Softening of concrete loaded in compression

Citation for published version (APA):

Vonk, R. A. (1992). Softening of concrete loaded in compression. [Phd Thesis 1 (Research TU/e / Graduation
TU/e), Built Environment]. Technische Universiteit Eindhoven. https://doi.org/10.6100/IR375705

DOI:
10.6100/IR375705

Document status and date:

Published: 01/01/1992

Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)

Please check the document version of this publication:

• A submitted manuscript is the version of the article upon submission and before peer-review. There can be
important differences between the submitted version and the official published version of record. People
interested in the research are advised to contact the author for the final version of the publication, or visit the
DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page
numbers.
Link to publication

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.

If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please
follow below link for the End User Agreement:
www.tue.nl/taverne

Take down policy

If you believe that this document breaches copyright please contact us at:
[email protected]
providing details and we will investigate your claim.

Download date: 17. Mar. 2024

 SOFTENING
OF CONCRETE LOADED
IN COMPRESSION
/ I
I

Rene Vonk
 SOFfENING

OF CONCRETE LOADED IN COMPRESSION

CIP- GEGEVENS KONINKLIJKE BIBLIOTHEEK, DEN HAAG

Vonk, Rene Alfred

Softening of concrete loaded in compression / Rent~ Alfred

Vonk.- Eindhoven: Technische Universiteit Eindhoven.-
m.
Proefschrift Eindhoven. - Met lit. opg. - Met samenvatting
in het Nederlands
ISBN 90-386-Q062-3
Trefw.: beton.

Copyright R.A. Vonk, Eindhoven, 1992

Printed by Wibro dissertatiedrukkerij. Helmond

 SOFTENING
OF CONCRETE LOADED
IN COMPRESSION

PROEFSCHRIFT

ter verkrijging van de graad van doctor aan de

Technische Universiteit Eindhoven, op gezag van
de Rector Magnificus, pro f. dr. J .H. van Lint,
voor een commissie aangewezen door het College
van Dekanen in het openbaar te verdedigen
op woensdag 24 juni 1992 om 16.00 uur

door

RENE ALFRED VONK

geboren te Leiderdorp
Dit proefschrift is goedgekeurd door de promotoren

prof. dr. ir. H.S. Rutten

en
prof. dr. -ing. K. Willam

en copromotor

dr. ir. J.G.M. van Mier

Voor mijn ouders

Voor Monique
ACKNOWLEDGEMENTS

The author wishes to express his thanks to pro£. dr. ir. H.S. Rutten, dr. ir. J.G.M. van
Mier and ir. H.J. Fijneman for their support and stimulating discussions during the
research. Especia.lly prof. dr. -ing. K. Willam is thanked for accepting to be second
promoter and carefully discussing and commenting the thesis.

The author is indebted to all eo-workers of the group of structural design, the,co--workers
of the Pieter van Musschenbroek Laboratory and the computer experts of the university for
their valuable assistance:

The financial support of the Netherlands Technology Foundation (STW) is gratefully

acknowledged.
CONTENTS

1. INTRODUCTION

1.1 Subject of research 1

1.2 Aim, scope and method 3
1.3 Summary of contents 4

2. SOFTENING OF CONCRETE

2.1 Introduction 6
2.2 Microcrack formation 10
2.3 Localization of deformations 16
2.4 Influence of material parameters 20
2.5 Influence of a. multiaxial stress state 22
2.6 Structural behaviour and boundary conditions 26

3. TESTING TECHNIQUE

3.1 Loading technique 33

3.2 Measuring technique 40
3.3 Preparation of specimens 45
3.4 Crack-detection technique 47
3.5 Summary of tests 49

4. MODELLING OF SOFTENING

4.1 Introduction 51
4.2 Continuum models 52
4.3 Fracture mechanics 54
4.4 Cracking in finite-element analysis 56
4.5 Crack behaviour under shear loading 59
4.6 Strain softening in finite-element analysis 62
4. 7 Simulation of heterogeneity 64
5. MICROMECHANICAL MODEL

5.1 Introduction 68
5.2 Idea behind the micromechanical model 69
5.3 UDEC 71
5.4. Constitutive model describing interface behaviour 75
5.5 Simulation of softening with UDEC 80

6. PROCESS OF SOFTENING

6.1 Introduction 84
6.2 Discussion of a compression test 84
6.3 Simulations of tensile tests 92
6.4 Simulation of a compression test 99
6.5 Conclusions 109

7. PARAMETERSTUDY FOR THE MICROMECHANICAL MODEL

7.1 Introduction 111

7.2 Influence of random mesh 112
7.3 Influence of material parameters 114
7.4 Influence of a variation of strength 121
7.5 Influence of autogeneous shrinkage 122
7.6 Conclusions 123

8. INFLUENCE OF NONUNIFORM DEFORMATIONS

8.1 Introduction 127

8.2 Observations in tests 129
8.3 Cause of nonuniform deformations 133
8.4 Simulations with a simple numerical model 137
8.5 Comparison of tests and simulations 144
8.6 Simulations of nonuniform loading 148
8. 7 Conclusions 152
9. INFLUENCE LATERAL BOUNDARY RESTRAINT

9.1 Introduction 154

9.2 Test results 157
9.3 Micromechanical simulations 163
9.4 Conclusions 169

10. INFLUENCE SIZE

10.1 Introduction 171

10.2 Test results 172
10.3 Micromechanical simulations 185
10.4 Conclusions 190

11. RETROSPECTIVE VIEW 192

APPENDIX A

Mesh generation procedure

REFERENCES

SAMENVATTING
 CHAPTER I
INTRODUCTION

1.1 SUBJECT OF RESEARCH

Softening of concrete
This thesis deals with research into softening of concrete loaded in compression. Softening
is a gradual decrease of mechanical resistance due to a continuous increase of deformation
forced upon a material specimen or structure. Quasi-brittle materials show softening.
They fail due to a process of internal crack growth. Well-designed reinforced concrete
structures show, in addition to yielding ofthe reinforcement, a considerable crack growth
before attaining peak load. Therefore, softening is important in the assessment of
structural safety.

Finite element afllllysis

During the last few decades, the computational power of computers has increased
enormously. This has stimulated the use of finite element analysis in research and design
of concrete structures. The increase of computer power made it possible to calculate the
total load-deformation relation up to failure in the case of arbitrary structures. The
nonlinear behaviour due to crack growth under multiaxial stress states could be taken
into account. This has created the demand for knowledge of nonlinear material behaviour
and numerical techniques to take the nonlinear behaviour into account. Knowledge in
both has progressed significantly and has resulted in adequate models and methods.

Fracture mechanics
Fracture mechanics, studying the behaviour of cracks under load, was given a new
impulse by the growing possibilities of numerical analysis of concrete structures
(Hillerborg [1976], Rots [1988], Elfgren (ed.) [1990], Hordijk [1991]). It concentrated
mainly on the description of the growth of cracks under tensile loading. Although concrete
loaded in compression fails due to crack growth too, knowledge on tensile failure did not
yet include compressive failure, possibly because tensile cracking was considered more
important, compressive failure was not recognized to be a fracture-mechanics problem
and because compressive failure is a much more complex phenomenon than tensile failure.
In 1984, test results of Van Mier at Eindhoven University of Technology gave the first
strong indication that fracture mechanics governs compressive failure of concrete and that

1
it can not be described adequately by continuum mechanics. Therefore the present
research is directed to a fracture-mechanics approach to compressive failure of concrete

Stnu:tv.ral beh.afliov.r
Recently, the statement "compressive failure is a structural behaviour" (Van Mier [1984])
is encountered more and more in literature. Two reasons can be mentioned. First, it is
caused by the fact that continuum mechanics falls short of adequately describing
softening of concrete. The mechanical response of a softening concrete specimen is
governed by a local fracture process which dominates the behaviour of the surrounding
continuum. The heterogeneity of concrete plays an important role in this process. The
response of the specimen is a combined action by several components, which can be
looked upon as the behaviour of a structure. Second, compressive softening of concrete
proves to be highly dependent on the boundary conditions. They play an essential role in
the softening process, which should not be overlooked. Our knowledge of the phenomenon
is limited, which has made test results ambiguous and their interpretation difficult. In the
present research special attention will be given to the structural behaviour found for
compressive softening.

Growing interest
Progress in the field of fracture mechanics has generated a growing interest in the subject
and its applications in structural design. Results are finding their way to the building
codes (ACI [1990], CEB-FIP [1991]). For the moment, this is mainly as far as the
application of tensile properties of concrete are concerned. The CEB-FIP Model Code
states that there is a lack of experimental data. and lays down a "conservative"
stress-strain relation for compressive softening. However, if fracture mechanics is
applicable to compressive softening, no such softening stress-strain relation is correct.
Then the softening depends on the size of the structure. In the worst case no softening
will be found and a brittle failure will occur at peak stress. It appears that compressive
failure is as important in the failure of concrete structures as tensile failure. Investigators
have recently published work to show the importance of a better knowledge of the real
nature of compressive failure for the rotational capacity of beams (Hillerborg [1988,1991],
Rokngo & Koyanagi [1992]).

