantage in using formulation (a) in finite element analysis. This
advantage stems from the fact that, usually, strains are used as primary
variables in section analyses or are derived first from strain-displacement
relations. In this context it is advantageous not to have to solve a non-linear
equation of type (b), but to be able to directly determine the steel stresses
from the strains. Because of this fact, attention is focused in this report on
two particular models which follow the formulation (a). The first model is a
modification of the original model of Giuffre, Menegotto and Pinto. 8081

59
 RC ELEMENTS UNDER CYCLIC LOADING

The second is a new simplified model which follows both formulations (a)
and (b) simultaneously and thus can be used with great efficiency in
explicitly calculating the steel stress from a given strain (formulation (a)) or
in explicitly calculating the steel strain from a given stress (formulation (b)).
The latter problem occurs often in connection with the solution of the stress
transfer problem based on bond stress-slip relations.

3.2. Giuffre - The original model of Giuffre, Menegotto and Pinto takes on the form
Menegotto - Pinto
model

where
e - er
€ =• (83)

and
a—
a = (84)
aa — ar
Equation 82 represents a curved transition from a straight line asymptote
with slope EQ to another asympote with slope Ex (lines (a) and (b),
respectively, in Fig. 51). &0 and e0 are stress and strain at the point where the
two asymptotes of the branch under consideration meet (point A in Fig. 51);
similarly, aT and er are stress and strain at the point where the last strain
reversal with stress of equal sign took place (point B in Fig. 51); b is the
strain-hardening ratio, that is the ratio between slope E\ and EQ and R is a
parameter which influences the shape of the transition curve and allows a
good representation of the Bauschinger effect. As indicated in Fig. 51
(eo, cr0) and (er, at) are updated after each strain reversal.
R is considered dependent on the strain difference between the current
asymptote intersection point (point A in Fig. 52) and the previous load
reversal point with maximum or minimum strain depending on whether the
corresponding steel stress is positive or negative (point B in Fig. 52). The
expression for R takes the form

600

a.

Fig. 51. Parameters of -600

-20
Giuffre, Menegotto and Pinto
Steel strain c.: mm/m
model

60
 REINFORCING STEEL

Fig. 52. Definition of

curvature parameter R(£) in
Giuffre, Menegotto and Pinto
model

-10 10
Normalized steel strain t'

(85)

where £ is updated following a strain reversal (Fig. 52). Ro is the value of

the parameter R during first loading and a\, a2 are experimentally
determined parameters to be defined together with Ro. The definition of £
remains valid in case that reloading occurs after partial unloading.
If the analytical model had a memory extending over all previous
branches of the stress-strain history, it would follow the previous loading
branch as soon as the new reloading curve reached it (curve (a) in Fig. 53).
This would require that the model store all necesary information to retrace
all previous reloading curves which were left incomplete. This is clearly
impractical from a computational standpoint. Memory of the past stress-
strain history is therefore limited to a predefined number of controlling
curves, which in the present model include
(a) the monotonic envelope
(b) the ascending upper branch curve originating at the reversal point
with smallest e value

Fig. 53. Reloading curve

after partial unloading: (a)
theoretically correct
behaviour; (b) behaviour of (a)
the model with limited
memory

61
RC ELEMENTS UNDER CYCLIC LOADING

(c) the descending lower branch curve originating at the reversal point
with largest e value
(d) the current curve originating at the most recent reversal point.
Due to the above restrictions, reloading after partial unloading takes place
along curve (b) rather than curve (a) as shown in Fig. 53. However, the
discrepancy between the adopted analytical model and the actual behaviour
seems acceptable, since the case depicted in Fig. 53 is the most un-
favourable one.
The above implementation of the model corresponds to its simplest form:
elastic and yield asymptotes are assumed to be straight lines, the position of
the limiting asymptotes corresponding to the yield surface is assumed to be
fixed at all times and the slope Eo remains constant (Fig. 51). More complex
asymptotes have been proposed for the original model by Stanton and
McNiven.82
In spite of the simplicity in formulation, the model is capable of
reproducing well experimental results with strain histories typical of
structural steel, i.e. strain excursions of equal magnitude in tension and
compression. Its major drawback stems from its failure to allow for
isotropic strain hardening. This fact can be of importance when modelling
cyclic behaviour of reinforcing bars in RC members.
The model proposed by Filippou et al.79 retains the bilinear asymptotes of
the original model in the interest of simplicity. It introduces, however, a
shift of these asymptotes to account for isotropic strain hardening. This is
accomplished by shifting the position of the yield asymptote before
computing the new asymptote intersection point following a strain reversal.
The shift is effected by moving the initial yield asymptote through a stress
shift <rst parallel to its direction. This idea was firtst introduced by Stanton
and McNiven. The imposed stress shift of the yield asymptotes ideally
depends on several parameters of strain history. Stanton and McNiven have
suggested that the main parameter is the sum of the absolute values of
plastic strains up to the most recent strain reversal. Filippou et al. decided to
choose the maximum plastic strain as the main parameter on which the yield
asymptote shift depends. The proposed relation takes the form

= Q!3 Oi4 (86)

where emax is the absolute maximum strain at the instant of strain reversal,
ey, (Ty are, respectively, strain and stress at yield, and 0:3 and 0:4 are
experimentally determined parameters. It is interesting to note that in a
recent study it has been found that the model of Giuffre, Menegotto and
Pinto as modified by Filippou et al. is even capable of accurately
representing the hysteretic behaviour of steel members.
Comparison of analytical with experimental results obtained from cyclic
tests on reinforcing bars84 is shown in Fig. 54. Fig. 54(b) depicts a stress-
strain history typical of top reinforcing bars where yielding in compression
is very limited, while Fig. 54(a) shows a stress-strain history typical of
bottom reinforcing bars. The accuracy achieved with the model which
includes isotropic strain hardening is seen to be very satisfactory. Since no
isotropic strain hardening of the tension envelope was introduced in the
model of Fig. 54(a) the discrepancy between analytical and experimental
results is quite significant. This problem can be solved by also allowing for
istropic strain hardening of the tension envelope. It should be noted that the
model which includes isotropic strain hardening is almost as simple and
computationally efficient as the one which does not allow for isotropic
strain hardening.

62
 REINFORCING STEEL

100

50

-50
Analytical
Experimental
-100
-001 001 002 003 004 005 -001 001 002 003 004 005
Steel strain: in/in Steel strain: in/in
(a) (b)

Fig. 54. Comparison of model with experimental results: (a) bottom reinforcing bar; (b) top reinforcing bar

An even simpler model has been recently proposed by Zulfiqar and

Filippou.86 While the computational advantage of this model over the
modified model of Giuffre, Menegotto and Pinto is only minor, its main
feature is the ability to express both stress as an explicit function of strain,
and strain as an explicit function of stress. This makes it quite useful in
applications where both types of formulation are needed. One such
application arises in connection with models of beam-column joints.

3.3. Simple non- The hysteretic behaviour of the model is described by monotonic envelope
linear stress-strain curves in tension and compression (Fig. 55), a load reversal curve for
model unloading and reloading in the opposite direction and a set of rules for
general strain or stress histories. The ingredients of the proposed model are
presented in detail below.

3.3.1. Behaviour under monotonic loading

The proposed steel stress-strain relation under monotonic loading consists
of three regions: a linear elastic region with slope equal to the initial elastic
modulus, a plastic yield plateau with very small slope and a non-linear
strain-hardening range (Fig. 55).

Fig. 55. Parameters of

monotonic envelope curve

63
RC ELEMENTS UNDER CYCLIC LOADING

Linear functions are used to describe the behaviour of the model in the
linear elastic and the yield plateau range, while two options are available in
the strain-hardening region. In the first option the strain-hardening behav-
iour is described by a non-linear exponential equation. This equation is
selected so that the steel stress can be expressed as an explicit function of
strain and vice versa. Since the same equation is also adopted for the load
reversal curve of the model, further discussion is deferred until the
description of the load reversal curve. In the second option the strain-
hardening behaviour is approximated by a linear function.
The trilinear monotonic envelope is defined by the following parameters:
the yield stress ay\, the initial elastic slope Eo, the slope E{ of the plastic
yield plateau, which is always assigned a very small value, the secant slope
E2 in the strain-hardening range and the strain at the onset of strain
hardening esh. These parameters are determined from experimental results
of monotonic tension tests on reinforcing bars.
The yield strain ey and the yield stress at the onset of strain hardening ay2
are determined from these parameters

= ~- and ay2 = ayl + (esh - ey)E{

The steel stress can be expressed as a function of strain using the following
relations
a = EQ€ for e < ey (87)
a = <7yl + (e — ey)E\ for ey < e < ey2 (88)
G = ayl + (esh - ey)Ei + (e - esh)E2 for e > esh (89)
Equations (87)-(89) can be inverted to obtain the strain as a function of stress

f o r
= 7T cr<^yi ( 90 )
E
e = (^-^-) + ey for <ryi < a < ay2 (91)

-(esh-ey)^- + esh for a > ay2 (92)

t2 t2
Since it is desirable to have a one-to-one correspondence between steel
stress and strain in order to be able to solve equation (88) in terms of the
unknown strain (equation (91)), the necessity of using a small value for
slope E{ becomes clear. The same parameters are used in describing the
monotonic behaviour of reinforcing steel in compression.
Experiments show that under cyclic load reversals the yield strength
increases beyond the initial yield value under monotonic loading. This
phenomenon of isotropic strain hardening is accounted for in the proposed
model by shifting the monotonic tension and compression envelope curves
after each strain reversal. The amount of shift in stress Acrst and the
corresponding shift in strain Ae st of the tensile envelope curve are
determined from equations (93) and (94) respectively
- (ht - \)<ryl (93)
Aest = (ht-l)ey (94)
ht is given by the relation proposed by Filippou et al.79

l.O (95)

64
 REINFORCING STEEL

where emax is the absolute value of the maximum previous compressive

strain at the instant of strain reversal, and ay\ and ey are the yield stress and
strain, respectively.
Similarly, the amount of shift in stress A<TSC and the corresponding shift
in strain Aesc of the compression envelope curve are determined from
equations (96) and (97), respectively

= ~{hc - (96)
Ae.* = -(Ac - (97)

where

(98)

and emax is the maximum previous tensile strain at the instant of strain
reversal. The constants au a2, a3 and a 4 in equations (95) and (98) are
determined from cyclic loading tests of reinforcing bars.

3.3.2. Behaviour under load reversals

To satisfy the requirement that the steel stress-strain relation can be
expressed as an explicit function of either stress or strain the following non-
linear equation describes the behaviour of the model under load reversals
(Fig. 56)
= a - be~Xe' for 0 < e* < 1 and 0 < a* < 1 (99)
Equation (99) also permits that the steel strain is formulated as an explicit
function of stress. This relation takes the form
1, (100)
e= ln
-x
where e* and a* are normalized values of stress and strain which are defined
as follows

a = and e* =

S(0,0)

Fig. 56. Load reversal curve

in normalized coordinates

65
RC ELEMENTS UNDER CYCLIC LOADING

Point 5 is the last point of load reversal and point E is the point with
maximum previous strain in the same direction of loading as the current
load reversal curve (Fig. 56). Under istropic strain hardening point E is
located on the shifted envelope curve. If no previous yielding has taken
place, point E is the yield point of the shifted envelope curve.
The elastic modulus of the normalized stress-strain relation takes the
form

The simplicity of equations (99) and (100) permits the use of only three
parameters a, b and A. These parameters need to satisfy two obvious
conditions, namely, that the load reversal curve passes through points 5 and
E in Fig. 56. These conditions imply that a* = 0 for e* = 0 and a* = 1 for
e* = 1. From these conditions a and b can be determined

a=b =
-e~x
Substituting the values of a and b in equations (99) and (100) gives

l — e~*

e* = -\ln[l - (1 - e~x)a*\ (102)

A
In order for the load reversal relation to be able to satisfactorily describe
the actual steel stress-strain relation, it is desirable that two further
conditions be satisfied; the slope of the proposed relation should be equal to
the normalized elastic modulus at point S and to the normalized modulus of
the monotonic envelope curve at point E. Because of the simplicity of
equations (101) and (102) only one constant, A, is available. Preliminary
studies revealed that better results are obtained if A is chosen so that the
slope of the proposed load reversal curve is equal to the normalized elastic
modulus at point 5, i.e. at e* = 0, da* /de* = E*. From equation (101)

^ =^ 1 (103)

Setting the left hand side of equation (103) equal to E* results in a non-
linear equation for parameter A

- ^ - , = E* (104)

which can be solved for A using Newton's iteration scheme.

To define a particular load reversal curve the value of A is established
iteratively, generally requiring three to four iterations. The value of A is
required for each new load reversal curve, but as long as no unloading takes
place before the monotonic envelope curve is reached A remains constant.
The proposed stress-strain relation is thus computationally very efficient.
The definition of the basic load reversal curve in equations (101) and
(102) needs to be complemented with a set of rules which describe the
behaviour of the model under a general stress-strain history. Ideally, the
model should have a memory which extends over all previous branches of
the stress-strain history. In this case it would follow the previous loading
branch, as soon as the new reloading curve reached it (curve (a) in Fig. 53).
Since this requires that the model store all necessary information to be able

66
 REINFORCING STEEL

to retrace previous reloading curves which were left incomplete, it is clearly

impractical from a computational standpoint. Memory of the past stress-
strain history is limited in the present model to the following information
(a) the current tension and compression envelope curves, which result
after isotropic strain hardening
(b) the load reversal points of maximum and minimum strain
(c) the most recent load reversal points
(d) the value of A of the most recent load reversal curve.
If it is desirable to better simulate the strain-hardening behaviour of
reinforcing steel, a non-linear strain-hardening curve can be used in the
proposed model. This relation follows the same exponential formulation in
equations (101) and (102) used for the load reversal curve. In this way steel
stress can be expressed as a function of strain and vice versa. In the case of a
non-linear strain-hardening relation, the following equation replaces
equation (89)

1 -e- A c -
for a > (105)
-e~x
where

and e* =

and A is determined from equation (104) with

E* =E2
(rj u — cr sh )

esh is the strain and ay2 the stress at the onset of strain hardening,
respectively. eu and au are the strain and stress corresponding to the tensile
strength of a reinforcing bar under monotonic loading. These values need to
be specified as input along with the other five parameters described earlier
for the trilinear case. In this case, however, the instantaneous tangent
modulus Et at the onset of strain hardening replaces the secant slope E2 as a
parameter (Fig. 55).
Comparison of analytical results with experimental data from cyclic tests
on reinforcing bars conducted by Ma et a/.84 are shown in Figs 57 and 58.
The analytical predictions in Fig. 57 are based on a trilinear monotonic

100

50-

-50-
• Proposed model
-——- Filippou's model
-100
-001 001 002 003 004 005 -001 001 002 003 004 005
Steel strain: in/in Steel strain: in/in
(a) (b)

Fig. 57. Comparison of simple model with trilinear envelope curve with experimental results: (a) bottom reinforcing
bar; (b) top reinforcing bar

67
RC ELEMENTS UNDER CYCLIC LOADING

100

50-

8
I 0

-50-
Analytical
Experimental
-100'
-001 001 002 003 004 005 -001 001 002 0-03 004 005
Steel strain: in/in Steel strain: in/in

Fig. 58. Comparison of model with non-linear envelope curve with experimental results: (a) bottom reinforcing bar;
(b) top reinforcing bar

100

50-

S o

-50-
Analytical
Experimental
-100-
-001' 001 002 003 004 005 -001 001 002 003 004 005
Steel strain: in/in Steel strain: in/in
(a) (b)
Fig. 59. Comparison of proposed model with modified model of G-M-P: (a) bottom reinforcing bar; (b) top
reinforcing bar

150

Fig. 60. Comparison of

proposed model with
experimental results from 001 002 003 004
specimen R-03 by Su 87 Steel strain: in/in

68
 REINFORCING STEEL

envelope curve, while those in Fig. 58 are obtained with a non-linear

monotonic envelope. The analytical predictions of the proposed model are
also compared with the predictions of the model which is an extension of
the Giuffre, Menegotto and Pinto model in Fig. 59. The stress-strain
histories in Figs 57(a), 58(a) and 59(a) are typical of bottom reinforcing
bars, while those in Figs 57(b), 58(b) and 59(b) are typical of top reinforcing
bars where yielding in compression is very limited in cases where the ratio
of bottom to top reinforcement is about 50%. The predictions of the
proposed model are compared in Fig. 60 with experimental data from tests
conducted in Ref. 87.
The correlation studies show that the proposed model is, in spite of its
simplicity, capable of describing quite accurately the hysteretic behaviour of
reinforcing steel under strain histories typical of reinforcing bars. The model
thus offers itself as an excellent alternative to more sophisticated models of
the hysteretic behaviour of reinforcing steel, particularly, in cases where
many evaluations of the steel stress-strain relation are required in each load
step.

69
4. Bond between concrete and steel
The bond behaviour under alternating loading is analysed following a short
description of the behaviour under monotonically increasing slip and
unidirectional cyclic loading, respectively. Various proposals for modelling
are discussed and evaluated.

4.1. Introduction The behaviour of reinforced concrete structural elements, subjected to large
alternate actions, is highly dependent on the interaction between steel and
concrete. In earthquake-resistant design of structures, economical require-
ments usually lead to the need for large seismic energy input absorption and
dissipation through large but controllable inelastic deformations. The
structural and non-structural damage incurred from an earthquake should
not lead to collapse of the structure or to the endangerment of human life.
Therefore the sources of potential structural brittle failure must be
eliminated and degradation of stiffness and strength under repeated loadings
must be minimized or delayed long enough to allow sufficient energy to be
dissipated through stable hysteretic behaviour.
In reinforced concrete, one of the sources of brittle failure is the sudden
loss of bond between reinforcing bars and concrete, which has been the
cause of severe local damage to, and even collapse of, many structures
during strong earthquakes. Even if no anchorage failures occur, the
hysteretic behaviour of reinforced concrete structures, subjected to alternate
actions, is highly dependent on the interaction between steel and concrete.
Cyclic loadings are generally divided into two categories. The first
category is the so-called iow-cycle' loading, or a load history containing
few cycles (less than one hundred) but having very large bond stress ranges.
Low-cycle loadings commonly arise in seismic and high wind loadings. The
second category is the so-called 'high-cycle' or fatigue loading, which is a
load history containing many cycles (typically thousands or millions), but at
a low bond stress range. Bridge members, offshore structures, and members
supporting vibrating machinery are often subjected to high-cycle or fatigue
loading. High-cycle loadings are considered a problem at service load
levels, while low-cycle loadings produce problems at the ultimate limit
state.
The bond behaviour under cyclic loading can further be subdivided
according to the type of stress applied. The first is repeated or unidirectional
loading, which implies that the bar stress does not change sense (tension to
compression) during a load cycle, the usual situation for fatigue loading.
The second is reversed cycle loading, where the bar is subjected
alternatively to tension and compression. Stress reversals are the typical
cases for seismic loading. In this Chapter, the behaviour of bond under low-
cycle loading is described.

4.2. Bond behaviour Bond behaviour is a combination of adhesion, bearing of lugs and friction.
under monotonically Adhesion is related to the shear strength of the steel-concrete interface and
increasing slip is the result primarily of chemical bonding. Bearing forces perpendicular to
the lug faces arise as the bar is loaded and tries to slide. In this phase
microcracking and microcrushing 89 of concrete in front of lugs are
produced. Friction is produced by bearing force on the interactional surface
and by shearing off the concrete between the lugs on the cylindrical
concrete surface at the tip of lugs.

70
 BOND BETWEEN CONCRETE AND STEEL

As both experimentally and theoretically proved,90 bond stress is a basic

function of slip where the latter indicates the relative displacement between
the steel and concrete cross-section. This relationship is called the bond
stress-slip relationship.
Type, rate or duration of loading influence bond behaviour as well as slip,
concrete strength, concrete cover, bar size, rib pattern of bar, confining
effects, casting position with respect to loading, distance to cracks, tem-
perature and surface coating of bar. Extensive guidance on the influencing
factors is available.91
Under monotonic loading two types of bond failures are typical. The first
is a direct pull-out of the bar, which occurs when ample confinement is
provided for the bar. The concrete immediately surrounding the bar fails due
to the shearing of the concrete between the lugs. Pull-out failures depend
primarily on the concrete strength and the pattern and geometry of the
deformations. The second type of failure is a splitting of the concrete
cover when the cover or confinement is insufficient to obtain a pull-out
failure. In this case the failure is due primarily to the tensile radial stresses
caused by the lug-bearing forces. Splitting propagates to the edges of the
member resulting in loss of cover and bond. The orientation of the splitting
cracks depends on the number of bars and their configuration within the
member.
Failure modes under low-cycle loading are very similar to those under
monotonic loading. Under high-cycle loading similar failure modes can
occur, but fatigue failure of both reinforcing bar and concrete must be
considered.
Bond stress plotted against slip curves for a bar loaded monotonically and
failing by pull-out are shown in Fig. 61. 9 2 The relationship shown is
dependent on the degree of confinement as well as the state of stress in the
concrete surrounding the bar (tension or compression).
For analytical applications, several linear and non-linear approximations
of the bond stress-slip relationship are available. The one which CEB
Model Code 199093 incorporated is based on the proposal by Eligehausen et
al.92 It consists of an initial non-linear relationship r = T\(s/s\)a valid for
s < si, followed by a plateau r = T\ for s\ < s < $2 (Fig- 62). For s > S2,r
decreases linearly to the value of the ultimate frictional bond resistance T3 at
a slip value of s3. This value s3 is assumed to be equal to the clear distance
between the lugs of the deformed bars. The same bond stress-slip
relationship is assumed regardless of whether the bar is pulled or pushed.
Different values of the st, s2, S3, a, r max parameters are proposed for
unconfined concrete regions (failing by splitting of concrete cover) and for
confined concrete regions (failing by pull-out) (see in Fig. 62).

100

Fig. 61. Monotonic bond Region 1: unconfined concrete in tension

stress plotted against slip Region 2: confined concrete
curves (Eligehausen et al.92,
yz
) Region 3: unconfined concrete in compression

71
RC ELEMENTS UNDER CYCLIC LOADING

Parameters for defining the bond stress—slip relationship

2 3 4 5

Value Unconf ined concrete* Confined concretef

Bond conditions Bond conditions

Good Allother Good Allother

cases cases

Si 0-6 mm 0-6 mm 10 mm

s2 0-6 mm 0-6 mm 30 mm

S3 10 mm 2-5 mm Clear rib spacing

a 0-4 0-4

Tmax 2-OVAk 2-WU 1-25v//ck

Slips T
f 0-15rmax 0-40rmal
Fig. 62. Analytical bond stress-slip relationship for 'Failure by splitting of the concrete
monotonic loading (CEB93) f Failure by shearing of the concrete between the ribs.

1-0 Pulled out —

1—y* 0-77^--
\

<r max r / ru = 0-65 .

^ <
0-5- ^ ^
0-50 ,
£ _^o --«
_ - ^~~
^ ^

01
1 0
0-40
0- 1

Fig. 63. Slip at the unloaded

end plotted against the 1—""
number of load cycles, (j>14,
lh = 3(j>, fc = 23 MPa (Rehm 005
10' io2 10 3 10 4 10 5 10 6
Eligehausen96) Ig n

001

4.3. Bond behaviour Cyclic loading produces a progressive deterioration of bond. The
under unidirectional deterioration can be observed as a slip increment.94"96 This process can
cyclic loading lead to failure at a cyclic stress level lower than the ultimate stress under
monotonic loading.

4.3.1. Behaviour
yo
According to the pull-out test results of Rehm and Eligehausen,96 with
constant amplitude cyclic loading, if no fatigue failure of bond occurs the
slip against number of load cycles relationship in double logarithmic scale is
approximately linear (Fig. 63). Rehm and Eligehausen also concluded that if
a fatigue failure of bond does not occur, then previous load cycles do not
negatively affect the bond stress-slip behaviour near ultimate load
compared with the monotonic loading behaviour (Fig. 64). Otherwise the
comparison of test results with cyclic and sustained loading show that cyclic
loading can be considered as a time accelerator compared with a sustained
load.
Based on pull-out test results with cyclic load Balazs97'98 distinguished
three different phases of the bond fatigue process (Fig. 65). During the

72
 BOND BETWEEN CONCRETE AND STEEL

Fig. 64. Effect of cyclic load

on static bond strength, Rehm
and Eligehausen96

1 short time loading

2 cycling loading
3 short time loading after
load cycles

0-2 0-4 0-6

Slip at unloaded bar end: mm

beginning of the load cycles (first phase) the slip rate is decreasing then
(second phase) keeps constant and is finally followed by a rapid increase
(third phase) producing pull-out failure (Fig. 65(a)). The slip versus number
of load cycles relationship in normal scale contains concave, linear and
convex portions (Fig. 65(b)). The intermediate linear portion ends at S(T U ),
i.e. the slip at monotonic bond strength which marks the beginning of the
failure branch. Therefore, S(TU) is proposed97 as a safe fatigue failure
criterion.
Load history has a significant effect on slip. For instance93 a periodical
increase of the maximum value of cyclic force produces a higher slip rate of
the intermediate linear portion (Fig. 66).
The accumulation of bond damage is supposed to be caused by the
progressive growth of microcracks and concrete crushing in front of the
ribs. Their effect is observed by increasing slips. Thus, increase of slip can
demonstrate the damage accumulation of bond.
Test results by Balazs show three different phases of slip against number
of load cycles for constant amplitude cyclic bond tests and not a linear
relationship as the Palmgren-Miner law assumes (Fig. 67). During the first
and third phase the fatigue damage proceeds more quickly than predicted by
the hypothesis. In the intermediate linear stage the damage is slower.

4.3.2. Model by Fehling

A physically-oriented bond model was developed by Fehling99 in order to
model the bond behaviour under cyclic loading and the dissipation of
energy due to bond hysteresis. The model consists of a strut-and-tie
structure representing the concrete around the reinforcing bar taking into
account
(a) local plasticity of concrete at the lugs of rebars
(Jb) internal cracks (cone shaped)
(c) splitting cracks (longitudinal cracks)
{d) deformation of rebars in transverse and longitudinal direction
(e) friction effects between concrete and steel
(/) creep effects considering general load histories
(g) damage due to cyclic loading.

73
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 65. Bond fatigue process

test results: <j>8 def. bar, 1/, =
12<j>, C16 (Baldzs*7)

20 40 60
Number of load cycles
Pull-out
failure

20 40 60
Number of load cycles n Bond stress: MPa
(b) (c)

For the physical modelling of bond the stress-strain relation of concrete

under triaxial stress conditions and for high values of strain is required. As a
result of tests with high local compressive bearing stresses, Lieberum100
found a relationship between the porosity of the cement matrix and the
amount of possible plastic strains (Fig. 68(a)). The unloading stiffness is
assumed to be equal to £ c3 . A simple assumption of the distribution of
stresses (Fig. 68(c)) in the range where plastic strains occur (Fig. 68(b)) in
combination with the stress-strain law mentioned above, enables the
calculation of plastic deformation of concrete contributing to the total slip
using a strut-and-tie model (Fig. 68(d)).
For modelling the unloading and reloading branches, the plastic zone near
a lug of the rebar may be described by the system consisting of springs and
friction elements as shown in Fig. 68(e). Friction is modelled as being
composed of a constant term (self-induced radial stresses F v and cohesion)
and a term due to radial stresses in the steel-concrete interface due to bond
stresses (radial component of force D in Fig. 68(e)).
The strut-and-tie model enables consideration of the influence of steel

74
 BOND BETWEEN CONCRETE AND STEEL

• Palmgren-Miner hypothesis
Loaded end (test result)
Unloaded end (test result)

I Pull-out
I failure
100-
Load history

75-
Slip:
93%
(62%)
0-6-

-002
75%
(50%)
s
0-4- <d
Q

60%
(40%)
r' 25-
0-2-

o-S— 100 000 200 000 300 000 25 50 75

—I—
100
10 000 Number of load cycles n
P-M is P-M is
unsafe safe

Fig. 66. Slip increase under periodically increasing Fig. 67. Bond damage accumulation with respect
cyclic load, <j>8 def. bar, lh = 12<j>, C16 (Baldzs97) to the Palmgren and Miner hypothesis, test result:
4>8 def. bar, lh = 12<f>, C16 (Baldzs97)

strains in longitudinal as well as in transverse direction. The longitudinal

tensile strains in the rebar are associated with transverse contraction
contributing to the total amount of slip. Furthermore, the local steel strain in
the direction of the rebars is influential with respect to the width of internal
cracks. The inclination of internal cracks is considered in the strut-and-tie
model leading to differences in modelled frictional behaviour near
transverse cracks and in more distant regions.
Transverse concrete stresses influencing the frictional behaviour as well
as the capacity of the concrete cover (with respect to longitudinal cracking)
can easily be included into the model. Assuming fully plastic distribution of
circumferential tensile concrete stresses around the rebar, the limit of bond
stress due to longitudinal cracking is checked for each load step.
In order to model the damage due to cyclic loading, the hypothesis is put
forward that damage is caused by abrasion and transport of particles into the
open internal cracks. Thus, the width of internal cracks as well as the sum of
the absolute values of slip increments form the basic parameters of the
damage model. In addition to the damage associated with slip reversals the
effect of bond-creep, which also contributes to bond softening, is modelled
using the bond-creep law proposed by Rohling and Rostasy.101 The damage
model is described in more detail by Fehling."
The results obtained with this simple model were adequate for the loading
and unloading branches but not for the reloading branch. The test results
showed a more pronounced increase of bond stress when the actual slip
during reloading approaches the maximum slip reached before. This may be
explained by the necessity of compacting a volume of particles in the plastic

75
RC ELEMENTS UNDER CYCLIC LOADING

D D D D

ac

= o2 = o3 = ac

NSC

P
(a)

— —— + --rR(x)

(c) (d)

i
(e)

Fig. 68. Model by Fehling": (a) stress-strain curves of concrete of different strength under 3-axial hydrostatic
compression; (b) modelling of local bond behaviour; (c) assumed stress distribution near lug ofrebar; (d) strut-and-
tie model; (e) unloading and reloading in plastic zone (model and simplification)

zone, which have been loosened during unloading, leading to a non-linear

force deformation curve which can be incorporated into the model. Also,
aggregate interlock effects in internal cracks may contribute to the observed
effect.
The comparison of the computed local bond behaviour with the test

76
 BOND BETWEEN CONCRETE AND STEEL

results showed a significant influence of the distance from the crack on

frictional bond stresses, which could not be fully explained by varying the
parameters of the strut and tie model. It is assumed that aggregate interlock
in internal cracks plays an important role for the frictional bond stresses.
Thus, the local value of frictional bond stress, originally assumed to be con-
stant, is modified depending on the distance from the next transverse crack.
The proposed analytical bond model needs the parameters for the damage
model as well as the frictional parameters, which could be taken as constant
for all types of concrete; beyond that, it requires only the input of geometric
variables describing the rebars and their lugs as well as strength and strain
values for the concrete. It could be shown that the model is able to predict
the bond behaviour for different concrete strengths and bar diameters satis-
factorily. The computational effort to follow the local strains within the
strut-and-tie model is limited and — considering the speed of modern
personal computers or mainframes — does not play an important role. A
comparison of test results and the results of the theoretical model is
presented in Fig. 69(a) and (b).

1
DK152: experiment
I
|
/ /!
i / A MmL/ /// i4
1' m%///!mm
// //Iif!ml t ¥ I'M
/ iJmAwo
M mm
If
Sir
r ' _
i
1
005 0-10 0-15 0-20 0-25
Slip: mm
(a)
DK152: model
1 / /t iff
1 r^ i mil
. — •

1 Ilkr' h
it
////
m
D-

{
UJiL
1 w
JiMWlJiw
w
Fig. 69. Bond stress-slip -1
relationships; Fehling9 :
(a) test result; (b) result of 005 0-10 0-15 0-20 0-25
calculation using model in Slip: mm
Fig. 68 (b)

77
RC ELEMENTS UNDER CYCLIC LOADING

10-

12-
ir ~-^v ^ Monotonic loading

1 8-
After 10
cycles
4-

J
i
-10 -8 -6 -4 -2 4 6 8 10 12
s: mm
--4

—8
1

--12
Series 2-4
Monotonic loading - ^ cycling between
— 16 s = ± 0-44 mm

(a)

. Monotonic loading

After 1 cycle

-10 -8 -6

Series 2-8
cycling between
Fig. 70. Bond stress-slip s = ± 4-57 mm
curves for cycling at different
maximum slips (Eligehausen
et al92)

78
 BOND BETWEEN CONCRETE AND STEEL

Although the model in the present formulation is oriented towards

repeated (or unidirectional) loading it is possible to include the mechanisms
required for the modelling of reversed loading.

4.4. Bond behaviour In the case of reversed cyclic loading the type of loading (load or slip
under alternating controlled), and the rate of loading (frequency) are important parameters for
loading bond strength.102103 Seismic loading, the most common case of reversed
cyclic loading, represents an intermediate step between load and slip
controlled cycles and is characterized by a wide frequency content.
Previous state-of-the-art reports closely linked to the present field are
available for reference.91'103106 Extensive studies considering several
parameters are also available. 78 ' 7992107 " 115 The main characteristics and
some modelling possibilities are summarized below.

4.4.1. Behaviour
Cycles with reversed loading produce degradation of bond strength and
bond stiffness that is more severe than for the same number of cycles with
unidirectional loading. 106112 Degradation primarily depends on the peak
slip in either direction reached previously (Fig. 70).9 Other significant
parameters are the number of cycles and the difference between the peak
values of slip between which the bar is cyclically loaded. Under otherwise
constant conditions the largest deterioration will occur for full reversals of
slip. Whenever the load cycles are limited to produce slip in one direction
only, there is not very significant degradation of the bond strength (Fig.
71 (a)).110
If the peak bond stress during cycling does not exceed 70-80% of the
monotonic bond strength r, the ensuing bond stress-slip relationship at first
loading in the reverse direction and at slip values larger than the one at
which the specimen was cycled, is not significantly affected by up to ten
repeated cycles (Fig. 70(a)).92 The bond resistance at peak slip deteriorated
moderately with increasing number of cycles.
Loading to slip values inducing a bond stress larger than 80% of the
monotonically obtained bond strength in either direction leads to
degradation in the bond stress-slip behaviour in the reverse direction
(Fig. 70(b) and (c)). As the peak slip increases, deterioration of bond
resistance is increased. Deterioration also increases with the number of
cycles, and is more extensive for full reversals of slip than for half-cycles
(Fig. 71).

-2 15
6000

4000-

2000-

o
m

-2000-
-40
-100 100 200 300 400 500 -600 -400 -200 0 200 400 600
Slip: x10~ 3 in Slip: x10 3
in
(a) (b)

Fig. 71. Comparison of monotonic and reversed cyclic bond stress-slip curves: (a) without change of slip sign;
(b) with change of slip sign; Hawkins et al."°

79
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 72. Slip-controlled load

reversals: <j)16 def. bar If, =
Loa d listory
2+ 4+
II
2<t>, C16 (Baldzs"2) (a) test s 1 + 3* 5 + -F:kN|-r:MPa —
AA
result; (b) deterioration of
peak bond stress; \/ A j-15
~7_ 1001
(c) deterioration offrictional VVVVV
I " 3" 5 20
bond stress - n— \ f ji
2- 4" i = + 50-
j-10/
/
4+

n
1+
1-2+2-3+3-4+4-5+5-

> (b)

^^ 1 i/i
Slip: mm

/
/
life 15

J 1/4 ~ -|

1
--5

4
10

r 7
/ 20-
I—10
If n
4+
3~4 4"
/
/ —15
V

(a)

Slip-controlled load reversals produce deterioration both of the peak bond

stress and of the frictional bond stress (Fig. 72). Force-controlled load
reversals produce a remarkable slip increase which demonstrates a more
pronounced damage in concrete matrix around the steel bar (Fig. 73). The
higher the load, the higher the slip increment.
Cycling a specimen at different increasing values of slip has a cumulative
effect on the deterioration of bond stiffness and bond resistance as indicated
in Figs 70, 71(b), and 74). Some additional cycles between slip values
smaller than the peak value in the previous cycle do not significantly
influence the bond behaviour at larger peak values.
The interactional behaviour under reversed cyclic loading (including also
monotonic loading) may be understood through the following reasoning
given by Eligehausen et al.92 When loading for the first time, the assumed
bond stress-slip relationship follows the 'monotonic envelope', which is
valid for monotonically increasing slip (path OABCD in Fig. 75). The initial
slope, all other variables being equal, depends on the related rib area. At low
bond stresses (point A), inclined cracks begin to propagate from the top of
the ribs; their growth and size is contained by the confining pressure
provided by, for example, transverse reinforcement. Transfer of forces will
be mostly by bearing, with a shallow angle of inclination (a = 30 degrees,
Fig. 75). Increasing the load, local crushing of concrete in front of the lugs
produces reduction of the tangent of the bond stress-slip curve (point B).
When the maximum bond stress is attained the concrete key is sheared off,
forming a cone with a length of about four times the lug height. At this point
the line of action of the force is at an angle of about 45°. With increasing
slip the bond stress will begin to drop slowly. As the bond shear cracks

80
 BOND BETWEEN CONCRETE AND STEEL

Fig. 73. Force-controlled _l/ 4

load reversals on four
different load levels, test
result: <f>16 def. bar, 1^ = 2<f>,
Ff r
Load history
i
F:k N
f; >a
I
i i A
C25; Baldzs"2
—
jiiuuiillllllllU >
mmhMMA\\ \ \
I
10liWMlWmMl J \
i
Ummmmiil \
- Jimm J

0-8
— f

0-6/1/
II 11 iT/7
4
w
wV
1 M
-UL TO 111
L
- - 0-6"
/ k
j1
/ I °' U-' ' g
- 4 - If 1Iff n)flF- —p
>w>
-ii b

Ml
Ml 111
ml
it
tut,
n mi
[it n 0-
— 10

\\
TAT
— 15
/ s
3- rti
0-
T

reach the bottom of the adjacent lug (point D) the bond stress begins to
drop, and by the time the slip has reached the lug spacing only the frictional
component remains (point E).
Under load reversals (Fig. 76), the initial loading follows the monotonic
curve, but the cyclic load behaviour is sensitive to the level of slip at which
the reversal occurs. Three possible qualitative models have been proposed,
depending on whether or not inclined cracks have formed. In the first case
(Fig. 76(a)) imposing a slip reversal at an arbitrary slip value below the
level where inclined cracking occurs, results in a stiff unloading branch
(path AF) because only a small part of the slip is caused by inelastic
concrete deformations. As soon as slip in the opposite direction is imposed,
the friction branch is reached (path FH). The slope of this portion of the
curve is small because the surface of the concrete surrounding the bar is

si
I-oad h story F- k
!0-
r: N/mm^
IP Mon(,lonic
• env slotje
-15 + 2.-
y
V'1/1/1/1/ " • 1J
7
y
\/,
0-
/\ / y
/ , ^-' T

//
/
i
/T
if
•5

/
/
- — •
-M

^ ^
H}
1 UJ
pi

0-
--5
mm
25

-X
4-
>

+ T^.Sliprm m
1 ]
Monotonic
/-envelope
jjT~ ^ Reversed

i
Fig. 74. Bond deterioration /
-" i
f 10 X; ; •*<•-cyclic
under increasing reversed
cyclic slip values, test result:
r
7 /
/
s*
envelope
s
lb ru)
<p8 def. bar, lb = 6<j>, C25;
Baldzs10*

81
RC ELEMENTS UNDER CYCLIC LOADING

_± 1
Gaps A Concrete •-
Bond
crack

(a) (b)

Fig 75. Bond mechanism under monotonic

(c) loading; Eligehausen et al.91

smooth. As soon as the cracks close, the stiffness differs little from that of
the monotonic envelop (point I).
Unloading from point I, where the slip in the two directions is about
equal, the curve (path IKL) is very similar to that from the initial unloading
curve (path AFH). The major difference is that due to previous cracking and
crushing of the concrete in front of the ribs, the point where the bond
stresses begin to pick up again (point L) will be shifted to the right of the
origin. The lug will not be bearing fully until point M is reached. Further
loading follows the bond-slip curve up to the monotonic envelope.
If unloading occurs after the inclined cracks have formed (Fig. 76(b)),
and therefore near the slip at which ultimate bond stress has been attained,
the unloading path is similar to that of the first case up to point F. Since
there is more damage to the concrete, a higher frictional resistance is
mobilized (point G). When the loading is reversed the lug presses against a
key whose resistance has been lowered by inclined cracks over a part of its
length that were induced by the first half-cycle. The splitting cracks created
in the first half-cycle close at a higher load than those of the first case (point

l i l t i I 1 1 1 t t
C
r Closed old r Monotonic
Gaps Gap sF, Old crack loading
crack
M
Kf- TL
10
'Crushed concrete New crushing/ C r u s n i n 9
/ / f r o m first half cycle' /I Monotonic and new / / / s h e a r cracks < /
Sorne_crushed concrete /// loading . Monotonic
shear cracks ' f r o m loadingOC
loading
(a) (b)

i 1 i i I I 1 t

Monotonic
loading Fig. 76. Bond mechanism under reversed cyclic
loading; Eligehausen et al.