"Oofll!rde Mechanics" project

Research into the nonlinear analysis of the mechanical behaviour of concrete structures is
organized in The Netherlands in the CUR "Concrete Mechanics" project. It aims at the
development of numerical tools for the analysis of the behaviour of concrete structures
and fundamental research towards the formulation of reliable constitutive models. The

2
research described in this thesis is part of the CUR "Concrete Mechanics" project. The
Technical Universities of Delft and Eindhoven, TNO Building and Construction Research,
Dutch building companies and structural engineering consultants cooperate in this
project, which is coordinated by the CUR (Centre for Civil Engineering Research and
Codes).

1.2 AIM, SCOPE AND METHOD

Aim
The aim of this research is to increase the knowledge of compressive failure of concrete in
order to enable reliable modelling of this phenomenon.

Scope
The research concentrates on mechanical analysis and modelling of the structural
behaviour of softening concrete loaded in compression. The three-dimensional process of
crack formation is analysed as it is governed by the heterogeneity of concrete, the
boundary conditions and dimensions of concrete specimens. A link is made between
fracture mechanics and modelling of compressive softening of concrete.

The behaviour of normal-strength concrete with relatively stiff and strong aggregates is
considered. The influence of drying shrinkage, creep or loading rate is left out of
consideration, although these aspects certainly play an important role in concrete
softening.

Method
The structural behaviour of softening concrete is investigated by carrying out extensive
deformation measurements on specimens loaded in unia.xial compression. Dimensions and
boundary conditions of the specimens are varied to study their influence on the recorded
behaviour. The essence of softening is that the process determining this behaviour is
found to be inside the specimen. Therefore, interpretation of the test results is supported
by inspection of internal crack patterns. A simple but effective method has been
developed to record crack patterns inside the specimens after a test.

In addition to laboratory tests, numerical investigation of compressive softening by means

of rnicromecha.nica.l simulation is undertaken. The mechanical behaviour of concrete is
simulated by modelling its internal structure on a lower scale level and calculating the

3
response of this structure. This is done with the aid of finite element analysis including
fracture mechanics. This had recently proven to be promising for the description of tensile
softening and had not yet been applied to the description of compressive softening. Only
recently a similar approach, partly based on fracture mechanics, was used by Willam et
al. [1989]. The tool was found adequate for the investigation of the behaviour of
heterogeneous materials and can be nsed as a basis for the formulation of more simple
models. The advantage of a micromechanical approach is that the process taking place
inside the material, which is determinative of the behaviour at the macrolevel, can be
followed continuously in time and space.

1.3 SUMMARY O:F CONTENTS

Bu.ckgrourul
This thesis starts with a number of chapters which create the background for the present
work. An outline of the present knowledge on compressive softening is given in Chapter 2.
It describes mainly experimental results and introduces the phenomena which are
investigated in the present research. Where it is thought to be useful, tensile softening is
discussed. In Chapter 9 the details of the experimental technique applied in this research
are given. The simple crack-detection technique which has been developed is described.
The chapter ends with a summary of all the tests. Chapter 4 treats the subject of
modelling nonlinear concrete behaviour, with special attention to softening and its
representation in finite element analysis. It closes with a number of micromechanical
models used in describing concrete softening. In Chapter 5, the new micromechanical
model is presented. The computer program UDEC, which is used for the micromechanical
simulations, is discussed. A new constitutive interface model has been implemented to
describe the behaviour of cracks under the influence of combinations of tensile and shear
loading.

Softening process
In Chapter 6, the process of compressive softening is investigated in general. The results
of tests and micromechanical simulations are compared. A uniaxial tensile test is
simulated to show that the micromechanical model is perfectly capable of describing this
kind of test. Chapter 1 gives the results of a parameter sensitivity study. It shows the
performance of the micromechanical model with respect to a variation of a number of
parameters.

4
Boundary conditions
The influence of the boundary conditions on compressive softening is subsequently
evaluated. Chapter 8 discusses the influence of a nonuniform distribution of axial
deformations over the cross-flection of a specimen due to rotation of the loading platens.
The phenomenon is explained by means of an analytical model and studied further by
means of a simple numerical model. A comparison of numerical and test results is made.
Chapter 9 treats the influence of the restraint by the loading platens of the lateral
deformations of the specimen. The influence is shown of the use of various loading platens
in tests. Micromechanical simulations are compared with the test results.

Size effect
The size effect in compressive softening caused by localization of cracking and
deformations is studied in Chapter 10. Results of tests on specimens of different height
and width are compared and discussed. Special attention is given to the influence of the
lateral boundary restraint by the loading platens. The performance of the micro-
mechanical model in describing localization of cracking and deformations, and
consequently of the size effect, in softening is discussed.

Retrospective view
A review of the total research is given in Chapter 11.

5
 CHAPTER2

SOFTENING OF CONCRETE

2.1 INTRODUCTION

Heterogeneity of concrete
For a proper understanding of concrete softening, it is very important to realize that
concrete is a heterogeneous material. This is often not of obvious importance, but it will
be shown that this characteristic of concrete is far from negligible, when softening is
investigated.

(a) (b) (c)

Fig. 2.1.1 Scale levels for concrete as distinguished by Wittmann [19'87]:
(a) micro level, (b) meso level and (c) macro level

It has often been shown that analysis of a phenomenon on a small scale can serve to gain
insight into the phenomenon and model it on a larger scale. For concrete research,
Wittmann [1987] has proposed a practical subdivision in three scale levels:

1. Micro level
This takes into account the physical and chemical processes at a molecular level in
hardened cement paste, which is itself highly heterogeneous (Fig. 2.1.1a).

2. Mesa level
Here the behaviour ofthe complex structure of concrete, a composite comprising
aggregates, pores and cracks in a hardened cement paste matrix, is taken into account
(Fig. 2.l.lb).

6
3. Macro le'Vel
This considers a homogeneous material whose behaviour is described by simplified
phenomenal laws, which are intended for application in structural engineering (Fig.
2.l.lc).

For the sake of simplicity, the term microlevel will only be used in this thesis as the
opposite of macrolevel, although most of the discussion will be about processes at the
mesolevel as defined by Wittmann. Softening of concrete is observed and defined at the
macrolevel. It can be explained by taking into account processes at the microlevel.

Definition of softening
Softening is a gradual decrease of mechanical resistance due to a continuous increase of
deformation forced upon a material specimen or structure.

Softening is found in a deformation-controlled test on a specimen of concrete. A

characteristic stress-deformation relation in such a test is shown in Fig. 2.1.2. Under
increasing deformation, the tangential stiffness of the specimen Kt shows a continuous
decrease until it is zero for peak load. The post-peak decrease of mechanical resistance
due to the continuous increase of deformation is called softening. It is characterized by
the descending branch of the stress-deformation curve which has a negative tangential
stiffness Kt. In this thesis Kt will be used as the measure for brittleness. The behaviour of
a specimen will be called more brittle, when the minimum value of the tangential stiffness
Kt during softening is smaller.

CT

Oj,eok

I
peak :post peak
. ...
Wpeok w

Fig. 2.1.2 Characteristic nominal stress-deformation relation of a concrete specimen

loaded in compression under deformation control

7
Cause of softening
The cause of concrete softening is found at the microlevel in a continuous crack-growth
process (Fig. 2.1.2). Due to the heterogeneity of concrete, a uniformly distributed external
stress or deformation results in a highly nonuniform distribution of internal11tresses and
deformations. This, combined with the fact that internal strength is nonunif9rmly
distributed too, and that initial internal stresses a.re present, causes progressive failure of
internal bonds when concrete is subjected to progressive deformation (Hsu et al. [1963]).
The material is gradually weakened because a diminishing number of internal bonds is
capable of opposing the externally applied deformation. This finally results in softening at
the macroscopic level.

Localization of deformations
In view of the progressive crack growth, there seems to be no use in distinguishing
between the pre-peak nonlinear behaviour and the post-peak softening. There seems to
be a continuous process. Initially, the cracking starts as distributed microcra.cking. They
a.re stable, which means that they grow only when the load is increased. Around peak
load, the formation of macrocracks starts. They are unstable, which means that the load
has to decrease to avoid an uncontrolled growth. In a deformation-controlled test, this
macrocrack growth results in softening and localization of deformations, which is the most
salient feature connected to softening. Localization of deformations means that all further
deformations concentrate in the macrocracks, while the concrete parts in series with these
macrocracks show decreasing deformations due to unloading. Only very recently,
researchers have started to recognize that this localization of deformations takes place in
concrete loaded in compression and very liitle is known about it (Van Mier
[1984,1986ab]).

Material property or structural property?

In a homogeneous continuum every part of the volume has the same properties and reacts
in the same way to an external action. The notions stress and strain then have a meaning.
The stress can then be determined from the response of a test specimen by dividing the
force by the total area transmitting this force and the strain by dividing the deformation
by the distance over which the deformation is found. When the loading is uniform and the
boundary conditions a.re chosen correctly, the relation between stress and strain is unique
and does not depend on the size and shape of the specimen. This unique relation can then
be called a material property.

The theory of the homogeneous continuum is applicable to concrete, even though it is a

heterogeneous material, when the size of the concrete volume is much greater than the

8
characteristic size ofthe heterogeneity. Within a small stochastic scatter, the relation
between stress and strain can be determined as a material property. Sometimes the terms
nominal stress and nominal strain are nsed to remind one that concrete is a heterogeneons
material and that the calculated values are only average values for the total specimen.

During microcrack growth, the theory of the homogeneous continuum is still applicable to
concrete becanse the microcracks are at the scale level of the heterogeneity of concrete.
When macrocracks start to grow, the scale level of the cracks quickly rises to the scale
level of the structure. The continuum theory is no longer applicable. Stress and strain lose
their meaning. The cracks have to be taken into account as discontinuities which,
together with the continuum, determine the behaviour of the structure. Due to the
discontinuity, the response of the structure becomes dependent on its size. This is why
softening has been termed to be a structural response (Bieniawski [1967], Van Mier
[1984,1986ab]). As a fact, softening cannot be measured as a property of concrete
irrespective of the dimensions. It is a combined response of a local and a continuous
process. The local process is governed by fracture mechanics, which describes the
behaviour of cracks. The structural response can only be analyzed by making assumptions
as to the local and the continuum components of the behaviour and verifying whether
they give the correct response for the total structure. It is important that this is done by
taking the correct boundary conditions into account. It has been found that compressive
softening is highly sensitive to the exact boundary conditions. The boundary conditions
have bedevilled the discussion about softening being a structural property owing to their
dominant and often unexpected role in test results.

Contents of fll.is chapter

Further in this chapter, the broad lines sketched above will be illustrated and put in
perspective to provide a basis for the research presented in this thesis. In Chapter 2.2 the
process of microcrack formation will be discussed in detail. Localization of deformations
due to unstable macrocrack growth will be treated in Chapter 2.3. To obtain a better
understanding of the failure process inside the concrete, the influence of the internal
concrete parameters will be discussed in Chapter 2.4. This will be done too on the
influence of multiaxial loading conditions in Chapter 2.5. Finally, the structural
behaviour of softening concrete and how it is influenced by boundary conditions in
experiments will be discussed in Chapter 2.6. Because of the lack of knowledge of
compressive softening, the knowledge of tensile softening will be used where it is thought
to be useful.

9
2.2 MICROCRACK FORMATION

Influence ofheterogenei.ty
Many years of concrete testing have shown that tensile cracking is the most important
and perhaps the only cause of crack formation in concrete. It seems that other types of
crack formation such as shear cracking, compressive cracking and crushing are
macroscopic phenomena which can be explained by tensile crack formation at the
microlevel. The heterogeneity of concrete disturbs a macroscopically uniform stress field
in such a way that tensile stresses are always found at the microlevel, even "Q.nder uniaxial
and multiaxial compressive loading. Moreover, it has been shown that tensile cracks are
already present in concrete before a load is applied (Hsu et a.l [1963]). The application of
an additional load makes them grow further.

Role of voids
Griffith [1924] was the first to recognize the important role played by voids ~n the failure
process of a material. In order to predict crack growth he calculated the stress field
around an elliptic void or crack (Fig. 2.2.1, see also Carpinteri [1986]). He took the
maximum hoop stress (tension= positive) along the contour of the void at the most
critical angle {3 as the criterion for crack initiation. For a compressive stress field ( u 1~0,
u2~0, u,~u:z) he found for the maximum hoop stress

Umax=------- (2.2.1)
4eo(Ut+u2)

Fig. 2.2.1 Elliptic crack in a biaxial stress field

10
The hoop stress can be many times the macroscopic stress, depending on the factor eo,
which is a function of the size and shape of the void. This factor decreases when the void
increases in size, which means that higher stress levels are found around larger voids. The
formula shows that the maximum stress increases with the difference between the
principal stresses, which indicates some kind of shear failure. On the other hand, the
maximum stress decreases with increasing confinement, which is represented by the sum
of principal stresses in the numerator. Crack growth always occurs in the direction of the
maximum compressive stress. This crack growth is stable. It is slowed down by lateral
compression and stimulated by lateral tension (Horii & Nemat-Nasser [1985,1986]). For
uniaxial compression the critical void orientation is 30o.

Role of aggregate
For normal concrete, the stiff aggregates act as stress concentrators. Internal forces are
mainly transferred from aggregate to aggregate. Fig. 2.2.2 shows this to mean that, in the
case of uniaxial compression, tensile forces must be present in the lateral direction to
balance the lateral components of the compressive forces. At a critical load level these
lateral tensile forces cause the formation of cracks parallel to the direction of compressive
loading, which weakens the concrete. This splitting mechanism is suppressed when a
lateral confining force is present.

!F
Q -tensile cracking

d-++-b
tJ
l
Fig. 2.2.2 Force transfer between aggregates in concrete

Role of bond between aggregate and mortar

Important work, linking the compressive stress-strain relation up to peak stress to the
internal crack-growth process, was carried out at Cornell University. A review of this
research is given by Slate & Hover [1984]. Concrete specimens were loaded to a certain
level of deformation, unloaded and subsequently sawn into slices for inspection of the

11
internal crack patterns. Hsu et al. [1963] used a dyeing technique to make cracks visible
for inspection with a microscope. Slate & Olefski [1963] developed an x-ray technique.
They divided cracks into three types:
1. mortar cracks
2. aggregate cracks
3. bond cracks

unstable growth of combined cracks

formation of combined crocks: mortar

cracks bridging between bond crocks
a. 7 O'peak
growth of bond cracks

0.3 O'peok

preexisting bond cracks

Fig. 2.2.3 Relation between the compressive stress-strain curve

and crack formation (Hsu et. al [1963])

Their research showed that a significant number of cracks already exists prior to loading.
Most of these cracks are bond cracks between mortar and coarse aggregates. They are a
result of volume changes in the cement paste during hydration, bleeding and drying
shrinkage. The interface between the two components has a different structure (Scrivener
& Pratt [1987], Mindess & Shah (eds.) [1988]) and is the weakest link in normal concrete
(Hsu & Slate [1963], Alexander et al. [1968]). Up to 30% of peak stress, the increase in
number and length of cracks is negligible (Fig. 2.2.3). At higher stress levels, new bond
cracks are formed and the existing bond cracks start to grow round the aggregates. At 70
to 90% of peak stress cracks appear through the mortar. The mortar cracks form bridges
between adjacent bond cracks, generally at points of shortest distance between the
aggregate particles. As the stress is further increased, more and more cracks coalesce and
crack growth becomes unstable. Hsu et al. [1963] observe almost no cracks through the
aggregates for normal-strength concrete. This is only found in the case of high-strength
concrete (Carrasquillo et al. [1981bJ). An example of a crack pattern found in a plane

12
perpendicular to the compressive load for normal concrete is given in Fig. 2.2.4a. An
example of a crack pattern in a plane parallel to the direction of the compressive load is
given in Fig. 2.2.4b (Shah & Sankar [1987]). Both crack patterns were recorded just past
peak load, when unstable crack growth appea.rs.

(a) (b)

Fig. 2.2.4 Crack patterns in a plane (a) perpendicular (Hsu et al. [1963]) and (b) parallel
{Shah & Sankar [1987]) to the compressive load

It must be mentioned that the choice of relating the nonlinear behaviour of concrete to
the crack process visible with the naked eye or at slight magnification seems quite
arbitrary. The investigation of concrete by means of a scanning electron microscope shows
that crack densities are a magnitude higher than the densities recorded by the methods
described above (Darwin & Attiogbe [1983]). On the other hand, it seems reasonable to
expect that the most significant effect will be found from the large combined cracks which
influence a much larger zone than small microcracks do. Carrasquillo et al. [1981b]
conclude from their research that the combined cracks are the most important indicators
for concrete failure.

Crack arrest
Concrete is not a perfectly brittle material, but shows some ductility. Crack initiation is
not identical to failure, but stable crack growth can take place. Crack arrest takes place
due to the heterogeneity of concrete. Cracks grow into areas of lower stress or higher
strength and are arrested there until an increase in load makes them grow further. Voids,
aggregates, but also the cracks themselves play an important role in this process of crack
arrest. In normal concrete the cracks are forced to grow round the aggregates, which

13
results in the formation of tortuous cracks, which consumes more energy than the
formation of a straight-line crack.