82
 BOND BETWEEN CONCRETE AND STEEL

H), and lead to an earlier formation of splitting cracks in the opposite

direction. Splitting cracks, combined with the existing inclined cracks along
the bar, result in a reduced envelope (path HI) and a reduction of bond
capacity in the second direction (point I). Unloading from this peak (path
IKLMN) and reversing the load results in a reduced stiffness and strength
because only the remaining uncrushed concrete between the lugs must be
sheared off. The bond strength (point N) is substantially lower than that of
point C, and lower than that of point I.
If unloading occurs after the slip has reached a value much larger than the
slip at maximum strength (point C), the behaviour will be very poor (Fig.
76(c)). Since more damage has occurred, the frictional resistance (point G)
will be larger than for either of the previous cases. However, since the
concrete between the lugs is completely sheared very little force can be
transmitted by bond when direction of loading is reversed (path HIJ).
Unloading and reloading in the opposite direction (path JKLMN) results in
very little additional bond capacity beyond that provided by friction since
most of the mechanical anchorage has been lost.

4.4.2. Analytical models

Analytical models for bond behaviour under reversed cyclic actions are
presented below in the order of increasing efficiency in use. An approach
for the monotonic envelope has been presented in section 4.2.
4.4.2.1. Model by Morita and Kaku.108 The first analytical model of
the bond stress-slip relationship for reversed cyclic loading was proposed
by Morita and Kaku. The monotonic envelopes, which are different for
loading in tension or compression and for confined or unconfined concrete
are given by two successive straight lines which follow closely the
experimentally measured curve (Fig. 77). The assumed bond stress-slip
relationship for the first cycle coincides relatively well with the behaviour
observed in experiments. However, the observed deterioration of the bond
resistance at peak slip and of the frictional bond resistance with increasing
number of cycles is not taken into account. After cycling between arbitrary

s L = (sB + Sj)/2
IN = <"H su = (Sj + s P )/2
= ar B TQ = arp

Fig. 77. Reversed cyclic bond K3 = 400 N/mm 3

stress-slip relationship a =0-18
proposed by Morita and /S =0-9 s s= 0 0 5 mm
108 fi = 0-9 - 0-44 (s - 005) 0-05 =S s « 0-5 mm
Kaku'""

83
RC ELEMENTS UNDER CYCLIC LOADING

i = 1 - Vn

Monotonic envelope

Fig. 78. Bond model proposed by Tassios104

slip values, it is assumed that the monotonic envelope is reached again at

slip values larger than the peak value in the previous cycle and followed
thereafter.
The model is sufficiently accurate for a small number of cycles between
relatively small slip values with corresponding bond stresses smaller than
about 80% of monotonic Tmax- However, it is inaccurate for several load
cycles, and it is not valid for slip values larger than the one corresponding to
0-8 T,max*
104
1U4
4.4.2.2. Model by Tassios. At the beginning of loading, up to point
A in Fig. 78, the bond resistance is developed by adhesion. Further loading
produces high compressive stresses in the concrete in front of the lugs and
hoop tensile stresses around the bar leading to inclined microcracks (point
B). When these cracks reach the concrete surface, say at point C, the bond
resistance may drop to zero in case of lack of confining reinforcement.
However, if concrete is well confined, the load can be increased till point D,
where shearing off the concrete between the bar lugs initiates. The same
bond stress-slip relationship is assumed regardless of whether the bar is
pulled or pushed.
During cycling, loading to a slip value s > sB, the values of r of the bond
stress-slip relationship for loading in the reversed direction are reduced by
one-third compared to the monotonic envelope. The bond stress-slip
relationship for reloading and for subsequent cycles between fixed slip
values is somewhat simplified compared to the real behaviour. However,
the deterioration of the bond resistance at peak slip and of the frictional
bond resistance is taken into account. When increasing the slip beyond the
cyclic peak value (s > sG in Fig. 78), it is assumed that the monotonic
envelope is reached again.
Tassios' model is an improvement compared to the older one of Morita
and Kaku in so far as the descending branch of the local bond stress-slip
relationship is given and the influence of load cycles on bond deterioration
for slip values smaller than or equal to the peak slip value of the previous
cycle is taken into account. However, the assumption that for slip values
larger than the peak value in the previous cycle the monotonic envelope is
reached again and followed thereafter, while the bond stresses in the
reversed direction are reduced by one-third compared to the monotonic

84
 BOND BETWEEN CONCRETE AND STEEL

envelope, is not sufficiently accurate. For monotonic loading, the model is

useful for the total slip range. However, for cyclic loading it is valid for slip
values s <C .Wax on ly- The deterioration of bond resistance was studied in a
following paper by Plaines et al.lu
4.4.2.3. Model by Viwathanatepa et a/.113 This model is summarized
in the five points listed below.
(a) A four-stage piecewise linear approximation is used as a monotonic
envelope (Fig. 79). The physical meaning of the controlling points
are the same as described in section 4.4.1. However, points B and C
(occurrence of internal bond cracks and splitting cracks) are omitted.
Different monotonic envelopes are assumed for unconfined concrete
in tension, confined concrete and unconfined concrete in com-
pression, which simulate the behaviour observed in experiments
(compare Fig. 61).
(b) Cycling between points A and A[ or unloading and reloading only
(paths GIG or KLK) do not deteriorate the envelope.
(c) Unloading from a point beyond A or A! and following the friction
path for an arbitrary small slip value produces reduced envelopes
(OAD'E'F' and OAiD'1E'1F'1) by reducing the characteristic bond
stress TD, TDI, T&, TEI and the slip values 5 E , 5 E i by a reduction
factor. The latter depends on the cumulative slip having magnitude
larger than those of the previous cycle. Therefore, no further
reduction of the envelope is assumed for subsequent cycles between
slip values smaller than or equal to the previous peak slips. As an
example, equation (106) gives the reduction of TQ. Similar equations
exist for the reduction of the other characteristic values which
describe the model.

Monotonic
loading

Monotonic '
loading

T
i = TE, rn = 0-9 i> r D ' = a0 r 0 To; = «b

= rE s M = 0-4 SG = aE r E = aE
Fig. 79. Reversed cyclic bond »L TE- TE;

stress-slip relationship SE' = PE S E SE; = /J E s E |

proposed by Viwathanatepa
etal."3 a, a', / i , / ( ' : function of loading history

85
RC ELEMENTS UNDER CYCLIC LOADING

a
TD (106)

where
SSJ, Ssji: sum of the peak slip values having a magnitude larger than
in the previous cycles for loading in tension or compression,
respectively;
sD: slip at point D;
C*TD) A-D> £TDI PD'. constants, evaluted from test results.
(d) The frictional bond resistance is assumed to be equal to TE of the
monotonic envelope and independent of the number of cycles.
(e) The bond stress-slip relationships for the reloading branch (path
MRN) and for additional cycles between fixed slip limits are very
similar to those proposed by Morita and Kaku. l08
The above model is a major improvement, because it takes into account
several features observed in experiments and it is approximately valid for
cycling between arbitrary slip values. However, in spite of being rather
complicated, it is not general. Some 20 parameters are needed to describe
the bond stress-slip relationship for cyclic loading, which have no clear
physical meaning and must be evaluated from test results. Furthermore, the
assumptions on which the calculation of the reduced envelope is based need
improvement. For example, an arbitrary number of cycles (> 1) in well-
confined concrete between s max = 2sD and smin = —2sD reduces TQ,
independent of the number of cycles, by 13%. TO, however, is reduced to
zero after eight cycles between almost the same peak slip values if only the
value of 5max is slightly increased in each cycle.
4.4.2.4. Model by Hawkins et a/.110 The monotonic envelope is
idealized by a trilinear bond stress-slip diagram. The characteristic values
of slips, bond stresses and tangents are obtained by statistical evaluation of
the test data yielding to three linear equations.
During load reversals, deterioration in the loading envelope does not
begin until the slip range exceeds s'o, where s'o is the slip for r max for
monotonic loading. The deteriorated capacity in the positive direction

Bond
r
stress

Cycle o'abzcdef: loading to a given slip for first time

rb = rz = - « r a , Sb = S. - (1 + a) TJK (109)
rd = re = - a r c , S d = Sc - (1 + a) TJK (110)
e = Sc + L; L = 0-5L; L = slip range = \S, + Sc| (111)
r. = fir,, S, = S, - (1 - /() TJK (112)

Cycle fghiklop: subsequent cycle response for r > r'c

Fig. 80. Local bond stress- T, = /hc. S, = Sc - (1 - P) TJK
rh = T, = - «Tg, S h = Sg - (1 + a) TJK
slip rules for cyclic loading r, = ro = - «r k , S, = Sk - (1 + a) rk/K
proposed by Hawkins et Si = Sg- U: U = 0-5 |S, + Sc\
So = Sk + L';L = 0-5|S g + S k |
aU'°

86
 BOND BETWEEN CONCRETE AND STEEL

follows the line M'X^N' where that line is derived from the line M'XN" that
parallels line MN (Fig. 80). The post-peak response envelope for the
negative direction mirrors that for the positive direction. Rules for deriving
the post-peak response envelope for the positive direction are

% = V = SN" = °-5(slug + W)
sx = sx, = 0-45slug = 0 - 5 ^ - W/2)
TX, = 045r x but TX> > 0-l/c' or 2-8 N/mm2
sY = 0-50siug; r Y > 0-1 f'c or 2-8 N/mm2
T = Ty — 0-5(s — Sy), Sy < S < Sff

where siug = lug spacing, and W = lug width.

The rules for unloading and the construction of loading loops are also
shown in Fig. 80. The response is defined by the unloading stiffness, K and
by the coefficients a and /3, which define, respectively, the stress level for
marked slip under reversal of loading, and the degradation in capacity for
one loading cycle between given slip limits. Statistical analyses of the test
data, and round-off of the resulting coefficients gave the following values
K = 2 6 0 N/mm2
a =115
j3 = 0-9 when r has not exceeded r c (107)
0-9 > P = 5(s - s'c) > 0-65 when r has exceeded TC (108)
In Fig. 80 k, c, z, o", a, g, M, M, and N are points on the monotonic
loading envelope. The loop o abzcdef defines the response for loading for
the first time between given reversing slips. Equations (109) to (112) (see
Fig. 80) define critical stress and slip values. The quantity L in equation
(111) defines the magnitude of the marked slip under constant reversed
stress. (3 is defined by equation (107). The loop efghiklop defines the
response for loading in a second or subsequent cycle to increased slips when
r has exceeded r c in a previous cycle. Initially, the response for r > T{ is
defined by the extension of line ef to the monotonic envelope. Unloading
rules from g to h and k to 1 are the same as those for a to b. Loading reversal
rules from h to i and 1 to o are the same as those from d to e. Since r has
exceeded T'C, (3 is defined by equation (108).
The foregoing rules for loop construction must be modified slightly for
type 1 cyclic loading (as in Fig. 71 (a)). If the slip range has not exeeded s'o
and the slips for negative loading have not exceeded —0-08 mm, then the
response in the negative loading direction is of the friction type before
intersection with the monotonic envelope, and follows the monotonic
envelope after intersection. If the slip range has not exceeded s'o but the slips
for negative loading have exceeded —0-08 mm, then the loop construction
rules are the same as those for type 2 loading (as in Fig. 71(b)) except that /?
values for the negative loading direction must be modified as follows

f3 = 0-9 when s has not exceeded so

P = 0-65 when * has exceeded sso
o

According to the proposal, for cyclic loading the bond stress-slip

envelope is similar to that for monotonic loading prior to attainment of the
maximum capacity. At the maximum capacity, the bond effectiveness for
cyclic loading is less than for monotonic loading with the decrease in
capacity being greater for fully reversed cyclic loading than for zero to a

87
RC ELEMENTS UNDER CYCLIC LOADING

maximum slip cyclic loading. After attainment of the maximum capacity,

the bond stress for a given displacement is always less than that for
monotonic loading although the rate of decrease in capacity with increasing
slips is similar to that for monotonic loading. The relatively complicated
model intends to prescribe the characteristic bond stresses and the selected
slip values during cycling with rules and factors obtained by tests applicable
for small and also for high slip values. The ascending branch of the trilinear
approach for the cyclic envelope give reasonable values for slips higher than
in the previous cycle but a relatively rough approach for lower ones owing
to the low loading models. Frictional bond stresses formulated as a function
of the previously reached maximum bond stress does not satisfactorily
follow force-controlled reversed cycles.
4.4.2.5. Model by Baldzs.107 First loading produces microcracking
and microcrushing in the concrete matrix around the rebar. When loading in
the subsequent cycles, the previous state is reproduced in a softer matrix
followed by a more stiff concrete response. The physical change is
represented by points of inflection. The points of inflection belong to the
intersection of reloading branch and previous unloading branch close to the
highest slip value before unloading. Hence, the first loading curves are
concave in both directions, all subsequent loading curves are convex
(following the frictional portion) until the intersection point mentioned
above, then become concave. As a consequence, the shape of the ascending
branch of any possible reversed cyclic bond stress-slip curve indicates
whether higher slip values during the previous load history occurred or not.

Load
histories:

Slip

I A1"
IM'lll

Primary loading direction

Monotonic envelope

Reversed cyclic
envelope

x, = 0-9 to 1 0
= 0-7to 1 0
= 0-5to10

Fig. 81. Bond fatigue

behaviour under reversed
Secondary loading direction
cyclic loading; Baldzs 107
 BOND BETWEEN CONCRETE AND STEEL

The answer is yes if a point of inflection is available, otherwise no.

Reversed cyclic loading produces reductions of both the bond strength
(r u , obtained by monotonic test) and the related slip, S(T U ). The reduction
depends on the actual load history, the number of reversed load cycles and
the maximum previous slip (Fig. 81). Ranges of reduction factors from the
test results are; K\ = 0-9 to 1-0 for the monotonic envelope in the secondary
loading direction, Ki = 0-7 to 1 0 for the bond strength and K3 = 0-5 to 1 0
for the slip at monotonic bond strength.
Assuming a symmetrical reversed cyclic load history, the maximum bond
stress 0-7 r u , is observed at a slip of 0-5 S(T U ). Hence 0-5 *(ru) provides a
fatigue failure criterion.
The softening of bond during a slip-controlled test is represented by the
decrease of the maximum bond stress itself. During a force-controlled test the
softening is indicated by a remarkable slip increase. Following slip-controlled
load reversals, the maximum bond stress decreases asymptomatically.
Frictional bond stress increases during force-controlled reversals and
decreases during slip-controlled reversals, consequently the frictional bond
stress is a function of the actual slip and of the number of previous load
cycles. Its maximum value is approximately 70% of the residual bond stress
obtained after the concrete between bar lugs is sheared off.
The slip rate due to force-controlled load reversals is approximately four
times higher in comparison to that of a unidirectional cyclic loading, both
having the same maximum value of cyclic force.112
The model summarizes the characteristics observed in bond behaviour
under slip-controlled or force-controlled load reversals. A reversed cycle
envelope is defined considering deterioration of both the bond strength and
that of the related slip which can be a suitable solution to take into account
the reduction in bond capacity. The frictional bond stress is considered as a
function of actual slip reaching a maximum value related to the residual
bond stress. Guidance,is given to compare the increase of slip values
produced by cyclic or by reversed cyclic loading, respectively.
4.4.2.6. Model by Ciampi et a/.78 Their extensive experimental
investigation concentrated on reversed cyclic bond behaviour for relatively
large slip values.78'92'109 The assumed bond model is illustrated in Fig. 82.
Although it simplifies the observed real behaviour, it takes into account the
significant parameters that appear to control the behaviour observed in the
experiments. The model's main characteristics, illustrated by following a
typical cycle, are described below.
When loading the first time, the assumed bond stress(r)-slip(5i)
relationship follows a curve valid for monotonically increasing slip, which
is called herein monotonic envelope (paths OABCD or O A ^ Q D i ) .
Imposing a slip reversal at an arbitrary slip value, a stiff unloading branch is
followed up to the point where the frictional bond resistance, T{, is reached
(path EFG). Further slippage in the negative direction takes place without an
increase in r up to the intersection of the 'friction branch' with the curve
OAi (path GHI). If more slip in the negative direction is imposed, a bond
stress-slip relationship similar to the virgin monotonic curve is followed,
but with values of r reduced as illustrated by the path IAiJ, which is part of
the curve OAiBiQD! that is, the 'reduced envelope'. When reversing the
slip again at J, first the unloading branch and then the frictional branch, with
r = Tf, are followed up to point N, which lies on the unloading branch EFG
(path JLN). At N the 'reloading branch' (same stiffness as the unloading
branch) is followed up to the intersection with the reduced envelope
OAB'C'D (path NE'), which is followed thereafter (path E ' B ' S ) . If instead
of increasing the slip beyond point N more cycles between the slip values
corresponding to points N and K are imposed, the bond stress-slip

89
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 82. Proposed analytical

model for local bond stress-
slip relationship by Ciampi et
78
al.""

relationship is like that of a rigid plastic model, the only difference being
that frictional bond resistance decreases with an increasing number of
cycles. A similar behaviour as described is followed if the slip is reversed
again at point S (path STU). To complete the illustration of the model,
details about the different branches referred to in the above overall
description are given below.
The description of 'monotonic envelope' is given in section 4.2 and by
Figs 61 and 62 respectively. 'Reduced envelopes' are obtained from the
monotonic envelope by decreasing the characteristic bond stresses T\ and r?
(see Figs 62 and 82) through reduction factors, which are formulated as a
function of one parameter, called herein the 'damage parameter a". For no
damage, d = 0, the reloading branch reaches the monotonic envelope. For
full damage, d= \, bond is completely destroyed (r = 0). The rationale for
this assumption is given by Fig. 83, which shows that reloading curves for
similar specimens subjected to different loading histories appear to form a
parametric family of curves.
The deterioration of the monotonic envelope seems to depend on the

16

- Monotonic loading
• After N = 1 cycle
12- Cycling between
±Smax

1 8-

Fig. 83. Effects of number of

cycles and of peak values of
slip snulx at which cycling is
performed on ensuing bond
stress-slip relationship for 6 10 12
/ft s: mm
s>sinax; Ciampi et al.

90
 BOND BETWEEN CONCRETE AND STEEL

Fig. 84. Damage parameter d

as a function of the o First slip reversal
O After N = 1 cycle
dimensionless energy • After W = 5 cycles
dissipation; Ciampi et al.7S • After N = 10 cycles

damage experienced by the concrete, particularly the length of the concrete

between the lugs of the bar that has sheared off. This, in turn, is a function
of the magnitude of the slip induced in the bar in both directions, the larger
the s max and the difference between peak slip values, the larger the damage.
Another influencing factor is the number of cycles. These parameters can be
related to the energy dissipated during the loading and unloading processes.
Therefore, it was assumed that the damage parameter d is a function of the
total dissipated energy only. However, it has also been taken into account
that only a fraction of the energy dissipated during subsequent cycles
between fixed peak slip values appears to cause damage, while the other
part appears to be used to overcome the frictional resistance and is
transformed into heat.
Figure 84 illustrates the correlation between the measured damage factor
d, for tests with full reversal of slip as a function of the computed
dimensionless dissipated energy factor E/Eo. The proposed function for d is
shown as well. In the computation of E, only 50% of the energy dissipated
by friction is taken into account. The normalizing energy Eo corresponds to
the absorbed energy under monotonically increasing slip up to the value s3.
Although there is some scatter, the agreement between the analytical and
experimental results seems acceptable.
No reduction of the current envelope (monotonic or reduced) is assumed
for unloading and reloading only. If a cycle is not completed to the current
values of smax or smin (e.g. path GHM), the damage parameter is linearly
interpolated between the values valid for the last slip reversal and for the
completed cycle (point E and point P in this example).92
It should be observed that the proposal for calculating the damage
parameter as a function of the total dissipated energy is theoretically correct
only in the range of the low-cycle fatigue; that is, when a small number of
cycles at relatively large slip values is carried out. In fact, if a high number
of cycles at small slip values is performed, the energy dissipated can be
relatively large, but no significant damage is produced and the reloading
branch reaches the monotonic envelope again. 6 On the other hand, when
limiting our attention to a small number of cycles (< 30), as in the present
study, the energy dissipated for cycles between small slip values is rather
small and the calculated damage, as a consequence, insignificant.
The frictional bond resistance after first unloading {n in Fig. 82)
depends upon the peak value of slip s max , and is related to the value of the
ultimate frictional bond resistance of the corresponding reduced envelope

91
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 85. Relationship 1-25

O First slip reversal O
between the frictional bond
• After N = 10 cycles
resistance Tf(n) and O
corresponding ultimate
frictional bond resistance 10-
T3(N); Ciampi et al.78

10 1-5

(T3 in Fig. 82). The relationship found in the tests is shown in Fig. 85.
However, if cycling is done between fixed values of slip (e.g. between fixed
5 max and smin in Fig. 82), n is reduced more rapidly than the ultimate r 3 of
the corresponding reduced envelope. Therefore, the analytical function in
Fig. 84 is used only for the calculation of the frictional resistance for the
first slip reversal. For subsequent cycles, T{ (e.g. Tf in Fig. 82) is reduced
from this initial value by multiplying it with an additional reduction factor
which depends on the energy dissipated by friction alone. If unloading is
done from a larger slip value than the peak slip in the previous cycle (path
STU), the new frictional bond resistance (rfu) is linearly interpolated
between two values. The first value is related to r^ of the corresponding new
reduced envelope using the analytical function given in Fig. 85, and the
second value is the n reached in the last cycle (Tf in Fig. 82). This
interpolation is done in order to have a smooth transition in the values Tf.
Note that the concept of relating damage to one scalar quantity, like the
normalized dissipated energy, has provided a basis for a relatively easy
generalization of the bond behaviour for random excitations. The bond
model selected can easily be extended to cover bond of bars under
conditions different from those reported herein, such as different bar
diameter, pattern of deformation (lugs), concrete strength, degree of
confinement, effect of transverse pressure, etc. This requires that the
pertinent experimental data necessary for computing the different
parameters, in particular the monotonic envelope, be obtained. If these
are not available, the suggestions given by Eligehausen et al.9 could be
used for choosing the required parameters.
The bond conditions in a joint vary along the embedment length. For an
interior joint, three different regions have been identified (see Fig. 86).
They show differences both in the shape of the monotonic envelopes,
different for positive and negative slip, and in the rate at which degradation
occurs. Of course, there is a gradual variation in the behaviour proceeding
from an unconfined region to a confined one.
The possibility of extending the analytical model shown here for confined
concrete to the unconfined regions, using information presented by
Viwathanatepa et a/.113 has been detailed in Reference 92. The analytical

92
 BOND BETWEEN CONCRETE AND STEEL

Fig. 86. Different bond

regions in an interior joint 113 Region 2
Loading 1
Loading 2

bond model can be generalized by two points. First, instead of only one, two
different monotonic envelopes are specified, one for positive and one for
negative slip values (compare Fig. 86). Secondly, the normalizing energy,
Eo, used in the computation of damage is chosen as the larger one between
EQ and EQ. These quantities define the corresponding areas under the
monotonic envelopes for positive and negative slip values up to slip value
s3. To take into account different rates of damage in the two directions of
loading, the pertinent dissipated energy, E, used for computing the reduced
envelopes, is multiplied by an amplification factor, b, which is different for
the upper and lower curve. The factors b+ and b~ are specified as input
values. Similar rules for the computation of damage apply to the friction
part of the curves.
More details regarding the quantification of the various parameters
involved, and of their distribution along the anchorage length in an interior
beam-column joint, are available.78'92 A comparison between experimental
and analytical results concerning a bar with a bond length of 25 bar
diameters is presented in Fig. 87.
4.4.2.7. Model by Pochanart and Harmon.115 The following rules
define the hysteresis rules for bond under generalized excitations (Fig.
88(c)).
(a) Under monotonic loading, the bond-slip relationship follows the
monotonic envelope.

300- i 18E
12F 15E
9D S - 21E
^-
JL—
A*
* 0- • " ^

1
e
(L-
-6K
21L

18L.
-300- ' 15L

0 -2
i: mm
(a)

Fig. 87. Comparison between (a) experimental and (b) analytical results; Ciampi et al. 78

93
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 88. Bond model for - Monotonic envelope

Strength envelope
generalized excitations: (a)
failure envelope; (b) reduced
envelope; (c) bond stress slip
Unloading and reloading
law; Pochanart and
Harmon"5
Reduced envelope

Frictional stress path

10

Bond stress

Monotonic envelope
Reduced envelope Reduced
envelope

Crushed concrete powder

—Ti Remaining concrete

key
1/4 ex
ex-Jt /;
(b)

(b) The unloading path is very stiff (Fig. 88(c)).

(c) As long as there is no slip to the opposite direction, reloading will be
on the same path as unloading. When the stress reaches the point
where the unloading took place, the bond-slip relationship will still
follow the previous envelope (Fig. 88(c)).
(d) If the reversed load is higher than the developed frictional bond
stress, there will be slip to the opposite direction, and the reduced
envelope based on the remaining concrete key will be used as the
new envelope (Fig. 88(c)).
(e) If reloading takes place after there is slip to the opposite direction,
the reloading path follows the unloading stiffness relationship until
either a reduced envelope or frictional envelope is crossed.
(/) If loading continues to the opposite direction, the loading path will
follow the frictional stress path until it crosses the new envelope and
then will follow the new envelope.

The reduced monotonic envelope is shown in Fig. 88(b). After cyclic

loading, during which the maximum total excursion was ex, the monotonic
envelope is reduced as shown in the figure and is defined by T\ and s,, the
new peak bond stress and corresponding slip value; by r 3 , the new residual
frictional stress value; by s'3 = /'3, the undamaged length of the concrete key;

94
 BOND BETWEEN CONCRETE AND STEEL

and by r{, the current value of the developed frictional stress. Also needed is
an offset value that tells where the strength component of the reduced
envelope begins. In Fig. 88(b), this offset value is set to one-quarter of ex.
T[ is composed of two components: the strength component and the
friction component. The strength component is obtained from Fig. 88(a)
given the total excursion. The friction component is determined from the
friction model shown in Fig. 88(a) by keeping track of the developed
friction lost during load cycling. The slip s\ corresponding to T[ can be
obtained easily since experimental results show that s{ minus the positive
excursion is a constant percentage of (, the undamaged length of the
concrete key which equals ls-ex. r 3 and rf are obtained by keeping track of
the developed friction and the loss of friction due to load cycling.
The deterioration of the frictional bond stress can be determined by
assuming that the frictional bond stress reduces by about 18% from the
frictional bond stress of the previous cycle, assuming no new frictional bond
resistance is developed during the cycle; that is, that the test is performed
under slip control. The deterioration of the bond strength is obtained given
the maximum total excursion and the strength model of Fig. 88(a). These
two rules define the damage law for bond.
The model appears to be able to predict a wide range of bond-slip
behaviour, including the results of both slip-controlled and load-controlled
tests from 100 to 40000 cycles (the limits were given by the authors). The
behaviour under cyclic loading can be predicted reasonably well by
repeated applications of a simple damage law applied to the reduced
envelope. The shifting of the ordinate axis of the bond-slip co-ordinate
system is a visual interpretation of the slip increase without an increase of
bond stress during the frictional bond behaviour; however, the value of
shifting and the shape of the ascending branch may need more specific
studies. The basic difference of the models proposed by Ciampi,
Eligehausen, Botero and Popov and by Pochanart and Harmon is the
definition of the bond damage during cycling.

4.4.3. An application by Filippou1 i4

Filippou recently presented a model describing the hysteretic behaviour of a
single reinforcing bar anchored in an interior beam-column joint. An
extension of the model to exterior joints is relatively straightforward. The
model focuses attention on the stress transfer between reinforcing bars and
surrounding concrete through bond. The portion of single reinforcing bar
between the beam-column interfaces of an interior joint is depicted in Fig.
89(a). The model proposed is as follows.
(a) The bar is subdivided by points A, B, C, D and M into four basic
segments. Points A and D are located at the ends of the bar. Point M
is located at the bar midpoint, while points B and C mark the
transition between the confined and unconfined concrete region at
the end of the pull-out cone. Additional points E and F can be
inserted in order to increase the numerical accuracy of integrations.
(b) The bond stress is always zero at the ends of the bar (points A and D
in Fig. 89(a)). As the magnitude of bar pull-out increases, bond is
gradually destroyed at the ends of the bar. This region of zero bond
stress spreads into the joint with increasing magnitude of bar pull-out
(segment A-AQ in Fig. 89(a)). At the bar end that is pulled out a
linear bond stress distribution is assumed between point Ao and point
B located at the transition between the confined and the unconfined
concrete region at the end of the pull-out cone. At the bar end that is
pushed in, a linear bond stress distribution is assumed between the

95
 RC ELEMENTS UNDER CYCLIC LOADING

AA. B E M F C

(a), (b) Monotonic envelopes

Confined core region e) Reduced envelopes
'/////A 7///// (c) Unloading
(d) Reloading
*- Reinforcing bar (f) First reloading
1 -20
-10
(a)

Fig. 89. Assumed bond stress distribution (a) and hysteretic bond-slip relationship (b); Filippou79'"4

end of the bar, point D, and point C which is located at the transition
between the confined and the unconfined concrete region.
(c) The bond stress distribution in the confined concrete region between
points B and C is assumed piecewise linear and is established
iteratively satisfying the equilibrium and compatibility conditions
using the bond model shown in Fig. 89(b).

80
Finite element model
Proposed model
60

-1
100 200 300 400 500 100 200 300 400 500
x: mm
(c)

Fig. 90. Distribution of (a) steel strain; (b) steel stress; (c) slip and (d) bond stress (in frame joint with increasing
magnitude of end slip); Filippou"4

96
 BOND BETWEEN CONCRETE AND STEEL

A typical hysteretic bond stress-slip curve is illustrated in Fig. 89(b). This

curve describes the stress transfer between reinforcing steel and confined
concrete as a function of the relative slip between the reinforcing bar and
the surrounding concrete. The effect of bar diameter on the monotonic
envelope curve is also shown in Fig. 89(b). During each half-cycle
following the first unloading, the monotonic envelope is reduced based on
the total energy dissipated during previous half-cycles. This establishes the
reduced envelope (curve (e) in Fig. 89(b)). The applied bond stress-slip
curve is practically the same as proposed by Ciampi et a/.78 (Fig. 82) with a
slight modification in the ascending branch during cycling taking a non-
linear increase of bond stress after the frictional branch instead of an abrupt
change reaching the previous maximum slip.
The model is capable of predicting the behaviour of bars in well-confined
interior or exterior joints under cyclic excitation. Experimental and
analytical results are presented in Fig. 90 for comparison.
Note that the solution scheme proposed in this study bypasses the
problem of having to define a bond stress-slip relation for portions of the
reinforcing bar embedded in unconfined concrete. The bond stress-slip
relation is only satisfied at points B, M and C in Fig. 89(a) which lie in the
confined region of the joint. The assumed linear bond distribution between
points A and B at one bar end and points C and D at the other end removes
the need for defining a bond stress-slip relation in the unconfined concrete
regions at the ends of the bar (segments A-B and C-D in Fig. 89(a)). This is
an advantage of the proposed model, since the hysteretic bond stress-slip
relation in unconfined concrete regions is not well established.

4.5 Conclusions Cycles with reversed loading produce degradation of bond strength and
bond stiffness that is more severe than for the same number of load cycles
with unidirectional loading. Degradation primarily depends on the peak slip
in either direction reached previously. Other significant parameters are rib
pattern, concrete strength, confining effects, number of load cycles and peak
value of slip between which the bar is cyclically loaded. Under otherwise
constant conditions, the largest deterioration will occur for full reversals of
slip. The accumulation of bond damage is supposed to be caused by the
progressive growth of microcracks and concrete microcrushing in front of
the protruding lugs.
Although the models simplify the real behaviour, they intend to take into
account parameters which appear to control the behaviour observed in the
experiments, these being: deterioration of related slip, increase of slip under
load-controlled reversals, decrease of bond stress under slip-controlled
reversals, deterioration of frictional bond stress and deterioration of residual
bond strength. Using the described rather sophisticated bond models, the
behaviour of bars embedded in concrete under cyclic excitations can be
predicted with sufficient accuracy for practical purposes.

97
5. Interface behaviour
This Chapter presents a review of the models that can be used in finite
element analysis of RC structures, for the description of concrete-to-
concrete and steel-to-concrete interfaces. The attention is focused on
aggregate interlock and dowel action. Whenever possible, both cyclic and
monotonic behaviours are considered. Whenever necessary, test data are
recalled and commented on.

5.1. Introduction Reinforced concrete elements are characterized by several interface

problems, which concern either the two constituent materials (concrete
and steel) or the concrete itself. With reference to concrete and steel, the
transmission of longitudinal forces and of transverse forces (shear forces) is
possible because of the activation of two potentially efficient interface
mechanisms, namely bond in the former case and dowel action in the latter.
Since the existence of RC elements is based on bond, bond-related
problems (tension stiffening included) require a comprehensive analysis
which is beyond the scope of this section. Consequently, bond will not be
considered here.
With reference to the transmission of shear forces along a well-defined
plane (either a cracked plane or a contact surface between two concrete
subelements cast at different times), a leading role is played by the shear,
flexural and kinking actions locally activated in the bars (Fig. 91 (a) and (b))
because of the relative displacements along the above-mentioned plane: this
type of shear transfer is termed dowel action and will be analysed in the
second part of this Chapter. Of course dowel action is ultimately related to
the relevant bearing forces which develop in the concrete close to the
interface with the steel (Fig. 91 (c)), where the concrete is very confined;
consequently, dowel action combines some important microaspects —
mostly related to interface properties — as well as many macroaspects
related to the specific structural environment. With regard to this, dowel
action can be active in the working load stage, as well as at impending
failure, depending on the specific structural situation.
With reference to plain concrete, the interfaces consist of local

|B| Bending
. Crack
Vdu = 2MJL =

[¥] Shear

\K\ Kinking

[¥] Concrete area influenced by

dowel action

Fig. 91. Dowel action

activated by (a) crack h = 2db
opening and slip; (b)
bending, shearing and
kinking actions; (c)
confinement in concrete M
underneath dowel (c)

98
 INTERFACE BEHAVIOUR

discontinuities such as cracks and surfaces separating two contiguous

subelements (cast at different times and/or having different properties) or
two different structures. Local discontinuities in concrete are so common
that their importance can never be overestimated, all the more so becuase
they are instrumental to the different transfer mechanisms, such as
aggregate interlock for the transmission of shear forces across cracks or
rough interfaces (Fig. 92), and aggregate debonding for the transmission of
tensile forces across rough through cracks (Fig. 93(a)).
Fig. 92. Aggregate interlock Aggregate interlock is generally associated with already formed,
across through crack continuous, mostly linear (at the macrolevel) through cracks, which are
locally rough; consequently, because of the relative displacements at the
interface (crack opening and slip), a shear force can be transferred via the
bearing and overriding of the contact points. Nevertheless, a substantial
interface shear transfer can only be attained if a restraining normal force is
provided along the crack, either by the reinforcing steel or by the boundary
restraints. Should the restraining stiffness be limited, as is usually the case,
crack opening would tend to increase with the slip and with the shear to be
transferred (crack dilatancy).
Aggregate interlock is essentially a material problem, since the behaviour
of a single crack is strictly related to the properties of concrete, such as
maximum aggregate size, aggregate shape and concrete tensile strength; on
this point, should cracking be regular with mostly parallel and closely-
spaced cracks, aggregate interlock constitutive laws could be written in the
form of stress-strain relationships, as for any material. Aggregate interlock
is analysed in the first part of this Chapter.
As to aggregate debonding, this mechanism is responsible for the last part
of the falling branch of the stress-strain curve in direct tension, since the
enucleation of aggregate particles and local friction (Fig. 93(a)) make it
possible to transfer small but not negligible tensile stresses even across open
and through cracks. Aggregate debonding may be better dealt with within a
fracture mechanics approach, although in certain cases with already
developed and stabilized cracks (such as in lightly reinforced beams
subjected to bending, in the second cracked stage) aggregate debonding is a
real interface problem, which requires the formulation of suitable stress-
displacement relationships. Since research on this topic is still in progress,
aggregate debonding will not be analysed in this Chapter. In fact, once
microcracking coalesces into blunt, continuous cracks and new interfaces
form, the tensile response of cracked concrete beams becomes a really
challenging problem, since the study must be extended to crack-width

' ' • . • • • ' • • • - • • . • . ' ' • • I f . " ' ' - • • . . - . • • • .