Failure mechanism
The crack growth in concrete loaded in compression results finally in a continuous crack
pattern, which divides the concrete volume into a number of triangular or conical pieces
which can shear off. This is schematically shown in Fig. 2.2.5. This kind of mechanism
was reported by Van Mier [1984] for both uniaxial and multiaxial compression. A single
shear fracture dividing a slender specimen in two has also been reported (Newman &
Sigvaldason [1965-1966], Kotsovos [1983], Torrenti et al. [1989]).

·- _.,.
·- _.,.
·- _.,.

·-
.,._
_.,.
_.,.
.,._ _.,.

Fig. 2.2.5 Schematized failure mechanisms of concrete loaded in compression

The tensile splitting described above cannot account for the inclined cracks needed for the
formation of the mechanism shown in Fig. 2.2.5. The cracks resulting from t~ose
mechanisms can have a slight inclination to the direction of the maximum compressive
load, but will mainly grow in the direction of this load or otherwise be arrested by this
load. However, often only the axial splitting cracks are reported by researchers, without
questioning the kinematics of the failure mechanism. It can be argued that, after the
concrete is split up into many slender columns, failure will take place due to bending and
buckling of these columns. This failure mode is not likely to be found when lateral
compression is present and cannot account for failure in multiaxial compression. It is
more likely that the shear failure, which is found in multiaxial compression tests, is also
present in uniaxial compression tests, but is modified by the formation of many splitting
cracks. The tests on slender specimens, mentioned above, which fail due to the formation
of one shear crack, indicate that the formation of splitting cracks is not essential to failure
in uniaxial compression.

14
Shear crt.r.d:mg
The above indicates that a shear-crack mechanism is present in concrete compressive
failure. Observations on crack formation in rock (Paterson [1978], Read&; Hegemier
[1985]) show that in this type of material shear cracks are formed on the microlevel. It
seems that the shear crack forms through an array of 11en echelon" splitting cracks, which
becomes unstable at a certain load (Fig. 2.2.6a). n is not unlikely that this kind of
mechanism can also account for concrete compressive failure, because rock and concrete
show many features in common in their failure behaviour. Stroeven [1973] reported the
same arrays of splitting cracks on top of the large aggregates in concrete loaded in
compression at the location where splitting cracks along the aggregate surface grow into
shear cracks through the mortar (Fig. 2.2.6b ). When the density of these splitting cracks
is critical for the state of loading, failure occurs by a complex process of bending, buckling
and crushing. This process is not well known. Horii &; Nemat-Nasser [1985,1986] have
shown by model calculations and model tests on brittle resin specimens that weakening of
a material by splitting cracks can result in shear failure under compressive loading.

)
/)1
tensile
shear area~ S"
I f .I 1 -E--- crack array 1
/,' I ll
//-....--.....
I
I

-
I

(a) (b)

Fig. 2.2.6 (a) Shear loading of an array of splitting cracks in a compression test
and (b) crack pattern reported by Stroeven [1973]

11
Direct 11 shear tests
Recently researchers have directed their attention to the problem of crack behaviour
under shear loading (Bordijk et al. [1989]). Many cracks in structures are initiated due to
tensile loading, but propagate in a field governed by combinations of tensile and shear
loading. Calculations have shown that the modelling of the behaviour of cracks under
shear loading can have a significant influence on the predicted structural behaviour (Rots

15
[1988]). The "direct" shear tests have shown that it is difficult, if not impossible, to create
pure shear loading without introducing any tension into the test specimen. Concrete is so
much more sensitive to tensile failure than to shear failure that tensile failure always
seems to dominate as soon as the slightest tensile stress is present. Finite element analysis
of the tests carried out have shown that they can be simulated well by only using tensile
crack models with crude assumptions for the behaviour under shear loading (Ingraffea &
Panthaki (1985], De Borst [1986], Rots [1988]}. It is now clear that these tests can not be
used to investigate the phenomenon of direct shear failure. They only prove 'that concrete
is many times more sensitive to tensile failure than to shear failure.

2.3 LOCALIZATION OF DEFORMATIONS

Unstable cracl:ing
For increasing deformations or load, the number of cracks and their length increases. The
zones influenced by the cracks grow and start to influence each other significantly. This
results in instability of crack growth and a coalescence of cracks. The process of crack
growth accelerates. This is clearly shown by the crack-detection experiments of
Krishnaswamy [1968]. Other indicators are the increasing nonlinearity of the
load-deformation curve and the significant increase of acoustic emission near peak stress
(Weigler & Klausen [1979], Diederichs et al. [1983]). Soon the complete specimen becomes
unstable, which means that the overall load has to decrease to avoid explosive failure.
Under deformation control, this process can be kept stable because the reduction of
mechanical resistance due to crack formation automatically results in a decrease of the
load transferred by the concrete, recorded as softening. Further in this section it will be
shown that loading concrete in deformation control is not sufficient to measure softening.

Localization of deformati0118
The most important consequence of cracking instability is the localization of' deformations
caused by it. Only in a small zone will cracks grow further and coalesce, concentrating all
further deformations in this zone. All cracks in series with this softening zone will stop
growing and deforming due to the decrease of load caused by the development of the
softening zone (Fig. 2.3.1). Localization of deformations goes hand in hand with
localization of cracking.
Initially, softening was modelled as the behaviour of a continuum by means of stress and
strain, as is done for pre-pea.k behaviour. Then it is called strain softening. Bazant [1976]
showed that softening cannot exist as strain softening in a homogeneous continuum. Then

16
strain softening will always localize in the smallest volume possible, because that results
in the smallest amount of energy for total failure. The energy needed for failure decreases
in proportion to the strain-softening volume while, on the other hand, the work done by
the unloading volume around the softening zone increases (Fig. 2.3.1). The most brittle
failure mode is found for localization of strain softening in a zone with no volume, viz. a
plane, and thus no energy dissipation at all. This means that a homogeneous continuum
cannot show strain softening and fails brittlely at peak stress. Tests show that softening
of concrete exists. This can be attributed to the stabilizing effect of its heterogeneous
structure.

Strain
Unloading< softening

Specimen
behaviour

Fig. 2.3.1 Specimen behaviour due to localization of strain softening

Localization in tension
Localization of deformations in tensile tests on concrete comes close to localization of
deformations in a zero volume (Gopalaratnam & Shah [1985], Cornelissen et al. [1986],
Hordijk [1991]). Therefore, it has become common use to describe tensile cracking by
means of a crack model with no volume (ffillerborg et al. [1976], Elfgren (ed.) [1989]).
The behaviour of this crack is described by a means of a stress-crack-opening relation
including softening. This model will be used to demonstrate the consequences of
localization of deformations for the structural behaviour of a. concrete specimen. In this
example the continuum behaviour will be described as linear elastic.

The consequences of localization at peak stress are shown in Fig. 2.3.2. The main
consequence is that concrete behaviour becomes size dependent. The crack stress-opening
relation of the fictitious crack is given in Fig. 2.3.2a as curve 0. It shows a. gradual
softening with crack opening,which starts when the tensile strength ft is reached. The

17
figure shows further the influence of the measuring length on the shape of the measured
stress-deformation curve. For increasing measuring length, or specimen size, the elastic
deformations of the continuum around the crack become more and more important, both
in the pre-peak region and in the post-peak region. The post-peak curve becomes steeper
due to the increasing influence of volumetric unloading for an increasing measuring
length. Even snap-back behaviour is found. The snap-back behaviour cannot be followed
in deformation controlled tests. This then results in a dynamic snap through as is
indicated by the dotted lines in Fig. 2.3.2a.. To have a stable softening test, short
specimens should be chosen or the test control should be coupled to measuring devices
giving a monotonica.lly increasing signal (Rokugo et al. [1986a], Taerwe [1991], Glavind
and Stang [1991]). Also a stiff loading frame and a fast servocontrol mechanism are
advantageous. In Fig. 2.3.2b the influence of localization of deformations on measured
stress-strain behaviour is given. It is clear that the post-peak behaviour is size
dependent.

(T

a
16d

(a) 8d

Wmaxw
/Irj
(T

(b)

Fig. 2.3.2 Consequences of localization of deformations in a tensile test on

(a.) stress-deformation relation and (b) stress-strain relation

18
Researchers report discontinuous crack growth and crack propagation through the
specimen even under macroscopic uniform loading, resulting in the gradual descending
branch (Van Mier [1986b,1991], Hordijk et al. (1987]). How this influences the localization
of deformations is not yet clear.

Size effect for strength

When macrocrack growth is present in a structure before nominal peak stress is reached,
the above discussed size effect extends to peak stress and the part of pre-peak behaviour
from the start of macrocrack growth onwards. This phenomenon has been observed for
many concrete structures which fail due to tensile crack formation (Bazant [1984,1986]).

Locd~uonincomp~n

In the last few years researchers have started to realize that softening of concrete loaded
in compression is probably also accompanied by localization of deformations like softening
of concrete loaded in tension. Very little is known about this localization.

ajapeok
1.0
0.8

H=100mm
H=200mm
(a)
-2 -4 -6 -8 -10
£[%o]
afapeak

0.6
0.4 H=SOmm
H=100mm
0 2
· H=200mm
00~--~----~--~----~--~~
(b) -0.2 -0.4 -0.6 -0.8 -1.0
W-Wpeok[mm]
Fig. 2.3.3 (a) Stress-strain curves and (b) post-peak stress-deformation curves for
concrete specimens of different height loaded in uniaxial compression
(Van Mier [1984,1986ab])

19
An important indication that localization of deformations is present in uniaxial
compression is found in experimental results ot Van Mier [1984,1986ab]. He carried out
compression tests on specimens of different height and found that the softening depends
on the height of the specimen (Fig. 2.3.3a). On comparing the p_ost-peak stress-
deformation curves, a striking resemblance was found (Fig. 2.3.3b). This indicates that
post-peak behaviour of concrete loaded in uniaxial compression is also governed by a
local fracturing process as found in uniaxial tension.

Others have also carried out tests on specimens of different height for rock (Hudson et
al.[1972]) and concrete (Rokugo & Koyanagi [1992J). However, these researchers did use
rigid loading platens which are known to influence compressive failure significantly,
depending on the shape of the specimen. In these tests the size effect caused by the
influence of the loading platens is much more significant than the size effect caused by
softening. It is very difficult to unravel the influence of both size effects. The brush
platens used by Van Mier were known to influence compressive failure significantly less.

2.4 INFLUENCE OF MATERIAL PARAMETERS

There is not much known about the influence of material parameters on softening of
concrete loaded in compression. Many data can be found about compressive strength, but
data on stress-deformation behaviour including softening are scarce.

Influence. of h.eteroge.ne.itg
In general test results indicate that concrete becomes more brittle when the heterogeneity
decreases. This is when
1. strength and stiffness of hardened cement paste come closer to those properties of
the aggregate, thus when the water-cement ratio decreases or the age of
concrete increases for gravel concrete or when lightweight aggregate is used for
normal-strength concrete.
2. the size of the aggregates decreases. The large aggregates have the most important
influence on the disturbance of a stress field. They influence crack formation
significantly, because they act as stress concentrators and crack arresters. They
force the cracks to go round them, which makes the crack pattern more tortuous
and discontinuous.
3. the bond between aggregate and hardened cement paste is better. This can be

20
 obtained by a higher strength of the cement paste, additions to the cement such as
puzzola.ns, and also by another mineralogical structure of the aggregate as in the
case of limestone.

lnjluenee of strength
The influence of strength, or indirectly the water-cement ratio or the amount of voids, is
found in the experimental results of Wischers (1978] and Carrasquillo et al. [1981a].
Characteristic results obtained by Wischers are shown in Fig. 2.4.1. The brittleness of the
concrete is found to increase with increasing strength. The increase of fracture energy, the
· surface under the stress-strain diagram, lags behind the increase in strength.

(1

--------- J3.§Q

_________ Bli~

_________8~~

_________ 6_3~

_______ 6_2~

______ _Bj;?

0 -2 -4 -6 -8 -10-12
e[%o]

Fig. 2.4.1 Influence of strength on compressive softening (Wischers [1978])

Carrasquillo et al. [1981b] report that the high degree of homogeneity of a high-strength
concrete results in less crack formation of a high-strength concrete than a low-strength
concrete at the same relative stress level(= u/fc). This results in a more linear pre-peak
stress-strain behaviour. The specimen becomes unstable very soon after the first larger
cracks are created. This can be explained by the low redundancy of the homogeneous
structure, which only allows for a small redistribution of forces before the specimen
becomes unstable. The crack planes are found to become steeper and smoother when the
concrete strength increases. For a high-strength concrete, cracks regularly pass through
the aggregates and no longer go round them. This results in a smaller lateral expansion

21
during softening. Very-high-strength concrete even shows snap-back behaviour for
normal-size specimens (Rokugo et al. [1986], Glavind and Stang [1991], Taerwe [1991]).

4:'ensile tests
For tensile testing, some more results are available, which indicate the influence of
material properties on softening of concrete (see Hordijk et al. [1989]). These tests show
that it is difficult to find the unbiased influence of one parameter, because it is often
impossible to change one parameter of concrete mix properties without changing others.
This seems to be the cause of some conflicting test results. For tensile tests a tendency
can be found for fracture energy to increase with an increase of maximum aggregate size
(see also Van Mier [1991b]). Many researchers report that hardened cement! paste itself
acts in a very brittle manner, but almost no proof can be found in literature. This is
probably due to the difficulty of carrying out stable softening tests on hardened cement
paste. Only a rare example can be found in two tensile softening tests by Van Mier
[1991a]. Further, it is found that limestone concrete is more brittle than gravel concrete,
because of the better bond with hardened cement paste and the lower stre~th of
limestone, which allows the cracks to pass through the limestone (Gopalara;tnam & Shah
[1985]). The same is found for lightweight concrete (Cornelissen et al. [1986]).

2.5 INFLUENCE OF A MULTIAXIAL STRESS STATE

Compressive failure of concrete is highly sensitive to the presence of lateral confinement.

A lateral tensile stress speeds up the process of cracking. Many researchers have
investigated the influence of biaxial and triaxial states of stress on stress-strain behaviour
up to peak stress. Examples can be found in the work of Nelissen [1972), Kupfer [1973],
Schickert & Winkler [1977], Linse [1978], Gerstle et al. [1978,1980], Newman [1979] and
Nojiri et. al. [1984]. A lateral compressive stress slows down the process of cracking
(Krishnaswamy [1968]). This can be explained by the fact that the lateral tensile stresses,
introduced on the microlevel by the heterogeneity by a macroscopic compre8sive load are
reduced by the lateral confinement. Also the formation of shear cracks becomes more
difficult. When cracks are present, a confining stress will slow down shearing of the cracks
due to an increase of the frictional stresses and stresses due to a;ggregate interlock.

Biazial strength
Figure 2.5.1 shows the influence of a biaxial stress state on the strength of concrete
(Kupfer [1973]). The presence of a lateral tensile stress has a. significant influence on the

22
compressive strength. The influence of a lateral compressive stress is significantly less.
This can be explained by the fact that in these tests a zero stress is present in the third
direction. And what is more important, the concrete is allowed to fail in this direction.
The lateral compressive stress plays a double role: on the one hand it is preventing failure
in its own direction, on the other, it also stimulates failure in the other lateral direction.
This probably explains the shape of the curve in Fig. 2.5.1.

I I I
I I I
-11'j1.o -o.e -0.6 -0.4 -0.2.
1 -0.2 -
r
0.4 !-

-0.6. -
a,
~-0.8 -

~ ,-:z
I
a2
-
-

Fig 2.5.1 Biaxial failure envelope (Kupfer [1973])

7'riazial stre:n.gth
Figure 2.5.2 gives the influence of a compressive triaxial state of stress on concrete
strength. In Fig. 2.5.2a the strength envelope is shown in the Rendulic plane, which is the
plane for equal lateral stresses. The influence of the lateral confinement is clearly shown
by the significant increase of strength with increasing lateral compression. No compressive
failure is found in the neighbourhood of the hydrostatic axis (Bazant et al. [1986]), which
is the axis with equal triaxial stresses. The strength envelope is positioned at a certain
distance of the hydrostatic axis. This is also shown in Fig. 2.5.2b, which gives the
strength envelope in the deviatoric planes, which are perpendicular to the hydrostatic
axis. It infers a shear failure mechanism. The maximum shear stress is proportional to the
difference in principal stresses. The increase in size of the strength envelope again points
to the influence of the confining pressure. The failure mode in these tests is a typical shear
failure. Kupfer [1973], Van Mier [1984] and Jamet et al. [1984] report shear cracks with an
inclination of 20o to aoo to the direction of the maximum compressive stress. The strength
envelope is asymmetric to the hydrostatic axis. Its distance to the hydrostatic axis is

23
greater when one large driving compressive stress is balanced by two smaller confining
stresses (compressive meridian) than when two compressive driving stresses are balanced
by one smaller balancing stress (tensile meridian).