Friction

Debonding

• • • • - . • ' • ' • •. }
A
Xbonding

••'.'•• i t l Friction

w
1
(a) (b)
Fig. 93. Through crack with: (a) aggregate debonding; (b) partially formed crack with contour lines of longitudinal
displacement in process zone; Cedolin et al."6

99
RC ELEMENTS UNDER CYCLIC LOADING

f,
• - /

1
1 [] (a)

^Bars
(b) (c) (d)

Fig. 94. Dowel action: (a) working load situation; (b) ultimate load situation; (c) dowel action and aggregate
interlock at collapse (vertical and hoop rebars); (d) dowel action in side-by-side elements

values of a few tenths of a millimetre, well beyond the values that are
generally examined with regard to the process zone at the tip of a
developing crack (a few thousandths or hundredths of a millimetre, Fig.
93(b)).
Before examining the general philosophy of the different approaches that
can be adopted in the formulation of the constitutive relationships, it is
necessary to look at the different effects that variable loads have on the
various interface mechanisms. Bond is very often required to withstand
either alternate forces (such as in beam-column joints subjected to seismic
loads) or pulsating forces (such as in a beam subjected to dead weight and to
live loads): consequently, bond degradation due to variable loads is of
paramount importance, with regard to strength and stiffness.
In monolithic structures (which may consist of different parts cast at
different times) dowel action is generally required to transmit pulsating
forces: this situation occurs both in the working load stage (Fig. 94(a)) and
at collapse (Fig. 94(b)). Alternate loads may occur in a few cases, such as in
RC secondary containment shells of nuclear reactors, subjected to seismic
loads and to internal pressurization (Fig. 94(c)): in these cases dowel action
and aggregate interlock are the only means for shear transfer across
horizontal and vertical cracks, if the shell is only orthogonally reinforced
(the former cracks run mostly along the horizontal construction joints, the
latter cracks are due to the hoop stresses). In side-by-side structures
connnected by dowels, such as in RC slabs resting on a deformable soil
(Fig. 94(d)), alternate forces are the rule, rather than the exception.
Since aggregate interlock and dowel action are very often activated by the
same forces and are strictly related to the same crack pattern, the
considerations made so far for dowel action also hold for aggregate
interlock. One must, however, bear in mind that for small crack widths
aggregate interlock dominates over dowel action, while at increasing
interface displacements (as is generally the case with load cycling) the role
of dowel action becomes prevalent, also because interface deterioration
builds up more quickly than concrete deterioration under the dowels.
Finally, the interaction between the two shear transfer mechanisms is even
more complex because of the variable restraint stiffness provided by the
reinforcement crossing the crack (bond-induced tension stiffening
deteriorates under load cycles).
As to aggregate debonding, which is related to highly deteriorated
concrete in tension, only monotonic loads are relevant, because variables

100
 INTERFACE BEHAVIOUR

(c) (d) (e)

Fig. 95. Tests for the study of aggregate interlock: (a) at constant confinement stiffness; (b) at variable confinement
stiffness; (c) at constant crack opening (infinite confinement stiffness); (d) at constant confinement action (zero
confinement stiffness); (e) at constant crack dilatancy

loads tend to immediately destroy the relatively weak mechanisms which

prevent aggregate debonding (chemical adhesion between the largest
aggregate particles and the mortar, and mortar-to-aggregate friction).
Interface problems, and namely aggregate interlock and dowel action,
involve so many different mechanical and geometrical parameters that
mathematical models must be based on sound experimental evidence.
Among the many different tests performed so far, the following have to be
recalled

(a) Aggregate interlock: direct shear tests at constant confinement

stiffness (Fig. 95(a)),117"119 at variable confinement stiffness (Fig.
95(b)), 120121 at constant crack opening (Fig. 95(c)), at constant
confinement action (Fig. 95(d)),122"124 at constant crack dilatancy
(Fig. 95(e)).125
(b) Aggregate debonding: direction tension tests (Fig. 96).
(c) Dowel action: tests on direct-shear specimens,117'1 beam-end
specimens,124 divided-beam specimens 1 1 2 8 and block specimens
129-133
(Fig. 97(a), (b), (c) and (d)).

Empirical equations based on test results were developed in order to

describe the role of a limited number of variables relevant to the design of
RC members (for instance: the bond stress value causing concrete splitting
in a short anchorage, the initial stiffness of aggregate interlock and dowel
action, the maximum mobilized bond stress in an anchorage, the maximum
mobilized shear stress along a reinforced rough interface). Of course

Onset of cracking
• • Test results

Gradual
removal of
Restraining bars restraining
Fig. 96. Example of direct bars
tension test for the study of Fa - ^ ^ F°.
concrete behaviour in
tension, with aggregate
debonding; Giuriani and
Rosati134

101
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 97. Tests for the study of

dowel action: (a) direct shear
test specimen; (b) beam end
test specimen; (c) divided
beam test specimen; (d) block
specimen -Reinforcement
Reaction
-Shear plane

(a)

P/2 P/2

'empirical models' have a limited validity and applications to cases not

covered by the tests are often questionable.
Once the leading parameters have been identified through experimental
evidence and the relative role of each of them has been assessed, rational
models can be worked out on the basis of a few assumptions regarding the
number and the role of the parameters to be introduced, and the morphology
of the physical phenomenon in question. Needless to say, rational models
always include a certain number of constants to be adjusted (once and for
all) by fitting a limited number of basic test results. Last but not least,
rational models should in principle describe one specific resistant
mechanism, from the initial elastic or quasi-elastic stage, up to failure.
Between the two extremes of empirical and fully rational models, other
models are possible, as for instance the so-called 'engineering models'
which are often based on a very simplified but basically sound view of the
phenomenon, and lump together different mechanisms, such as aggregate
interlock and dowel action. In this way, sufficiently accurate equations of
limited validity become available to engineers, who generally need a tool
for a better design and not a research tool, however precise and reliable it
may be.

5.2. Aggregate The shear transfer mechanism based on aggregate interlock has been known
interlock for a long time in its behavioural aspects, owing to the many test results
obtained in the 1960s, 1970s and early 1980s. The different test results shed
light on the many parameters involved, either kinematic (crack opening and
slip, crack dilatancy, initial crack width), mechanical (confinement across
the crack plane, aggregate and matrix strengths) or morphological (overall
and local roughnesses, maximum aggregate size and aggregate type, i.e.
rounded, crushed, light-weight). Unfortunately, scant attention has been
devoted so far to path dependency, either in the displacement field or in the
stress field, to cyclic behaviour and to the three-dimensional aspects of a
crack, while the statistical aspects of aggregate distribution along a crack
have been explored by very few researchers.
Remarkable experimental effort made it possible to understand such basic

102
 INTERFACE BEHAVIOUR

concepts as crack dilatancy (i.e. the coupling between shear stress and crack
opening, which also involves the coupling between normal stress and crack
slip) and shear-confinement interaction, but no comparable effort was
devoted to the formulation of rational constitutive laws until recently,
becuase only recently have the properties of aggregate interlock been
recognized as material properties. In fact, the formulation of suitable
constitutive laws for cracked concrete became necessary after the concept of
smearing the cracks over an entire element was shown to be highly suitable
for finite element analysis.
So far, the greatest attention has been devoted to planar cracks subjected
to monotonic loading, in order to formulate the incremental stiffness matrix
of a crack, which may be generally assumed in the form (see Fig. 98 for the
symbols)
Bat\ fd8n\

where Bnn, Bnt, Btn, Btt are crack stiffness coefficients, which depend on
8n, 8t, o^n, a^t and possibly also other state parameters. The stress-
displacement relations in equation (113) are analogous to the stress-strain
relations of incremental plasticity. Note that crack stiffness matrix is neither
symmetric, nor positive definite, so that crack response tends to be unstable.
However, the response is usually stabilized by the restraint provided by the
reinforcement.
In order to formulate the stiffness coefficients of matrix B, test results at
constant crack opening (Fig. 95(c)) and at constant crack slip should be
available; since tests at constant crack slip have scant inherent meaning,
tests at constant confinement (Fig. 95(d)) have been performed, but these
tests are far less numerous than the tests of constant crack opening and at
constant (or variable) confinement stiffness, the latter being of very limited
use in the formulation of matrix B, since all parameters (cr^, o£ t , 8n, 8t)
evolve during each test.
Due to the relative scarcity of test data, simpler and less general
formulations have been adopted so far, according to the total deformation
theory. Generally the functional relationships have been expressed in a
direct stress-displacement form or in a mixed form
*) (114)
n, fit) (115a)
fin) fit = A(<4, «„) (115b)
Of course the weakness of a total deformation approach is the path-
independency of the response, while path-dependency must normally be
expected of inelastic behaviour.
In finite element analysis, once an element becomes cracked, it is
convenient to adopt a 'smeared crack' approach in order to introduce the
properties of cracks, aggregate interlock included. By replacing the actual

Fig. 98. Main variables of

planar crack

103
RC ELEMENTS UNDER CYCLIC LOADING

Bars
py sin 0

(a) (b)

Fig. 99. Cracked reinforced concrete Fig. 100. Equivalent steel ratio pn

interface displacements with averaged (smeared) strains, the constitutive

laws of the cracks become the laws of a 'new material' (cracked concrete)
and the incremental stiffness matrix D CR can be worked out. Of course, the
smeared crack approach applies in principle to the analysis of regularly
cracked fields, where the concept of 'average crack spacing', s, applies (Fig.
99). In this case cracked concrete constitutive laws can be expressed as
follows
firm 0 Bnt
= s 0 0 0 l detCR \ i.e. dac = D C R de C R (116)
Btn 0 Btt v nt /
where de™ = d6n/s, de™ = d7° R /2 = dSt/2s, de™ = 0 and 5(0°, a3) is
the average crack spacing attained at the stress condition (a*1, 0s). The
spacing may be considered as a local variable, related to a number of
factors, such as concrete tensile strength; bond-slip relation, which in turn
depends on the type of bars and on concrete grade; or geometrical
parameters (bar diameter, bar spacing and concrete cover).
The generally accepted formulation for the average crack spacing in a
stabilized crack pattern (where no more cracks can be formed) is

(117)

where c is the concrete cover, <f> is the bar diameter and p is the geometrical
steel ratio; k\, k2 and ^3 are constants, related to bond efficiency, type of
deformed bars and type of loading.
Equation (117) is valid only for one array of parallel bars at right angles
to the crack pattern, which is seldom the case, since generally two arrays of
bars are present, neither of them at right angles to the cracks (Fig. 100(a)).
In these cases an 'equivalent' steel ratio, pn, (Fig. 100(b)) must be worked
out and introduced into equation (117). With regard to this, a criterion for
establishing the equivalence between pn and (px, py) should be introduced,
but so far nothing is available, except in the limit case with completely
yielded bars; in this case a simple equilibrium condition leads to the
expression for pn
2
9 + pyfsySm20 i.e. pn = pxcos20 + pysin20 (118)

where 6 is the angular deviation between the x-bars and the axis at right
angles to the cracks.
In order to work out the stiffness matrix, Dc, of the concrete as a whole
(cracked concrete + solid concrete between the cracks, called in the

104
 INTERFACE BEHAVIOUR

following 'plain concrete'), it is necessary to add the flexibility of solid

concrete, F , to the flexibility matrix of cracked concrete, F C R
Fc _ (119)
where

(120)
Since the stresses and the strains in the solid concrete are usually small, a
of

linear elastic behaviour can often be assumed for the formulation of F

1 -v
1
0
F s c = - 1 -Vc 1 0 or
0 0 1+

1 — Vc 0
F sc
= 1 1 0 (121)
-Vc
3(1 - 2 v c
0 0 3(1 - 2vc)Kc I2G
where Kc = Ec/[3(\ - 2vc)].
Should the simplification of linear elasticity be dropped, one could
determine the flexibility matrix F s c by differentiating some total stress-
strain relations for concrete, or just assume incremental isotropy and
evaluate the tangent shear modulus, GCT, and the bulk modulus (or modulus
of volume expansion), KCT, as functions of octahedral normal and shear
strains, using some of the models presented in Chapter 1.
Once the tangent stiffness of plain concrete has been evaluated, the
influence of the reinforcement can be introduced, by adding the tangent
stiffness matrix of the reinforcement, Ds, to Dc. Because of steel-to-concrete
compatibility at the macrolevel, the average strains of the reinforcment are
assumed to be equal to those of plain concrete; consequently, the stress-
strain relations for the reinforcement in the crack reference system (n,t)
(Fig. 98) can be written as follows
dcf = D s de where de = de8 = dec (122)
Note that the effects of tension stiffening are taken care of in the stress-
strain relation of the steel of each array, and ultimately are implied in the
formulation of the stiffness coefficients Df-.
Since crack orientation does not generally coincide with the orientation of
either of the bar arrays, the matrix Ds must be obtained by transformation
P2 Q2 PQ
D*R,, R, = Q2 P2 -PQ
i=l,2
-2PQ 2PQ P2-Q2_
""• B / Y f,
0 0
0 0 0 (123)
0 0 0
with Q = sinl?,, P = cos0(, / = 1 for x-bars (0| = 9, Fig. 100(a)) and / = 2
for y-bars (02 = \ir + 0).
OABYR: with tension stiffening
OYR: without tension stiffening
In D^, px and py are the geometrical steel ratios of the two arrays, and
Fig. 101. Stress-strain law of Es(ex), £s(cy) are the tangential stiffnesses of the steel, tension stiffening
steel with and without tension included (Fig. 101). Note that Es(ej) is never smaller than the elastic
stiffening Young's modulus of the steel, and can even be a few times larger.

105
RC ELEMENTS UNDER CYCLIC LOADING

The response of the material as a whole (reinforcement + plain concrete =

cracked reinforced concrete) is represented by the vector of the stress
increments, which is obtained by adding the vectors of the concrete and of
the reinforcement
do- = d(7c + da8, da = (D c + D s )de, i.e. da = Dde (124)
where D is the tangential stiffness matrix of cracked reinforced concrete.
The matrix D must be further transformed according to the element
coordinates.
Before reviewing the mathematical models available at the moment, it
may be useful to remember that these models have been, or are being,
developed with the primary purpose of providing the necessary information
to be used in non-linear finite element computer programs for the analysis
of RC structures. Nevertheless, once the relationships among the most
important parameters become available and are shown to be reliable, useful
information can filter into everyday design activity, which is not necessarily
associated with large finite element linear or non-linear programs. In fact,
several important parameters may be easily evaluated either through a direct
use of most of the constitutive laws or by means of simple derivations.
Among others, the following parameters can be quoted: shear transfer
capacity (r max ) of a crack for a given crack width, initial shear stiffness and
initial friction coefficient of a crack for a given crack width or a given
restraint stiffness, shear stiffness of a crack after a few load cycles, shear
retention factor (a = ^CRy^ELASTic^ a n ( j t a n g e n t shear modulus of cracked
concrete as a function of crack width, shear stress-to-confinement stress
envelopes either at constant crack width or at constant restraint stiffness.
Summing up, most of the models are sufficiently easy to handle to be of
interest, not only to researchers and analysts, but also to professional
engineers. In the following, each mathematical model will be introduced
with a specific denomination in order to make it easier to quote the model in
the text. An asterix will mark those denominations which have been
introduced here, but which do not appear in the original papers.

5.2.1 General roughness model* ~ ' ~

The rough and irregular shape of a crack, which is stochastic in nature, is
considered as a superposition of high frequency components upon low
frequency ones, the latter being determined by the relative location and size
of the largest aggregate particles ('general roughness'), the former being
related to the smallest particles and to the mortar asperities protruding from
the crack surfaces ('local roughness') (Fig. 102).
Because of crack opening and slip, the two faces of a crack come into
contact along some finite areas. These areas are considered as idealized
'contact points' along the general roughness, which is assumed to retain its
shape during the relative movement at the interface (perfect stiffness of the
general roughness): at each contact point frictional forces develop because

w Upper
" block

Fig. 102. General and local

roughness of crack faces: 8n
+ 8°n = 6y(x) + 8°y; 8, = 8x(x)

106
 INTERFACE BEHAVIOUR

Fig. 103. (a) Cracked panel

Cycle 1
in uniaxial tension and shear;
(b) model prediction of cyclic
response

(a) (b)

of either material internal friction and/or local interface roughness, the latter
being subjected to degradation due to possible cyclic loading. The shape of
the general roughness is introduced by means of suitable, prefixed functions
of the coordinate x, such as arcs of second order parabolae.
In its original form, the general roughness model is only a 'functional'
model and not a 'predictive' model, since the many geometrical parameters
involved are a priori unknown and a posteriori unmeasurable. These
parameters may be determined by means of a multiple regression analysis,
starting from a sufficiently large data base. As regards qualitative predic-
tions, the model seems to work well (Fig. 103)135'136 and to be suitable to
cyclic loading.
With reference to the shear stiffness of a reinforced concrete block
cracked in the direction x (Fig. 103(a)), the following expression can be
derived for the relationship between the shear stress and the strains
1 C\
7yx (125)
1 1
+ _
UC Cty
where: Gc = elastic shear modulus of the concrete; c\ - constant; ay, (3y =
functions of the extensional and dowel stiffnesses of the cracks, and of crack
spacing. Expressions for G011 (shear secant modulus of cracked concrete)
are given by Leombruni et a/.136 Equations (125) was worked out by means
of a multiple regression analysis performed on test data by Gergely,
Jimenez-Perez, Laible and White at Cornell University.137'138
l39 142
5.2.2. Rough crack model ~
The rough crack model is based on some general properties that are to be
expected for rough cracked surfaces, and that can be deduced by
considering a few simple micromechanical models (Fig. 104(a) and (b)):
the interface stresses are assumed to be mostly dependent on the
displacement ratio r = 6t/6n (wedge effect); for small crack widths, the
confinement stress vanishes (aggregate debonding); for large crack widths,
the stresses tend to vanish due to concrete microcracking and crushing, and
eventually to the loss of contact between the crack faces. The general
roughness is disregarded and only the local roughness is introduced;
consequently, since the number of contact points along the crack interface
may be regarded as infinite, the resulting stress-displacement relations may
be considered continuous and smooth.
The interface stresses are formulated in the following way
<t = Fl(6n/dtk)F2(r) (126a)

107
RC ELEMENTS UNDER CYCLIC LOADING

where r = 6t/6a with 6t, 6n crack slip and opening.

By optimizing the fits of Paulay and Loeber's test data121 (constant crack
opening, infinite restraint stiffness) and of Daschner and Kupfer's test
data122 (constant confinement stress, zero restraint stiffness), and by
assuming for crack roughness the aggregate grading suggested by Fuller's
curve, the following equations were formulated

+a2\r\
<t

x0-25
(126b)

where r 0 = 0-25/c'; a\, a2, «3 = constants related to /c'; dA = maximum

aggregate size.
By derivation of the relationship between o^t and the displacements, both
the 'crack shear stiffness', KCR, and the 'equivalent shear modulus', GCR, of
cracked concrete can be worked out (see also the curves of Fig. 105)142

GCR = = sK'CR (127)

where r u = r o [l - \/(28n/d&)].
The rough crack model can also be extended to cyclic loads, provided
that suitable stress displacement relationships are formulated both for
reloading and unloading.
As regards pulsating loads, the tests at constant crack opening by Paulay
and Loeber121 show that the shear stiffness of a crack is actually greater in
the following cycles than in the first cycle, because of some contact between
the general roughness of the crack faces (Fig. 104(a)), after the deterioration
of the local roughness in the first cycle. Though different from the first
cycle, the following cycles are practically coincident (at least up to / = 20, i
being the number of load repetitions).
Since test data are really limited and even non-existent for confinement, a
suitable and reliable constitutive law can hardly be formulated. A tentative
formulation for pulsating loads is presented by Gambarova;140 for loading
and reloading the constitutive laws are very similar to equation (126), while
completely different formulations are presented for unloading. Typical
hysteretic cycles under pulsating loads are shown in Fig. 106.

Aggregate particles

Crushed mortar
Fig. 104. Crack morphology
(F = local interface forces):
(a) wedge effect; (b)
aggregate debonding; (c)
microc racking and concrete
crushing

108
 INTERFACE BEHAVIOUR

Fig. 105. Equivalent tangent

shear modulus of cracked 12-103 6-103
/ \
concrete: (a) at constant
shear stress/crack opening E 4 / \
= 0-25 mm
ratio; (b) at constant crack
opening, according to 1 / X

50\
Equation (127); f'c = 31 MPa,
da (maximum aggregate size)
i 2

• - —

= 19 mm n
0-5 10 1-5 20
r = d,ldn
(b)

ru = 7-83

17
A <°Sr>" = 5-80
An = 0-125 mm

I -'A • I J\ dn = constant
I / /
/
(d,)Ud,y,
Shear displacement d,
(a)

121
Fig. 106. (a) Typical hysteretic cycles at constant crack opening under pulsating loads (b) fit of test results
with the loading and unloading equation presented by Gambarova'40; f'c = 31 MPa, da (maximum aggregate size)
= 19 mm

5.2.3. Two-phase model' »8.»9.»«,i45

The two-phase model is based on the fundamental assumption that concrete
is a two-phase system, consisting of very stiff inclusions (medium and large
aggregate particles, d& > 0-25 mm) embedded in a very soft matrix
(hardened cement paste, fine aggregate particles included). Since the
plastic deformations of the matrix are expected to predominate over the
elastic deformations, the stress-strain relation of the matrix may be regarded
as rigid-plastic (Fig. 107(a)).
As regards concrete and interface morphology, since in normal concretes
(neither high strength nor light-weight concretes) the weakest link of the
system is the interface between the two phases, the cracks run along the
surface of the aggregate particles and through the matrix; consequently,
cracks are rough due to the particles projecting from the two faces and the
'microroughness' of a crack dominates over the 'macroroughness', since
aggregate maximum dimensions are far larger than crack width (Fig.
107(b)). Finally, the inclusions and the matrix are assumed to be in partial
contact because of relative displacements at the crack interface (Fig. 107(c)
and (d)) and the contact areas are at the onset of sliding; a Fuller's type
aggregate distribution is adopted for the asperities along a cracked plane.
By writing the equilibrium conditions in the crack reference system («, t)
and by evaluating the 'averaged' contact areas An, At for a unit surface of
the crack plane (An = Sa n , At = T.at, Fig. 107(c)), the stress-displacement
relations can be formulated as follows

109
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 107. (a) Two-phase

model"9: rigid-plastic
stress-strain relation
assumed for the matrix
(cement mortar); (b)
simplified local roughness of
crack plane; (c) contact area
between matrix and
aggregates; (d) stresses at
matrix-aggregate contact

(c)

<„ = - (128)
in which A, and An are very complex functions of the relative displacements
6n, 6t), maximum aggregate size (da) and relative aggregate volume (pk =
ratio between the total volume of the aggregates and the total volume of the
concrete: in most cases, the value 0 • 75 was adopted by the authors)" 8 " 119
Ax and An have an integral formulation, and the integrations have to be
performed numerically, with respect to the aggregate diameter which is
considered as a variable related to the statistics of the contact points at the
interface and to aggregate grading. A linearization of equation (128) is
proposed by Walraven et al. The agreement between the values predicted by
the two-phase model and the test results is fairly good (Fig. 109).
The extension of the two-phase model to cyclic loads (either pulsating or
alternate) and to sustained loads has been recently attempted144 taking
advantage of the inherent capacity of the model to follow and to quantify
the damage process (Fig. 108): for any couple of values 6n, St), both the
contact areas and the contact forces can be directly evaluated (Fig. 108(a)).
The fit of some of Laible's test results is presented in Fig. 109:138 a
(a) reduction of the friction coefficient and of the contact areas had to be
introduced because load recycling makes crack interfaces smoother,
particularly in concretes with poor aggregate strength.

5.2.4. Dilatancy factor model123

T = O%,
As in the previous two models, only planar cracks and monotonic load
histories are considered. As well as the four main parameters (crack opening
and slip, shear and confinement stresses) concrete compressive strength and
initial crack opening are also introduced. The development of the model

r= IV = 1 N = 15
\
E
10:
//
/'
| h 10 10-
05: / 05- E 0 5-
-10
/,'J--
-0-5
/- "b-5 ' ' Vo 1 5 10 15

f-
• / / ' -0-5
d
' -0-5- - -0-5-
N

f/ \ -
(b)
-10 - Test
-10- -10-
/ -- - Calculation
Fig. 108. (a) Definition of
new points along the crack
interface; (b) various stages Fig. 109. Fit ofLaible et al. 's test results (RC specimens with constant confinement
of the crack damaging stiffness)138; f'c = 26MPa, da = 32 mm, /x = 0-20, apu = 30MPa, £° = 0-75mm
process (initial crack opening)

110
 INTERFACE BEHAVIOUR

was based on a set of measurements regarding the shear stress as a function

of the slip and as a function of crack opening (or dilation), at constant
confinement (four values were considered for confinement).
The functional relationship among the four main parameters
6n, 6t, a^, alt is expressed as
o-cnt = V(SU (TU 6n = W(8t, aU (129)

which can be written in incremental form

dW .„ dW
(130)

Introducing the following 'service' variables: a (shear stiffness) =

dV/dSu (3 (dilatancy factor) = -dW/dSt, /z (friction coefficient) =
dV/da^n, \IR (normal compliance) = dW/da^, the incremental equation
(18) becomes

where D CR is the tangent stiffness matrix of cracked concrete. In principle

the parameters a, (3, (j., R are functions of the stresses and of the
displacements, as well as of other physical and mechanical properties of the
material.
As regards the first constitutive law, o^t = V(St, o^n), the following
empirical formulation was proposed by the authors

< t = Ko6t (\ + ^ \ exp(-Mt) (132)

where Ko = initial shear stiffness, function of 6° (initial crack width) and / c '
(concrete compressive strength); b = empirical parameter, function of Ko, f[
and CT^JJ (confinement stress).
As for the second constitutive law, 6n = W(8t, o^n), the formulation is
still empirical and based on the authors' test data

(3 = -^jr- = a\ exp[—a2 — «3(o"nn)~a4] (133)

where a\, a2, a^, a4 are constants.

According to equation (133), the dilatancy factor /? is practically
constant, but this is an approximate assumption since f3 tends to be large at
small crack openings and even to have a singularity for 6a = 0.
The dilatancy factor model is shown to be able to describe different
loading histories reasonably well, both at infinite and finite transverse
stiffness (see Fig. 110). No extensions to cyclic loads have been
attempted so far.

5.2.5. Incremental slip-confinement model*146

The framework of the model is close to those of the rough crack model and
of the dilatancy factor model, but two of the crack stiffness matrix
coefficients are formulated in a direct way, making the model path-
dependent. In order to work out the coefficients of the crack stiffness
matrix, the authors focus their attention on crack slip and confinement
6t = A(acnt,Sa) acDn = F(acat,SD) (134)
which can be written in incremental form

ill
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 110. Comparison of the

fc = 35 MPa
dilatancy factor model with 40 da = 15 mm Jimenez-Perez ef a/.
other existing models and
Houde and Mirza
with test data137 Fenwick
Oivakar ef a!.
Gambarova ef al.

025 0 50 0-75 100 1-25

Initial crack width: mm

dA dF
dSt = (135)

Introducing the following 'service' variables

kt (crack shear stiffness) = I ——

dF
kn (crack normal stiffness) = ——

ii{ (friction coefficient) = — ( ——

-i
= -rr^- (dilatancy ratios)
d6Q

the incremental equation (135) becomes

j[dot
daC
nt\ A
\daj or

1 -
WJ ~ 1 d6n
(136)

where ^ is a non-dimensional parameter £ = ^f/3d kn/kt-

The four variables A:t, kn, /if and /?d are in principle state variables, since
they depend on the stresses and on the displacements, and possibly on
concrete properties and crack roughness. For these variables the empirical
formulations are proposed
, 6n, J (137)
(138)
Mf = ) (139)
/c) (140)

Numerical tests at infinite, finite and zero restraint stiffness were

performed, generally in close agreement with the predictions of the rough
crack model (Fig. 111).

112
 INTERFACE BEHAVIOUR

o Numerical [139]
• Tests [121, 138, 146]

0-2 04 0-6 0-5 10 1-5

Slip <V Tim Crack width dn: mm
to
(b) (c)

Fig. 111. Incremental slip-confinement model:146 (a) shear stress; (b) shear stiffness plotted against crack slip
(shear displacement); (c) frictional coefficient plotted against crack opening (normal displacement)

Comparisons with other test results seem to be extremely successful, but

the physical consistency of the model should be demonstrated in some way.
The extension to cyclic loads has not been attempted so far.

5.2.6. Contact density model U1

The contact density model is based on the assumption that the crack surface
consists of a set of differently oriented contact planes (Fig. 112(a)), whose
directions are characterized by a frequency function p(Q) (contact density
function) and which are subjected to a mostly compressive stress a (contact
stress) accompanied by friction (which is neglected in the applications
shown by Li and Maekawa).147
With reference to a unit area of crack surface, the shear and normal
stresses across a crack can be formulated as follows

a"nt — j At a k p sin© d©,

-i: Atak p cos© d©, for 0 > 0 (141)

= 39 MPa 10 y.
= 0-3 mm
<>„

Al
a Jw It

I 5
\d -0-4 0
' '/ ' 0-2 04
A,', mm
- 5 - — Experiment
— Analysis

-10-
(a) (b)

1 1 1

fc = 23 MPa bn = 1-Ov^

Analysis 4 ' - 4
Fig. 112. Contact density
model — cyclic shear
- - unloading
— loading j t A0
0
1
1
- 2
transfer at constant crack 1
1 d

width:147 (a) contact planes o ol <

o/i ' Q
and contact stresses; (b) 0-4 0-8 0 0-4 0-8 1-2
alternate cycles; (c) pulsating dt~. mm ,: mm
cycles (c)

113
RC ELEMENTS UNDER CYCLIC LOADING

tT 0 for
nt=°'nn = © < 0, (no local contact) (142)
where At is the total area corresponding to the nominal unit area of crack
surface (A, is considered as an interface — or material — constant) and k is
a reduction factor which introduces the effects of crack displacements on
the contact area; a and k are functions of crack displacements 6n and 6t and
of contact orientation 8 ; (At p(Q)) is the effective contact area along the
planes whose orientation falls in the interval 6 ± dQ/2.
Assuming that friction is negligible ( 6 is independent of crack dis-
placements), that crack displacements are very small with respect to crack
roughness (k = 1), adopting a reasonable formulation for />(= 0-5 cos 8 )
and for/l t (= 20 500f^'^/E c ), giving a a simple formulation based on crack
displacements via geometrical considerations, equation (141) can be worked
out, as well as the crack stiffness matrix B (equation (113)), by derivation of
the same equation (141). By integration, according to any prefixed
displacement path, the crack response curves (a^v o£n) can be obtained.
In this way, several test results (either at constant or at variable crack width)
were successfully fitted, under monotonic as well as cyclic loads (pulsating
or alternate loads, Fig. 112(b) and (c)).
The proposed model is very attractive for its ingenuity and simplicity;
some basic ideas go back to the microplane model148 (contact planes,
contact stresses) and the two-phase model (plastic behaviour in the contact
zones at the interface).

5.2.7. Generalized rough crack model*149

This model is a really general and sophisticated approach developed for the
analytical description of shear transfer via aggregate interlock, and for the
modification of concrete stiffness matrix in order to include the effects of
cracking, besides the strains in the solid (uncracked) concrete between the
cracks. The proposed approach can in principle account for crack
degradation under cyclic loading. In order to avoid possible confusion
with the rough crack model developed by Bazant and others (see section
2.2.), the denomination 'generalized rough crack model' is used here.
The constitutive relationship of the concrete (cracked + solid), da = D de,
(where D is the tangent stiffness matrix, cracking included) is worked out
with either a flexibility or a stiffness approach, based on the few funda-
mental assumptions that
(a) any strain increment can be divided into an uncracked strain and an
added crack strain
(b) the crack strain can be divided into as many component vectors as
there are crack orientations, each component being associated with a
'flow' direction and with a generalized equivalent strain
(c) the stress increments are related to the uncracked strain increments
through the uncracked constitutive relationships
(d) the increments of the generalized stresses across a crack (micro-
stresses) are related to the actual stress increments (macrostresses) by
a suitable transformation
(e) the increments of the generalized crack strains are related to the
increments of the generalized stresses by the tangent flexibility and
stiffness matrices of the crack system which may or may not be
diagonal.
In order to work out the coefficients of the matrix D, it is necessary to
adopt one of the available crack theories which lead to the development of
different mathematical models. Since most of the models regard planar

114
 INTERFACE BEHAVIOUR

(a) (b) (c)

Fig. 113. Generalized rough crack model — idealizations of crack morphology:149 (a) castellated crack; (b) rigid
saw-tooth crack; (c) deformable saw-tooth crack

cracks, the generalized rough crack model has been limited so far to plane
problems, though extensions to three-dimensional behaviour is possible.
Since this is based on crack smearing, the transformation from crack
displacements (which appear in the constitutive laws of the models) to
'equivalent' strains requires the introduction of a 'reference' length, le,
which is closely related to crack spacing.
Constitutive crack models (based on constitutive relationships), as well as
physical models (based on direct modellization of a rough interface (Fig.
113)) can be introduced into the generalized rough crack model: so far, the
so-called deformable tooth model (Fig. 113(c)) has been adopted, while the
solid concrete has been modelled by means of linear elasticity or by the
endochronic theory. The stiffness approach was used by Riggs and Powell
to calculate the matrix D. 149
In the deformable tooth model the initial shape of the crack is a saw-tooth
and under loading a quasi-static sliding of the two surfaces occurs, with
friction and surface deformations (one face of the crack is assumed to be
deformable, while the other is assumed to be stiff). Since the model is
characterized by nine constants, which depend on the mechanical and
geometrical parameters, and the determination of the crack state is complex
in itself, the implementation of the model in a pre-existing finite element
program proved to be rather difficult. However, Paulay and Loeber's test
results are well fitted, while the fitting of cracked concrete cyclic behaviour,
although promising, requires further improvements (Fig. 114).

5.2.8. Microplane model148

The concrete, either solid or cracked, is considered as a system of randomly
oriented planes — microplanes — (Fig. 115(a) and (b)) which are
characterized by a uniaxial normal stress-normal strain law; the shear
stiffness of the microplanes is disregarded. Although originally developed for
the description of the non-linear behaviour and fracture of concrete and rocks

Slip: mm Slip: mm
0-5 10 1-5 0 05 10 1-5
1 i i
- 20
S. — Experiment
— Analysis 15 |
_
1 A
i i
\A i
10 2
^5
1
/
Fig. 114. Cracked concrete
II

/ hy i
i - 0-5
cyclic response: (a)
analytical results for 15
it r
002 004 006 002 004 006
cycles; (b) fit of Laible et Slip: in Slip: in
al. 's test results'38 (a) (b)

115
 RC ELEMENTS UNDER CYCLIC LOADING

Fig. 115. Microplane

model:148 (a) concrete
microstructure; (b)
microplanes; (c) constitutive
law of a microplane; (d)
fitting of Paulay and
Loeber's test results121

Microplanes
Paulay and Loeber's tests
—- 75% of test results
0-4 0-8
<V mm
(c) (d)

subjected to triaxial or biaxial stresses, 150151 the microplane model has been
applied also to describe shear transfer across blunt cracks, by modelling the
cracks as 'crack bands', with a width equal to maximum aggregate size.148
Within a crack band, the microplanes stand for the thin mortar layers
(Fig. 115(a)) which transmit either compressive stresses (contact planes)
between two contiguous aggregate particles, or tensile stresses (planes with
strain softening); Bazant and Gambarova give the microplane constitutive
law a hypoelastic formulation, with different branches for loading,
unloading and reloading in compression and tension148
a = Fie )e (143)
The tangent stiffness matrix of the microplane system is obtained by
means of an integration performed over a unit-sphere (spatial cracks) or
over a unit-circle (planar cracks), representing all possible microplane
orientations in the space or at right angles to the reference plane (Fig.
115(b)). In the latter case
2
(144)

with aijkm = nitijtikUm, where n are the director cosines of each microplane
(1,2 = references axes, x and y in Fig. 115(b)).
Simple asymptotic or exponential laws have been adopted for the
microplane constitutive relationship (Xn(en), both for loading and unloading
(Fig. 115(c)). The model can describe the path-dependency of the concrete
(even if the constitutive relationship is of the total stress/total strain type)
and the reorientation of the principal stresses with respect to principal
strains.

116
 INTERFACE BEHAVIOUR

Fig. 116. Comparison

between experimental and
computational results for
tests with constant crack
width (6n = 0-51 mm): (a)
shear stress-confinement
stress envelopes with dashed
area enclosing 75% of
Paulay and Loeber's test
results (Sn = 013-
0-51 mm);121 (b) shear
stress-crack slip and crack
slip-confinement stress 0-2 0-4 0-6 0-8
d,: mm
curves (b)

RCB : Rough crack model

(Bazantera/.139)

RCG : Rough crack model

(Gambarovaefa/.141)

T P : Two-phase model
(Walravenefa/. 118120 )

CD : Contact density model

Most available test results have been successfully fitted (see for instance
Fig. 115(d)), and the extension to cyclic loads — though not yet attempted
— may be easily achieved, because of the inherent simplicity of the model,
which is based on a single, many-branched relationship, valid for each
microplane (equation (143)).
Recently the shear stiffness of the microplanes has been introduced for a
better description of concrete non-linear behaviour, and the model has been
extended to cyclic loading (cyclic compression).152
Three of the aforementioned models have very recently153 been intro-
duced into the DIANA computer code and a very comprehensive comparison
has been carried out in order to assess the performance of the models with
regard to the description of a few tests performed at constant crack opening
and at constant restraining stiffness, and to the numerical stability once the
models are implemented in a finite element program.
The study regards the rough crack model in two versions,139'141 the two-
phase model and its linearization,118119 and the contact density model.147
The three models perform fairly well, on the whole, but the contact density
model seems to have the edge, particularly from the numerical point of
view. An example of the thorough analysis performed by the Dutch
researchers is shown in Fig. 116.