,_ • Compresslve
...,16'' meridian

Tensile
"'meridian

(a) (b)

Fig. 2.5.2. Triaxial failure envelope in (a) the Rendulic plane

and (b) the deviatoric plane (Nojiri et al. [1984])

Softening
A multiaxial state of stress influences softening as significantly as it influences strength.
Besides its influence on the crack-growth process, it also has a great influence on the
behaviour of the already cracked concrete. A limited number of tests has beenI carried out
to investigate softening of concrete under multiaxial loading conditions. Some researchers
have carried out compressive softening tests by means of standard cylinder tests in which
the axial stress is applied in deformation control and the lateral confining stress by means
of a fluid pressure through a flexible membrane (Jamet et. al. [1984], Willam .et al. [1985],
Smith et. al. [1989]}. The influence of a constant lateral confining pressure is shown in
Fig. 2.5.3. Concrete behaviour becomes more ductile when the confining pressure
increases. It is remarkable how sensitive softening is to a slight confining pressure. It
seems that the softening disappears at higher confining pressures. This is often referred to
as the brittle-ductile transition. Jamet et al. [1984] report that the failure mode then
changes completely. At low confining pressures, resulting in softening, localized shear
cracks are found. At high confining pressures, not resulting in softening, they report that
no visible sign of mechanical damage is present. The residual stress level, observed in Fig.

24
2.5.3, indicates that, besides crack formation, an important role in post-peak behaviour is
played by aggregate interlock and friction. Shear cracks have a significant mechanical
resistance when a confining stress is present {Walraven [1980], Bazant & Gambarova
[1984]).

a,- 0'3
[N/mm 2] a3 =-100 N/mm 2
-150
0'3 = -50 N/mm 2

-100

-so

~---a:,= -3 N/mm 2
~= 0 N/mm 2
o~~--~------~~
0 -25 -50 -75 -100
E1[%o]

Fig. 2.5.3 Influence of confining pressure on concrete behaviour (Jamet et. al. [1984])

u3 = -3N/mm 2
50 o:, =-10 N/mm 2
0'3 = -25 N/mm
2

25 u3 =-50 N/mm 2

OJ =-100 N/mm 2
0
0 -2s -so -75 -1 oo
s,[%o]
Fig. 2.5.4 Lateral deformations as influenced by the confining pressure
(Jamet et al. [1984])

The lateral deformations decrease for an increasing confining pressure (Fig. 2.5.4). It was
shown in Fig. 2.2.5 that the ability of lateral expansion is of vital importance for the
creation of a failure mechanism. When lateral deformations are prevented completely no

25
failure can take place. Bazant et al. [1986] carried out tests on cylinders which were
completely locked up in the lateral direction. Initially the specimens showed a decreasing
stiffness, probably due to the collapse of the pore system. Due to closure of the pores, the
resistance to further deformations again increased and a loading of 2000 N /mm2 was
reached without an indication of failure.

True triax:ial softening tests, with independent stress-deformation regimes in three

directions, have been carried out by Van Mier [1984,1986ab]. His test results show that
two failure modes can be found in triax:ial compression. A distributed fractUfe is found in
regions near the compressive meridian, which includes uniax:ial compression. In those tests
equal deformations are allowed in the two lateral directions. This results in the formation
of crossing shear cracks without a preferential direction. This is a more ductile failure
mode. A clear localization of deformations is found when a preferential direction of failure
is present. This happens in the case of different lateral confining stresses or in
plane-strain tests in which deformations in one direction is prevented completely. Then
the failure mode is more brittle.

2.6 STRUCTURAL BEHAVIOUR AND BOUNDARY CONDITIONS

The measured softening in a test is due to structural behaviour of a specimen. This is

obvious when we observe localization of deformations. The reaction of a specimen to
loading conditions is a combined reaction of local and continuum components, which
makes it dependent on the size of the specimen. According to the theory of the
homogeneous continuum, concrete behaviour is initially determined by the continuum. It
is now important to know where, when and how the continuum behaviour is disturbed by
the local behaviour. It has been found that this process of localization is very sensitive to
the boundary conditions. This has complicated the discussion about localization and
concrete softening being a structural property. The structural behaviour of sdftening
concrete as influenced by the boundary conditions in tests will be discussed here.

Methods of load application

Testing concrete in order to determine material properties is often done by measuring the
behaviour of a specimen which is as homogeneous as possible, under loading conditions
which are as uniform as possible. Two extreme choices can be made to load a. specimen,
1. to prescribe the boundary stress of a specimen by means of a flexible loading
platen (Fig. 2.6.la) and

26
 2. to prescribe the boundary displacement by means of a rigid loading platen (Fig.
2.6.1b).

w~constant u~constant

(a) (b)

Fig. 2.6.1 Loading of a specimen by means of (a) a flexible loading platen

and (b) a rigid loading platen

Due to the heterogeneity of concrete, a uniform boundary stress results in a nonuniform

distribution of boundary displacement& and a uniform boundary displacement results in a
nonuniform distribution of boundary stresses. In the linear-elastic range the average
response to both methods of load application is not essentially different and both methods
result in the same stress-strain relation when the specimen is large enough.

I
11
11
11

(a) (b)

Fig 2.6.2 (a) Influence of lateral restraint and (b) loading by means of a steel brush

Lateral dejomuJtions
Lateral deformations of loading platen and specimen are often not the same. When a rigid
loading platen is used, this results in shear forces restraining the lateral expansion of the

27
specimen. It is well known that these shear forces cause triaxial compressive stress fields
in the specimen near the loading platens {Fig 2.6.2a), which makes the specimen stronger
and more ductile. This method of load application results in the well known hour-glass
failure mode. In uniaxial tests the influence of the lateral boundary restraint on the
compressive strength can be minimized by increasing the specimen length (Schickert
[1980]).

Reduction of 1.o.Ural rutrain.t

Other loading platens have been developed to reduce the lateral restraint of the boundary
displacement& by the loading platen. In particular multiaxial compressive testing, where
the specimen height cannot be increased, has stimulated this development (Newman
[1979]. Gerstle et al. [1980a}). Measures that were taken are
1. friction reducing measures, which limit the magnitude of the shear stresses, by
means of polishing the loading platen surface, adding grease or talc powder and
using interlayers with low coefficient of friction (Mills & Zimmerman [1970],
Launay & Gachon [1971), Bertacchi & Bellotti [1972], Kobayashi & Koyanagi
[1972], Erdei [1980], Nojiri et al. [1984]).
2. reduction of the lateral stiffness of the loading platen by using brush platens to
load the specimen as first developed by Hilsdorf [1965) and later applied by
Nelissen [1972], Liu et al. [1972], Kupfer [1973], Linse [1978] and Van Mier
[1984]).
3. the use of flexible loading platens. Schickert [1980] developed a flexible loading
platen consisting of steel rods fixed on a sheet of rubber, which guarantees uniform
stresses along the boundaries of the specimen. Another example is the application
of fluid pressure through a flexible membrane. This is done to apply the lateral
pressure in standard triaxial tests on cylinders, but has also been applied in a
cubical triaxial cell (Andenaes et al. [1977], Sture & Desai [1979]).

Injluence of lateral rutrain.t

Kotsovos [1983] carried out a number of uniaxial compression tests on cylinders with a
height-width ratio of 2.5, varying the lateral restraint of the loaded boundaries. He found
that although peak stress was not influenced by the boundary restraint, the softening
differed significantly. Fig. 2.6.3 shows that post-peak behaviour of two types of concrete
becomes more brittle when the boundary restraint decreases. Kotsovos assumes that
cracks are initiated in the central uniaxially stressed zone of the specimen and that their
propagation is delayed by the triaxial compressive stress state near the loading platens.
Based upon these test results, Kotsovos comes to the conclusion that the compressive
softening found in tests is probably almost completely due to the boundary restraint and

28
that it would be most realistic to describe concrete post-peak behaviour by a complete
and immediate loss of load-carrying capacity. It seems that this is an extreme standpoint
in view of the results for the low-strength concrete. Besides, there are indications that
the concrete acted more brittlely due to the introduction of splitting stresses by the soft
interlayers. However, it was shown by these tests that compressive softening is highly
sensitive to the lateral boundary conditions created by the loading platens.

F/Fpeak F/Fpeak
1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2

00 0.5 1.0 1.5 2.0 2.5 00 0.5 1.0 1.5

w/wpeak w/wpeak

(a) {b)
Fig. 2.6.3 Influence oflateral restraint on softening of (a) a low-strength and {b) a
normal-strength concrete: {1) active restraint, {2) plain steel platen, {3) MGA pad used
before, {4) brush platen, {5) rubber layer, {6) MGA pad not used before (Kotsovos [1983])

Nonuniform deformatiom
Flexible loading platens can only be used to load concrete up to peak stress. Then loading
with a flexible loading platen becomes unstable. By using rigid loading platens, it is
assumed that deformations of a specimen are under control and that softening can be
measured properly. However, highly nonuniform opening of the crack in unia:xial tensile
tests has been found {Reinhardt [1984], Van Mier [1986a], Hordijk et al. [1987]). Hordijk
et al. (Hordijk et al. [1987], Hordijk & Reinhardt [1989] showed that, although the
boundaries of a specimen are subjected to a uniform displacement, this does not have to
mean that the specimen deforms uniformly over any cross-section parallel to the loaded
boundaries during softening. They showed that this phenomenon can explain the irregular
shape of the softening curve found in these tests (Fig. 2.6.4a). Assume that a specimen is
loaded uniformly up to peak stress {Fig. 2.6.4b ). A rotation of the boundaries of the
softening zone results in unloading at one side and softening at the other. This causes a
drop in the nominal stress, which is preferred by the specimen. A moment opposing this
rotation is built up due to the rotational stiffness of the specimen outside the softening
zone. This results further in the softening process in a neutralization of the rotation
causing the second drop in the nominal stress-deformation relation.

29
 Front

Rear

-20 20 60 100 140

w(JLm)
-Nonuniform
deformations (test)
3.0 ····Uniform
2.0 deformations

1.0
0.0 [..__......._:::::::=:::---~
0 40 80 120w ( JLm )
(a) (b)

Fig 2.6.4 Influence of nonuniform deformations on softening (Hordijk et al. [1987])

Hinged loading pl4tem

Hinged loading platens have been used in compressive softening tests (Kotsovos [1983],
Van Mier [1984], Torrenti et al. [1989]). They are used because it is easier to load a
specimen uniformly by means of a hinged loading platen than by means of a fixed loading
platen. However, they do not resist the nonuniform deformations discussed above, where
it was assumed that the loaded boundaries were completely fixed. It is not known how
significant this influences compressive softening behaviour.

Where is localization found 1

Loca.lization of deformations in one shear crack seems to be stimulated by the use of a
slender specimen and a hinged loading platen (Fig. 2.6.5a, Newman and Sigvaldason
[1965-1966], Kotsovos [1983], Van Mier [1984], Torrenti et al. [1989]). Newman and
Sigvalda.son showed that the use of one hinged loading platen can change a more
distributed failure mode into a failure mode with one shear crack. Van Mier (1984] found
that the increase of the height of the specimen increases the chance of finding a failure
mode with one shear crack. The tests by Kotsovos [1983] show that the incli~ation of the
shear crack decreases when the lateral boundary restraint decreases. This shows that the
macrocrack pattern, determining the failure mode, depends highly on the test setup.

30
When concrete specimens are loaded uniformly with fixed loading platens which cause a
minimal lateral restraint to the concrete deformations, then a more distributed type of
failure is found which contains more vertical splitting cracks (Fig. 2.6.5b, Schickert
[1980], Van Mier [1984], Shah & Sankar [1987]). This distributed nature of the cracking
has caused the lack of agreement about localization of deformations for uniaxial
compressive softening.

(a) (b)

Fig 2.6.5 Failure modes found in uniaxial compression

When is localization of tleformatiou found?

Generally, it is assumed that localization starts around peak stress. Cracks larger than
the maximum aggregate size are already found before peak stress is reached (Hsu et al.
[1963], Carrasquillo et al. [1981b]). Shah & Sankar [1987] state that localization starts
approximately at peak stress, which is shown by the formation of continuous cracks at the
edges of the specimen (Fig. 2.2.4b ). However, they conclude that further in the post-peak
regime cracking is more distributed again. It is more often reported that cracking starts
from the edges of the specimen growing inwards (Van Mier [1984], Read & Hegemier
[1984]) and that softening is the mechanical resistance of a constantly reducing, more or
less intact core of the specimen. It is not known whether this is due to the restraint to
lateral deformations by the loading platens, shrinkage stresses, a wall effect as a result of
casting or the discontinuity found at the edge of the specimen.

A sophisticated method which can be used for recording the localization of deformations
during a test is stereophotogrammetry. This technique is regularly used in aerial mapping
and has recently been applied by Torrenti et al. [1989] to follow the displacements of
points on the surface of a concrete specimen loaded in uniaxial compression. Figure 2.6.6
shows an example of the deformations at several stages of loading which reveal the growth

31
of a localized shear crack. The size of the squares in this figure is a measure of the local
deformations. There is a.n indication that the deformations already localize in the
middle-left part of the specimen before peak stress is reached. The discontinuity in the
displacement field expands gradually from left to right. One must be cautious when
interpreting these results, because the shear crack formation a.nd growth have certainly
been influenced by the lateral restraint of deformations by the loading platens and the
rotation of the loading platen during softening.

F"(kN)
-80

-60
-40

-20 8
•photograph
0
0 -0.2 -0.4 -0.6 -0.8 -1 .0
w(mm)

. ..
~ '

:~' .·
· •. c;:, .:~o o-
,. • c c • . . . .
~

• • • • 0
2 • • • •

0-5
Fig. 2.6.6 Deformations at several stages of a uniaxial compression test
(Torrenti et al. [1989])

32
 CHAPTER3

TESTING TECHNIQUE

3.1 LOADING TECHNIQUE

Loading apparatus
Both uniaxial and triaxial compression tests have been carried out with the triaxial
apparatus at Eindhoven University of Technology (Fig. 3.1.1, Van Mier [1984], Vonk
[1989]). This apparatus consists of three identical loading frames with a capacity of 2000
.kN each. One loading frame is shown in Fig. 3.1.2a. Each of the three frames is provided
with its own hydraulic servomechanism, which can be used to load a specimen either in
force control or in deformation control. For this purpose the load cell or the LVDTs
attached to the loading platens are used to supply a feedback signal. The electric
regulation circuit enables a constant force or deformation ratio to be maintained between
the different axes of loading during a test. All the tests have been carried out in
deformation control to be able to measure softening. A constant strain rate of 1x1o-s 1/s
was used. In the triaxial tests, the main loading axis was used in deformation control,
while the other axes were used in force control, prescribing for each a constant load ratio
with respect to the main loading axis.

Fig 3.1.1 Triaxialloading apparatus

33
 (j) Concrete specimen
® Steel brushes
@ LVDTs
@ Hinge

® Load cell

® Hydraulic actuator

84 Rods
(5x5mm2 )

(a) (b)

Fig. 3.1.2 (a) Single loading frame ofthe triaxial apparatus and (b) short brush

34
Lotuling plateM
Five different loading platens have been used to investigate their influence on compressive
softening (Vonk [1989]):
1. Short brush (BS)
2. Long brush (BL)
3. Tefl.on platen (T)
4. Dry platen (D)
5. Hinged tefl.on platen (HT)

·Hinged loading platens are often used because they facilitate a good contact between
loading platen and specimen when the surfaces of the specimen and the loading platen are
not parallel. In the tests presented here, the hinge in the loading frame (No. 4 in Fig.
3.1.2) was set free when the loading platen was placed against the specimen. During
loading, the hinge was fixed. The hinged tefl.on platen was supplied with another hinge,
because the hinge in the loading frame did not function optimally.

Steelbru.V&u
The short brush (Fig. 3.1.2b) is the same as that used by Van Mier [1984] in his triaxial
tests. The brush rods (5x5 mm2) are separated at the clamped side by means of a grid
consisting of 0.2-mm-thick phosphor bronze strips. The rod packet is clamped to a solid
steel block. The effective length of the rods is 84 mm. Steel brushes are used in
compression tests, because their restraint to the lateral expansion of the concrete
specimen is slight (Hilsdorf [1965], Linse [1978]). The brush rods can easily follow the
lateral expansion of the specimen by bending. Peak stress in a compression test is not
influenced significantly by steel brushes (Van Mier [1984]). The shear forces induced in
the specimen can be minimized by increasing the length or decreasing the width of the
brush rods. This is limited by the buckling load of the rods.

Calculation of ahear 6tresaes

The shear stress induced in the specimen and the buckling stress of the rod can be
estimated with simple mechanical formulas. However, it is not clear what boundary
conditions are present for the brushes used in the present experiments. Most researchers
assume a complete fixation at the clamped side of the rod (Hilsdorf [1965], Linse [1978]).
Whether this is true can be queried when inspecting the construction of the brush in Fig.
3.1.2b. The phosphor-bronze grid and the clamping system are relatively weak. At the
rod end in contact with the concrete specimen a state between free rotation and a perfect
fixation can be assumed. Due to an eccentricity of the force in the rod, a moment can be
transferred by the brush-rod top. However, the stresses in the contacting area will

35
become significant, which will cause a penetration of the rotating brush rod in the
concrete surface and limit the moment tra.nsfeJred by the brush-rod top ..