5.2.9. Simplified models for plain concrete

With reference to plain cracked concrete, the simplest way to describe the
shear-stiffness deterioration and introduce the contribution of aggregate
interlock is by reducing the shear modulus Gc of uncracked concrete, by
means of a suitable shear-retention factor a (JR = aGc with a < 1. The
development of the various formulations given to the shear-retention factor
has followed the continuous improvement and the ever increasing
sophistication of finite element analysis. Nevertheless, any formulation —
no matter how complicated — has to be regarded as a very simplified and
empirical representation of the shear behaviour of cracked concrete since

117
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 117. Shear-retention

factor a = GCR/GC: (a),(b)
shear-stiff cracks; (c),(d)
shear-deformable cracks; (c)
two-step; (d) variable
formulations for a

(a) (b) (c) (d)

such relevant aspects as crack roughness and aggregate shape and size
(which are principally responsible for crack dilatancy), concrete strength,
aggregate-to-mortar friction are ignored.
Furthermore, there is general agreement among researchers that in many
structural problems the value of the shear-retention factor (either constant or
variable) is not critical, provided that a sufficiently large value is adopted
(a > 0 • 1). Otherwise the system of equations in finite element analysis
becomes ill-conditioned and the iterative solution has to be stopped because
of numerical instability (divergence). Whether divergence is just a numer-
ical problem or has a physical meaning is still an important question to be
answered, as observed by Rots et al.64
In the earliest period (1970-1975) the factor a was generally given a
constant value (between 0, smooth cracks, and 1, fully locked-up cracks
(Fig. 117(a)); 0 - 4 - 0 - 5 in most cases). Later on, a two-step or a
continuously decreasing formulation was adopted (see for instance Cedolin
and Dei Poli154), after crack opening was recognized to be a major factor in
reducing crack shear-stiffness (Fig. 117(d)).
In most formulations the crack opening is replaced by the equivalent
strain e ^ ( = Sn/s, where s is the crack spacing) at right angles to the cracks.
Very often the solid concrete contribution to normal strain is neglected:
e m = e^ 1 . For the literature on the shear-retention factor see Walraven and
Reinhardt,119 Rots et al.,64 Walraven and Keuser.155
Among the most recent formulations,155'156 can be quoted

with k = 4447 (145a)

a— (145b)
k = F[da, dm,i 7nt]
a = exp ( - * 4 ) > 0 • 7 with k = 140 000 (145c)
Rational but generally more complex formulations can be obtained by
deriving the constitutive laws of the models presented in the previous
sections 2.2 to 2.5. In this case a can be formulated as follows155 (see
equation (116))
-i
a = +1 (145d)

Finally, no formulation of a including cyclic effects has been attempted so

far, but some useful information comes from the tests by Tassios and
Vintzeleou,157 on shear transfer by means of friction, at a given confinement
stress (rough cracks)

-1 ir
7fr,n = 0 • 002(n -

118
 INTERFACE BEHAVIOUR

where
= cycle number
<$t = factional response and crack slip at the end of the loading
branch of the nth cycle
= frictional response in the first cycle, for a crack slip equal to
= confinement stress acting on crack interface
6tn = crack slip corresponding to the maximum mobilized shear
stress under monotonic actions.
By definition the secant shear moduli are

7^R = / '? , GfR = /c \ with 5 = crack spacing

s/ \s
and the following expression can be worked out for an/a\

^ = 1 - (145e)
QCR
1
For n = 5-15, aclf'c = 0-25-0-5, St/6tu =0-5-1-5 the values of an/a\ fall
within the range 0-4-0-8, according to equation (145e)

5.2.10. Engineering models for reinforced concrete

By the term 'engineering models' we mean those models which do not
describe each particular resistant mechanism in its very nature, but consist
of design-oriented equations resulting from a macrolevel description of one
or more resistant mechanisms, whose effects are often lumped together, on
the basis of equilibrium and compatibility conditions valid 'on average'.
Five different engineering models are briefly presented in the following; the
formulation of these models spans the period from 1970 to 1987.
5.2.10.1. lsenberg and Adham's model}59' The two-material system
consisting of two arrays of perpendicular bars and concrete is replaced by an
equivalent orthotropic material as soon as the principal tensile stress or both
principal stresses reach the strength in tension and concrete cracks (1, 2 =
crack directions (Fig. 118). At this stage the amounts of steel are resolved
into the amounts Asl and As2 parallel and perpendicular to the cracking
directions, and the mechanical properties (such as the stiffnesses Ex, E2 and
G\i = G0*) depend on the orthotropy in the directions 1, 2. The model
accounts for cracks, progressive failure of bond (with reference to the
resolved amounts of steel), concrete multiaxial behaviour (transverse
pressure included, direction 3), crack-induced orthotropy (fixed cracking).
In the reference system of the orthotropy, the formulation of the tangent
shear modulus is given by equation (146)

-cl Esp\
(1-A.) +•
2(l+v 2 1 ) 2(1+v 8 )

+ (1- + • (146)
Cracked
sections
2(l+v 1 2 ) 2(1+v 8 )
where Ai, A2 (bond-slip moduli) are the ratios of the debonded lengths of
Fig. 118. Orientation of the rebars to crack spacing (Ai and A2 depend on crack width); Eci and Ec2
orthotropic axes 1, 2, 3, are the tangent moduli of concrete; p\ and p2 are the 'resolved' steel ratios;
defined by initial
cracking;1'8 x, y, z global the reinforcement is suppoed to be elastic. No comparison with test data is
coordinate system presented. No extension to cyclic loading is performed. Aggregate interlock

119
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 119. (a) Typical RC

element; (b) typical shear
strain-shear stress curves
(p, = 1%; p2 = 1-5%;
Pi =P4=P) 0-004
P = O^--
• 0 0 5 ^ . ——-

id
— —
—

1 2 3
Applied shear stress: MPa

Rebars
(b)

and dowel action are neglected. In most practical cases the value of a varies
between 0-75 and 1.
5.2.10.2. Duchon's model.159 Reinforced concrete elements subjected
to membrane forces are modelled as a system of vertical, horizontal and
two-way inclined bars, plus a system of concrete struts, after concrete
cracking (Fig. 119). By writing equilibrium and compatibility equations
along the cracks (where the internal forces are absorbed entirely by the
reinforcement), the stresses and the strains in the steel and concrete, as well
as the moduli of the resultant material, can be worked out. No closed-form
expression for G01* is presented, but an expression similar to the one given
by Collins160 can be derived. Aggregate interlock and dowel action are
neglected. No extension to cyclic loading is performed. The values
predicted for G 0 * are very low (CfR/Gc = 0-1-0-2).
5.2.10.3. Collins' model.160 Reinforced concrete and plain concrete
planar elements subjected to prevailing shear are modelled by means of
diagonal compression fields (Fig. 120), after concrete cracking. Considering
the equilibrium requirements (both in the longitudinal and transverse
direction) and the compatibility requirements (the directions of the principal
tensile strain and stress are assumed to be coincident) the behaviour of a
member in shear is analysed, after the average stress-average strain
relationships for both the steel (bars and stirrups) and the concrete struts are
given suitable formulations. The angle a^ of the diagonal compression field
turns out to be a function of the steel ratios in the horizontal and vertical

(a)
Strut

(d)

(b)

Steel
o, = 0, a2 * 0

Fig. 120. ad, ad: diagonal Fig. 121. Saw-tooth idealization of single crack: (a) crack opening; (b)
compression field; 1, 2: principal crack opening and slip (aggregate interlock and dowel action); (c)
directions in cracked concrete, at behaviour of diagonal concrete struts; (d) compressive stress-strain
onset of cracking relationship for concrete in strut

120
 INTERFACE BEHAVIOUR

directions. As a result, the stresses and the strains, as well as the effective
shear modulus are evaluated prior to yield of the stirrups, while the ultimate
shear stress is evaluated after yield of the stirrups. Dowel action is
neglected.
The effective shear modulus, CFR, is as follows
s h e a r
? ) (147)
where n = EJEC is the modular ratio and pt, p\ are the steel ratios in the
transverse and longitudinal directions. No application to cyclic loading has
been attempted so far.
5.2.10.4. Perdikaris and White's model}61 A saw-tooth type
idealization is adopted for the cracks (Fig. 121 (a)), which are assumed to
run parallel to the two-way reinforcement, if the tensile stresses prevail
(Tky 5: 0-7 MPa with x,y = reinforcement directions). Furthermore, diagonal
cracking is assumed to be the prevalent cracking mode for sufficiently large
shear stresses (T^ > 0-7 MPa (Fig. 121(b)). Based on the local equilibrium
and compatibility requirements, and on largely empirical formulations for
the shear stiffnesses, KlSr (interlock shear transfer) and KDA (dowel action),
three expressions are worked out for CJ~R

1 I

IADA + ^fsiOfy (^DA + ^IST)** ^

for r xy < 0-7 MPa
(KfST = KfST = 0 for Txy < 0-35 MPa) (148a,b)
CR 0-5Em
Gm =
l + (l+—Vin2/?+fl+ —
n n
\ Px/ \ Py
for r xy > 0-7 MPa (148c)
where the symbols sK, sy and stand for 'average crack spacing' and 'steel
ratio', Go is the initial shear modulus of concrete, £ cu is the compressive
secant modulus of concrete at ultimate, n is the modular ratio (=ES/ECU) and
/? is the angle of the diagonal cracks to the x-direction. Both aggregate
interlock and dowel action are neglected for Txy > 0-7 MPa. The expression
(148) does not apply to cyclic loading.
5.2.10.5. Tanabe and Yoshikawa's model.162 On the basis of a model
developed for tension stiffening in a RC member subjected to uniaxial
tension, the stiffness matrix of a RC panel in plane stresses is worked out. In
pure shear, the modulus G^1* has the following formulation

1 + F, + F2
with F, = (SnEc + 1)-', F 2 = -SnEcvc{SnEc + 1)"' (149)
where Ec, Gc, vc are the elastic moduli, and 5 n is a function of the 'tension
stiffening factors' Ax and Ay:
K
S - '
On = — "

where /Xx = /z/cos 2 0 CR , /iy = /x/sin2^CR, /x = normalized crack spacing,

0 CR = crack orientation at the onset of cracking {a\ =fct)-

121
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 122. Curves of

normalized shear stiffness
(a) GCR/GC; (b) dilation T? =
ey/jxy,; (c) applied shear
stress TX>, plotted against
shear strain •yxy px = p v
= l-3%,fc, = 2MPa, n =
E/Ec = 7'*'

5.3. Dowel action Even more than aggregate interlock, dowel action has been known for a
long time as an effective shear-transfer mechanism, which is present in
different types of connections and whose importance goes beyond RC and
PC structures. Stud and pin connectors, anchor bolts and reinforcing bars
develop dowelling forces in the surrounding concrete, which produce highly
localized compressive stresses accompanied by less localized tensile
stresses. The latter, combined with the wedging action of the surface
deformations of the steel, produce splitting cracks, particularly if concrete
cover, bar free interspacing and transverse reinforcement are not properly
designed.
Contrary to aggregate interlock which is essentially a material property
(crack roughness depends on aggregate type and shape, on cement paste
strength and on mutual adhesion), dowel action is essentially a structural
property, because its effectiveness is strongly related to the detailing of the
reinforcement, to the shape of the section and even to the loads or to the
constraints, since transverse pressure can markedly reduce or increase
dowel stiffness and strength.
Because of the many parameters involved, the majority of the tests
performed so far regard the ultimate capacity of dowels, for different values
of the geometrical parameters of the section and reinforcement, while
several basic aspects such as cyclic loading, fatigue, concrete deterioration
and also mathematical modelling have received a great deal less attention.
On the other hand, in RC members dowel action becomes active when
cracking is of major importance, but this stage is reached often in the ulti-
mate load situation, when the ultimate capacity and not the other afore-
mentioned aspects are relevant. Moreover, dowel action is a very localized
phenomenon in itself, unless a regular one-way or two-way reinforcement is
accompanied by a regular system of cracks, which is seldom the case,
except in planar elements, with regularly distributed loads and no diffusive
processes.
With reference to RC and PC members, dowel action has several specific
aspects. The local bending and shear of the bars is always accompanied by a
complex triaxial state of stress in the concrete. Depending on the geometry
of the concrete mass interacting with the bar, tensile stresses (Fig. 123(a)) or
compressive-shear stresses (Fig. 123(c)) prevail. In both cases, the dowel

122
 INTERFACE BEHAVIOUR

Fig. 123. Stresses around

dowel: (a) with limited I I
concrete cover; (b) splitting
cracks; (c) concrete flake and
failure surface in shear- i^rr^q^rm
HD-
compression underneath a
dowel; (d), (e) strong (model v,v,
(a)
I) and weak (model II)
mechanisms of dowel action
in a RC beam
V V

W,, A
•T

(c) (d) (e)

effectiveness may be reduced due to the formation of splitting cracks (Fig.

123(b)) or concrete flakes underneath the bar (Fig. 123(c)). On the other
hand, the effectiveness of the dowel is enhanced by the lack of lever arm of
the shear force with respect to the crack faces (Fig. 123(d)): the eccentricity
e can be neglected with respect to bar diameter). Moreover, the overall
effectiveness of a dowel depends on two different mechanisms (Fig. 123(d)
and (e)), which are activated when the dowel pushes against concrete core
(mode I) or against concrete cover (mode II). Generally mode I is stronger
and stiffer than mode II, but in specific circumstances the two modes tend to
converge (Fig. 123(e)): for instance, mode I is impaired by the skewness of
the cracked plane, while mode II is enhanced by the closeness of a stirrup.
As regards modelling, the beam resting on a cohesionless foundation
(either elastic or inelastic) is still the most viable model, but the formulation
of the foundation modulus (or subgrade stiffness) k[F/L3] is critical, if a
refined incremental analysis of dowel action is required (initial stiffness,
non-linear behaviour, ultimate strength, softening, displacements under
working and ultimate loads).
For a single bar embedded in an unlimited mass of concrete (no cover and
interspace effects, no concrete splitting), the results of past experimental
research lead to somewhat scattered values for the foundation modulus,
from 75 to 450 MPa/mm.129'163 Since many researchers were also interested
in the evaluation of dowel strength within a limit analysis approach, based
on the formation of a plastic hinge in the bar and on local crushing of
concrete (Fig. 124), the assumption of a localized hinge was generally
introduced. 129163164 As for the 'bearing strength' of concrete (i.e. the
compressive strength of the highly confined and ductile concrete under the
dowel) values as large as (l-8-6-5)/c' were obtained through different types
of tests. l65(a)
For a single bar close to a free surface (small cover, with or without
stirrups) or to other bars (small free interspace) the interaction between the
limited surrounding concrete and the bar becomes critical, since the collapse
is always preceded by concrete splitting (Fig. 123(b)), either in the
horizontal plane (limited concrete width or bar free interspace) or along
oblique planes (limited concrete cover with respect to concrete width and/or
bar interspace). The many tests devoted to the study of the role of concrete
cover and bar free interspace, 127132133 of the stirrups, and of the interaction
between bond and dowel forces,120 made it possible to work out a number of
rational or semi-rational equations for the prediction of ultimate dowel

123
 RC ELEMENTS UNDER CYCLIC LOADING

Fig. 124. Internal forces (m =

bending moment' V = shear),
location of plastic hinge (Mu
= ultimate plastic moment
developed by the section of
the dowel) and applied forces
in a dowel.126 Vj = applied
shear; f*c. = concrete bearing
strength in triaxial
compression (crushing
strength)

Dowel

capacity. These equations trace back very often to relatively simple, but
effective limit anlaysis models. 127164l65(b) Finally, the long-established
rational model by Johnston and Zia166 is still the most comprehensive
attempt to describe the different resistant mechanisms, which are activated
as soon as the concrete cover cracks and the stirrups (if any) come into
action. An extensive review of the literature on dowel action pushing
against concrete cover is presented in a paper by Dei Poli et al. together
with new test data.130

5.3.1. Regularly cracked and reinforced fields

In a regularly cracked and reinforced field (cracking in one or two
directions, one-way or two-way reinforcement, generally with rectangular
mesh), the tensile and shear forces induced in the bars by the loads and by
the subsequent cracking may still be considered as forces applied to the
composite material (cracked concrete + reinforcement) on condition that the
crack displacements are smeared over a length equal to crack spacing
(resolved in the direction at right angles to the cracks), Young's modulus of
the steel is given an effective value taking into account tension-stiffening
effects and the embedment of the bars is given a suitable foundation
modulus.
The most simple case is represented by regularly spaced and linear cracks
running parallel and at right angles to the bars of a reinforcing net (Fig.
125). It is easy to demonstrate that the stiffness matrix of the composite
material has the form shown in equation (150)
0 0
0 Ec(l+npy) 0
(150)
0 0
na(k)db
•IS •k-s where the strains in the solid concrete have been neglected with respect to
the equivalent strains obtained by smearing the crack displacements (except
in the ^-direction, where there is no crack contribution and the concrete
Fig. 125. Regularly cracked behaviour may be assumed to be linearly elastic, at least in compression). In
and reinforced field with one
this simple case there is no coupling between the shear (normal) stress at the
system of cracks aligned with
y-bars crack interface and the normal (shear) stress in the bars, but both tension

124
 INTERFACE BEHAVIOUR

(a) (b) (c) (d) (e)

Fig. 126. Regularly cracked and reinforced field with one system of cracks at an angle to the y-bars: (a) forces
developed in the bars; (b), (c) forces and displacements in the crack reference system; (d), (e) effects of crack
skewness on dowel action (x-bars): (d) negative and (e) positive (Nx, Ny, V^,, Vyx = normal and shear forces per unit
length of a crack)

stiffening and dowel action are present: the former is introduced through
E*(exR) ^ Es and the latter through the foundation modulus k(T^R).
In equation (150), n is the elastic modular ratio (=EJEC) and a is the
parameter governing the elastic foundation theory
kdh \
a =
4EsJhJ
where Jb and db are the moment of inertia and the diameter of the section of
the bars.
Whenever the cracks have another direction with respect to the bars (Fig.
126(a)), not only the tensile and shear forces developed by the bars would
have two resolved components in the crack reference system (Fig. 126(b)),
but for each bar array the foundation modulus k would be different,
depending on the skewness of the crack plane (tension stiffening in the bars
is little affected, or totally unaffected by the orientation of the crack plane).
For instance, for 6t > 0 (Fig. 126(c)) the x-bars push against a limited cover
due to the skewness of the crack (Fig. 124(d)), while for 6t > 0 (Fig. 126(c))
the same bars push against an unlimited cover (Fig. 126(e)) and the
foundation modulus is larger than in the previous case (see also section 3.2).
In equation (150), the modulus El(e^R) should be given one of the
'tension-stiffening' formulations available in the literature.16? As for dowel
action, the non-linear behaviours of the concrete and of the reinforcement
are lumped into £ ( 7 ^ ) , although the two non-linearities should in principle
be dealt with separately. The equivalent strain 7 ^ is the smeared
component of the crack displacements, at right angles to the x-bars (Fig.
126)
(5tcos0)
"xy

VcosflJ

5.3.2. Dowel action against concrete core

When a dowel pushes against concrete core (Fig. 123(d), mode I), the force-
displacement response curve V(Wi) — with Wx - displacement of the dowel
in the shear plane — is mostly linear up to a load level close to 40% of the
ultimate dowel capacity (see for instance Di Prisco and Gambarova.131
Beyond this level, the response becomes non-linear because of the com-
bination of the non-linearities of the embedment and of the bar. Beyond a
load level close to | - | , a concrete flake forms in the microcracked and

125
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 127. (a) Formation of a

concrete flake underneath the
dowel; (b) typical force-
displacement curves;129 f'c =
32MPa,fsy = 440 MPa

A\\ mm
(b)

crushed zone underneath the bar and close to the shear plane. Once the
concrete flake gets detached, the non-linearity of the embedment becomes
even more pronounced, since a non-negligible lever arm accompanies the
dowel force (Fig. 127(a) and (e)* = 0-4-0-6 db). Typical force-displacement
curves are shown in Fig. 127(b):129 as a rule, the response curves are
bilinear, with an elasto-plastic behaviour, but a limited softening for small
diameter bars (db= 14 mm) and a limited hardening for large diameter bars
(db = 24 mm) is often found in the tests. With regard to this point, high
strength concretes with basaltic aggregates exhibit a definitely elasto-
softening behaviour </c' > 80 MPa).131
In order to describe dowel action at increasing loads (monotonic loading)
the model of the beam resting on a cohesionless foundation is still
acceptable, but different formulations should be adopted depending on the
purpose of the modellization. Should the purpose be the description of the
load-displacement curve limited to the loaded section of the dowel (i.e. in
the shear plane), a relatively simply unidimensional formulation for k would
be sufficient; even a constant value, if the load level is sufficiently low with
respect to the ultimate capacity (service loads).
Five different formulations are cited below and the corresponding force-
displacement curves are compared with some of the results shown by Dei
Poli et al. (Fig. 128).129 The equations are presented for only two
formulations, since in the other cases the details are easily found in the
literature referenced.
5.3.2.1. Walraven and Reinhardt.119 The force-displacement
response traces back to the theory of the beam resting on an elastic
foundation (see also Section 5.3.2.4): for the foundation modulus an
expression already worked out by Paulay and Loeber is adopted (k = 188
JVC0'85), with some adjustments in order to introduce the effects of concrete
stiffness in a more consistent way and to improve the fitting of the authors'
own test results on dowel action.

126
 INTERFACE BEHAVIOUR

Fig. 128. Plots of several

equations regarding the V
response of a dowel, and fit / G
-
80
of test data by Dei Poll et
I
al.:'29 W = Walraven and
Reinhardt; V - Vintzeleou B MS
and Tassios; S = Soroushian = 30MPa
et at; G = Di Prisco and 60 S ! 1 1

V
• ^ . 0-2 0-4 0-6 0-8 10
Gambarova; B - Brenna
et al. db = 24 mm
^ ^ -

\\B>
40 40 -
Ibffx30 v ^>•"(3 < S_

fa s = 18 mm
20
Bv 10 w db ~ 14 mm
20 -
results

W 0 i i 1

i i
1 2 1 2
Ay. mm / ] , : mm

5.3.2.2. Vintzeleou and Tassios.164 The displacement-force curve is

described by means of a four-order polynomial, which is the best fit of many
test results. Both the displacements at the end of the linear behaviour and at
collapse (Wlin and Wu) appear in the formualtion, as well as the ultimate
capacity Vu. The expressions for Vu, Wlin and Wu are available for
reference.164
5.3.2.3. Soroushian et a/.163 The force-displacement response is
described by means of two 'best fit' equations, which require the previous
evaluation of the ultimate dowel capacity Vu and of the corresponding
displacement Wu. The equations for V and W are also presented.16
131 168
5.3.2.4. Di Prisco and Gambarova,'il and Brenna et al.ioa Both
formulations are based on the well-known force-displacement relationship
given by the theory of the beam resting on an elastic foundation

= — Wi with /3 = , a = (151)
4EsJh
where keq is the so-called 'equivalent foundation modulus', which is a non-
linear elastic modulus instrumental in the description of the force-
displacement response of the loaded end-section of the dowel, but unable
to describe the behaviour of the dowel in the other sections. The formulation
proposed by Di Prisco and Gambarova131 is based on the ratio V/Vu (which
is considered here as a 'damage index'), where Vu can be evaluated by
means of the available equations (see for instance Dulacksa126); as for the
roles of f and db, the proposal by Soroushian et al. is adopted165(a)

with ks = 127VU0AC6
V
for — < 04 7 = 2-12

V
for — > 04 7 = /0-544 + 0-026 cos h 8 — -
Ml

The formulation proposed by Brenna et al. is based on the ratio W\ldb

127
 RC ELEMENTS UNDER CYCLIC LOADING

&eq = 7&o, with ko = 600 f®'1

7 = {l-5[a + v (J 2 (40 W\/c - bf
where
a = 0-5875 -0-01125 / c '; b = 0-0015 £ - 0-225
c = 0-00375/'+ 0-4375 d = 0-0025/' + 0-575
The formulation also describes reasonably well the response of a dowel
embedded in high strength concrete (£ = 70-90 MPa). A more realistic
formulation for k, with the introduction of the local displacement W and of
the coordinate x (measured from the shear plane) has been recently
proposed by Di Prisco et al., but checks are still in progress: in this way,
the entire displacement field along the dowel can be described, including
the curvature, and the contact forces between the dowel and the concrete
can be analysed.
As for the ultimate dowel capacity Vu, several reliable equations are
present in the literature: in Fig. 129(a) the equations by Dulacska126 and
Soroushian et a/.163 are plotted, together with some of the test results
presented by Dei Poli et al. Both equations consider the interaction between
the dowel force and the tensile force in the bar, as well as the lever arm of
the dowel force with respect to the forefront of the concrete mass (see Fig.
129(b), where the equation proposed by Walraven and Reinhardt is also
plotted"^.
As regards dowel action under cyclic loading, limited test data are
available,164 and in most cases the tests have been load-controlled, with the
load peaks well below the dowel bearing capacity under monotonic loading.
The systematic tests carried out by Vintzeleou and Tassios 133164 under
imposed transverse displacements (Fig. 130) show that cycling causes a
degradation of the dowel response, mainly due to concrete, since steel
response is practically unaffected by cycling. On the basis of plain concrete
degradation under imposed strains, a recurrent formula is presented (Fig.
131
130)'"
VI.
= l-Xy/(n-\) (152)
V1
max
1/7 for full reversals (alternate cycles)
with A =
1/14 for repeated loading (pulsating cycles)

-Dulacska
-Soroushian
f'c = 30 MPa ':'••': : ' • • ' • • ' ' • • )
. -•:•:•:•.••••• Walraven — Reinhardt (pa = 001)
200 - • Walraven - Reinhardt (eM : 002)
1 Tests A and B
••>.•' :•••'.'-.•
I Walraven — Reinhardt (oa 003)
Fig. 129. Ultimate dowel N
capacity with no tensile • Slanted shear plane 6 = 45° J^
160 - N
force (a) and with tensile
force (b): (a) fits of test
120 "
data by Dei Poli et al.;'29
(b) failure envelopes:
V*u = ultimate capacity 80 -
according to Dulacska ; Soroushian
Vu(Soroushianetal.163) **
40 "
= 0-82 V*u; V (Walraven Dulacska f' = 30 MPa fsv = 440 MPa
and Reinhardt'w)= 0-72
ii i i
V* • N * = A u f •
12 16 20 24 28 32 10
psi = steel ratio of the db'. mm
longitudinal reinforcement (a)

128
 INTERFACE BEHAVIOUR

Specimen A5 - d b = 18 mm

fc= 31-2MPa

50 I Cycles

Ay > 0
40

30

20

10
Fig. 130 (right). Dowel response under reversed transverse
displacement (Vintzeleou and Tassios164): empirical model
Fig. 131 (far right). Typical hysteresis loops for pulsating loads
(Dei Poli et al.v). dh = 18 mm, fc = 31 -2 MPa
i /29i

where V ^ . = dowel response after the nth cycle

Knax = dowel response at the onset of unloading during the first
cycle ( V ^ > 0-3 V^, V^ = ultimate capacity under
monotonic loading)
n = number of the cycle
Under repeated loading, with full exploitation of dowel capacity at the end
of each cycle (Fig. 131), the limited test results available so far show a small
degradation compared to the response curve under monotonic loads and a
negligible energy dissipation due to hysteresis.

5.3.3. Dowel action against concrete cover

When a dowel pushes against concrete cover (Fig. 123(d)), the non-linear
behaviour is triggered by the splitting of the cover (side splitting for bars
placed at corners, or bottom splitting) and/or bar free interspace (side
splitting), but later on the response is strongly affected by the transverse
reinforcement (if any) and by the characteristics of the dowel. As a

Section B

(t f . . . • • • • . . : . •
' < • • •
* > •

51)
\

(a)
Section A

*.
(b)

Fig. 132. (a) Dowel action

against concrete cover; (b) !!
-I-'I I''' 'V : : ;
'k - - -

model based on the beam B /\ V (c)

resting on a cohesionless
foundation; (c) limit cases for SB = shear-bending crack

the composite section SP = splitting crack

129
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 133. Typical load- D11

displacement curves for long 56 ^ ^
D9
dowels (Dei Poll et al.li0):
the first stirrup is (a) close to
48
and (b) far from the shear /db = 24 mm +

plane; fc - 24MPa,fsy =
440 MPa, s - distance 40
: . : : • ; • .

between the shear plane and

the first stirrup (see Fig. 123) / /
D5
J2

11 / O'b = 18 mm
-" D7
24

16 D1
— —

• —

D3
8 \// db= 14 mm

n
f Cover
f H

2 3
A,: mm
(a)

consequence, the ultimate dowel capacity depends on the stirrup ratio, on

the position of the first stirrup with respect to the shear plane and on the
limit moment of the dowel section. Moreover, the transition between the
initial, mostly elastic behaviour (unsplitted concrete) and the final, mostly
plastic behaviour (after stirrup and/or dowel yielding) is governed by the
thickness of the cover and by the spacing of the dowels; both parameters
govern the 'lift-off of the cover.
Typical force-displacement curves are shown in Fig. 133: l3 ° these curves
were obtained by testing block-type specimens provided with a single long
dowel and a few stirrups (stirrup section-to-dowel section ratio 0-20-0-25),
for three bar diameters (db = 14, 18, 24 mm), two concrete covers (c = db,
2db), two stirrup arrangements (first stirrup close to or far from the shear
plane, s = db, 3db), one stirrup spacing (= 6db). The linear regressions
regarding the ultimate dowel capacity and the initial stiffness are shown in
Fig. 134.130
In order to describe dowel action at increasing loads (monotonic loading),
the model of the 'composite beam' resting on a cohesionless foundation and
consisting of the bar plus bottom and side cover has been largely used in the
past: the foundation is the concrete mass placed between the bar and the
neutral axis (a section susbjected to positive bending and shear is considered
in Fig. 132). Since the composite beam is suspended to the concrete core
and tends to be detached from it by the dowel action, the non-linear
behaviour of the dowel is strongly affected by the concrete strength in
tension.
The prediction of dowel strength when the dowel pushes against concrete
cover is largely based on empirical formulae, which refer (in most cases) to
the situation with no transverse reinforcement (no stirrups). In this case the
dowel strength is assumed to coincide with the critical force causing
concrete splitting, i.e. Vu = Vcr, where Vcr depends on bar diameter, bottom
and side cover, concrete tensile strength.
Among the many formulations of Vu = Vcr available in the literature for
side splitting, six were examined by Dei Poli et al.

130
 INTERFACE BEHAVIOUR

Fig. 134. Linear regressions

regarding the ultimate dowel First stirrup: Concrete cover: /
capacity (a) and the initial - » • - close to • o c — 2d b •
dowel stiffness (b)uo shear plane • o c = db
o-°- far from \ \
60
shear plane E60
/
50 S 50 -

40 - / s =db
° 40
>
J 30 30 -
= 3d b
S
20 20

10 10 - S 50-85%
I 1

12 16 20 24 28 32 12 16 20 24 28 32
da: mm d^. mm
(a) (b)

Krefeld and Thurston128

0-83 10- 4
y — (153)
W
• Taylor (see reference 130)
Vn = 9078 + 0-1 blU. (154)

Soroushian «?r a/165(b)

Vu = 0-83(fcn - 4)/ct7 (155)

Houde and Mirza (see reference 130)

Vu = 37 *„{/(£) (156)

Baumann and Riisch (see reference 130)

V u = 1-64 (157)

• Jimenez-Perez er a/.170
3-79c
Vu = M d 3-25 + (158)

The symbols are explained in Fig. 135; in equation (155),

with k = 272MPa/mm;/c' = cylindrical strength in compression;/cc = cubic

strength in compression;/ct = 0 • 27/ c ' 2//3 = strength in tension according to
CEB.

131
RC ELEMENTS UNDER CYCLIC LOADING

/>•
Tests D IO Concrete cover c = 2d0
(refs 130, 131) I • Concrete cover c = da
• Krefeld and Thurston
(•*)-
60 • Taylor
• Soroushian er al.
A Houde and Mirza
• Baumann and Rusch
50 Jimenez-Perez ef al.
(*)

40

i
30

a = 140 mm b = 300 mm c = 1-5db

/>„ = 6d b b' = 6 n + db = 7d b 20
d = <3/2)a = 210 mm 1. = 10d
x, = £/4 = 2-5d
p = AJdf = 0075 (<Va) 10
(*) Specimen section (tests D13° 131
)

(") Equivalent beam-section:

compressive zone
3 4
("*) Reference beam s/d b
(a) (b)

Fig. 135. Ultimate dowel capacity: (a) assumptions and symbols adopted to make the predictions of equations (153)
to (158) comparable to test results by Dei Poll et al.;'3 (b) effects of the position of the first stirrup on dowel
capacity

The predictions obtained with equations (153) to (158) are shown in Fig.
135(b), where the test results by Dei Poli et al. are also presented.130 As
regards cyclic loading, the same general considerations made in section 3.2
(dowel action against concrete core) hold for dowel action pushing against
concrete cover.
Under fully reversed transverse displacements (Fig. 136)133'164 the dowel
response becomes asymmetrical. Although Vintzeleou and Tassios do not
commit themselves, equation (152) seems to hold also for dowel action
pushing against concrete cover (with no stirrups).
In Fig. 137 the curves of stiffness versus number of cycles are shown:
under repeated loading the degradation turns out to be less severe than under
full reversals.

5.4. Concluding As shown in the previous sections, the behaviour of concrete-to-concrete

remarks interfaces (aggregate interlock) and concrete-to-reinforcement interfaces
(dowel action) in shear transfer problems under cyclic loading has not been
adequately investigated and needs further systematic research.
It is clear from the limited test data available so far that a remarkable
degradation occurs with respect to both strength and stiffness, but working
out comprehensive and consistent models and manageable equations is
rather difficult, because of the intricate crack patterns in aggregate interlock
and the many parameters in dowel action.
On the other hand, to what extent refined models regarding each single
mechanism affect the overall analysis of a structure still has to be assessed,
because the real behaviour of the whole results from the contributions of
several parallel mechanisms, which often overshadow each other's role.

132
 INTERFACE BEHAVIOUR

n th cycle

3rd

Fig. 136. Typical hysteresis loops for fully reversed Fig. 137. Degradation of normalized stiffness due to
transverse displacements: I + A l m a x | = | - Ai m a x | cycling:133 RL = repeated loading (pulsating loads); AL
(Vintzeleou and Tassios)13'' IM (dimensions in mm) = alternate loading (full reversals); AL* = alternate
loading after a series of cycles at low force levels

Summing up, only the joint effort of the people involved in basic research
on each single mechanism and of the analysts who introduce the different
models into their comprehensive computer programs, trying to fit the
various structural behaviours, will allow the code-makers to accomplish a
synthesis, and to help the professional engineers to design better and safer
structures, which is the ultimate scope of the whole chain of events
discussed.

133
6. Finite element modelling of reinforced
concrete

6.1. Introduction At the macroscale, composite materials are typically modelled as homo-
geneous, isotropic or not, with effective or equivalent properties. This is the
case for plain concrete (a composite of aggregates and cement paste), and
for fibre reinforced concrete. In reinforced concrete the mechanical
behaviour of its two constituents is very different, and the distribution of
reinforcement is typically non-uniform. So, a homogeneous description of
reinforced concrete is rarely used, and if so, only in regions where the
reinforcement is uniformly distributed. Instead, the two materials are
usually modelled individually. Such a separate representation of the two
materials, which fits very well with the finite element method, rests on the
(possibly questionable) assumption that the sum of the two constituent
materials describes reliably and without bias the behaviour of the actual
composite.
Models for the individual materials, concrete and steel (like the ones
previously presented in Chapters 1-3), for their interaction through bond
(Chapter 4) and for the behaviour of their interface (Chapter 5), can be used
as building blocks to construct fairly sophisticated models of reinforced
concrete at the microlevel. In models of this type separate finite elements
are used for the concrete (three- or two-dimensional ones, depending on the
dimensionality of the analysis) and separate ones for the steel (mono-
dimensional 'truss' elements representing a certain length of a single bar, if
the problem is analysed in three dimensions, or of a layer of bars with the
same two-dimensional position, if it is analysed in two dimensions).
Moreover, bond interaction between steel and concrete is modelled through
special elements, which connect nodes on steel elements to nodes on the
adjacent concrete elements, with the same initial position in space. Finally,
discrete cracks may be inserted between neighbouring concrete elements, by
splitting each node on their interface into two, and connecting the latter
through special contact or crack link elements, which model the behaviour
of the interface.
Integration of all the individual models into a comprehensive microlevel
model of reinforced concrete, capable of reliably predicting its behaviour up
to and beyond ultimate strength, under all possible loadings, including
cyclic, is not an easy task. Moreover, even if such a powerful tool is
available, its applicability to realistic reinforced concrete structures may be
limited by its prohibitively large requirements in computer time and
memory. For this reason, in recent years considerable effort has been
devoted in developing macrolevel composite steel/concrete models for
reinforced concrete. Models of this type treat reinforced concrete as a
continuum, with properties and behaviour which are the integrated outcome
of those of its constituent materials and of their interaction. Development of
macrolevel models has been promoted also by the belief of many leading
researchers that the interaction between steel and concrete is so strong and
complex that our real material is reinforced concrete, a composite with
properties which are not simply a superposition of those of its constituent
materials.
In the following, we will first present an overview of non-linear finite
element modelling of reinforced concrete with separate elements for the
concrete and the steel, and we will then examine 'macrolevel' composite
steel/concrete elements for uniform distribution of the reinforcement.

134
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

6.2. Finite element For discrete modelling of single bars (or of groups of bars with the same
modelling of the location, as for example the top and the bottom reinforcement of a beam) in
reinforcement and of two-dimensional or three-dimensional analyses, we use one-dimensional
its interaction with truss elements. These elements are typically two-noded, but if compatibility
concrete of displacement expansions with higher order concrete elements (e.g. eight-
noded isoparametric ones in two-dimensional analyses) is desired, higher
order one-dimensional elements (e.g. three-noded ones) must be used for the
steel bars. In axisymmetric analyses single reinforcing bars or groups of
concentrated bars in the hoop (circumferential) direction are modelled by
special single-node hoop elements, whereas constant density layers of hoop
and/or meridional steel within a surface of revolution, are modelled using
two- (or three-) noded axisymmetric membrane elements. In two-dimen-
sional or three-dimensional analyses, uniform density distributions of the
reinforcement (e.g. closely spaced stirrups in a beam, or reinforcement
meshes in a shear wall) may be modelled with two-dimensional or three-
dimensional elements, respectively, which are superimposed to (i.e. have
the same nodes as) the concrete elements. Such smeared reinforcement
elements can, of course, co-exist with one-dimensional elements for
concentrated reinforcement.
In all cases above, the constitutive relations of the steel elements are
derived in a rather trivial manner, considering that the normal stress and
strain in the direction of the reinforcing steel are related through the uniaxial
models presented in Chapter 3 for reinforcing steel.
When the overall macroscopic behaviour is mainly of interest, the nodes
of the steel and the concrete elements which have the same coordinates are
also assigned the same degrees of freedom. Then the interaction between the
two materials, such as bond-slip and dowel action effects, are modelled
implicitly, by appropriate, though phenomenological, modification of the
constitutive relations of steel and/or concrete. Incorporating tension stiff-
ening in the steel or concrete model is an example of such an implicit
representation of interaction effects. If the detailed and explicit description
of the interaction response at the microlevel is required, then nodes which
are connected with the steel and the concrete elements and have the same
(or practically the same) coordinates are assigned different degrees of
freedom, and special contact elements are inserted between them. Such
contact elements can be used only in association with the discrete, not the
smeared, representation of reinforcement.
The early version of these contact elements are the dimensionless 'bond
link' elements, connecting a single concrete node to the corresponding steel
node. The constitutive relation of these elements is equivalent to two non-
linear springs (three for three-dimensional analyses), relating lumped nodal
forces to the corresponding relative displacements between these two nodes.
One spring is in the direction of the steel bar(s) and models the bond-slip
effects. Any of the cyclic bond-slip models presented in section 4.4.2. of
Chapter 4, can be used for this purpose. The other spring(s) is normal to the
bar, and may model dowel action effects. If modelling of these latter effects
is not considered necessary, then infinite stiffness is assigned to these latter
spring(s), or the degrees of freedom of the concrete and steel nodes normal
to the bar are constrained to be the same. Coupling between the concrete-
steel interaction in the longitudinal and transverse to the bar directions, can
also be included in the stiffness matrix of the bond link element. Although
dimensionless, bond link elements require information regarding the
orientation of the steel bars to which they are connected. In this respect,
it is worth mentioning that bond link elements can very conveniently be
composed of standard two-noded truss elements, one of them along the
direction of the bar, and another (or two, if a three-dimensional analysis is

135
RC ELEMENTS UNDER CYCLIC LOADING

performed) normal to it. Each of these truss elements is connected with one
of the two nodes to be linked (the steel or the concrete node), and with an
extra node rigidly connected to the other one of the two linked nodes. So,
without requiring more degrees of freedom than a special bond link element,
the truss elements provide automatically the required orientation informa-
tion. One only needs to implement in the uniaxial constitutive relation of the
truss element the appropriate cyclic bond-slip or dowel action model.
To avoid the artificial lumping of interaction forces at the linked concrete
and steel nodes, and to improve the flexibility and the efficiency of
modelling the steel-concrete interaction, continuous contact elements have
been developed.171"174 These elements are arranged along the entire steel-
concrete interface, and they are typically isoparametric, with order com-
patible with that of the concrete and steel elements which they connect.
One-dimensional and two-dimensional (surface) contact elements in three-
dimensional space are used to connect one-dimensional discrete steel
elements or two-dimensional surface ones, respectively, to the surrounding
concrete. The elements have double nodes with the same coordinates, one
connected to a concrete element and the other to a steel element. The
constitutive model relating the three stress components in the contact
element (for the general case of three-dimensional analysis) to the three
deformation measures, offers the possibility of modelling the effect of
normal stress on the steel-concrete interface upon the bond-slip behaviour.
It is sometimes convenient to combine the continuous contact element with
the corresponding steel element into a single element, which incorporates
both the reinforcing steel and its interaction with the concrete.174
Nearly all widely used general purpose non-linear finite element
programs, which have a reinforced concrete modelling capability along
with their capabilities for metal plasticity, geomaterial and rubber
modelling,175"1 7 do not have special bond link or continuous contact
elements. So, in effect, they can account for bond between concrete and
steel only through modelling the tension stiffening effect. It is noteworthy
that Mehlhorn and his co-workers have implemented their contact element
into ADINA. 1 7 2 1 7 3

6.3. Modelling of the 6.3.1. Distinction between 'concrete in compression' and 'concrete in
concrete component tension'
It has become clear in Chapters 1 and 2, that the behaviour of plain concrete
under 'predominantly compressive' stresses exhibits fundamental differ-
ences from that under 'predominantly tensile' ones. For relatively low stress
levels, under which the behaviour of uncracked plain concrete is essentially
linear-elastic, such differences do not exist. However, when stresses ap-
proach and reach the failure criterion, the failure mode, and more impor-
tantly, the post-failure behaviour differ significantly: under predominantly
tensile loading, attainment of the failure criterion leads to cracking and
localization of extensional deformations along a well defined crack plane.
From then on the behaviour becomes strongly anisotropic: normal to the
crack plane the material strain-softens in the manner described and
modelled in Chapter 2, whereas parallel to the crack plane the behaviour is
essentially that of uncracked concrete with the dimensionality of the
problem reduced by one, so under triaxial stress conditions it becomes
biaxial, and under biaxial, it becomes uniaxial. (As discussed in detail in
section 6.4.3., in reinforced concrete the behaviour parallel to a crack plane
is not unaffected by what happens normal to the crack plane, due to the
tensile stress component which develops in this latter direction because of
the transfer of forces from the reinforcement crossing the crack through
bond, as well as other effects.) On the contrary, under predominantly

136
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

compressive stresses, the post-failure behaviour is less anisotropic, with the

material exhibiting gradual strain-softening in all three directions, until a
certain ultimate deformation, at which the material 'crushes' in all three
directions, releasing its stresses. Such post-failure behaviour corresponds to
the progressive shrinkage of the failure surface in stress space, and then its
sudden collapse, to the zero stress point.
A proper concrete model implemented in a finite element program should
be general and integrated, possibly incorporating as special cases submodels
for the different behaviour of concrete under predominantly compressive or
predominantly tensile stresses. To do this, the model should be capable of
differentiating between these two cases, and identifying when to apply the
submodel for concrete in tension or the one for concrete in compression.
Without such a generality and differentiating capability, a finite element
solution may run into numerical and other difficulties, in situations in
which, for example, the behaviour in one region (or even worse, in one
direction of an integration point) of the analysed structure is governed by
concrete cracking, and that in another is controlled by the non-linearity of
concrete in compression or by crushing. 178179 The approximation suggested
by Wang,179 of specification by the user of different concrete models, to be
used at different regions of the same structure or element depending on the
anticipated behaviour, is unlikely to be a viable solution.
As stated above, the behaviour of concrete in tension becomes markedly
different from that in compression, when and after the failure surface is
reached. So, the location in stress space of the point where the failure
surface is reached, offers a convenient criterion for the type of failure mode
and post-failure behaviour. Three examples of failure mode identification
through the failure surface are given here, drawn from three finite element
programs (two of them having widespread use) for the non-linear analysis of
reinforced concrete in three dimensions. 175 ' 177178 In all three programs,
concrete is modelled separately by 20-noded isoparametric solid elements.
An isotropic failure surface is defined in the space of the principal stresses
0ci > 0c2 > crC3 (index c is for concrete). The failure surface in the triaxial
tension region is defined by a tension cut-off175'180

/ct = O-cl > <7c2 > (Tc3 > 0 (159)

(fct is the uniaxial tensile strength of concrete), and in the two tensions-one
compression region
0"ci > o"c2 > 0 > crc3

by the plane

in which a'fct = (1 — a)fct is the strength in the direction of principal

tension, ac\, for aci = 0 andCTC3= —/c, with a = 1 • 0 (a1 = 0) 18° or a = 0-75
{a1 = 0-25).175 In the two compressions-one tension region
0"ci > 0 > ac2 > <Tc3

the failure surface is described by a hyperbolic paraboloid

<W/ct = (1 + a <W/c) 0 + a <rc3//c) (161)

In all three regions above, the concrete is considered to be in predominant
tension, and attainment of the failure surface signals the formation of a
crack normal to the direction of ac\. The concrete is considered to be in
compression, only when all three principal stresses are non-positive

137
RC ELEMENTS UNDER CYCLIC LOADING

0 > (Tc\ > O"C2 > CTc3

The failure surface in this latter stress subspace is 180 a Drucker-Prager cone,
described by the simple expression

(2/bc - / c ) Ah) + (/be -fc) h +/bc = 0 (162)

in which J2 is the second deviatoric stress invariant, /] is the first stress
invariant, and /t,c > 0 is the strength of concrete under equal biaxial com-
pression, (jci = 0 <rC2 = crC3- On the contrary ADINA offers to the user the
possibility of specifying himself the failure surface in the all-compression
region. The user specifies the value of the minor principal stress, <rC3, at
three points on each intersection of the failure surface by planes of crci//c =
constant. At O& = ac\ > aC3~, at aC2 = (3 o<% > ac\ (where (5 is a user
specified fraction, e.g. 0-5 or 0-75); and at <rC2 = acs < <rci.175 The failure
surface is then defined in a piecewise manner by (hyperbolic-paraboloidal)
interpolation between sets of four such neighbouring points.
In the early version of ABAQUS the failure surface follows the Chen and
Chen model.181 For concrete in compression the failure surface is a
paraboloid of revolution around the hydrostatic axis (linear in J2 and /j),
passing through the points (0, 0, -fc) and (0, -fbc, -fbc) of uniaxial and equal
biaxial compression. For concrete in tension it is a hyperboloid of revolution
(linear in J2, quadratic in /,), through the points (fct, 0, 0) and (0, 0, -/ c ) of
uniaxial tension and compression. The two regimes are separated by the
cone 3J2 + /i = 0, which passes through the uniaxial compression axes, —CTC,
(i = l,3).
In the more recent version of ABAQUS, both above segments of the failure
surface are replaced for simplicity with Drucker-Prager cones, the one for
concrete in compression through the uniaxial and equal biaxial compression
points, and the one for concrete in tension through the uniaxial tension point
and through the point (a'/ct, 0, —fc). This point, introduced above in
conjunction with the concrete in tension failure envelope, 175180 is one
towards which the failure envelope heads in the biaxial compression-
tension quadrant, without passing through it, as it intersects the concrete in
compression part of the failure surface shortly before.177 It is worth
mentioning that close to this latter point of intersection lies also the point
with aci/aC2 = —1/15 that separates the concrete in tension from the
concrete in compression part of the biaxial failure envelope in the two-
dimensional finite element program MICRO developed at Delft.

6.3.2. Finite element modelling of the concrete in compression

Constitutive relations of the type reviewed in Chapter 1 for plain concrete in
compression are implemented at the finite element integration point level
and used within an incremental iterative solution scheme, for three
purposes.
(a) To construct the tangent, D*, or secant, Dsc, rigidity matrix, which
relates the incremental stress and strain vectors dac = D* de, or the
total: <rc = D'e, respectively. This rigidity matrix is then used to
form the corresponding (tangent or secant) stiffness matrix of the
element

K(tors) =
= ffgT D (tors)

in which B is the familiar (integration point) strain-nodal displace-

ment matrix.

138
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

(b) To compute the (total) stress vector ac, which corresponds to the
strain vector e, calculated at the integration point of interest from the
most recently computed nodal displacements. The total stress vector
is used then to form the vector of internal nodal forces of the element

T
<7cV (164)

Assembly of the element internal force vectors into a structure

internal nodal force vector, and subtraction from the effective loads
vector (which also includes terms due to acceleration and damping in
dynamic analyses), produces the unbalanced nodal force vector, that
needs to be equilibrated through equilibrium iterations.
(c) To calculate and output the final stresses at each step of the
incremental analysis, after convergence of the iterative procedure.

For cyclic loading, or, in general, for load paths which are not monotonic/
proportional, differential (rate) type of constitutive relations only, such as
the ones reviewed in Chapter 1, are appropriate. Constitutive models of the
total (secant) type, such as those based on non-linear elasticity, should be
limited to monotonic, nearly proportional loading, as they can describe only
path-independent behaviour. It should be mentioned, though, that secant
type formulations allow full error control via equilibrium iterations, as the
total stress vector ac, and hence the vectors of internal nodal forces and of
unbalanced loads, is calculated exactly through the usually algebraic
expressions giving ac in terms of e. (Another advantage of the secant
formulations is that they can treat softening behaviour without numerical
problems and without resorting to special techniques, such as the arc-length
method). On the contrary, in formulations of the rate type, computation of
the total stress vector requires step-by-step numerical integration of the
(path-dependent) constitutive equations. Therefore, equilibrium iterations
through the unbalanced load vector can correct the solution only up to the
accuracy of the operator for the integration of the incremental constitutive
relation. This means that, unless a very fine step is used for this integration
operation, significant linearization errors may accumulate, leading to
considerable drift in the solution.
Among the incremental (rate) type of stress-strain models for concrete in
compression, the hypoelastic ones (either the orthotropic or the invariant-
based models) reviewed in section 1.3. seem most convenient, as far as
numerical implementation is concerned. A word of caution is due at this
point: as the loading/unloading criterion of such models is rather arbitrary
and lacking a sound theoretical basis, these models may violate the
continuity requirement for nearly neutral loading. However, models based
on the theory of plasticity and employing convex yield and loading surfaces
and an associated flow rule, satisfy Drucker's stability postulate which
guarantees uniqueness of solution. Nevertheless, as noted in section 1.4.1.,
experimental evidence and theoretical arguments regarding the non-linear
behaviour of frictional materials, suggest that validity of the normality rule
for concrete is neither a realistic nor a necessary assumption. The
alternative, i.e. non-associated plasticity (as applied in DIANA for
example), 182 gives rise to non-symmetric rigidity and stiffness matrices,
and requires a non-symmetric equation solver.
The criteria that seem to govern selection of the constitutive model for
concrete in compression to be implemented in a non-linear finite element
code, especially a general purpose three-dimensional one, are the following.
The model should be simple and numerically stable, yet it should be
theoretically sound and capable of reflecting the important features of the

139
RC ELEMENTS UNDER CYCLIC LOADING

experimental behaviour. Examples of three-dimensional models which

seem to meet these criteria are the hypoelastic orthotropic model in
ADINA175 and the associated plasticity model in ABAQUS.177 These models
are briefly reviewed below.
The hypoelastic model in ADINA refers to the three current principal stress
directions, and is based on the modified Saenz equation (equation (3) in
section 1.3.1.), which extends up to a crushing stress and strain on the
softening branch in compression (cr,/ and e,j in Fig. 7 of Chapter 1). It also
uses the tangent rigidity matrix of equation (163) in this latter section (u =
constant). The tangent moduli E, (i - 1, 3) to be used in this rigidity matrix
are taken as the slopes of this uniaxial a-e curve, at the current value of the
corresponding strain, e,. To take into account the effect of multiaxiality, the
coordinates of the ultimate strength (peak) point, aic, e,c, in Fig. 7 of
Chapter 1, and those of the crushing point, a,f and e,y, are scaled-up. The
scaling factors are different for the two axes but the same for all three
principal stress directions. The scaling factor for the cr-axis is computed as
follows. If ac\ > ac2 > <7C3 are the current values of the principal (pre-
dominantly compressive) stresses, the value && of the failure stress in the
direction of the most negative principal stress, a^, is computed, for constant
values of ac\ and <rc2. Then the scaling factor along the a-axis is taken as
equal to 7 = <rzc/fc (in other words, the 'ultimate strength' is increased to
and that along the e-axis equal to 7(017 + C2). Values of 14 and —0-4
174
are suggested174 for c, and c2, resulting in a somewhat larger enhancement
of ductility than of strength due to multiaxiality. The value of the initial
tangent modulus is unaffected by multiaxiality, and can be considered as the
modulus of the initially elastic triaxial behaviour. Unloading is linear-
elastic, with this latter modulus and the constant Poisson's ratio. Unloading/
reloading is signalled not independently but simultaneously in the three
principal stress directions, on the basis of the value of J2 (this is equivalent
to using an isotropically hardening loading surface with the shape of a
cylinder normal to the 7r-plane). The maximum value of J2 ever reached
during the entire previous response is saved, and used to distinguish
reloading from virgin loading.
The a-e model for concrete in compression used in the earlier special
purpose (oriented exclusively to reinforced and prestressed concrete) finite
element program,180 differs little from the one in ADINA, being closer to the
general orthotropic model reviewed in section 1.3.1. (e.g. in following
Darwin and Pecknold's equivalent uniaxial strain concept, in the rules for
unloading/reloading, etc.).
The associated plasticity model in the earlier version of ABAQUS is
identical to that of Chen and Chen.181 In addition to the failure surface
mentioned in the previous section 6.3.1., it incorporates a similarly shaped
initial yield or 'first loading' surface, that signals the departure from
linearity of the a-e response of concrete in compression at relatively low
stress levels. The plastic modulus for isotropic hardening between the initial
yield and the failure surfaces is such that the monotonic uniaxial a-e curve
in compression is reproduced. In the most recent version of ABAQUS, there
is no initial yield or first loading surface. Hence concrete is considered to
exhibit non-linear behaviour from the very beginning of loading. The
loading surface is similar to the failure surface, i.e. a Drucker-Prager cone.
In both versions of ABAQUS, unloading/reloading is linear elastic. 77
For convenience, the a-e relations implemented for concrete in a finite
element code are usually applied uniformly for all combinations of principal
stress values, even for those in the all-tension or tension-compression
sectors of stress space prior to cracking. Because of the low level of tensile
stresses for such stress combinations, the a-e models typically yield

140
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

effectively linear-elastic response in tension. Of course, once the failure

surface is reached within the part identified with concrete in tension, the
modelling of the post-failure response differs radically from that of concrete
in compression, as described in the following section 6.3.3.

6.3.3. The treatment of post-cracking behaviour: discrete versus smeared

cracks
As already stated, when the stress point reaches the failure surface over its
part identified with concrete in tension, a crack is considered to occur. So,
for crack initiation a strength criterion is invariably used. However, for
crack representation two different approaches have been proposed and used:
the discrete crack and the smeared or distributed crack approach. According
to the former, once the criterion of crack initiation is attained, an individual
discrete crack is introduced between two concrete elements. In early work
the (discrete) crack location was predetermined. To reduce the bias
introduced in this way in the solution, in later work cracks were allowed to
form along the common side of any two adjacent concrete elements, when
the average of the stresses in these two elements satisfied the crack-
initiation criterion (Fig. 138). If this criterion was satisfied by the stress
average in more than two mutually adjacent concrete elements, introducing
some ambiguity regarding the crack location, the crack was formed along
the sides of the elements which were most nearly normal to the major
principal tensile stress. Although more general than that of predefined crack
location, this approach still lacks in generality, as it forces the cracks to
follow interelement boundaries. In addition, a stiffer overall response is
produced.
In more recent work, the location and orientation of cracks is not limited
by the finite element geometry, as new elements are allowed to form, and
nodes are added at new locations, along the normal to the direction of the
major principal stress <rci at crack formation.174183 For example, a crack is
considered to start within a step of the analysis from a node of the element
where the cracking criterion is first fulfilled, and to follow the normal to the
direction of ac\ up to its intersection with another side of the element in
question. This latter point of intersection becomes a node of the finite
element mesh, to which a neighbouring existing node is moved (Fig. 139).
In this way the introduction of the crack does not, generally, increase the
number of concrete elements. A contact element is automatically introduced
between the nodes on the two faces of the crack. The constitutive relation of
this contact element expresses the dependence of the shear stress along, and
of the cohesive stress across, the crack, on shear slip and crack width, and is
based on Hillerborg's fictitious crack approach,1 4 and on Walraven and
Reinhardt's aggregate interlock model.119

Fig. 138. Discrete crack

modelling by nodal
separation on interelement
boundaries

141
RC ELEMENTS UNDER CYCLIC LOADING

Fig. 139. Mesh

rearrangement for Uncracked concrete element

propagation of discrete Crack line \

cracW74 \ 4 \ 3'\
\
NODA-
Bond 1
1 2/I 9
—>£•

6l
4 1
J10I
ir
NODF-
I 7 8\
Steel element Crack element
/

ELF
A Dowel-action element

All these variants of the discrete cracking approach require pairs of nodes
at opposite sides of a crack. So, either new nodes have to be added to the
model upon formation of a crack, or double nodes have to be provided from
the beginning along all interelement boundaries where cracks may poten-
tially develop, and allowed to separate when the crack forms and opens up.
In the first case every time a new crack is formed or an existing one
propagates, the finite element topology changes, the band width has to be
reminimized, etc. In the second case, a large number of extra nodes, and
corresponding linkage or contact elements along interelement boundaries,
need to be added to the model from the very beginning.
The discrete cracking approach of the two-dimensional special purpose
program MICRO developed at Delft,176178 deserves special mention. The
program uses special hybrid triangular elements for uncracked concrete,
with linear interpolation of stresses over the element and linear interpolation
of displacements along the element sides. (The stress and displacement
distributions are related through certain conditions that guarantee
equilibrium within the element and minimize lack of interelement equili-
brium along the boundaries.) When the cracking criterion is reached at the
centre of gravity of an element, a discrete crack is introduced through this
latter point. A discontinuous displacement interpolation is added to the
linear ones along the two sides intersected by the crack, amounting to three
new degrees of freedom: one slip translation along the crack, and a crack
width that varies linearly along the crack (Fig. 140(a)). The same is done to
the stress field in the element, which also becomes discontinuous (Fig.
140(b)). Appropriate relations are introduced between the new displacement
and the new stress degrees of freedom across a crack according to
Hillerborg's fictitious crack model,184 see section 2.3, and to Walraven's
model for aggregate interlock.'18 This is equivalent to introducing a contact
element along the discrete crack, which connects the two faces of the crack
through the normal stress-crack width (see Figs 33, 34, 45 and 46 of Chapter
2) and the shear stress-shear slip-crack width relations. Only one point of
intersection with cracks is allowed along each side of a concrete element.
So, if such a point of intersection exists already, a new crack, in the same or
in the neighbouring element, is forced to pass through the same point. In this
way a multilinearly continuous crack may form, crossing several elements
(Fig. 140(c)). The occurrence of cracking imposes no mesh rearrangement
or disconnection between elements. Moreover the degrees of freedom are
arranged in two sets: the first one, containing the original degrees of
freedom prior to cracking, is fixed, while the other, with the new ones, is
augmented every time a new crack forms. The two matrix equations for the
two sets of degrees of freedom are coupled through the unbalanced load

142
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

(b)

(c)

Fig. 140. Discrete crack modelling in program MICRO: (a) discontinuous displacement interpolation along crack;
(b) discontinuity in stress fields, for one and two cracks; (c) possible configuration of cracking

vectors, and are solved sequentially in an iterative fashion, according to the

initial stiffness (strictly speaking to the initial strain) approach. So cracking
of new elements does not require either renumbering of nodes or
reformulation of the stiffness matrix for the original system. It affects only
the equations for the additional degrees of freedom. To allow redistribution
of stresses due to cracking, only one crack is permitted to form at each step
of the analysis, in the element where the cracking initiation or propagation
criterion is violated most. Then, a new element is allowed to crack only
after the cohesive stress across the crack drops significantly.
Application of the models by Miguel et al.i74 and Grootenboer et al.n6
produces a few (or even one) dominant cracks, in problems with modes I
and II fracture. A reason for this is that both programs account explicitly for
the steel-concrete interaction through bond.
In the early applications of the discrete crack model, cracks were
considered brittle, i.e. stresses normal to the crack were reduced to zero
right after cracking. Obviously then, computed stresses at the crack tip were
mesh size-dependent, and crack propagation based on a strength criterion
was not objective. It has been shown, though, that if discrete crack
propagation is based on the stress intensity factor or on the fracture energy
release rate, Gf, then it is mesh independent and objective.185 A numerically
convenient way to achieve this is by employing Hillerborg's fictitious crack
model, i.e. by using a relation between the cohesive stress normal to the
crack and the crack width, the area under which equals the fracture energy
G{ (see Chapter 2 for such relationships). So, if the nodes on opposite faces
of the discrete crack are connected through either linkage elements or a
continuous contact element, with a normal stress-crack width relation
(model opening component) of the type above, the discrete crack approach
is objective. Such crack interface elements also lend themselves to
modelling of the interface shear phenomena, i.e. of aggregate interlock
and dowel action (modes II and III sliding components), without having to
resort to the semi-empirical 'shear retention' factor. Indeed, the two rela-
tively recent finite element programs based on the discrete crack approach

143
RC ELEMENTS UNDER CYCLIC LOADING

mentioned above 174176 employ such crack contact elements, with crack
cohesive and shear stresses dependent on crack width and shear slip.
The discrete crack approach lends itself better to the modelling of the
interface shear phenomena, and allows direct calculation of crack width. So,
whenever the behaviour is governed by one or by a few dominant cracks,
and/or the local behaviour is of interest, the discrete crack approach is the
natural way to go. However, in its general form, in which cracking is
allowed to form anywhere in the finite element mesh, this approach is
numerically inconvenient, as it requires redefinition of the mesh and
introduction of new degrees of freedom. From the computational point of
view, this is a serious drawback. In addition, up to the present time the
discrete crack approach has been applied only to two-dimensional problems
under monotonic loading, and its extension to three-dimensional conditions
and to cyclic, or generalized loading is not immediately obvious. Regarding
cyclic loading, it should be noted that although cyclic versions of the
individual constituent models of the discrete crack approach are available
(see the cyclic fictitious crack model presented in Chapter 2, and the
interface shear models with cyclic capabilities in Chapter 5), finite element
implementation of a general cyclic discrete crack model still has to wait for
the solution of important issues regarding closing of cracks and opening of
new ones in other directions, reopening of cracks, etc.
Because of its computational inconvenience and its essentially fixed-
crack character, the discrete crack approach enjoys today limited acceptance
and application. Nevertheless, when a major discontinuity in the geometry
(e.g. at the junction of two different structural elements) predetermines the
location of dominant cracks the best way of modelling the problem seems to
be to combine discrete cracks at the predefined location with a smeared
crack modelling for the rest of the structure. It is worth mentioning that a
special interface contact element for cyclic loading has been developed by
combining cyclic models of interface shear along the crack and of bond-slip
along the inclined bars crossing it.186 This special element has been
successfully used,186 at the junction of a cyclically loaded shear wall with
its foundation, the rest of the wall and the foundation being modelled using
composite steel/concrete elements with smeared cracking. The reader
should be cautioned at this point that continuous interface contact elements,
although apparently more general than the lumped ones inserted between
nodes in the early applications of the discrete crack concept, have been
found to introduce artificial fluctuations in the fraction profile along the
interface, when the crack is closed (a situation simulated by means of a very
high dummy stiffness of the interface element, in the direction normal to the
crack).187 So, in this respect the more primitive lumped interface contact
elements seem to be superior.
The discrete crack approach fits the physical concept of fracture, as
cracks are introduced via discontinuities in the displacement fields. For
problems with distributed fracture, however, as for example in shear walls
with densely distributed reinforcement and diffuse cracking, the smeared
crack approach seems closer to physical reality. According to this latter
approach, cracks are assumed to be distributed (smeared) over a concrete
element, or over the tributary volume of an integration point, and the local
displacement discontinuity at crack locations is smoothed out over this latter
volume, through the displacement interpolation functions of the finite
elements. Only the concrete tangent rigidity matrix at the (cracked)
integration point or element has to be modified upon and after cracking,
without changing the finite element topology. Due to this latter feature, and
as it imposes no restrictions on the crack direction, the smeared crack
approach has gained much wider acceptance than the discrete crack

144
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

approach, especially for situations in which the overall behaviour is of

interest, and not the localized one at the cracks.
Initiation of cracking is governed by the strength criterion. However, its
propagation to yet uncracked neighbouring elements suffers from the same
lack of objectivity problem as the discrete cracking approach: if the mesh is
refined, tensile stresses in the elements ahead of the cracked one increase,
possibly triggering the strength criterion for cracking. This mesh sensitivity
can be avoided, if Hillerborg's fracture energy based fictitious crack model
is used to establish the falling branch of the cr-e curve of concrete in tension,
to be used in the direction normal to the crack. More specifically, mesh
objectivity is achieved if this latter cr-e curve of concrete in tension is
obtained from the normal stress-crack width relation, that incorporates both
the bulk behaviour (linear elastic up to crc =/ c t) and the post-cracking
softening branch (see Figs 32, 45 and 46 of Chapter 2), by dividing the
crack width ordinates by an effective element length, A. 185189 Since the area
under this total stress-crack width relation should be equal to the fracture
energy release rate Gf, if the softening branch is approximated as linear,
then the terminal strain of this linear falling branch will be equal to 2Gf/hfct.
The effective element length, h, is the maximum length of the cracked
element region in the direction of the normal to the crack plane, i.e. the
projection of the cracked region on this latter direction.182 If cracking is
monitored at the integration point level, the cracked element region is the
tributary region of an integration point at which cracking has occurred (Fig.
141). For example, for 2 x 2 integration in the plane, or 2 x 2 x 2 in a
three-dimensional element, the cracked element region is one-quarter or
one-eighth respectively of the element area or volume. This means that the
most common practice of taking the equivalent length h equal to the side of
a finite element, is appropriate only for square elements with cracking
parallel to the sides, and with a single integration point. Taking h as equal to
the square root of the area or to the cubic root of the volume,189 does not
guarantee mesh-independent results.
The procedure outlined above for implementation of the fictitious crack
model into the smeared crack approach, imposes an upper limit on the
maximum dimension of a finite element or of an integration point tributary
region, in any (oblique) direction. As the total area under the a-e curve of
concrete in tension should be equal to Gf/h, the area under the initial elastic
branch up to the cracking stress should not exceed this value. It is
Fig. 141. Equivalent length h conservative, in this respect, to consider the case of uniaxial tension, in
in cracked concrete which cracking occurs at <r=/ct. Then
elements'7H
f?t/2E<Gt/h i.e. K2Gf£//] (165)

The right-hand side of equation (165) is usually termed characteristic

length, A, of the material. For concrete A varies between 0-4 m and 0-8 m. l84
At the limit, i.e. when h is equal to A, the curve of concrete in tension
exhibits an ideally brittle behaviour upon cracking. For h > A, this curve
will exhibit an undesirable snap-back behaviour.
The issue of mesh objectivity for smeared cracking is of practical
importance only for elements or regions which are very lightly reinforced.
In moderately or heavily reinforced elements or regions, results will show
very little dependence on mesh size, even when the approach outlined above
is not followed. This is made clear by considering the modification effected
on the falling branch of the <r-e curve of concrete in tension, to model
tension stiffening. If a linear falling branch is chosen for both cases, the one
modelling softening of plain concrete ends at a terminal strain of X/h times
the cracking strain fctIE, whereas the one for tension stiffening is typically

145
RC ELEMENTS UNDER CYCLIC LOADING

taken to end at 20-80 times/ ct /£, l79-190>191 with higher values corresponding
to heavier reinforcement. Therefore, for the values of h commonly used in
modelling, tension stiffening completely overshadows softening of
concrete.
The (uniaxial) a-e curve of concrete in tension, with its mesh size-
dependent falling branch for mesh objectivity, is used in the multiaxial a-e
model of concrete in tension. It is illustrative to refer once more, for
example purposes, to the two major general purpose three-dimensional
programs, ADINA and ABAQUS.175'177 In the latter's associated plasticity
model with isotropic hardening, the user is expected to provide as input a
multilinear a-e curve of the concrete in uniaxial tension, including the
falling branch. This user-supplied curve is used then by the program to
establish internally the dependence of plastic modulus on strain, for the part
of the model that applies for predominantly tensile stresses. ADINA, on the
other hand, constrains the falling branch of the cr-e curve of concrete in
uniaxial tension to be linear, with a user-supplied terminal strain. After
cracking, the tangent modulus of concrete normal to the crack is reduced to
a very small positive value, in the orthotropic tangent rigidity matrix
equation (6) in Chapter 1, and equilibrium iterations are performed using the
stress value computed from the falling branch of the a-e curve. This
approach is preferred over using the negative stiffness of the falling branch
in the tangent rigidity matrix, to avoid numerical problems. In both ADINA
and ABAQUS, tensile unloading and reloading from a point on the falling
branch is along a straight line from the origin to the point of maximum
previous tensile strain.
A word of caution is due at this point: as neither of these two smeared
crack-based programs include any bond link or contact elements to simulate
the bond-slip interaction of steel with concrete, and as both of them use a
bilinear a-e curve for steel in uniaxial tension, which does not lend itself to
modification to account for tension stiffening, this latter effect can be taken
into account only through the falling branch of concrete in tension.
Therefore, in these two programs this latter falling branch should be used to
model the tension stiffening effect of reinforced concrete, rather than the
mesh size-dependent softening of the concrete component. This means that
the terminal strain of the user-supplied falling branches should be assigned a
fairly high value.
The smeared crack approach has been criticized as incapable of pre-
dicting localization of fracture. However, it produces proper strain localiza-
tions for mixed mode cracking in both plain and reinforced concrete
elements,64 regardless of whether one adopts a fixed or a rotating crack
model.

6.3.4. Fixed (orthogonal or non-orthogonal) versus rotating cracks

As stated above, all smeared cracking models assume the first crack at an
integration point to occur normal to the direction of the maximum principal
tensile stress of concrete, ac\, when the failure surface is reached for the
first time. Then, stress-induced anisotropy should be introduced in the
model, with the behaviour in direction 1, normal to the crack, considered to
be fundamentally different than the one in the orthogonal direction parallel
to it. Within the framework of a three-dimensional orthotropic model for
concrete, described by the rigidity matrix Dc given by equation (6) of
section 1.3.1., right after cracking tangent modulus E\ is that of concrete in
tension, the other terms in the first row and column Dc in equation (6) are
zero, and moduli E2, £3 in directions 2 and 3 are those of uncracked
concrete, in the biaxial stress subspace within the plane normal to direction
1. This is the approach of both three-dimensional finite element programs

146
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

we have repeatedly mentioned above: ADINA, which uses a genuinely non-

linear orthotropic model for concrete, and ABAQUS, in which the biaxial
version of the associated plasticity concrete is used for the behaviour
parallel to the crack plane.
At the instant of crack formation, these directions of principal stresses,
aci (i = 1,3), are also principal directions of strain. So, right after cracking
the values of the shear moduli in directions \i (i = 2,3) are unimportant, as
shear stresses and strains in these directions are zero. Actually, an isotropic
plasticity type of model, such as the concrete in compression model in
ABAQUS, relates only the increments of principal (normal) stresses and
strains, and hence does not include any shear moduli. The role of the shear
modulus Gyi (i = 2,3) may be important, however, during the post-cracking
loading and response, if the stress-induced anisotropy caused by cracking is
taken into account. This is the case for the so-called 'fixed crack' approach
followed by all smeared cracking models which we have mentioned up to
now. l 7 5 1 7 7 1 7 9 In this approach, direction 1 of the normal to the first crack at
an integration point is fixed as an orthotropy direction for the rest of the
response, and non-zero effective shear moduli of cracked concrete, Gu (i =
2,3) are introduced. This permanent memory of damage orientation is a key
feature of the fixed crack approach.
Due to the non-zero shear moduli, shear stresses rj, (/ = 2,3) may build up
on the crack plane during subsequent loading and response, and therefore
directions 1, 2 and 3 may stop being principal directions of concrete stress.
The principal stresses, <rci, aC2, &&, in the three new principal stress
directions may violate the cracking criterion (i.e. reach the failure surface
over its concrete in tension part) anew, indicating cracking in a new
direction, 1'. However, most programs follow the orthogonal fixed crack
approach in which subsequent cracking is forced to be orthogonal to the
first. So, for two-dimensional analyses a second crack may form only in
direction 2, normal to direction 1, whereas in three-dimensional ones, two
principal stresses, &&, and acs are computed only, in the plane normal to
direction 1, and the cracking criterion is checked only in terms of these two
stress values. Therefore, violation of the cracking criterion in a direction 1'
other than directions 2 or 3 will remain unnoticed, and no action will be
taken. This has been observed to result in a much stiffer predicted response
than the actual. 187191192 To remedy this, one could follow the non-
orthogonal fixed crack approach, 178190192 " 195 in which the actual principal
stresses in directions 1', 2' (and 3') are computed and the second crack is
introduced in direction 1', non-orthogonal to 1. De Borst and Nauta,194 and
accordingly the three-dimensional Delft code DIANA, allow a second crack
to be formed only if the angle between directions 1 and 1' exceeds a certain
(empirically chosen) threshold value. Cervenka's two-dimensional non-
orthogonal fixed crack model190'195 does not include any threshold angle,
for the second crack. So, it does not suffer from the problem of discontinuity
reported by Crisfield and Wills191 for violation of the cracking criterion
slightly below or slightly above the threshold angle. The way that Cervenka
treats the orthotropic behaviour after formation of the second crack is as
follows.195 When a second crack is formed, the diagonal term of D c
corresponding to the normal modulus in direction 2 parallel to the first crack
plane is set equal to zero, and it is replaced by the addition of another
rigidity matrix to the tangent rigidity matrix of concrete, which results from
transformation to the global coordinate system of a rigidity matrix
introduced in the local coordinate system defined by the normal (1') and
the parallel (2') to the second crack plane. This latter rigidity matrix has
only one non-zero element, namely the diagonal term which corresponds to
direction 2', i.e. the parallel to the second crack. This term is set equal to the

147
RC ELEMENTS UNDER CYCLIC LOADING

normal tangent modulus of concrete in this latter direction, determined from

the cyclic uniaxial stress-strain law of concrete in compression.
The Delft approach for two or three dimensions, 178187194 and
Barzegar's 192193 approach for two dimensions introduce multidirectional
non-orthogonal cracking using the concept of strain decomposition in
cracked concrete. According to this concept, the strain increment tensor is
decomposed into one component for the intact concrete between the cracks
and as many additional components as the number of multidirectional
cracks at a point. The strain increment tensors associated with each one of
the multidirectional cracks is obtained by appropriate rotation trans-
formation of the corresponding crack displacement increment vectors (with
one mode I opening component and in general two modes II and III sliding
components) from the local coordinate system of each crack to the global
system, and by normalization through the corresponding crack band or
effective element length (see Fig. 141). The traction increment vectors of
the individual cracks are similarly transformed into a stress increment tensor
in the global system. Incremental crack traction and displacement vectors
are related through the crack rigidity matrix, having on the diagonal the
appropriate tangent moduli of the crack in the opening mode I and in the
sliding modes II and III, and possibly explicit off-diagonal coupling terms.
The incremental rigidity relation between the global smeared stresses and
smeared strains is obtained then in a form and a manner similar to those in
classical incremental plasticity. This computationally tedious approach
offers the potential for separate treatment of the interfaces and their
behaviour from the bulk concrete between the cracks.
In the Delft and Barzegar's multidirectional crack models, previous cracks
are allowed to unload, either in a secant manner, 192193 or elastically.18? As
the direction and the state of all previously formed cracks are retained in
memory, the number of cracks at a point should be limited, to keep memory
requirements from exploding. This is achieved by introducing as criterion the
aforementioned threshold value of the angle between a previous crack and
the new direction of crci. Rots et al. suggest the criterion for formation of a
new crack to be either the build up of a principal tensile stress ac\ at an angle
greater than 30° (the threshold value) from a previous crack, regardless of its
magnitude, or the exceedance of the tensile strength of concrete,/ ct , by ac\,
regardless of direction.187 Barzegar193 suggests a lower threshold angle (15°),
but considers its attainment as a necessary condition for new crack formation,
and imposes further the requirement that in the new direction of ac\, either
the crack initiation criterion on ac\ and <JCI has to be fulfilled again, or the
magnitude of ac\ should reach 80% of the corresponding value at first crack
formation. In accordance with the experimental results which suggest that
only the most recently formed crack is active,196 Rots et al. allows all
previous cracks to unload elastically, i.e. to close, becoming inactive and
retained in memory only for the possibility of reactivation. Barzegar and
Ramaswamy consider secant unloading of previous cracks, and find that the
latter close and become inactive by themselves, after formation of new
cracks.192'193 Then the non-orthogonal multidirectional crack approach
reduces to one of a single currently active crack, which differs from the
rotating crack approach described below, only in that it allows for different
directions of principal stresses and strains, through an externally defined
modulus of the interface in sliding shear. Indeed, Barzegar has shown that
under plane stress conditions, the incremental rotations of principal stresses,
&Qa, and principal strains, d9e in a load step are related as follows
dflg = 2(6, - e2) dr, 2 = /? (Ec2 - v2Ecl) (e, - e2)
d0e {a\ - a2) d7,2 1 + v £ci(l - v)e, - (Ec2 - vEc{]e2

148
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

In equation (166) dr^/d^n is the instantaneous effective shear modulus

of cracked concrete, the shear retention factor j3 is the ratio of this latter
modulus to that of uncracked concrete, and Ec\, Ec2 are the secant moduli of
concrete in the directions normal and parallel to the currently active crack,
respectively. So, the interface shear and dowel action behaviour and the way
they are modelled, control the deviation between the principal directions of
stress and strain. It is worth mentioning that in the multidirectional non-
orthogonal cracking models above the fracture energy, Gf available to new
cracks, is reduced by the amount consumed in all the previous cracks.
In a fixed (orthogonal or not) crack model the value of the effective shear
modulus of cracked concrete may be important. This latter modulus is
usually expressed via its ratio to the shear modulus of uncracked concrete,
Go, termed 'shear retention factor' (section 5.2.9.). This shear retention
factor is sometimes assigned a constant value less than 10, 179 or it is taken
to decrease with the strain e\ normal to the crack plane.175'177 In his model
for monotonic loading,190 Cervenka proposed a shear retention factor,
GX2IGO, equal to 1 — (e t /0-005) k , and a parameter k equal to 0-4 (although
the predictions for monotonic loading seem insensitive to the exact value of
k within the range from 0 to 2-0). In his recent proposal for generalized
loading,195 Cervenka has adopted Kolmar's197 expression for a shear reten-
tion factor equal to //i[(27 - 2e)/ei]/c, in which c = 10 - (lOOOp - 5)/6,
and p is the effective reinforcement ratio normal to the crack
(p = Hjpi cos2 6j with 0, denoting the angle between reinforcement layer
i and the normal to the crack plane).
Using a constant shear retention factor ft between 005 and 0-4, Wang et
al.179 have reached the same conclusion as Cervenka regarding the
insensitivity of the results to the value of the effective shear modulus of
cracked concrete. Barzegar192 has also found that although the values
0 = 001 or /3 = 0-8 x 10~4/ei produce an extremely flexible response,
results are close to the experimental ones for /? ranging from 0-20 to 0-99.
These conclusions seems to apply only for the non-orthogonal fixed crack
approach: using an orthogonal fixed crack model, Crisfield and Wills191
have found strong dependence of the post-cracking load-deformation
response on the value of the shear retention factor, considered as normal
strain-independent. Similarly, Rots et al. have found that values of (3
significantly greater than 0 (e.g. above 001) yield overly stiff response
when used in an orthogonal fixed crack model. 64187
In their proposal for a mesh-independent orthogonal fixed smeared crack
model,182 Dahlblom and Ottosen derive the effective shear modulus Gy in
directions ij, on the basis of addition of shear flexibilities

(167
»

in which Go is the shear modulus of uncracked concrete, the sum in the last
term extends over the cracked directions (to direction / only, or to both / and
j , if there is cracking only normal to direction i, or in both directions,
respectively) and the shear slip modulus C s is the ratio of 77, times the crack
opening wh over the resulting shear slip Sy, along the crack (assuming that
such a proportionality between 77/w, and Sy exists).
It is worth mentioning that even though in many cases response predictions
may seem relatively insensitive to the value of the shear retention factor, a
non-zero shear retention factor must be included in a fixed crack model, to
avoid numerical instabilities and spurious kinematic modes associated with
the displacement degrees of freedom in the direction normal to a crack, if all
concrete elements connected to a node crack in the same direction.

149
RC ELEMENTS UNDER CYCLIC LOADING

A disadvantage of the fixed crack approach is that, in spite of provisions

for a second crack, oblique to the first, at the same point, if the cracking
criterion is violated for the second time, the approach tends to produce a
crack pattern close to that at first cracking, and far from the one observed
near ultimate strength. Moreover, an orthogonal fixed crack model produces
a stiffer than actual response, and overestimates the ultimate load.191 As a
remedy, the rotating crack approach has been proposed by Gupta and
Akbar198 and adopted by many researchers. l99~202 In the early and still more
widespread version of this approach, directions of orthotropy 1 and 2 (or 1
to 3, in three-dimensional analyses) are chosen to coincide with the
instantaneous principal directions of smeared strain. So, a notional, but not
physical, crack is created, that follows the 'average' most predominant or
active crack direction at a point, without retaining any memory of the
previous history of cracking there. Being memoryless, the rotating crack
model is a secant, rather than a tangent approach. In a rotating crack
approach total stresses a, and strains e, in the global coordinates systems xyz
are obtained from the corresponding principal values, through appropriate
rotation transformations. The incremental (tangent) relation between a and e
in the global system is obtained by differentiation of these relations,
producing terms which include partial derivatives with respect to the angles
of the principal stress directions with respect to the xyz axes, as these angles
change. This differentiation operation produces a tangent rigidity matrix
relating principal stresses and strains, transformed into the xyz system. The
second term is due to the rotation of the principal axes, and has been
computed for two-dimensional problems as follows

1 -1 -cot 20
(o-cl — crc2) sin220
-1 1 cot 20 (168)
8(ex - €yy) 2
-cot 20 cot 20 cot 20
in which ex and ey are normal strains in global directions x and y, and 0 is the
angle from the x axis to the instantaneous direction 1, measured counter-
clockwise.
The value of the effective shear modulus is less important than in the
fixed crack approach, as the shear strains are zero in the instantaneous
directions of principal strain. Its continuously changing value is implicit in
the tangent rigidity relation between da and de in the global system. As
shown by equation (166), for two-dimensional stress conditions the
coincidence of the principal stress and strain directions yields an effective
shear modulus equal to (crcl — crc2)/2(ei — e-i). It has been shown187 that if
this latter value of the effective shear modulus is used, the threshold angle is
taken equal to 0°, and elastic unloading of the previous cracks is assumed,
then the multidirectional non-orthogonal crack model approach produces as
a limit the rotating crack model. In comparison to the general form of the
former, the rotating crack approach is much simpler and less demanding in
computer memory, at the expense, of course, of the potential for more
realistic modelling of the interface shear behaviour, offered by the
multidirectional non-orthogonal crack approach. Comparative applications
of the rotating crack model and of the non-orthogonal fixed crack one, have
been made on plain and reinforced concrete examples. In the fixed crack
model, various values of the threshold angle from 0° to 90° and of the shear
retention factor, /?, have been adopted. This factor has been taken to
decrease with the ratio a = ef/e™, of the strain normal to the crack to the
falling branch terminal strain in the fictitious crack normal stress-strain
model, as (J = (1 - a)2/[\ - (1 - a) 2 ]. The results have shown that the
rotating crack approach produces a slightly less stiff response than the non-

150
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

orthogonal fixed crack model with shear straining, even becoming negative.
In addition, the stress normal to the crack decreases with shearing. So, the
crack rotation introduces implicit shear softening and shear-normal stress
coupling in the rotating crack model, without the need to resort to the
complicated interface shear modelling required by fixed crack models.187
These same examples have shown the inadequacy of the fixed orthogonal
and non-orthogonal crack model with large values of the threshold angle.
So, overall the rotating crack approach seems to achieve the best
combination of simplicity, accuracy and convergence characteristics.
The smeared crack approach seems to suffer from stress-locking in the
vicinity of dominant discrete cracks. This problem is less serious if one uses
the rotating crack approach, or the fixed crack approach with a zero shear
retention factor ((3 = 0). In all other cases significant stresses are predicted
in directions non-parallel to the crack. As shown by the more realistic
predictions of discrete crack models for the same problem, these latter
stresses are spurious, and are responsible for the excessive stiffness of the
response. The problem of stress-locking seems to be inherent in the smeared
cracking approach, and is attributed to the imposed displacement continuity
between softening cracked elements and neighbouring uncracked ones: as
strains in the former elements increase normal to the crack, neighbouring
elements are forced to follow and their stresses increase spuriously, leading
even to the artificial extension of cracking into them, which works against
strain localization and is also undesirable. The stress-locking problem
can be reduced by employing the rotating crack approach (the fixed crack
approach with /3 — 0 leads to spurious kinematic modes and instability
problems), but it can be avoided only through use of a discrete crack model.
The fixed crack and the rotating crack approaches have been proposed
and used, the former far less successfully than the latter, for monotonic
loading, proportional or not. For cyclic loading in two dimensions, two
variants of these approaches have been developed, to deal with the
formation of cracks in two nearly orthogonal directions, and for their
opening and closing due to the reversals of loading and response. The
former approach, proposed by Okamura and his co-workers,203'2 is a fixed
crack approach with provisions for formation of a second (fixed) crack and
for shifting the directions of orthotropy to the latter (with direction 1 normal
to the new crack), if the smeared strain normal to this second crack becomes
greater than that normal to the first.
The shear stiffness of the Okamura model is based on a non-linear
interface transfer model with the following monotonic curve, which serves
as an envelope for cyclic response

/3 (169a)
TT^
in which a = 712/ei is the ratio of shear strain to that normal to the crack,
and fc is in MPa. The compressive stress component developed normal to
the crack due to the interface shear is equal to:

(169b)

This component should be superimposed to the concrete stress developed in

direction 1 due to t\ alone. Clearly equations (168) and (169) introduce in
the rigidity matrix D c in two dimensions, coupling terms between &Tn and
dei (term 31 of Dc), and between dac\ and d7i2 (term 13). Unloading from
the envelope curve, equation (169a), in the space of Tn and a is linear to a
point on the horizontal axis with abscissa equal to 0-85 of the strain ratio a

151
RC ELEMENTS UNDER CYCLIC LOADING

at the stress reversal, from where reloading starts on a straight line passing
through a point on the vertical axis at an ordinate equal to 010 of the value
of T\2 at reversal. Reloading follows this line up to its intersection with the
envelope curve in the opposite direction of loading, for first loading in this
latter direction, or up to the last unloading branch from it, for subsequent
cycles. From then on, the envelope curve or this latter branch is followed.
Finally, the value a — 0-5 is specified as a criterion for shear failure.
The second approach for cyclic loading is that proposed by Stevens et
al. as a modification of the rotating crack approach followed in the
numerical implementation of the modified compression field theory,202 to
cover cyclic loading in two dimensions as well. This approach has been
developed on the basis of the results of cyclic shear tests on reinforced
concrete panels, which have shown that during certain parts of the cyclic
response the principal directions of strain lag behind those of stress by
approximately 90°. This observation instigated the alternative assumption
that the principal directions of the concrete stress increment tensor da
coincide with those of the strain increment tensor de of the reinforced
concrete continuum. In the finite element implementation of the model, at
each step of the incremental analysis the strain increment tensor de is
computed at each integration point from the displacement increments of the
step, and the corresponding principal directions of the strain increment
tensor are determined. The stress and strain tensors computed at the end of
the previous step of the analysis are transformed to the local coordinate
system defined by the aforementioned principal directions. Then the normal
stress increments in these two directions are calculated from the corres-
ponding normal strain increments and from the values of normal stress and
strain in these two directions at the end of the previous step. (The shear
stress increment is zero, as these two directions are considered as principal
directions of the stress increment tensor.) Calculation of these two normal
stress increments, and determination of those terms of the tangent rigidity
matrix D c which relate the normal stress and strain increments, is performed
independently in each direction, on the basis of the cyclic law adopted for
the cracked concrete, described later on. The shear term on the diagonal of
Dc is continuously changing and equal to (dcrci — d<7c2)/ 2(de, — dt2) (see
the corresponding term for the rotating crack approach based on total
stresses and strains. For a linear-elastic isotropic material, this shear term of
both approaches equals one-half the modulus of elasticity). However, as this
shear term refers to principal directions of the stress and strain increments,
its exact value is immaterial. Finally, the tangent rigidity matrix D c is
transformed from the coordinate system of the instantaneous principal stress
and strain increments to the global coordinate system xy.
The ability of both approaches described above for cycling loading to
reproduce cyclic test results on two-dimensional reinforced concrete panels
has been verified.203"205 However this capability seems to be largely due to
features of the models other than their provisions for crack rotation.

6.3.5. Accounting for bond-slip effects in smeared cracking models,

through tension stiffening
In section 6.2. we saw how the bond-slip interaction between steel bars and
concrete can be explicitly modelled, through the use of special bond-link or
contact elements, interposed between discrete steel and concrete elements.
Such elements can be used in conjunction with discrete modelling of cracks,
in which case the variation of bond, steel and concrete stresses between
cracks is reflected in the results, at an accuracy which depends on the degree
of mesh refinement between consecutive discrete cracks. They can also be
used in conjunction with the smeared modelling of cracks. However, in this

152
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

case results reflect only the trend of the variation of average bond, steel and
concrete stresses along the bar, to which such bond-link or contact elements
are connected. Local variation between cracks is, obviously, unaccounted
for, since the model cannot provide any information regarding the location
of individual cracks. For smeared cracking, it is computationally more
efficient to assume perfect bond between steel and concrete and account
implicitly for the effect of bond-slip on the average stresses and strains in
the steel and the concrete, through appropriate, yet empirical, modification
of the properties of these two materials. So, at the expense of some accuracy
in the description of the variation of average steel and concrete stresses,
significant savings can be effected, as the extra degrees of freedom of the
bond-link or contact elements at the steel-concrete interface can be
eliminated, and these special elements do not need to be developed and
added to the finite element library. The modification of steel or concrete
properties for this purpose, is effected through appropriate incorporation of
the tension stiffening effect in their constitutive models.
At this point it seems opportune to remember the essentials of the tension
stiffening effect in reinforced concrete, on the basis of a prismatic axial
member, cracked due to concentric axial tension. At the cracked cross-
sections, concrete stress is zero and steel stress is maximum. Due to transfer
of stresses from the bars to the concrete through bond, tensile stresses
develop in the concrete between the cracks, attaining their maximum value
at crack mid-distance. So, the average concrete stress along the element is
non-zero. Because of the transfer of tensile stresses to the concrete, steel
stress decreases with distance from the crack, becoming minimum near mid-
crack. Equilibrium requires that at any cross-section steel stress is equal to
that at the crack minus the concrete stress divided by the steel ratio, p. The
smeared (i.e. average) axial stress in any cross-section of the reinforced
concrete element is equal to the sum of the corresponding concrete stress
plus the corresponding steel stress times p, and remains constant along the
element. The smeared axial strain of the element is equal to the average
value of the steel strain between two cracks, and is considerably less than
the steel strain at the cracked cross-sections. Actually, yielding of the steel
takes place only at these cross-sections and in their immediate vicinity.
Immediately after yielding at a cracked cross-section, the reinforcement
reaches strain-hardening, although at a short distance from the crack bond
may have reduced steel stresses and strains to below the yield values.
As the smeared axial strain is considerably less than the maximum steel
strain, which corresponds, through the stress-strain curve of steel, to its
stress at the cracked cross-section, the plot of this latter steel stress against
the smeared axial strain lies above the stress-strain curve of bare steel. On
the contrary, the plot of the average steel stress along the bar against smeared
strain lies below this latter curve. Finally, the plot of the average concrete
stress along the element against smeared strain exhibits a long post-cracking
tail, far above and beyond the falling branch of plain concrete in tension. All
these effects are demonstrated in the experimental results of Fig 142, taken
from Shima et al.,206 and constitute the well-known tension stiffening effect
in cracked reinforced concrete. The term is due to the stiffening of the bare
steel, due to the participation of concrete in tension between the cracks. Its
underlying reason is the bond between steel and concrete. The lower the
reinforcement ratio, the more pronounced the tension stiffening effect is, and
the more it affects the post-cracking deformations of reinforced concrete.
Extensive experimental studies of the tension stiffening effect have been
performed, under monotonic, repeated and reversed loading. The works of
Hartl,207 Guenther208 and Fehling209 from which the results in Fig. 143 are
drawn, are good examples.

153
RC ELEMENTS UNDER CYCLIC LOADING

Strain distribution

A A A y
55

0-5 10 1-5 20 2-5

Position along bar: m
(a)

Fig. 142. Concentric tension test results after Shima et al.,206

demonstrating the tension stiffening effect: (a) distributions of strain
and stress of steel bar in the post yield range for Specimen No. 4; (b)
load-average strain and average stress-strain relationships of the steel Average strain:
bar and the concrete for Specimen No. 1 (b)

In the one-dimensional case, such as that shown in Fig. 142, tension

stiffening can be incorporated in a smeared cracking model in two ways:
according to the first, and more direct, approach, a long-tailed average
concrete stress-smeared strain curve, such as that shown at the bottom of
Fig. 142(b), is used for the concrete, whereas for steel an average steel
stress-smeared strain curve is applied, which falls below that of bare steel
(middle of Fig. 142(b)). When this first approach is used, it is said that
tension stiffening is associated with the concrete. In the second approach,
that of associating the tension stiffening effect with the steel, the (cyclic)
stress-strain curve of plain concrete in tension is taken as the average
concrete stress-smeared strain curve, whereas the average steel stress-
smeared strain curve is obtained from that of the total smeared stress-
smeared strain curve (top of Fig. 142(b)) by dividing its ordinates by the
steel ratio p, after subtracting from the latter those of the adopted concrete
stress-smeared strain curve.
In the two approaches above, the tail of the average concrete stress-
smeared strain curve, or the total smeared stress-smeared strain relation are
obtained either purely empirically by curve-fitting to test results, or semi-
empirically. In the second case the differential equation of bond, a
simplified bond-slip law (as the rigid-plastic relation for monotonic loading
and the Coulomb-friction law for cyclic, adopted by Fehling,209 see also
section 4.3.2.) along with certain simplifying assumptions regarding the
probability distribution of crack spacing (hyperbolic, according to Fehling),
or the variation of slip along the bar (linear from the crack surface to crack
mid-distance, as assumed by Fehling), are employed, to develop a
'theoretical' total stress-smeared strain curve.
In the one-dimensional case the two approaches in principle give the

154
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

Fig. 143. Tension stiffening

effect in the cyclic uniaxial
tension experiments: (a) by
Guenther/"* (b) by
Fehling209

80000

60000

40000

20000

0-1 0-2 0-3 0-4 05 0-6

Deformation: mm
(b)

same end result, as both reproduce the desired total stress-smeared strain
curve. However, for concrete reinforced in two (usually orthogonal)
directions, loaded biaxially and cracked at an angle to the reinforcement, the
two-dimensional extensions of the two approaches are based on different
(influential) assumptions, and therefore yield different results. For the two-
dimensional problem Gupta and Maestrini183'210 have developed an 'exact'
analytical solution. Their work is based on the assumptions of

(a) a linear bond-slip relation

(b) rotating crack directions, which coincide, at least prior to steel
yielding, with the principal directions of smeared strain
(c) a linear-elastic isotropic law for concrete, with the same value of E in
tension and compression.

Cracking is signalled with the help of a biaxial failure envelope which is

linear in the tension-compression quadrant, between values of — cf and / c t
that may decrease with the value of the principal tensile strain, C|, as
exp[—c(e\ — ei cr )], where c is a non-negative constant and ei c r is the value
of ei at cracking. Moreover, the isotropic value of the concrete modulus, is

155
RC ELEMENTS UNDER CYCLIC LOADING

taken equal to the secant modulus in the direction of principal concrete

compressive stress, crC2, as this modulus is obtained from an orthotropic
equivalent uniaxial model of the type presented in section 1.3.1. It is
assumed in this latter model that the secant modulus at the peak of this
equivalent uniaxial stress-strain curve is independent of the stress ratio
0ci/0c2- Despite its linearity assumptions, the model is strongly non-linear
(the main non-linearity being in the calculation of the mean crack spacing,
knowledge of which is required by the model, through iterations) and very
complex. Nevertheless it fails to produce as good a fit to the test results as
the phenomenological approximate tension stiffening models described
below probably because of the many linearity assumptions. In an earlier,
more empirical and more practical approach to the two-dimensional
problem of a mesh-reinforced doubly (but not necessarily orthogonally)
cracked concrete in biaxial tension, Floegl and Mang211 have derived
multiplicative tension stiffening factors to be applied to the steel stress and
stiffness. These factors are given by algebraic expressions in terms of the
bar diameter, the crack distance, and the average bond and steel stresses
between cracks. These latter stresses are given by non-linear empirical
functions of the slip.
Between the two-dimensional extensions of the two phenomenological
approaches for treatment of tension stiffening, which have been presented
before, more popular is the former, i.e. the one which associates tension
stiffening with the concrete. Its various alternative options differ in their
falling branch of the concrete smeared stress-smeared strain law in tension,
and on whether or not they reduce the average stress-smeared strain curve of
steel below that of bare steel. Early work in this direction covered only
monotonic loading. In Cervenka's early contribution the average
principal tensile stress of concrete is given by the following expressions

crci = Ec e, for e.{ <fct/Ec (170)

*ci =/ct [l - ( 0 ^ ) k 2 ] > 0 f

or e, >fct/Ec (171)

Later,195 Cervenka has set k2 equal to 10, as suggested by his earlier

parametric studies. Equation (171) falls below the softening branch
proposed for the same purpose by Vecchio and Collins202

J & ("2)
Equation (172) has been fitted to a large number of experimental points,
which exhibit considerable scatter. An alternative to equation (172) has
been proposed more recently by Collins and Mitchell,212 yielding smaller
and more realistic values of <rci for large values of ei

f r e >
° ^ <173>
in which a, = 1-0 or 0-7 for high bond or smooth bars, respectively, and
a2 — 1-0 or 0-7 for short-term monotonic loading, or for sustained or
repeated loading, respectively.
In DIANA and according to Crisfield and Wills191 the falling branch is of
the form of equation (171) with k2 = 10 and with the value 0005 in the
denominator replaced by a strain parameter. In the application of DIANA by
Wang et a/.179 the value of this strain ranges from 0001 to 0003 and is
limited by the yield strain of steel. Crisfield and Wills state the terminal
strain as equal to 0-004. It is also to be remembered, that the user-specified

156
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

linear or multilinear falling branches of the <rci — e\ curve of concrete in

ADINA and ABAQUS respectively, can be used to incorporate the tension
stiffening effect in the analysis using these two three-dimensional programs.
Finally Hu and Schnobrich200 use the following descending branch

fore!>/ct/£c (174)
(1 +1000 c, (0/9O)15)
in which 0 is the angle in degrees between direction 1 (the normal to the
crack) and that of the reinforcement with the higher steel ratio. It is
noteworthy that for 6 approaching zero equation (174) yields stress values
close to or equal to fct, regardless of the strain value. These unrealistically
high values of CTCI are limited, however, by the requirement that the total
stress in the direction of any reinforcement layer should not artificially
increase beyond that corresponding to yielding of this layer. This
requirement is satisfied if

i/ (175)

in which asi is the average steel stress in reinforcement layer /, and ^, is the
angle between direction 1 and the bars of layer /. A similar limitation is also
imposed on the relatively high values of ac\ obtained from equations (172)
and (173) for large values of ei. 205 This is accomplished through a more
complicated procedure of checking whether shear stresses can be transferred
by the cracked interface, without exceeding the yield stress of steel.
All the aforementioned proposals have limited provisions for cyclic
loading. Unloading from the falling branch of the concrete ac\ — ei curve is
linearly oriented towards the origin, and reloading is again from the origin
towards the point of maximum e\ value in the previous response. 175177191
(It is noteworthy that in the rotating crack approach by Crisfield and
Wills,191 this latter maximum ever value of e\ may be associated with a
different direction of the principal tensile strain, e\.) Moreover, the above
proposals include no reduction of the average steel stress-smeared strain
curve below that of bare steel, due to the incorporation of the entire tension
stiffening effect in the stress-strain curve of concrete. On the contrary, the
two models described below are more complete in both these respects.
Okamura and his co-workers203'204 have proposed a relation between ac\
and ei which consists of a monotonic curve that serves as an envelope for
the cyclic response, and of rules for unloading and reloading. The
monotonic curve comprises the familiar ascending branch, equation (170),
a horizontal branch thereafter up to a strain of 2fct/Ec, and a falling branch of
the following form

ffci =/ct (~~) for 6, > 2/ ct /£ c (176)

\Cc Cl/

For the exponent c the values 0 • 2 and 0 • 4 are suggested for welded wire
mesh or deformed reinforcement, respectively. The curves of unloading
from the envelope and for reloading to it consist of a superposition (through
addition of the stress-ordinates) of two stress-strain curves. The first is
intended for modelling the interaction with the steel through bond, and has
parabolic unloading from the point a c i e — ei>e on the envelope to a point on
the vertical axis at an ordinate of —0 • 0016£ei)e, at which the parabola has
horizontal tangent. Reloading from that latter envelop point, or from any
other intermediate point on the unloading parabola, is linear, directed
towards the envelope point <7cie — ei,e- The second component of the

157
RC ELEMENTS UNDER CYCLIC LOADING

unloading-reloading model is a cyclic contact model, such as the ones

outlined in Chapter 2 for plain concrete in cyclic tension, with
normalization of the horizontal axis by the effective element length of
Fig. 141. Any one of those models can be used for the present purposes as
well. Nevertheless, the model by Shima et al.206 is outlined also for
completeness. It is activated only for 6| less than a contact strain value,
approximately equal to 000015, below which linear unloading occurs from
zero contact stress towards negative (i.e. compressive) values, at a slope of
Ec/3. This unloading branch merges with the monotonic stress-strain curve
of concrete in compression, or with a reloading branch of it, depending on
whether we have virgin loading in compression normal to the crack or not.
Reloading is elastic, with a slope equal to Ec, up to the horizontal axis and
horizontal (i.e. at zero stress) thereafter. On the basis of analytical
parametric studies, Okamura and his co-workers reduce the ordinates of the
steel average stress-smeared strain curve due to tension stiffening, only for
the monotonic (or envelope) branch, while leaving the rules for unloading
and reloading unaffected. The reduced envelope curve is taken elastic-
linearly strain hardening, with an effective yield stress equal to

/y,ef = /y - <7clCOS20, //>, > fy y/(250PxPy/fc) (177)

In equation (177)/ c is in MPa and the empirical coefficient 250 is replaced
by 10 for welded wire mesh. The effective strain-hardening modulus is
approximately equal to
067 025
/inn\ O-l-^
. - A /j A\ ° /m\
( 400^

Jy / \PxJ
/30N

\/c /
(|7g)

in which 6 is the angle of the reinforcement layer with ratio px with respect
to direction 1 (the normal to the crack) and / c , / y are in MPa. For welded
wire mesh coefficient 100 is replaced by 72.
The other model which associates tension stiffening with concrete but
covers cyclic loading as well, is the aforementioned model by Stevens et
al.205 In this model the envelope for concrete average stress-smeared strain
in tension, which is identical to the curve for monotonic tension, consists of
the linearly-elastic branch up to tensile strength, equation (170), and a
falling branch thereafter, that includes the entire effect of tension stiffening
and is given by

in which the asymptotic value a equals

2
cos20, -I- sin21 —- )

In equation (180) index / refers to different reinforcement directions, 0, is

the corresponding angle with direction 1, and p,, $, (mm) are the
corresponding steel ratio and bar diameter. The term 270/y/a in the
exponential should not be taken less than 1000.
According to equation (179) the average tensile stress of concrete falls off
rather fast with increasing smeared strain. By doing so equation (179)
accounts indirectly for the reduction in the value of the tensile stress ac\ due
to the incapability of cracks to transfer shear stresses by aggregate interlock
beyond a certain value, which depends on crack width and on the stress of
the reinforcement crossing the crack. °2

158
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

Region 1
Region 2
V7\ Region 3

Compressive
envelope
Ideal reloading curve
L \ N Cracks open
Y/ X Cracks closed Hypothetical unloading path

/ A(c c a n , fc
I A'fcm. U)
Cracking point

'cm Compressive
envelope E

) /
Hyperbolic
unloading path,
Equation 2 1 A— Ideal reloading curve

fen
Equation 1 v

Tensile
•

y
TV/
/A fen

Cracking point
envelope
envelope

Fig. 144. Rules for unloading—reloading from compression to tension in cracked reinforced concrete according to
Stevens et al.20

The envelope curve of equations (170) and (179) is supplemented with a

set of complex rules and corresponding curves for the unloading and
reloading behaviour, from tensions to compression and vice versa (see Fig.
144 and Stevens et al.205 for details).
Finally, as all the tension stiffening has been assigned to the concrete
through equation (179), the ordinates of the steel stress-strain curve beyond
yield are reduced by 75/ct/<I>, to counterbalance the asymptotic term a in
this latter equation.
Although less widespread than the approach of associating the entire
tension stiffening with the concrete, the alternative of assigning most of it to
the steel seems closer to physical reality, as tension stiffening arises from
bond stresses, and hence it should be associated with the steel bars and with
their direction. Actually this latter practice has been recommended on these
grounds by Wang et al.,119 who have used the other approach, and by
Kolleger et al.,2li who have implemented and used both approaches, to
discover that for pure shear the former approach significantly under-
estimates stiffness near ultimate strength, whereas the latter comes quite
close to the experimental behaviour. Mehlhorn171 has also found that if
tension stiffening is associated with the concrete, but is expressed in terms
of the strain in the direction of the reinforcing steel, results are the same as

159
RC ELEMENTS UNDER CYCLIC LOADING

those obtained when tension stiffening is assigned to the steel. On the

contrary, if tension stiffening associated with the concrete is expressed in
terms of the principal strain, e\, its effect vanishes too early.171
The main computational contribution in the direction of associating
tension stiffening with the steel is the recent work by Cervenka et a/. 195214
In his early work195 Cervenka has adopted equations (170) and (171) with
k2 = 1-0 as the average stress-smeared strain of concrete in tension, whereas
most recently214 he replaced equation (171) with a linear falling branch with
a softening modulus

f2h
£ = ^ (181a)

in which Gf is the fracture energy of concrete and h is the effective length of

the cracked element or integration point region. This falling branch does not
represent tension stiffening, but the softening of concrete in tension, as
dictated by the requirement for mesh objectivity of the smeared crack
approach (section 6.3.3.). If the value of Gf is not known, the ratio 2Gf/fct
can be approximated in equation (181a) by the crack width corresponding to
complete loss of contact, which is approximately equal to 0-05 mm. It is
obvious that the linear falling branch with the modulus of equation (181a)
can be replaced by any other which is based on the fictitious crack
approach, and describes better the cyclic stress-crack opening relation of
concrete in tension.
In good approximation the average stress-smeared strain curve of
concrete in tension, including tension stiffening, can be taken as the one
above, followed by a tail at constant stress ac\ = 0-4/ct, up to the strain
which corresponds to yielding of steel (see Fehling209). This tail is added to
the ordinates of the monotonic stress-strain curve of steel. The end result is
that the average steel stress that corresponds to a given smeared stress, is
taken equal to the corresponding value of the stress-strain curve of bare
steel, increased by

0-4/ct - Y l acJ cos2(9/ - °

in which <7C/ > 0 is the tensile concrete stress normal to crack j (j' = 1
denotes the first crack and j = 2 the second, if it exists), and 0j the angle
between the reinforcement layer in question and the normal to crack j .
However, this total steel stress is not allowed to exceed the yield stress of
steel. Finally, similarly to what is done by Shima et al.206 and in accordance
to the limited test data on tension stiffening under cyclic loading,213 the
effect of tension stiffening is included only in the stress-strain curve of steel
for virgin loading, but not in its unloading and reloading branches.
Application of Cervenka's latest tension stiffening approach, as well as of
Mehlhorn's approach of associating tension stiffening with the concrete, but
expressing it in terms of the strain in the steel, requires that within the same
element, information is available both for the concrete and for the steel. So,
these approaches can be implemented only within the composite steel/
concrete elements described in section 6.4. below. To circumvent this
problem, and to retain the rationality of associating the strain softening of
cracked concrete with the direction normal to the cracks, and the tension
stiffening due to bond with the direction of the reinforcement, Barzegar and
Ramaswamy193 introduced a hybrid approach in their two-dimensional
model with separate elements for the concrete and for the steel layers in the
two orthogonal directions x and y. After cracking, the concrete element
employs a ac\ — et relation in the direction normal to the first crack, the

160
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

falling branch of which represents pure concrete softening in tension (i.e.

the total area under this curve equals Gf/h). Unless direction 1 of the normal
to the crack coincides with one of the two reinforcement directions (JC or y),
the strain in either one of these directions will reach the cracking strain of
concrete, eci=fct/Ec, later. When this happens, the concrete <rci — e\
relation normal to the crack is de-activated (the value of ac\ drops to zero),
and two 'tension stiffening springs' are activated in the reinforcement
directions. Although they are in the steel directions, these springs relate
concrete smeared stresses, a^ and acy, to the corresponding smeared strains,
ex and ey. Their ac — e relation is softening and trilinear, starting from a
stress equal to the value of ac\ at the moment that the maximum of ex and ey
reaches ecr, and decreasing linearly to O-35crci for e(xoty) equal to 50% of the
steel yield strain, and further to 0-1 ac\ when the latter yield strain is
reached and exceeded. It must be noted that such a shift of concrete
softening to tension stiffening in the reinforcement directions can be
implemented within the same (concrete) element, only for uniformly
distributed mesh reinforcement in two given orthogonal directions, x and y.

6.4. Composite 6.4.1. Introduction

steel/concrete
modelling, for two- As already stated in section 6.1, macrolevel composite steel/concrete
dimensional elements have been proposed to analyse the overall non-linear behaviour of
homogeneously homogeneously reinforced concrete elements. In their present form the
reinforced elements corresponding models treat reinforced concrete, with or without cracks, as a
non-homogeneous continuum with smeared properties. In these properties
the effect of discrete reinforcing bars and cracks is locally averaged within
the volume of a finite element, or within the tributary volume of an
integration point of such an element. In addition to incorporating the
interaction between the two materials in an implicit, yet efficient and
relatively simple manner, this approach also offers computational
advantages. It lends itself to numerical implementation in standard finite
element codes, by simply introducing the constitutive relations of the
composite material, i.e. of reinforced concrete, in the two- or, in general,
three-dimensional elements of the code, isoparametric or not, and it does not
require either special bond-link elements between concrete and (discrete)
reinforcement or changing the topology of the finite element mesh after
crack formation, to model the interface.
To the present time macrolevel models fo the non-linear behaviour of
reinforced concrete have been proposed, developed and successfully applied
only for the common case of plate elements with fairly uniform distribution
of reinforcement, under two-dimensional stress conditions (plane stress),
monotonic or cyclic. Development of these models has been guided by tests
on simple reinforced concrete panels with reinforcement in one or, usually,
two (orthogonal) directions and subsjected to uniaxial or biaxial loading,
monotonic or cyclic. Biaxial loading is usually non-proportional, creating
conditions of biaxial tension and compression in directions not in general
parallel to those of the reinforcement. Phenomenological models of the
behaviour of the reinforced concrete continuum are then fitted to the results
of these tests. These models maintain the main features of the individual
material models which govern the behaviour for this particular combination
of geometry (planar) and loading conditions (biaxial tension and com-
pression). For example, the tension stiffening effect, which is due to the
interaction of steel and concrete through bond, enters as a modification to
either the uniaxial stress-strain curve of the reinforcement, or to the falling

161
RC ELEMENTS UNDER CYCLIC LOADING

branch of the uniaxial stress-strain curve of concrete in tension. As another

example, the reduction of compressive strength and stiffness of cracked
concrete due to simultaneous transverse tension is included as a
modification of the uniaxial concrete stress-strain curve in the direction
of principal compression.
Presently available macrolevels of reinforced concrete are applicable
only to walls, deep beams and other plate elements with rather uniform
distribution of reinforcement in one or two directions, provided that
cracking, inelastic action and, in general, deformations are likely to be
distributed in a fairly uniform manner, and not to exhibit strong localization,
for example, in a single crack. Connections of such plate elements to others
or to the foundation are identified as regions of probable localization of
deformations, perhaps in the form of a major crack. Therefore, different
modelling (e.g. using a special interface element, such as the one developed
by Bujadham et al.215 should be applied for these regions.
Development of three-dimensional macrolevel composite steel/concrete
models of reinforced concrete awaits accumulation of experimental data
regarding the behaviour of reinforced concrete under triaxial loading.
Extension of the presently available two-dimensional models to three
dimensions is not impossible, but without the support of such three-
dimensional experimental information such an extension would be rather
speculative. In the following, macrolevel models for reinforced concrete are
described, with emphasis on cyclic loading.

6.4.2. Smeared stresses and strains in cracked reinforced concrete

In a reinforced concrete model of the type described here, stresses and
strains are considered to be smeared, i.e. averaged, over the volume of a
finite element or over the tributary one of an integration point of an element.
Smeared stresses are total forces in a volume containing steel bars and
cracks in the concrete, divided by the corresponding geometric cross-
sectional area, and smeared strains are overall deformations of this volume
divided by the corresponding overall dimensions. So these smeared stresses
and strains cannot describe the variation of stresses and strains between
cracks, or the different deformations of steel and concrete due to bond-slip.
Increments of smeared stresses and strains referring to the global
coordinate system are related through the tangent rigidity matrix Dl of the
reinforced concrete continuum, defined in the global system. This tangent
rigidity matrix is constructed at each integration point of an element by
superposition of the tangent rigidity matrices of the individual materials, i.e.
that of concrete Dc, and those of the layers of reinforcement, DS(, in the
various directions indexed by /, after transformation to the global coordinate
system

D = T 1 D c T c + E P / T* D s/ TS(- (182)

In equation (182) T c and Ts, are the corresponding transformation matrices,

and pi the steel ratio of layer / within the tributary volume of the integration
point.
Rigidity matrix Ds, refers to a local coordinate system with the x-axis
parallel to the reinforcement of layer /, and its only non-zero element is the
first diagonal term, which is equal to the instantaneous tangent steel
modulus, E%, as determined from the cyclic uniaxial stress-strain law
adopted for the steel. This steel stress-strain model should take into account
also the effect of tension stiffening. If the entire tension stiffening is
assigned to the stress-strain curve of concrete in tension, then the ordinates
of the steel stress-strain curve have to be reduced over those of bar steel,

162
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

due to tension stiffening.205 If tension stiffening is assigned to the bars, as it

is usually done in Europe, however,190195 the ordinates of the stress-strain
curve of the reinforcement are increased. Details have been given in the
subsection on tension stiffening.
At low levels concrete is uncracked, and the construction of its tangent
rigidity matrix can be based on a model of concrete in compression defined
in any orthogonal coordinate system, but more conveniently in that of the
principal directions of concrete stresses. Typically for such stress levels a
linear, elastic, isotropic rigidity matrix can be used. For the plane stress
conditions for which the models presented herein apply, the model is
normally equipped with a biaxial failure envelope of uncracked concrete,
selected among the ones proposed in the literature specifically for biaxial
loading, or obtained from any triaxial failure surface by setting 03 = 0.
Cracking is considered to occur when the stress point reaches the failure
envelope within its concrete in tension region, i.e. in the tension-tension, or
most (or all) of the tension-compression quadrants, in the space of principal
concrete stresses (see section 6.3.2.). The combination of these principal
stresses usually reaches the model's biaxial failure envelope within the
tension-compression quadrant. A crack is then considered to be formed
normal to the direction of principal concrete tension. (If both principal
concrete stresses are tensile, the crack is considered to occur normal to the
direction of the maximum of these two stresses.) From then on, construction
of D c is based on the two-dimensional constitutive relations of cracked
reinforced concrete.
The models which have been proposed to date consider the concrete
component of cracked two-dimensional reinforced concrete as an ortho-
tropic material. Then, in reference to the axes of orthotropy, matrix Dc is, in
general, taken to be diagonal (see equation (6) of Chapter 1 considering only
axes 1 and 2 and neglecting Poisson's effect). The first two diagonal terms,
which relate each increment of smeared concrete normal stress, <rci or crC2,
to the corresponding increment of smeared normal strain, e\ or ei
respectively, are the tangent normal moduli of concrete, E{ and E2, in
directions of orthotropy 1 and 2, whereas the last diagonal term is the
effective tangent shear modulus G12 of cracked concrete (see section 6.3.4.).
The two-dimensional version of what is included in section 6.3.4. for
fixed against rotating cracks and in section 6.3.5. for tension stiffening
applies also to the orthotropic modelling of concrete in composite steel/
concrete elements, and covers completely the ac\ — e\ response and the
determination of the tangent modulus E\ in the cracked direction 1. For
direction 2, which is in compression, some additional effects are described
below. These effects are special to reinforced concrete and do not have a
counterpart in plain concrete.

6.4.3. Stress-strain and strength in compression in the presence of

transverse tension
After cracking, construction of the tangent rigidity matrix Dc of the concrete
component is based on the assumption of orthotropy, with orthotropic
direction 2 being taken parallel to the direction of the (fixed or rotating
crack). Hence, to determine the normal modulus in direction 2, the stress-
strain behaviour and the ultimate strength of concrete in the direction of
principal compression needs to be specified.
When reinforced concrete is subjected to biaxial compression-tension,
causing cracking of the concrete, the concrete strength in the direction of the
applied compression is reduced, below that in the absence of transverse
tension. At the level of local, as opposed to average, stresses, the tension

163
RC ELEMENTS UNDER CYCLIC LOADING

developed in the concrete between cracks, as a result of bond, creates a

biaxial stress condition in the compression-tension quadrant. As in this
quadrant the biaxial failure envelope of plain concrete gives a compressive
stress at failure less than the uniaxial compressive strength, the local, and
hence the average, compressive stress at ultimate concrete strength is
reduced. Besides, the concrete strips between cracks are relatively slender,
and consequently less resistant to compression. This slenderness effect is
more pronounced if small diameter bars are used, causing more closely
spaced cracks.
In recent years considerable research has been devoted to the
investigation of the strength of cracked reinforced concrete under biaxial
compression-tension. Among these investigations, the ones performed at
the University of Toronto 2 ' 2 0 5 differ significantly from the rest, which
were conducted at European institutions, mostly German. They are different
in terms of the test set-up, the specimen configuration, the range of
parameter values, and the conclusions drawn.
The European investigations have shown that for high bond bars and for
stress and strain conditions likely to occur in practice at the ultimate limit
state of strength (i.e. for values of the ratio of principal smeared tensile
strain, ex, to principal smeared compressive strain, e2, below 3), the strength
reduction due to simultaneous transverse tension does not exceed 15%. For
example, Schlaich et al.2i6 found that the reduction in compressive strength
is about 10%, if the reinforcement (stressed to about yielding) is at 90° to the
direction of applied compression, whereas when the reinforcement is in two
directions, at +45°, and —45° to the direction of the compressive stress, the
strength reduction is about 14% for reinforcement spacing equal to 50 mm,
or zero, for a spacing of 100 mm. Kolleger et al., however, observed no
significant change in compressive strength due to orthogonal tension, for
levels of transverse tensile strain from 007% to about 0-21%, i.e. up to
yielding of the reinforcement. More recently Kolleger and Mehlhorn have
concluded a series of many tests in Kassel and in Toronto, reaching the
conclusion that the reduction in compressive strength does not exceed 20%,
and that larger reductions found by Vecchio and Collins may be due to the
yielding of steel in direction 1 or to testing problems.171 On the basis of
their own results, Kolleger and Mehlhorn recommend a reduction of the
compressive strength and the corresponding strain in the direction of
o-c2, 62, according to equations (184a) and (185b) below, with (3 taken to
depend on the smeared concrete tensile stress in direction 1, aci

\-0>p= 1-1 - 0 - 4 ^ > 0-8 (183)

/ct

The University of Toronto tests were performed on panels reinforced with

small diameter smooth bars (welded wire mesh). So it has been commented
that these tests are not representative of the conditions usually found in
practice, since the lack of bond along the steel bars leads to less disturbance
of the concrete blocks between cracks, and to larger crack widths, i.e. to
greater principal tensile smeared strains, t\. Nevertheless, the results of
these tests have led to the following comprehensive proposal for monotonic
loading.202
In the presence of principal smeared strains, tensile ei > 0, and
compressive e2 < 0, the compressive strength of concrete in the direction
of e2 decreases to

f; = PA (184a>
and the corresponding strain at the peak, e*o, to

164
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

<xo =
(185a)

in which the values fc and eco > 0 are those corresponding to uniaxial
compression of concrete, and the coefficient fi equals
1
<1 (186a)
0-85-0-27-

The proposed relation between e2 < 0, and the compressive stress of

concrete, in the direction of e2 is

— —fc 2— - - (e for 0 > e2 > 4 (187)

fi \ co

= -fife fore* > e 2 > 2 e ! (188)

for 2eco > e2 (189)

in which coefficient fi is given by equation (186a). Equations (187) and

(188) give the ascending and descending branch, respectively, of the
monotonic relation between aC2 and e2, in the form of two parabolas, with
horizontal tangent at (f*, e*o). It is noteworthy that the initial tangent
modulus and the secant modulus at the top of the curve are constant,
independent of the value of e\.
More recently Vecchio and Collins202 simplified their original proposal
as follows
(184b)
(185b)
1
<1 (186b)
0-8-0-34-

and

for 0 > e2 > (190)

ac2 = for 2eco > e2 (189)

In other words, in the new proposal, the strain at ultimate strength f* is not
affected by the value of e\, the 0& — e2 relation is described by a single
parabola for eC2 between 0 and 2eco, and the value of t\ affects only the
strength value, f* and through it (in a multiplicative fashion), the entire
0c2 — ^2 relation, including the initial modulus and the secant at ultimate
strength. Barzegar and Ramaswamy have also adopted this more recent
Vecchio and Collins proposal, in the smeared cracking model of their
concrete elements.193
It is worth mentioning that the dependence of the relation between aC2
and e2 on t\, i.e. on the value of the smeared strain in the transverse
direction, results in a tangent rigidity matrix D c which, in general, has not
only diagonal terms, but also non-symmetric off-diagonal ones, which
introduce coupling between directions 1 and 2. Omission, for convenience,
of these non-symmetric off-diagonal terms is considered to affect only the
rate of convergence, but not the final result.191'205

165
RC ELEMENTS UNDER CYCLIC LOADING

The main feature of Vecchio and Collins proposal, which is the reduction
of compressive strength in the direction of ei with increasing value of the
transverse tensile strain e\, according to equations (184a) and (186a) or
(184b) and (186b), is not in major disagreement with the European
proposals for the reduction in the value of concrete compressive strength, in
the usual range of ei/e 2 values found in practice. For values of this latter
ratio below - 4 0 and down to - 2 0 , Vecchio and Collins' proposal gives
very large reductions of compressive strength, of the order of 80%, which
may have to be viewed with some caution. Nevertheless, such low values of
e\/e2 are very rarely associated in practice with ultimate strength, especially
when high bond bars are used.
Cervenka has also proposed equations (184a) (185a) and (187)—(189) but
with the following coefficient (3, for the reduction of strength and strain at
the peak

The optimal value of the parameter kx to fit the experimental results of

Vecchio and Collins was found to be k\ = 0-45. For the type of steel and
specimens used by Schlaich et al.,216 Cervenka suggested the value kx =
0-20, which gives less reduction of strength with increasing e\. Beyond
ultimate strength Cervenka adopted a linear falling branch for the aC2 — €2
relation. Unloading and reloading is linear, with the initial tangent modulus
of concrete, Ec, up to the e-axis, and then along this latter.195
It is worth mentioning that Crisfield and Wills191 had to use, along with
their rotating crack model, the early Vecchio and Collins strength
degradation procedure expressed by equations (184a)-(186a), in order to
achieve good agreement with the panel test results,202 as far as the failure
mode and load, and the load-deformation response is concerned. Adopting a
less drastic strength degradation, more in line with the European proposals,
would have led to significant overestimation of the failure load, and to
prediction of a different failure mode than observed in the tests.
The models with emphasis on cyclic loading, mentioned in the previous
sections on tension stiffening and fixed against rotating cracks, do not differ
much from the ones above, as far as the treatment of the effect of ej on the
behaviour in direction 2. The monotonic && — €2 relation (Stevens et al.205)
which serves as an envelope for cyclic loading, consists of the parabola of
equation (190), applied only up to the point of ultimate strength, given by
equations (184b) and (185b). For coefficient /?, equation (186b) is applied.
Beyond ultimate strength the monotonic <rc2 — 62 relation consists of an
asymptotic residual strength equal to 30% of ultimate strength, and a cubic
parabola as the falling branch between ultimate and residual strength. The
effect of confinement is reflected by applying a multiplicative factor on the
stress and strain at ultimate strength. Finally, the model is equipped with
complex unloading-reloading rules from compression towards tension, and
vice versa (see Fig. 144).
The model by Okamura et a/.186'203 has an envelope curve, applicable
also for virgin or monotonic loading, of the following form
<TC2 = -/?£cexp{0-73e2[l - exp(l-25e2)]} (191)
The reduction factor f3 assumes the following values
0=1 for 00012 > e, > 0 (192a)
0 = 1-15- 125e, for 0-0044 > e{ > 0-0012 (192b)
0 = 0-6 for e, > 0-0044 (192c)

166
 FINITE ELEMENT MODELLING OF REINFORCED CONCRETE

tne
Unloading from a strain €2,e> o° envelope is parabolic, towards a point
on the horizontal axis at a strain
€pi = -2-85[1 - exp(0-35e2,e)] (193)
and reloading is linear, towards the previous envelope point.

167
7. Solution strategies for non-linear structural
equations

7.1. Non-linear The discrete non-linear equations which express, at a given time t,
equilibrium equations equilibrium of a structure, modelled through finite element space discret-
ization, may be written as
= 0 (194)
F e x t contains the externally applied nodal forces, (equivalent to body,
surface and concentrated forces), and F i n t the internal nodal forces, that is
the structural reactions associated with the present configuration; more
explicitly F i n t = S[(u(t)] is a function of the evolution of the unknown nodal
displacements u. Time t has the meaning of physical time, in dynamic
problems, or of a parameter associated to load levels, in static problems. In
dynamic analysis the vector F e x t , besides given external loads P, would also
contain inertia and damping forces, that is

= P(t)-Mu(t)-Cu(t) (195)

7.1.1. Incremental formulation

In both static and dynamic problems, non-linear structural analysis is
usually conducted using an incremental approach, where the load
parameter, or physical time, t, is subdivided in steps and solution progresses
from step to step. If solution at step n, corresponding to time t, has been
achieved, that is un and all the other relevant quantities are known, solution
at subsequent step, n+l, corresponding to time t + At, is sought imposing a
new equilibrium state

In static problems, where F e x t = P, the unknown vector un+{ appears only in

the expression of the structural reaction vector F i n t = 5, so that linearization
of the latter about time t leads to

At)-u(t)]=Sn

and, by substitution, equation (196) becomes

K Au=R (197a)
where the residue vector, R, is defined by R = Pn+i — Sn, K = ^ is the
tangent stiffness matrix, and AM a vector of displacement increments.
Equation (197), the incremental approximation of equation (3), is a linear
equation in the unknown AM, which can be solved to produce an
approximate solution, at step n+l

= «n (198)

In dynamic analysis, provided that appropriate difference formulas are

used as an approximation for velocities and accelerations in the expressions
of inertia and damping force terms, linearization of equation (196) may lead
to equations formally similar to equation (197). In particular we may
anticipate that this happens if the difference formulas, used for expressing

168
 SOLUTION STRATEGIES FOR NON-LINEAR STRUCTURAL EQUATIONS

time derivatives in equation (195), define a so-called 'implicit integration

method'.
Although we are going to introduce integration methods for dynamic
analysis at a later stage, it is instructive to fix ideas from now through a
simple example. In this example, neglecting, for simplicity, damping forces,
we consider only inertia forces, and use one of the simplest implicit
formulations, the so-called trapezoidal rule. This consists of the expressions
At,. . .
«n+l = «n + - y («n + Kn+l)
. At,.. .. .
«n+l = «n + -r- ("n + )
from which it also follows

Substituting in equation (196) we get first

( ^2
4

(«n+l -Un)-—Un-
4

«
and subsequently, using the already given linearization of S[u(t + At)],
K*Au=R* (197b)
where K* and R* are, in this case, respectively defined by

Equations above show the stated similarities between the numerical

incremental formulation of static and dynamic problems.
Equation (197a) or (197b) can be used directly to advance solution from
one step to the next, but this solution, Mn+1 = un+Au, is in general affected
by an error due to the linearization. Because of this error equation (196) is in
general not satisfied exactly. If the step or load increments are very small
the error may be small and the solution acceptable. This solution strategy is
sometimes used and is referred to as 'incremental solution without iteration
corrections'.
In general, because of cost considerations, larger load increments or time
steps are preferred; in this case the errors produced by linearization may not
be negligible and their accumulations may produce gross overall errors.
Some procedures which include equilibrium iterations should be part of the
strategy, so that equation (196) is also satisfied to a given tolerance. These
procedures are the subject of the next paragraphs.

7.1.2. Newton-Raphson iteration methods

The most used iteration schemes for the solution of non-linear finite element
equation (196) are some form of Newton-Raphson iterations. If, within the
step n+\, M(I), represents the approximate solution of equation (196) at
iteration i, the residue vector of the unbalance of the solution, / ? w =
^ext(«(<)) — ^int(«(())» is not, in general, equal to zero. We may improve the
approximation and seek for a better solution, « ( ' + l ) = u(l)+Au^'\ by
imposing that the linearized part of R(l+l) = R(u(l+l)) be equal to zero.
Linearization about previous approximation, « (l) , leads to equations similar,
in form, to equation (197a) and (197b)

169
RC ELEMENTS UNDER CYCLIC LOADING

For static problems

K « A « w = JfW (199)

where K(l) is the tangent stiffness matrix correspondent to displacements

u('\ and R(() the vector of unbalanced loads, more explicitly defined as R(l) =
Pn+l - S(«(/)).
For dynamic problems

K*w A « w = R*w (199a)

where an effective stiffness matrix, K*(<), and a corresponding residue
vector, /?* w , are involved. With reference to the previously considered
example they become

V At2
and

- M [-ij (M(0 _ Mn) ^ i n _fin)_

Starting from the solution of equations (197) or (197a), iterations
continue, through repeated solutions of equations (199) or (199a) and
consequent updating of the solution vector.
«('+') = «(') + Au« (200)

up to the point where the residue vector is made as small as desired. The
described methodology implies that at every iteration, within a step, the
tangent matrix K(l) and the residue vector /? (l) be evaluated. Such a strategy
is referred to as 'full Newton-Raphson method'.
Since the updating and factorizing, for solution, of the tangent stiffness at
every iteration is a computationally expensive operation, different alter-
natives have been proposed; these may be considered as modifications of
the original Newton-Raphson method.
One such alternative, called 'initial stiffness method', is to use the initial
stiffness matrix KQ throughout the steps and the iterations, by operating
always with the equations
K o Au (/) = R{i)
In this case the stiffness matrix is formed and factorized only once during the
analysis; the method corresponds to a linearization of the response about the
initial configuration. For problems with significant non-linearities this can
lead to very slow convergence or even divergence, if the system stiffens.
In order to accelerate convergence or avoid divergence it may be
effective to use the so-called 'modified Newton-Raphson strategy', where a
new tangent stiffness matrix is calculated, from time to time, in order to
follow the evolution of non-linearities; within a step such matrix is always
kept constant during iterations. Without any a priori knowledge of non-
linearities it may be effective to reformulate stiffness at the beginning of
every step.
Figures 145-147 give a graphical representation, in the case of a one
degrees of freedom system, of the different alternative schemes considered.

7.1.3. Displacement updates and line search

In case of slow convergence of modified Newton-Raphson or initial

170
 SOLUTION STRATEGIES FOR NON-LINEAR STRUCTURAL EQUATIONS

Pn + 1 xT-^

Pn+

F/g. 145 (top left). Full Newton-Raphson method

Fig. 146 (top right). Initial stiffness method
Fig. 147 (bottom left). Modified Newton-Raphson
strategy

stiffness iterations, it may be effective to complement the strategy with

some sort of acceleration schemes; these act by modifying the displacement
updating equation (200).
If a line search strategy is used, the correction vector AM, computed
during iterations, is considered only as a direction vector; this is
subsequently scaled, through a scale factor /?, in order to produce the
updated solution.
u(i+\) = „(/) + p(0 A.M(0 (200a)

/?W is chosen, on the basis of a line sesarch algorithm, so that the out-of-
balance loads, R{l+l) = /?(M (0 + /?W A«W) have a nearly zero component in
the direction AM. This condition is satisfied to a given tolerance.
Formulas similar to equation (200a) are used, for updating displacements,
in other acceleration schemes, for example the so-called Aitken acceleration
scheme,217"219 with the only difference that (3 is no longer a scalar, but a
diagonal matrix; this is continuously updated on the basis of information
acquired during iterations.

7.1.4. Quasi-Newton methods

Another class of iteration methods, known as matrix update methods or
quasi-Newton methods, has emerged recently for the effective solution of
non-linear equations of the form of equation (196). These methods are based
on the updating of the stiffness matrix (or rather its inverse), in equation
(199), to provide a secant approximation to the matrix from iteration /—I to
iteration i, and avoiding the expense for a tangent matrix and its inverse.
Assume that we have already computed u ('-') H„('), R»('-')', R ; then the
(l (l>
(l)
updated matrix K ; to be used in equation (199), should satisfy the quasi-
Newton equation

171
RC ELEMENTS UNDER CYCLIC LOADING

where An*1"1) = «W - i/'" 1 ) Art*1"1) = R® - R^1^

For one degree of freedom systems Fig. 147 gives a geometrical
representation of K. In multidimensional cases the above equation does not
uniquely define K, but only places a restriction on its definition.
Of the many available procedures of this type the most widely used in
finite element analysis is the Broyden-Fletcher-Goldfarb-Shanno method
(BFGS) (inverse) update, based on the formula

(201)
where K is the last formed and factored stiffness matrix. Quasi-Newton
trix. Quasi-N
updates are often used with line searches, so that from

the quasi-Newton equation becomes

V (202)
The BFGS vectors v and w are defined
v = Att('- 1 7(A«('- 1 ) T A*'1'"1)) w = -A*''"1) + aR«-l\
where a = [(-0 AR^T Au^~• >)/(*('-'> T Au('-'>)] l/2
Recalling that K /? ( / - 1 ) = A« ( '~'), it is not difficult to show that K, as
implicitly defined by equation (201), satisfies the quasi-Newton equation
(202). Proper recursive application of equations (199) and (201) enables the
search direction to be computed at different iterations using only vector
operations, that is, back substitutions and forward reductions, without
explicit computation of updated matrices or costly factorizations; the
factored K is never altered during such calculations (see Matthies and
Strang220).

7.2. Static analysis So far external loading has been assumed to be a given function of time.
(special problems) This means that, in static analysis, the analyst prescribes the load levels for
which the equilibrium configurations are to be calculated; moreover, during
equilibrium iterations, such load levels have been considered to remain
constant. If load-displacement processes have to be traced in which there is
no a priori knowledge of the load-time function, special path-following
methods are needed. This is important, for example, when the structural
response about and after limit (collapse) points, (Fig. 149), is of interest.
Among these methods a short account is given here of the displacement
control methods and the arc-length methods. Both methods provide for
automatic step selection and non-constant load iterations.

7.2.1. Displacement control and arc-length methods

Let us assume that the structure is subjected to a proportionally varying load
XP, where A is a load factor which describes the intensity of the reference
load vector P; A is considered now as an additional unknown. At the same
time an extra constraint equation is introduced allowing determination of
the new load level A. The linearized equation to be solved at iteration i+\
becomes

AA W )P - S(« w )

172
 SOLUTION STRATEGIES FOR NON-LINEAR STRUCTURAL EQUATIONS

Limit point
Pn + 1

Fig. 148. Quasi Newton methods; updated K Fig. 149. Analysis beyond limit (peak) point

which can also be written, using the residue vector computed at previous
iteration, JfW = ® W
K(0 (203)

The solution of equation (203) may be split in two parts, corresponding to

the following two linear systems, having the same coefficient matrix

K w AMJ° = P

and expressed by

A H W = AAW A H I 0 + . .(')
,221
In the displacement control method suggested by Batoz and Dhatt "" a
single displacement component is selected as a controlling parameter during
a step. At the beginning of the step that component is given the prescribed
value AMJ0^ = 6 and this constraint determines the first value of the load
increment A\(°\ In subsequent iterations the same displacement component
is always given zero increments, A«| () = 0, and this determines subsequent
value of the load increment corrections AA^. Iteration is continued, at
varying load levels, until a new equilibrium position is found (R = 0).
In the arc length method similar conceptual lines are followed, with the
main difference being that the constraint which determines AA^ during
iterations is no longer a single displacement parameter, but a more complex
function of an arc length of the solution curve in force displacement space.
Both methods allow the use of different strategies for the stiffness
reformulation and solution of equations (equation (203)), (full Newton-
Raphson, modified Newton-Raphson, BFGS updates), and may include line
search or other acceleration schemes in the unknowns updating
stage.219'222'223

7.3. Dynamic In non-linear dynamic analysis of structures equilibrium equation (194) may
analysis be more explicitly rewritten, using equation (195) as
M u(t) + C ii(t) + S(u(t)) = P(t) (204)
where S(u(t)), the vector of structural reactions, contains the non-linear

173
RC ELEMENTS UNDER CYCLIC LOADING

behaviour. As a particular case, for a linear elastic structure, S(u(t)) is

expressed as Ku(t) and the system becomes
M u{t) + C u(t) + K u(t) = F(/) (205)
This is a system of linear differential equations of second order with
constant coefficients and, although we are mainly interested in the more
general non-linear system of equation (204), we shall give special
consideration to equation (205) in order to ease the introduction of all the
important concepts which need to be considered for the solution of equation
(204). Two methods of solution are possible for equation (205), namely
direct integration and mode superposition. In particular we shall concentrate
on direct integration, which can be very easily extended to non-linear
systems. Reference to mode superposition will be also made for clarifying
some problems which arise in direct integration.

7.3.1. Direct integration methods

In direct integration equation (205) is numerically integrated, using a step-
by-step procedure, without any prior transformation. Direct integration
seeks satisfaction of equilibrium equation (205) at discrete time points, ...,
n, n+\, ..., within the interval of solution, starting from time 0 up to a final
time tf, this implies a subdivision of the total time span in intervals At
which we shall consider, for simplicity, equal. Different numerical
integration schemes correspond to different finite difference expressions
for velocities and accelerations to be used in equation (205), or, also, to
different assumptions for the variation of displacements, velocities and
accelerations within each time step. These differences influence accuracy,
stability and cost of the solution procedure.
There is a large body of literature concerning numerical integration
schemes (see for example work by Hughes and Belytschko224) and this is
not the place to review them. Moreover only very few schemes have really
emerged and are in common use. So considerations shall be restricted to two
schemes only, which are representative of the two important categories of
the explicit and implicit methods. These are the central difference rule
(explicit) and the trapezoidal rule, or constant average acceleration rule
(implicit). The latter may also be considered as a member of a more general
parametric family of integration schemes, which is called the Newmark
method; this is very popular and used in many computer codes for dynamic
analysis of structures.
7.3.1.1. Explicit schemes. Explicit schemes can be introduced with
reference to the most popular of them, the central difference scheme, in
which it is assumed that

2 +
«n = ( )

and

After substituting the two expressions above in equation (205), written at

time n, the following equations are obtained

( 2 o 6 )

174
 SOLUTION STRATEGIES FOR NON-LINEAR STRUCTURAL EQUATIONS

from which it is possible to solve for the unknowns « n+1 , assuming that
displacements, velocities and accelerations at previous time steps are
available.
It is a common characteristic of explicit methods that they contain in the
expression of velocities and accelerations, at time n, the values of
displacements at time n+l. This permits the writing of equilibrium
equations at time n, in which the unknowns un+i only appear in the
expressions of inertia and damping forces and not in the terms which refer
to the structural reaction vector. Since the mass matrix is often diagonal and
damping forces are either completely neglected or simplified with the
assumption of a diagonal damping matrix, the solution may be advanced
without solving any system of equations. If this is the case, since no
procedure of triangularization of a matrix is required, there is no need to
assemble a matrix like K, but only force vectors representing structural
reactions, and the solution can be essentially carried out at the element
level, with important savings of high-speed storage.
A disadvantage of explicit methods is, on the other hand, that they require
the time step At to be smaller than a critical value which can be calculated
from the mass and stiffness properties of the complete structure. More
specifically this critical value is a fraction of the smallest period of the finite
element assemblage with m degrees of freedom. Integration schemes which
require the use of a time step smaller than a critical value are said to be
conditionally stable. Since the effective use of conditionally stable methods
is limited to certain problems, it is important to consider schemes which are
unconditionally stable. These can be found only among the implicit schemes
which belong to the category illustrated next.
7.3.1.2. Implicit methods. An important and simple example of this
category, already used as an example in 7.1.1., is the so called trapezoidal
rule or constant average acceleration rule. All the implicit methods are such
that displacements at time n+l depend not only on displacements, velocities
and accelerations at previous steps, but also on the quantities at step n+l.
In order to advance solution, equilibrium equations need to be written at
time n+l, where all the terms (inertia, damping and structural reaction
forces) depend on the unknowns un+i. This implies that at each step a
stiffness matrix has to be assembled, and a system of equations has to be
solved. The trapezoidal rule is defined by the following formulas

Un+l =U,,+ — (Un + Un+i)

At ..
M«+l — un + — \Un + nn+\)

from which, solving with respect to accelerations and velocities

_ 4 4 4 . . .
"n+l ~ ^ 2 "n+l "" ~A~5U" ~~U» — "n

Un+[ = —Uu+\ — -T:«fl — "n

and substituting in equation (205), at time n+l

M+ c
A7 ) " " + M ""
Numerical integration can now be advanced solving the above linear

175
RC ELEMENTS UNDER CYCLIC LOADING

system of equations in the unknowns « n+1 ; the coefficient matrix, given by

the combination of mass, damping and stiffness matrices in the first
parenthesis, is called 'effective stiffness' for the particular integration
scheme.
7.3.1.3. Stability and accuracy analysis. The two fundamental
concepts of stability and accuracy of integration schemes need to be
introduced. For the linear system of equation (205) we might use mode
decomposition prior to time integration, by changing basis from the finite
element coordinate basis to the basis of eigenvectors of the generalized
eigenproblem. There would follow a system of m uncoupled equations
where the /th row,

Xi + 2^i (Ji Xi + LJ^ xi = ru i=\,...m (208)

represents the equilibrium equation of motion of a single degree of freedom
system and may be solved using the same numerical integration schemes we
have considered for direct integration. If all m equations are integrated using
the same time step then the mode superposition is totally equivalent to a
direct integration with the same integration scheme and the same time step.
Therefore, to study the accuracy of direct integration, we may focus the
attention on the decoupled equations instead of considering the original set
of equation (205). In this way the variables to be considered in the stability
and accuracy analysis are only At, u>j, £, and not all elements of the
stiffness, mass and damping matrices.
In order to predict the dynamic response of the structure accurately it
would seem that all m decoupled equations (208) need to be integrated
accurately. Since in direct integration the same time step is used for each
equation, At would have to be selected corresponding to the smallest period
in the system. However, in many cases the predominant contribution to the
response is given by a few modes of vibration only, with negligible
contribution of the higher ones. So, if we are interested only in integrating
the first p equation of the m of equation (205), we might choose At only in
relation to the period Tp, which may be very much larger indeed than the
smallest period of the system Tm. As a consequence the important question
arises of what response is predicted in the numerical integration of one of
equation (208) if At is much larger than 7}.
Stability of an integration method has to do exactly with this problem. It
means that initial conditions, due for example to numerical errors in the
solution, must not be amplified artificially, specially for the equations with
large At/T values, thus causing the global response to blow up. Stability is
determined by examining the behaviour of the numerical solution of
equation (208), under zero modal force r, for arbitrary initial conditions. As
a consequence it depends only on the time ratio At/T, the damping ratio f
and possibly on adjustable parameters, if the method includes any. An
integration method is unconditionally stable if the solution for any initial
condition does not grow without bound for any time step At, in particular
when At/T is large. The method is conditionally stable only if the above
only holds provided that At/T is smaller than a certain value, usually called
the stability limit. An important result of the analysis of stability is that all
explicit methods are only conditionally stable and that unconditional
stability may only be achieved using certain implicit methods.
Accuracy can also be investigated with reference to the one degree of
freedom equation (208), in the particular case of zero force and zero
damping, for a given initial velocity.
The numerical solution of such a 'test problem' can then be compared
with the known exact one. For different integration schemes the errors can

176
 SOLUTION STRATEGIES FOR NON-LINEAR STRUCTURAL EQUATIONS

be measured in terms of period elongation and amplitude decay (in the exact
solution there is no decay, since there is no physical damping), as a function
of the time ratio At/T.
The analysis for various integration schemes shows that all methods are
accurate, that is, they show negligible period elongation and amplitude
decay, for At/T smaller than about 0 0 1 . The differences come out for
larger At/T.
The characteristics of the integration errors exhibited by the test problem
for different At/T may be used in the discussion of the simultaneous
integration of m equation (208) which is equivalent to the integration of
equation (205). We observe that the equations for which the time step-to-
period ratio is small are integrated accurately, but that the response in the
equations for which At/T is large is necessarily totally inaccurate. Using
explicit conditionally stable methods we are forced to use At smaller than a
critical value which is small enough to obtain accuracy in the integration of
practically all m equations. Using unconditionally stable schemes the time
step can be much larger and should only be small enough that the response
in all modes which contribute significantly to the total structural response is
calculated accurately. The other model components can not be accurate but
the errors are unimportant because the response provided by those
components is negligible. However, it is important to stress again that this
applies only if the integration scheme is unconditionally stable, i.e. if it may
be assumed that the amplitudes in the component which are negligible do
not spuriously grow.
Note that numerical damping can be considered a good feature of implicit
unconditionally stable integration schemes because it may be used to damp
out and practically suppress the response of those modes for which, since
the time step-to-period ratio is large, the response cannot be calculated
accurately anyway. A method with good damping characteristics is, for
example, presented by Hilber et al.

7.3.2. Solution strategies in non-linear dynamic analysis

Solution strategies in non-linear dynamic analysis differ according to the
use of explicit or implicit integration schemes. In particular, the use of an
explicit integration scheme avoids the solution of a system of non-linear
equations, while the use of an implicit scheme makes static and dynamic
analysis very similar, as already discussed.
7.3.2.1. Explicit integration. As in linear analysis the equilibrium of
the finite element system is considered at time t in order to calculate the
displacement at time t + At. By substituting in the equilibrium equations
the expressions of the first and second time derivative, the unknowns un+i
appear only in the inertia and damping terms. So, if these matrices are
constant and diagonal, there is no need to factorize a matrix and the solution
is a very direct forward-marching scheme. The shortcoming is in the severe
time step restriction. For example, using the most common explicit scheme,
the central difference, the time step size At must be smaller than Tm/7r
where Tm is the smallest period in the finite element system. This time
restriction, derived using a linear system, is also applicable to non-linear
analysis provided that the change in stiffness, and as a consequence the
change in Tm, is taken into account. In fact, since the value of Tm is not
constant during the response calculation, the time step At needs to be
decreased if the system stiffens; this time step adjustment must be
performed in a conservative way so that the condition At < Tm/ir is
certainly satisfied at all times.
7.3.2.2. Implicit integration. As in linear analysis, we have to con-
sider the equilibrium equations at time t + At. It has already been shown, in

177
RC ELEMENTS UNDER CYCLIC LOADING

section 7.1., that the problem becomes very similar to the corresponding
static problem (compare again equations (197), (197a), (199) and (199a)).
Instead of a tangent stiffness matrix, an 'effective tangent stiffness matrix',
K*, has to be considered, as, for example, in equation (207) or (197a), with
contribution also from the inertia and damping matrices; additional
contribution from these matrices appears also in the expression of the
residue vector R*. Expressions of both K* and R* depend on the particular
implicit method which is used; but it is important to notice that the
contribution of the inertia term in K* is always in the form (c/A/ 2 )M.
When using the same solution strategies, the inertia of the system renders
its dynamic response, in general, smoother than its static counterpart, and
convergence of the iterations can be expected to be quicker. This improved
convergence characteristic in dynamic analysis depends on the contribution
of the mass matrix to the effective tangent stiffness. Indeed this
contribution, being inversely proportional to At2, becomes dominant when
the time step is small.
The first solutions of non-linear dynamic analysis using implicit
integration schemes were performed without equilibrium iterations. It has
been subsequently recognized that the iterations are very important, since
any error allowed in the incremental solution at a particular time directly
affects, in a path-dependent manner, the solution at any subsequent time.
Indeed, because of the high path dependency of any dynamic response, the
analysis of a non-linear dynamic problem requires iterations, at each time
step, more stringently than static analysis does.

7.3.3. Fields of application for explicit and implicit methods

Equations describing the dynamic behaviour of structural systems are, in
general, characterized by the fact that their frequency spectrum is spread out
over a relatively large domain of the positive real axis, that is T\ is very
much larger than Tm, where Tm and T\ are respectively the smallest and the
highest preiods of the free undamped system. If the system were linear the
response could be considered as the superposition of all individual mode
response histories. However, in many cases, only low mode response is of
interest. Moreover discretization of spatial domain often introduces high
frequency modes which are physically meaningless and hence should not be
allowed to participate in the solution. On the other hand, in wave propaga-
tion type problems it may be very inaccurate to consider only the first few
modes in computing the response history.
When only low-frequency mode response is of interest, the use of implicit
unconditionally stable algorithms is generally to be preferred over explicit
conditionally stable algorithms. Explicit conditionally stable algorithms
require that the size of the time step employed be directly proportional to
the smallest period of the discrete system. In practice this is a severe
limitation since accuracy in the lower modes can be attained with time steps
which are very large compared with the period of the highest mode. For
unconditionally stable algorithms a step size may be selected independently
of stability considerations and thus can result in a substantial saving in
computational effort. In addition to being unconditionally stable, when only
low mode response is of interest, it is often advantageous for an algorithm to
possess some form of numerical dissipation to damp out any spurious
effects due to high frequency modes. When, as in wave propagation type
problems, a large number of frequencies of the system are excited, it may be
convenient to use an explicit method, since the time step has to be small
anyway and advantage can be taken of the greater simplicity of computation
associated to an explicit scheme.

178
References
1. Pramono E. and Willam K. Fracture energy-based plasticity formulation of
plain concrete J. EM, Am. Soc. Civ. Engrs, 115, June 1989.
2. Lubliner J. et al. A plastic-damage model for concrete. Int. J. Solids Struct.,
25, No. 3, 1989.
3 Ba2ant Z. P. and Oh B. H. Microplane model for progressive fracture of
concrete and rock. /. EM, Am. Soc. Civ. Engrs, 111, 1985.
4. Bazant Z. P. and Prat P. C. Microplane model for brittle-plastic material: I
theory, II verification. / EM, Am. Soc. Civ. Engrs, 114, Oct. 1988.
5 Chen W. F. Plasticity in reinforced concrete. McGraw-Hill, New York,
1982.
6. Finite element analysis of reinforced concrete. State of the art report.
American Society of Civil Engineers, New York, 1982.
7. Bazant Z. P. and Bhat P. Endochronic theory of inelasticity and failure of
concrete. J. EM, Am. Soc. Civ. Engrs, 102, Aug. 1976.
8. Elwi A. A. and Murray D. W. A 3D hypoelastic concrete constitutive
relationship. J. EM, Am. Soc. Civ. Engrs, 105, Aug. 1979.
9. Buyukozturk O. and Shareef S. S. Constitutive modeling of concrete in
finite element analysis. Computers and Structures, 21, No. 3, 1985.
10. Stankowski T. and Gerstle K. H. Simple formulation of concrete behavior
under multiaxial load histories. ACI J., 82, No. 2, Mar.-Apr. 1985.
11. Shafer G. S. and Ottosen N. S. An invariant-based constitutive model.
Structural Research Series 8506, Dept of Civil Environmental and
Architectural Engineering, University of Colorado, Boulder, 1985.
12. Vermeer P. A. and de Borst R. Non-associated plasticity for soils, concrete
and rocks. Heron Journal, 29, 1984.
13. Han D. J. and Chen W. F. A nonuniform hardening plasticity model for
concrete materials. J. Mech. Mat, 4, 1985.
14. Fardis M. N. and Chen E. S. A cyclic multiaxial model for concrete.
Computational Mechanics, 1, 1986, 301-315.
15. Chen E. S. and Buyukozturk O. Constitutive model for concrete in cyclic
compression. J. EM, Am. Soc. Civ. Engrs, 111, 1985.
16. Yang B. et al. A bounding surface plasticity model for concrete. J. EM, Am.
Soc. Civ. Engrs, 111, Mar. 1985.
17. Comite Euro-International du Beton. Concrete under multiaxial states of
stress. Constitutive equations for practical design. CEB, Lausanne, 1983,
Bulletin d'Information 156.
18. Krajcinovic D. Continuous damage mechanics. Appl. Mech. Rev., 37, 1984.
19. Lemaitre J. How to use damage mechanics. J. Nucl. Eng. & Design, 80,
1984.
20. Resende L. and Martin J. B. A progressive damage continuum model for
granular materials. Comp. Meth. Appl. Mech. Engng, 42, 1984.
21. Krajcinovic D. and Fonseka G. U. The continuous damage theory of brittle
materials. J. Appl. Mech., Am. Soc. Mech. Engrs, 48, 1981.
22. Mazars J. Description of the multiaxial behavior of concrete with an elastic
damaging model. Proc. RILEM Symp. on concrete under multiaxial
conditions. INSA-OPS Toulouse, May 1984.
23. Dougill J. W. On stable progressively fracturing solids. Z. Angew. Math.
Phys., 27, 1976.
24. Gerstle K. H. et al. Behavior of concrete under multiaxial stress states. J.
EM, Am. Soc. Civ. Engrs, 106, Dec. 1980.
25. Scavuzzo R. et al. Stress-strain curves under multiaxial load histories.

179
RC ELEMENTS UNDER CYCLIC LOADING

Report, Dept of Civil Environmental and Architectural Engineering,

University of Colorado, Boulder, 1983.
26. Kotsovos M. D. and Newman J. B. A mathematical description of the
deformational behavior of concrete under complex loading. Mag. Cone.
Res., 31, No. 107, June 1979.
27. Darwin D. and Pecknold D. A. Nonlinear biaxial stress strain law for
concrete. J. EM, Am. Soc. Civ. Engrs, 103, Apr. 1977.
28. Schickert G. and Winkler H. Results of test concerning strength and strain
of concrete subjected to multiaxial compressive stresses. Deutscher
Ausschuss fur Stahlbeton, 277, Berlin, 1977.
29. Traina L. A. Experimental stress-strain behavior of a low strength concrete
under multiaxial states of stress. AFWL-TR-82-92, Air Force Weapons
Laboratory, Kirtland Air Force Base, New Mexico, 1982.
30. Saenz L. Equation for the stress-strain curve of concrete. ACI J., 61, Sept.,
1964.
31. Kupfer H. et al. Behavior of concrete under biaxial stresses. J. EM, Am. Soc.
Civ. Engrs, 99, Aug., 1973.
32. Willam K. J. and Warnke E. P. Constitutive model for the triaxial behavior
of concrete. IABSE Proc, 19, 1975.
33. Eberhardsteiner J. et al. Triaxales Konstitutive Modellieren von Beton.
Report, Institute for Strength of Materials, Technical University of Vienna,
1987.
34. Sargin M. Stress-strain relationship for concrete and the analysis of
structural concrete sections. No. 4, Solid Mechanics Division, University of
Waterloo, 1971.
35. Van Mier J. G. M. Strain-softening of concrete under multiaxial loading
conditions. Dissertation, Eindhoven University, 1984.
36. Dafalias Y. F. and Popov E. P. Plastic internal variable formalism of cyclic
plasticity. Appl. Meek, Am. Soc. Civ. Engrs, 43, 1976.
37. Dafalias Y. F. and Herrmann L. R. Bounding surface formulation of soil
plasticity. Soil Mechanics — Transient and Cyclic Loads, Pande and
Zienkiewicz (eds). Wiley, New York, 1982.
38. Fardis N. M. et al. Monotonic and cyclic constitutive law for concrete. J.
EM, Am. Soc. Civ. Engrs, 109, 1983.
39. Spooner D. C. and Dougill J. W. A quantitative assessment of damage
sustained in concrete during compressive loading. Mag. Cone. Res., 27,
1975.
40. Resende L. A damage mechanics constitutive theory for the inelastic
behavior of concrete. Comp. Meth. Appl. Mech. Engng, 60, 1987.
41. Krajcinovic D. Constitutive equations for damaging materials. J. Appl.
Mech., Am. Soc. Mech. Engrs, 50, 1983.
42. Bazant Z. P. and Kim S. S. Plastic-fracturing theory for concrete. J. EM,
Am. Soc. Civ. Engrs, 105, June 1979.
43. Han D. J. and Chen W. F. Strain-space plasticity formulation for hardening
softening materials with elastoplastic coupling. Int. J. Solids Structures, 22
No. 8, 1986.
44. Lemaitre J. Coupled elastoplasticity and damage constitutive equations.
Comp. Meth. Appl. Mech. Engng, 51, 1985.
45. Ba2ant Z. P. and Prat P. C. Microplane model for brittle-plastic material —
parts I and II. J. EM, Am. Soc. Civ. Engrs, 114, No. 10, 1988, 1672-1702.
46. Ba2ant Z. P. and O2bolt J. Nonlocal microplane model for fracture, damage
and size effect in structures. J. EM, Am. Soc. Civ. Engrs, 116, No. 11, 1990,
2485-2505.
47. Bazant Z. P. and Oh B. H. Efficient numerical integration on the surface of

180
 REFERENCES

a sphere. Zeitschrift fur angewandte Mathematik und Mechanik ZAMM, 66,

No. 1, 1986, 37^9.
48. Stroud A. H. Approximate calculation of multiple integrals. Prentice Hall,
Englewood Cliffs, New Jersey, 1971.
49. Bazant Z. P. Parallel viscous element and damage element coupling as a
model for rate effect in concrete fracture. Note privately communicated to
M. Jirasek, S. Beissel and J. OZbolt. June, 1991.
50. de Borst R. Continuum models for discontinuous media. Proc. of Int. Conf.
on Fracture Processes in Brittle Disordered Materials. Noordwijk, The
Netherlands, 1991, 18.
51. Bazant Z. P. (T. A. Jaeger (ed).). Numerically stable algorithm with
increasing time steps for integral-type aging creep. First Int. Conf. on
Struct. Meek in Reactor Technology, West Berlin, 4, Part H. 1971, 17.
52. Baiant Z. P. and Chern J. C. Strain softening with creep and exponential
algorithm. J. EM, Am. Soc. Civ. Engrs, 113, No. 3, 1985, 381-390.
53. Pijaudier-Cabot G. and Bazant Z. P. Nonlocal damage theory. J. EM, Am.
Soc. Civ. Engrs, 113, 1987, 1512-1533.
54. Sinha B. P. et al. Stress-strain relations for concrete under cyclic loading. J.
Am. Concr. Inst., 61, No. 2, 1964, 195-211.
55. Reinhardt H. W. and Cornelissen H. A. W. Post-peak cyclic behavior of
concrete in uniaxial tensile and alternating tensile and compressive loading.
Cement and Concrete Research, 14, 1984, 263-278.
56. Eligehausen R. and Ozbolt J. Size effect in concrete structures. RILEM-CEB
Workshop on Application of Fracture Mechanics in Concrete. Turin, 1990,
38.
57. Ozbolt J. and Bazant Z. P. Microplane model for cyclic triaxial behavior of
concrete. J. EM, Am. Soc. Civ. Engrs, 118, No. 7, 1992, 1365-1386.
58. Hillerborg, A. et al. Analysis of crack formation and crack growth in
concrete by means of fracture mechanics and finite elements. Cement and
Concrete Research, 6, 1976, 773-782.
59. Duda H. Bruchmechanische Verhalten von Beton unter monotoner und
zyklischer Zugbeanspruchung. Thesis, T. H. Darmstadt, 1990.
60. Petersson P. E. Crack growth and development of fracture zones in plain
concrete and similar materials. Report TVBM-1006, Thesis, Div. of
Building Materials, University of Lund, Sweden, 1981.
61. Gustafsson P. J. Fracture mechanics studies of nonyielding materials like
concrete. Report TVBM-1007, Thesis, Div. of Building Materials,
University of Lund, Sweden, 1985.
62. Gopalaratnam V. S. and Shah, S. P. Softening response of plain concrete in
direct tension. ACI Journal, 82, No. 27, 1985, 310-323.
63. Cornelissen H. A. W. et al. Experiments and theory for the application of
fracture mechanics to normal and lightweight concrete. Fracture Toughness
and Fracture Energy of Concrete. F. H. Wittmann (ed.), Elsevier,
Amsterdam, 1986, 565-575.
64. Rots J. G. et al. Smeared crack approach and fracture localization in
concrete. Heron Journal, 30(1), Delft, 1985.
65. Gylltoft K. Fracture mechanics model for fatigue in concrete. RILEM
Materials and Structures, 17(97), 1984, 55-58.
66. Reinhardt H. W. et al. Tensile tests and failure analysis of concrete. J. Struct
Engng, Am. Soc. Civ. Engrs, 112(11), 1986, 2462-2477.
67. Yankelevsky D. Z. and Reinhardt H. W. Uniaxial behaviour of concrete in
cyclic tension. J. Struct Engng, Am. Soc. Civ. Engrs, 115(1), 1989, 166—
182.
68. Duda H. and Konig G. Rheological model for the stress-crack-width

181
RC ELEMENTS UNDER CYCLIC LOADING

relation of concrete under monotonic and cyclic tension, fracture behaviour

and design of materials and structures. ECF8 (European Group of
Fracture), Turin, Oct. 1990, 585-594.
69. Hordijk D. A. and Reinhardt H. W. A constitutive model for crack cyclic
behaviour of plain concrete, fracture behaviour and design of materials and
structures. ECF8 (European Group of Fracture), Turin, Oct. 1990, 579-
584.
70. Konig G. et al. SNAP — Ein nichtlineares Finite-Element-Programm,
Bericht zum Schlusskolloquium des DFG-Schwerpunktprogramms.
Nichtlineare Berechnungen im Konstruktiven Ingenieurbau, Mar. 1989
Hanover, S. 687-700, Springer-Verlag, New York, Berlin, Heidelberg,
London, Paris, Tokyo, 1989.
71. Hordijk D. A. Doctoral Thesis, Delft University of Technology, The
Netherlands.
72. Comite Euro-International du Beton, Response ofRC critical regions under
large amplitude reversed actions, Paris, Aug. 1983, Bulletin d'Information
161.
73. Agrawal G. L. et al. Response of doubly reinforced concrete beams to
cyclic loading. ACI Journal, 62, No. 7, July 1965.
74. Dafalias Y. K. and Popov E. P. A model for nonlinear hardening materials
for complex loading. Acta Mechanica, 21, 1975.
75. Finite element analysis of reinforced concrete. State-of-the-art report, Am.
Soc. Civ. Engrs, New York, 1982.
76. Park R. et al. Reinforced concrete members with cyclic loading. J. Struct.
Div., Am. Soc. Civ. Engrs, 98, No. ST7, July 1972.
77. Ramberg, R. and Osgood W. R. Description of stress strain curve by three
parameters. Technical Note 902, N.A.C.A., July 1943.
78. Ciampi, V. et al Analytical model for concrete anchorages of reinforced
bars under generalized excitations. Earthquake Engineering Research
Center, Report No. EERC 82-83, University of California, Berkeley, Nov.
1982.
79. Filippou F. C. et al. Effects of bond deterioration on hysteretic behavior of
reinforced concrete joints. Earthquake Engineering Research Center,
Report No. EERC 83-19, University of California, Berkeley.
80. Menegotto M. and Pinto P. E. Method of analysis for cyclically loaded
reinforced concrete plane frames including changes in geometry and
nonelastic behavior of elements under combined normal force and bending.
Proceedings, IABSE Symposium on Resistance and Ultimate Deformability
of Structures Acted on by Well Defined Repeated Loads, Lisbon, 1973.
81. Pinto P. E. and Giuffre A. Comportamento del Cemento Armato per
Sollecitazioni Cicliche di Forte Intensita, Giornale del Genio Civile, No. 5,
1970.
82. Stanton J. F. and McNiven H. D. The development of a mathematical model
to predict the flexural response of reinforced concrete beams to cyclic loads,
using system identification. Earthquake Engineering Research Center,
Report No. EERC 79-02, University of California, Berkeley, Jan. 1979.
83. Kaldjan M. I. Moment curvature of beams as Ramberg-Osgood functions. J.
Struct. Div., Am. Soc. Civ. Engrs, 93, No. ST10, Oct. 1967.
84. Ma S. Y. et al. Experimental and analytical studies on the hysteretic
behavior of reinforced concrete rectangular and t-beams. Earthquake
Engineering Research Center, Report No. EERC 76-2, University of
California, Berkeley, 1976.
85. Ciampi, V. and Nicoletti M. Parameter identification for cyclic constitutive
models with stiffness and strength degradation. Proceedings of 8th

182
 REFERENCES

European Conference on Earthquake Engineering (VIII ECEE), Lisbon,

1986.
86. Zulfiqar N. and Filippou F. C. Models of critical regions in reinforced
concrete frames under earthquake excitations. Earthquake Engineering
Research Center, Report No. EERC 90-06, University of California,
Berkeley, May, 1990.
87. Su T.-L. Behavior of reinforcing bars under inelastic cyclic reversed
loadings, Report No. SM79-I, Department of Civil Engineering, University
of Washington, Seattle, 1979.
88. Goto Y. Cracks formed in concrete around deformed tension bars. ACI
Journal, 62, No. 1, Jan. 1965, 71-93.
89. Giuriani E. Experimental investigation on the bond-slip law of deformed
bars in concrete. IABSE Colloquium Delft 1981, IABSE, Zurich 1981, 121—
142.
90. Rehm G. Uber die Grundlagen des Verbundes zwischen Stahl und Beton.
Deutscher Ausschuss fiir Stahlbeton, 138, 1961.
91. Comite Euro-International du Beton, Bond action and bond behaviour of
reinforcement. CEB, Paris, Dec. 1981, State of the art report, Bulletin
d'Information 151.
92. Eligehausen R. et al. Local bond stress-siip relationships of deformed bars
under generalized excitations. Earthquake Engineering Research Center,
Report No. UCB/EERC 83-23, University of California, Berkeley, Oct.
1983.
93. Comite Euro-International du Beton, CEB — FIP Model Code 1990 First
Draft. CEB, Paris, Sept. 1990, Bulletin d'Information 195.
94. Bresler B. and Bertero V. Behaviour of reinforced concrete under repeated
load. J. Struct. Div., Am. Soc. Civ. Engrs, 94, No. ST6, June 1968, 1567-
1590.
95. Edwards A. D. and Yannopoulos P. J. Local bond stress-slip relationship
under repeated loading. Mag. Cone. Res., 30, No. 103, June 1978, 62-72.
96. Rehm G. and Eligehausen R. Bond of ribbed bars under high-cycle repeated
loads. ACI Journal, 76, No. 2, Feb. 1979.
97. Balazs G. L. Bond behavior under repeated loads. Studie e Ricerche —
Corso Flli. Pesenti, Politecnico di Milano, 8, 1986, 395-^30.
98. Balazs G. L. Bond behaviour under repeated load. CEB, Treviso, May 1988,
Bulletin d'Information 195, 197-202.
99. Fehling E. Zur Energiedissipation und Steifigkeit von Stahlbetonbauteilen
unter besonderer Beriicksichtigung von Rissbildung und verschieblichem
Verbund. Dissertation, T. H. Darmstadt, 1990.
100. Lieberum K. H. Das Tragverhalten von Beton bei extramer
Teilflachenbelastung. Dissertation, T. H. Darmstadt, 1987, D 17.
101. Rohling A. and Rostasy F. S. Zum EinfluB des Verbundkriechens auf die
RiBbreitenentwicklung sowie auf die Mitwirkung des Betons aus Zug
zwischen den Rissen. Institut fur Baustoffe, Massivbau und Brandschutz,
Technische Universitat Braunschweig, Deutsche Forschungsgemeinschaft
Bonn, Fachseminar 19. und 20. May 1988, p. 133 ff.
102. Shah S. P. and Chung L. Effect of cyclic loading rate on response of model
beam-column joints and anchorage bond. Proceedings of the 3rd U.S. Nat.
Conf. Earthquake Eng., EERI, San Francisco, 1986.
103. RILEM, Dynamic behaviour of concrete structures. Report of the RILEM
65 MDB Committee, G. P. Tilly (ed.), Elsevier, Amsterdam, 1986.
104. Tassios T. P. Properties of bond between concrete and steel under load
cycles idealizing seismic actions. AICAP-CEB Symposium, Structural
Concrete Under Seismic Actions, Comite Euro-International du Beton, 1,

183
RC ELEMENTS UNDER CYCLIC LOADING

Rome, 1979, Bulletin d'Information 131.

105. Comite Euro-International du Beton. Response of RC critical regions under
large amplitude reversed actions. CEB, Prague, Oct. 1983, Bulletin
d'Information 161.
106. ACI Committee 408. Bond under cyclic loading. State of the art report.
107. Balazs G. L. Fatigue of bond. Paper accepted for publication by ACI
Materials Journal, 88, No. 6, Nov.-Dec, 1991.
108. Morita S. and Kaku T. Local bond stress-slip relationship under repeated
loading. Resistance and Ultimate Deformability of Structures Acted on by
Well Defined Repeated Loads. IABSE Symposium, Lisbon, 1973, 221-227.
109. Eligehausen R. et al. Hysteretic behavior of reinforcing deformed hooked
bars in R/C joints. Proceedings of the Seventh European Conference on
Earthquake Engineering, 4, Athens, Sept. 1982, 171-178.
110. Hawkins H. M. et al. Local bond strength of concrete for cyclic reversed
loadings. Bond in Concrete. P. Bartos (ed.), Applied Science Publishers
Ltd., London, 1982, 151-161.
111. Plaines P. et al. Bond relaxation and bond-slip creep under monotonic and
cyclic actions. Bond in Concrete. P. Bartos (ed.), Applied Science
Publishers Ltd., London, 1982, 193-205.
112. Balazs G. L. Bond softening under reversed load cycles. Studi e Ricerche —
Corso Flli. Pesenti, Politecnico di Milano, 11, 1989, 503-524.
113. Viwathanatepa S. et al. Effects of generalized loadings on bond of
reinforcing bars embedded in confined concrete blocks. Earthquake
Engineering Research Center, Report No. UCB/EERC-79/22, Aug. 1979.
114. Filippou F. C. A simple model for reinforcing bar anchorages under cyclic
excitations. J. Struct. Engng., Am. Soc. Civ. Engrs, 112, No. 7, July 1986.
115. Pochanart S. and Harmon T. Bond-slip model for generalized excitation
including fatigue. ACI Materials Journal, Sept.-Oct. 1989, 465^1-76.
116. Cedolin L. et al. Tensile behavior of concrete. J. EM., Am. Soc. Civ. Engrs,
113, No. 3, Mar. 1987.
117. Millard S. G. and Johnson R. P. Shear transfer across cracks in reinforced
concrete due to aggregate interlock and dowel action. Mag. Cone. Res., 36,
No. 126, Mar. 1984, 9-21.
118. Walraven J. C. Fundamental analysis of aggregate interlock. Journal of the
Structural Division, Am. Soc. Civ. Engrs, 107, No. ST11, Nov. 1981, 2245-
2270.
119. Walraven J. C. and Reinhardt H. W. Theory and experiments on the
mechanical behavior of cracks in plain and reinforced concrete subjected to
shear loading. Heron Journal, 26, No. 1 A, Dept. of Civil Engineering, Delft
University of Technology, Delft, 1981.
120. Houde J. and Mirza M. S. A finite element analysis of shear strength of
reinforced concrete beams. ACI Special Publication SP42, Detroit, 1974,
103-128.
121. Paulay T. and Loeber P. J. Shear transfer by aggregate interlock. ACI
Special Publication SP42, Detroit, 1974, 1-15.
122. Daschner F. and Kupfer H. Tests on shear transfer across cracks in normal
and light concretes (in German), Bauingenieur 57, 1982, 57-60.
123. Divakar M. P. et al. Constitutive model for shear transfer in cracked
concrete. /. Struct. Engng, Am. Soc. Civ. Engrs, 113, No. 5, May 1987,
1046-1062.
124. Nishimura A. et al. Shear transfer at cracked section in reinforced concrete.
Transactions of the Japan Concrete Institute, 4, 1982, 257-268.
125. Taylor H. P. J. The fundamental behaviour of reinforced concrete beams in
bending and shear. ACI Special Publication SP42, Detroit, 1975, 43-77.

184
 REFERENCES

126. Dulacska H. Dowel action of reinforcement crossing cracks in concrete.

ACI Journal, 69, No. 69-70, Dec. 1972, 745-757.
127. Furuuchi H. and Kakuta Y. Deformation behavior in dowel action of
reinforcing bars. Transactions of the Japan Concrete Institute, 7, 1985, 263-
268.
128. Krefeld W. J. and Thurston C. W. Contribution of longitudinal steel to shear
resistance of reinforced concrete beams. ACI Journal, 63, No. 14, Mar.
!966, 325-344.
129. Dei Poli S. et al. Dowel action as a means of shear transmission in R/C
elements: a state-of-art and new test results, (in Italian), Studi e Ricerche, 9/
87. School for the Design of R/C Structures, Politecnico di Milano, Milan,
1988, 217-303.
130. Dei Poli S. et al. Shear transfer by dowel action in R/C elements: the effects
of transverse reinforcement and concrete cover, (in Italian), Studi e
Ricerche, 10/88, School for the Design of R/C Structures, Politecnico di
Milano, Milan, 1989, 9-76.
131. Di Prisco M. and Gambarova P. G. Test results and modelling of dowel
action in normal, high-strength and fiber-reinforced concrete. Proc. of the
First Bi-Annual Environmental Specialty Conference of the Canadian
Society of Civil Engineers — CSCE, 2, Hamilton (Ontario), May 1990,
702-722.
132. Paschen H. and Schonhoff T. Investigations on shear connectors made of
reinforcing steel embedded in concrete (in German), DAfSt, Report 346,
Berlin, 1983, 105-149.
133. Vintzeleou E. N. and Tassios T. P. Behavior of dowels under cyclic
deformations. ACI Structural Journal, Technical Paper No. 84-S3, Jan.-
Feb. 1987, 18-30.
134. Giuriani E. and Rosati G. P. Behaviour of concrete elements under tension
after cracking. Studi e Ricerche, 8/86, School for the Design of RC
Structures, Politecnico di Milano, Milan, 1987.
135. Fardis M. N. and Buyukozturk O. Shear transfer model for reinforced
concrete. J. EM, Am. Soc. Civ. Engrs, 105, No. EM2, Apr. 1979, 255-275.
136. Leombruni P. et al. Analysis of cyclic shear transfer in reinforced concrete
with application to containment wall specimens. MIT-CE R79-26, June
1979.
137. Jimenez-Perez R. et al. Shear transfer across cracks in reinforced concrete.
Cornell University, Dept. of Structural Engineering, Report 78-4, Aug.
1978.
138. Laible J. P. et al. Experimental investigation of seismic shear transfer
across cracks in concrete nuclear containment vessels. ACI Special
Publication SP53, Detroit, 1977, 203-226.
139. Bazant Z. P. and Gambarova P. G. Rough cracks in reinforced concrete. /.
Struct. Div., Am. Soc. Civ. Engrs, 106, No. ST4, Apr. 1980, 819-842.
140. Gambarova P. G. Shear transfer by aggregate interlock in cracked
reinforced concrete subject to repeated loads (in Italian), Studi e
Ricerche, 1H9, School for the Design of R/C Structures, Politecnico di
Milano, Milan, 1980, 43-70.
141. Gambarova P. G. and Karakoc. C. A new approach to the analysis of the
confinement role in regularly cracked concrete elements. Transactions of
the 7th SMiRT Conference, H, Paper H5/7, Chicago, Aug. 1983, 251-261.
142. Gambarova P. G. Shear transfer in R/C cracked plates (in Italian),
Transactions of the 1983 Meeting of the Italian Society of R/C and P/C
Structures-AICAP, Bari, May 1983, 141-156.
143. Frenay J. Time-dependent shear transfer of cracked reinforced concrete.

185
RC ELEMENTS UNDER CYCLIC LOADING

Computational Mechanics of Concrete Structures: Advances and

Applications, Transactions of IABSE Colloquium Delft 87, Delft, Aug.
1987, 113-120.
144. Pruijssers A. F. Shear resistance of cracked concrete subjected to cyclic
loading. Computational Mechanics of Concrete Structures: Advances and
Applications, Transactions of IABSE Colloquium Delft 87, Delft, Aug.
1987, 43-50.
145. Walraven J. C. Aggregate interlock under dynamic loading. Darmstadt
Concrete, 1, 1986, 143-156.
146. Yoshikawa H. et al. Analytical model for shear slip of cracked concrete. J.
Struct. Engng, Am. Soc. Civ. Engrs, 115, No. 4, Apr. 1989, 771-788.
147. Li B. and Maekawa K. Contact density model for cracks in concrete.
Computational Mechanics of Concrete Structures: Advances and
Applications, Transactions of IABSE Colloquium Delft 87, Delft, Aug.
1987,51-62.
148. Bazant Z. P. and Gambarova P. G. Crack shear in concrete: crack band
microplane model. J. Struct. Engng, Am. Soc. Civ. Engrs, 110, No. 9, Sept.
1984, 2015-2035.
149. Riggs H. R. and Powell G. H. Rough crack model for analysis of concrete.
/ EM, Am. Soc. Civ. Engrs, 112, No. 5, May 1986, 448-464.
150. Bazant Z. P. and Oh B. H. Microplane model for progressive fracture of
concrete and rock. J. EM., Am. Soc. Civ. Engrs, 111, No. 4, 1985.
151. Gambarova P. G. and Floris E. Microplane model for concrete subject to
plane stresses. Nuclear Engng and Design, 91, 1986.
152. Bazant Z. P. and Prat P. C. Microplane model for brittle-plastic material: I.
theory; II. verfication. J. EM., Am. Soc. Civ. Engrs, 114, No. 10, 1988.
153. Feenstra P. H. et al. Numerical study on crack dilatancy. I: models and
stability analysis — II: applications. J. EM, Am. Soc. Civ. Engrs, 117, No. 4,
Apr. 1991, 733-753 and 754-769.
154. Cedolin L. and Dei Poli S. Finite element studies of shear critical R/C
beams. J. EM., Am. Soc. Civ. Engrs, 103, No. EM3, June 1977, 395^110.
155. Walraven J. C. and Keuser W. The shear retention factor as a compromise
between numerical simplicity and realistic material behavior. Darmstadt
Concrete, 2, 1987, 221-234.
156. Hanna Y. G. and Mirza M. S. Post-cracking behavior of planar structures.
Proceedings of the Canadian Society of Civil Engineers, Annual
Conference, Ottawa, June 1983, 107-127.
157. Tassios T. and Vintzeleou E. Concrete-to-concrete friction. /. Struct.
Engng, Am. Soc. Civ. Engrs, 113, No. 4, Apr. 1987, 832-849.
158. Isenberg J. and Adham S. Analysis of orthotropic reinforced concrete
structures. J. Struct. Div., Am. Soc. Civ. Engrs, 96, No. ST12, Dec. 1970,
2607-2623.
159. Duchon N. B. Analysis of reinforced concrete membranes subject to tension
and shear. AC! Journal, 69, No. 9, Sept. 1972, 578-583.
160. Collins M. P. Towards a rational theory for reinforced concrete members in
shear. J. Struct. Div., Am. Soc. Civ. Engrs, 104, No. ST4, Apr. 1978, 649-
666.
161. Perdikaris P. C. and White R. N. Shear modulus of precracked R/C panels.
J. Struct. Engng, Am. Soc. Civ. Engrs, 111, No. 2, Feb. 1985, 270-289.
162. Tanabe T. and Yoshikawa H. Constitutive equations of a cracked reinforced
concrete panel. Computational Mechanics of Concrete Structures:
Advances and Applications, Transactions of IABSE Colloquium Delft 87,
Delft, Aug. 1987, 17-34.
163. Soroushian P. et al. Analysis of dowel bars acting against concrete core.

186
 REFERENCES

ACI Journal, 83, No. 59, July-Aug. 1986, 642-649.

164. Vintzeleou E. N. and Tassios T. P. Mathematical models for dowel action
under monotonic and cyclic conditions. Mag. Cone. Res., 38, No. 134, Mar.
1986, 13-22.
165(a). Soroushian P. et al. Bearing strength and stiffness of concrete under
reinforcing bars. ACI Materials Journal, Technical Paper No. 84-M19,
May-June 1987, 179-184.
165(b). Soroushian P. et al. Behavior of bars in dowel action against concrete
cover. ACI Structural Journal, Technical Paper No. 84-S18, Mar.-Apr.
1987, 170-176.
166. Johnston D. W. and Zia P. Analysis of dowel action. / Struct. Div., Am.
Soc. Civ. Engrs., 97, No. ST5, May 1971, 1611-1630.
167. Giuriani E. On the axial stiffness of a bar in cracked concrete. Bond in
Concrete. Bartos P. (ed.), Applied Science Publishers, London, 1982.
168. Brenna et al. Studi e Ricerche, 11/89, School for the Design of RC
Structures, Politecnico di Milano, Milan, 1990.
169. Di Prisco et al. Non-linear modelling of dowel action of a bar immersed in a
concrete mass of infinite extent (in Italian). Studi e Ricerche, 10/88. School
for the Design of RC structures, Politecnico di Milano, Milan, 1989.
170. Jimenez-Perez R. et al. Bond and dowel capacities of reinforced concrete.
ACI Journal, Symposium Paper, No. 76-4, Jan. 1979.
171. Mehlhorn G. Some developments for finite element analyses of reinforced
concrete structures. Computer aided analysis and design of concrete
structures, Proc. 2nd Int. Conf. Zell-am-See, Apr. 1990, 1319-1336.
172. Mehlhorn G. et al. Nonlinear contact problems — a finite element approach
implemented in ADINA. Computers and Structures, 21, No. 1/2, 1985, 69-
80.
173. Mehlhorn G. and Keuser M. Isoparametric contact elements for analysis of
reinforced concrete structures. Finite Element Analysis of Reinforced
Concrete Structures. Am. Soc. Civ. Engrs, 1986, 329-347.
174. Miguel P. F. A discrete-crack model for the analysis of concrete structures.
Computer aided analysis and design of concrete structures, Proc. 2nd Int.
Conf. Zell-am-See, Apr. 1990, 897-908.
175. Bathe K.-J. et al. Nonlinear analysis of concrete structures. Computers and
Structures, 32, No. 3/4, 1989, 563-590.
176. Grootenboer H. J. et al. Numerical models for reinforced concrete structures
in plane stress. Heron Journal, 26, No. lc, 1981, 83.
177. Hibbitt H. D. et al. ABAQUS Manuals: V.I: User's Manual; V.2: Theory
Manual; V.3: Example Problems: V.4: System's Manual' Version 4.5,
1984, Version 4.8, 1989, Providence, RI.
178. Blaauwendraad J. Realizations and Restrictions. Applications of numerical
models to concrete structures. Finite Element Analysis of Reinforced
Concrete Structures. Meyer C. and Okamura H. (eds), ASCE 1986, 557-
578.
179. Wang Q. B. et al. Failure of reinforced concrete panels — how accurate the
models must be. Computer aided analysis and design of concrete structures,
Proc. 2nd Int. Conf. Zell-am-See, Apr. 1990, 153-163.
180. Same Y. Material nonlinear time-dependent three dimensional finite
element analysis of reinforced and prestressed concrete structures. Thesis
submitted to the Department of Civil Engineering, Massachusetts Institute
of Technology, in partial fulfillment of the requirements for the Degree of
Doctor of Philosophy, Cambridge, MA, 1974.
181. Chen A. C. T. and Chen W. F. Constitutive relations for concrete. J. EM,
Am. Soc. Civ. Engrs, 97, EM3, Aug. 1975, 4 6 5 ^ 8 1 .

187
RC ELEMENTS UNDER CYCLIC LOADING

182. Dahlblom O. and Ottosen N. S. Smeared crack analysis using a generalized

fictitious crack model. J. EM, Am. Soc. Civ. Engrs, 116, No. 1, Jan. 1990,
55-76.
183. Gupta A. K. and Maestrini S. R. Post-cracking behaviour of membrane
reinforced concrete elements including tension-stiffening. J. Struct. Engng,
Am. Soc. Civ. Engrs, 115, No. 4, Apr. 1989, 957-976.
184. Hillerborg A. Numerical methods to simulate softening and fracture of
concrete. Fracture Mechanics of Concrete: Structural Application and
Numerical Calculation. Sih G. C. and Di Tommaso A. (eds) Martinus
Nijhoff Publishers, Dordrecht, Netherlands, 1985, 141-170.
185. Bazant Z. P. and Cedolin L. Blunt crack propagation in finite element
analysis. J. EM Div., Am. Soc. Civ. Engrs, 105, No. EM2, Apr. 1979, 297-
315.
186. Okamura H. and Maekawa K. Non-linear analysis and constitutive models
of reinforced concrete. Computer aided analysis and design of concrete
structures, Proc. 2nd Int. Conf. Zell-am-See, Apr. 1990, 831-850.
187. Rots J. G. and Blaauwendraad J. Crack models for concrete: discrete or
smeared? Fixed, multi-directional or rotating? Heron Journal, 34, No. 1,
1989, 59.
188. Bazant Z. P. and Oh B. H. Crack band theory for fracture of concrete.
Materials and Structures, 16, 1983, 155-177.
189. Bazant Z. P. Mechanics of distributed cracking. Appl. Mechanics Rev., Am.
Soc. Mech. Engrs, 39, No. 5, 1986, 675-705.
190. Cervenka V. Constitutive model for cracked reinforced concrete. ACI
Journal, Proc. 82, No. 6, Nov.-Dec. 1985, 877-882.
191. Crisfield M. A. and Wills J. Analysis of R/C panels using different concrete
models. J. EM, Am. Soc. Civ. Engrs, 115, EM3, Mar. 1989, 578-597.
192. Barzegar F. Analysis of RC membrane elements with anisotropic
reinforcement. J. Struct. Engng, Am. Soc. Civ. Engrs, 115, No. 3, Mar.
1989, 647-665.
193. Barzegar F. and Ramaswamy A. A secant post-cracking model for
reinforced concrete with particular emphasis on tension stiffening.
Computer aided analysis and design of concrete structures, Proc. 2nd Int.
Conf. Zell-am-See, Apr. 1990, 1001-1016.
194. de Borst, R. and Nauta P. Non-orthogonal cracks in a smeared finite
element model. Engrg Computations, 2, 1985, 3 5 ^ 6 .
195. Cervenka V. Constitutive model for cracked reinforced concrete under
general load histories. CEB, Mar. 1987, 157-167, Bulletin d'lnformation
178/179.
196. Vecchio F. J. and Collins M. P. The modified compression-field theory for
reinforced concrete elements subjected to shear. ACI Journal, Proc. 83, No.
2, Mar.-Apr. 1986, 219-231.
197. Kolmar W. Beschreibung der Kraftubertragung uber Risse in nichtlinearen
Finite Elemente Berechnungen von Stahlbetontragwerken. Technische
Hochschule Darmstadt, Dec. 1986, 193.
198. Gupta A. K. and Akbar H. Cracking in reinforced concrete analysis. J.
Struct. Engng., Am. Soc. Civ. Engrs, 110, No. 8, Aug. 1984, 1735-1746.
199. Balakrishnan S. and Murray D. W. Prediction of R/C panel and deep beam
behaviour by NLFEA. J. Struct. Engng, Am. Soc. Civ. Engrs, 114, No. 10,
Oct. 1988, 2323-2342.
200. Hu H.-T. and Schnobrich W. C. Nonlinear analysis of cracked reinforced
concrete. ACI Structural Journal, Proc. 87, No. 2, Mar.-Apr. 1990, 199-
207.
201. Milford R. V. and Schnobrich, W. C. Numerical model for cracked

188
 REFERENCES

concrete. Computer aided analysis and design of concrete structures, Proc.

1st Int. Conf., Split, Yugoslavia, Sept. 1984.
202. Vecchio F. J. and Collins M. P. The response of reinforced concrete to in-
plane shear and normal stresses. Publ. No. 82-03, Dept. of Civil
Engineering, University of Toronto, Mar. 1982, 332.
203. Izumo J. and Okamura H. Ultimate strength and deformation of RC panels
subjected to in-plane stresses. Computer aided analysis and design of
concrete structures, Proc. 2nd Int. Conf. Zell-am-See, Apr. 1990, 177-188.
204. Okamura H. et al. Verification of modeling for reinforced concrete finite
element. Finite Element Analysis of Reinforced Concrete Structures. Meyer
C. and Okamura H. (eds.), Am. Soc. Civ. Engrs, 1986, 528-544.
205. Stevens N. J. et al. Analytical modeling of reinforced concrete subjected to
monotonic and reversed loading. Res. Report, Dept. of Civil Engineering,
University of Toronto, Jan. 1987, 201.
206. Shima H. et al. A constitutive model for tension stiffness of reinforced
concrete under reversed loading — post yield behaviour is included.
Computer aided analysis and design of concrete structures, Proc. 2nd Int.
Conf. Zell-am-See, Apr. 1990, 1079-1090.
207. Hard G. Die Arbeitslinie Eingebetteter Staehle unter Erst- und
Kurzzeitbelastung. Beton- und Stahlbetonbau, 8, 1983.
208. Guenther G. Verbundverhalten zwischen Stahl und Beton unter monoton
steigender, schwellender und lang andauernder Belastung. Dissertation,
Kassel, 1989.
209. Fehling E. Zur Energiedissipation und Steifigkeit von Stahlbetonbauteilen
unter besonderer Berucksichtigung von Rissbildung und verschieblichem
Verbund. Dissertation, T. H. Darmstadt, 1990.
210. Gupta A. K. and Maestrini S. R. Unified approach to modeling post-
cracking membrane behavior of reinforced concrete. Jour, of Struct.
Engineering, Am. Soc. Civ. Engrs, 115, No. 4, Apr. 1989, 977-993.
211. Floegl H. and Mang H. A. Tension stiffening concept based on bond slip. J.
Struct. Div., Am. Soc. Civ. Engrs, 108, ST12, Dec. 1982, 2681-2701.
212. Collins M. P. and Mitchell D. Prestressed concrete basics. Canadian
Prestressed Concrete Institute, Ottawa, 1987, 614.
213. Kolleger J. et al. Zug- und Zug-Druckversuche an Stahlbetonscheiben.
Forschungsberichte aus dem Fachgebiet Massivbau, Nr. 1, Gesamt-
hochschule Kassel, 1986, 190.
214. Cervenka V. et al. Computer simulation of anchoring technique in
reinforced concrete beams. Computer aided analysis and design of
concrete structures, Proc. 2nd Int. Conf. Zell-am-See, Apr. 1990, 1-19.
215. Bujadham B. et al. Cyclic discrete crack modeling for reinforced concrete.
Computer aided analysis and design of concrete structures, Proc. 2nd Int.
Conf. Zell-am-See, Apr. 1990, 1225-1236,
216. Schlaich J. et al. Druck und Querzug in bewehrten Betonelementen.
Forschungsbericht des Instituts fur Massivbau, Universitat Stuttgart, 1982.
217. Bathe K. J. Finite element procedures in engineering analysis. Prentice-
Hall, 1982.
218. Bathe K. J. and Dvorkin E. N. On the automatic solution of nonlinear finite
element equations. Computers & Structures, 17, 1983, 871-880.
219. Crisfield M. A. An arc-length method including line searches and
accelerations. Int. J. Num. Meth. Eng., 19, 1983, 1269-1289.
220. Matthies H. and Strang G. The solution of nonlinear finite element
equations. Int. J. Num. Meth. Eng., 14, 1979, 1613-1626.
221. Batoz J. L. and Dhatt G. S. Incremental displacement algorithms for non
linear problems. Int. J. Num. Meth. Eng., 14, 1979, 1262-1266.

189
RC ELEMENTS UNDER CYCLIC LOADING

222. Crisfield M. A. A fast incremental/iterative solution procedure that handles

'Snap-through'. Computers & Structures, 13, 1981, 55-62.
223. Ramm E. Strategies for tracing the nonlinear response near the limit points.
Wunderlich, Stein, Bathe (eds), Nonlinear Finite Element Analysis in
Structural Mechanics, Springer Verlag, 1981.
224. Hughes T. J. R. and Belytschko T. A precis of developments in
computational methods for transient analysis. J. Appl. Mech., 50, 1983,
1033-1041.
225. Hilber H. M. et al. Improved numerical dissipation for time integration
algorithms in structural dynamics. Earthq. Eng. & Str. Dyn., 5, 1977, 283-
292.

190