The aim was to measure the shear stress as a function of the lateral displacement of the
brush-rod top. The tests showed that the initial stiffness lies somewhere between the
theoretical value for a brush rod with one side fixed and one side hinged and the value for
a brush rod with two sides fixed. The behaviour was found to be highly nonlinear and to
be dependent on the preparation of the brush and the loading history in former tests. The
use of a brush in compression tests on concrete resulted always in a certain amount of
play in the clamping system due to the considerable lateral deformations of the specimens
and consequently of the brush-rod tops.

To give an indication of the shear stresses induced in the concrete by the steel brushes,
some values have been calculated for a zero normal stress. They are given in Table. 3.1.3.
The shear stress is proportional to the lateral displacement w of the brush-rod top. Also,
the buckling stress is given, as the shear stress decreases gradually to zero when the
normal stress increases from zero to the buckling stress. It has been assumed that one end
of the rods is completely fixed. For the other end, two cases have been assumed: a free
rotation and a complete fixation. The actual state in the tests will be somewhere in
between.

If shear
(N/m2} Ifbuckling
(N/m2)

end condition rotating fixed rotating fixed

BS 1::: 84 IRII 2.00 w 7.99 w 138 552
BL 1:::119 IRII 0.70 w 2. 81 1i 69 275

Table 3.1.3 Calculated values for shear stress (as a function of the lateral displacement w
of the rod top) and buckling stress for the steel brushes

The shear stresses increase when the lateral deformations of a specimen increase.
Compressive softening of concrete is accompanied by an acceleration of the lateral
expansion. This means that the lateral shear stresses will increase significantly and that it
is not certain whether the influence of the brushes is still as negligible as it was in the
behaviour up to peak stress.

In order to estimate the influence of the short brush on compressive softening, it was
decided to carry out tests not only with the brush used by Van Mier but also with a brush

36
with longer rods. Table 3.1.3 shows that the long brush, with an effective rod length of
119 mm, gives a significantly less lateral restraint. The shear stress has been reduced to
35% and the buckling stress to 50%. It was decided to reduce the lateral restraint by
increasing the rod length, because this influences the buckling stress relatively less than
reducing the lateral restraint by decreasing the cross-t~eetion of the rod. It was thought
dangerous to exceed the buckling stress in the tests. However, in the triaxial tests, axial
stresses were reached significantly exceeding the buckling stress of 69 Nfmm2 in Table
3.1.3 calculated for the rotating rod top. This indicates that the assumption of a free
rotation of the rod top is probably not correct.

Frictitnull8trusu
The restraint to lateral deformations due to friction is quite different from that due to the
bending of brush rods. Frictional stresses rise quickly as a function of the shear
deformation but are limited by a. maximum which depends on the frictional properties of
the surfaces of the loading platen and the specimen and the normal stress in their contact.
Frictional stresses increase with an increase of the normal stress. This means that the
restraining frictional stresses in a compressive softening test are maximal at peak stress
and decrease with softening. The restraining shear stresses due to the bending of the
brush rods grow more slowly but continuously as a function of the lateral deformations.
This means that, when softening takes place and the lateral expansion accelerates, the
lateral,restraint due to the bending of brush rods will become greater than that due to
friction. It was therefore decided to develop loading platens with a low coefficient of
friction to investigate the effect on compressive softening of the platen and compare it to
the effect of steel brushes.

Rei.tu:f.itm oI frictio'n.
As found in literature (for instance, see Dorinson and Ludema [19851) the most common
measures to reduce the frictional properties of a surface are:
1. polishing the surface
2. adding layers which reduce the frictional properties
3. lubricating the surface

In order to reduce the friction of the loading platens, talc powder and aluminium sheets
were used by Launay & Ga.chon [1970], teflon by Mills & Zimmerman [1970), rubber by
Kobayashi & Koyanagi [1972), polyethylene and molybdenumdisulphide by Bertacchi &
Belotti [1972], polished steel sheets by Erdei [1980), rubber or teflon and molybdenum-
disulphide by Kotsovos [1983], teOon and grease by Nojiri et al. [1984] and polyacetal and
teflon by Shah & Sankar [1987].

37
Tejltm. platen.
A loading platen with a very low coefficient of friction was developed for this research
(Vonk [1989]). The surface of a loading platen was first tempered and then polished. This
resulted in a surface roughness Ra ~ 0.05 fiJll. Further, an additional sheet of teflon was
applied. Teflon was chosen because it is the best solid lubricant. It is known that the use
of intermediate layers of a soft material can make it impossible to carry out a stable
softening test. Besides, there is the danger of introducing splitting stresses into the
concrete specimen when using a soft intermediate layer (Kotsovos (1983]). To avoid these
drawbacks, a very thin sheet of teflon 0.05-mm-thi~ was used. Further, it was found
profitable to slightly grease the interface of loading platen and teflon. Only very little
grease was added. The effect of several kinds of grease was tested. The lowest coefficient
of friction was found for an ordinary bearing grease (Molykote BR2 plus).

0.03

6
+'
0 0.02
E
...0 ll1
i
:!:!
l
(.)
0.01
v--
0.00 L - - - 1 - - - - ' - - - - ' - - - - ' - - - - ' - - - - - - '
0.00 0.25. 0.50 0.75 1.00 1.25 1.50

Distance of sliding [rrrn)

Fig. 3.1.4 Result of tests to determine the friction of a teflon platen

Mt.GSUremem of frietio'n.al strusu

A series of tests was carried out to determine the frictional properties of the teflon platen
(Vonk [1989]). A concrete cube was loaded with a constant for~ in the vertical axis of the
triaxial apparatus using two teflon platens. One of the horizontal axes was used to make
the concrete cube slide between the two teflon platens. The slidi~ speed was about 1
,.an/s, which is representative for the sliding speed in the cc;>mp~sion tests on concrete.
;
This is important because teflon shows some creep, which inflqences the measured
friction. Fig 3.1.4 shows a characteristic experimental result. Only a very small
displacement is enough to activate the shear stresses. The peak value is already found for
a displacement of few microns. Subsequently, a significant stick-slip behaviour takes

38
place. This behaviour is known to be strongly influenced by the characteristics of the
loading apparatus (Bartenev and Lavrentev (1981]}. If a test setup of greater stiffness had
been used, the stick-tJlip behaviour would have been smaller and a more gradual curve
found. The shape of the curve during the stick-tllip behaviour is not a material property,
but a property of the testing technique. It is expected that this stick-tllip behaviour will
be less pronounced in compression tests on concrete because in those tests only very small
parts of the surface will slip at one time over a short distance. The decrease in friction in
Fig. 3.1.4 is considerably, which is probably due to the self-lubricating effect of teflon
(Bartenev and Lavrentev [1981]).

020
....

-~
....
(J
0.15
...
.
. .... ..... cry platen
first peak
;t: •·········• ....... .
.... ··•· cry platen
0 w=1.5rrm
0.10
1: ··+ · tefl. platen
~
....
first peak

8 0.05 •.. ... .... ....... .. ··• · tefl. platen

w=1.5rrm

•·········•·········•
0.00
0 -10 -20 -30 -40

Fig. 3.1.5 Coefficients of friction for the teflon platen and the dry platen

Figure 3.1.5 shows the coefficients of friction found in the tests with the teflon platens.
The values vary from 0.026 to 0.048 for the first peak and from 0.008 to 0.013 after 1.5
mm of sliding. The normal stress was limited to -30 N/mm2 because of the danger of
failure of the concrete cube. No aluminium or steel specimens can be used instead of
concrete specimens because it was found that they do not provide reliable results. Those
materials have a smoother surface, which infiuences the measurements of the friction
significantly. In tests with a steel cube, coefficients of friction were found which were 50%
lower than those found in the tests with concrete cubes (Vonk [1989]). The coefficients of
friction are low compared to values mentioned by other researchers. Shah & Sankar [1987)
report a coefficient of friction of 0.07 for tefion. Nojiri et al. [1984) also carried out friction
tests. They report a coefficient of friction varying from 0.025 to 0.03 for a loading platen
with two tefl.on sheets with some grease in between. The low values in the tests presented

39
here are due to the great care given to polishing the surface of the loading platen, the
addition of the grease and the low sliding speed, which allows for some creep.

Hinged tajlon platen and dry platen

The hinged teflon platen was identical to the teflon platen except for the hinge (Vonk
[1989]). The dry platen was identical to the polished steel platen used for the teflon
platen. Fig. 3.1.5 also gives coefficients of friction for the dry platen. This platen also
shows a stick-slip behaviour. The coefficients of friction decrease from 0.138 to 0.180 for
the first peak to 0.128 to 0.132 after 1.5 mm of sliding. Values reported in literature are
scarce. Nojiri et al. [1984] report a coefficient of friction of 0.18. It is expected that the
coefficient of friction of untreated steel platens used by other researchers is greater than
that of the dry platens used in this research, because these last platens were polished.

3.2 MEASURING TECHNIQUE

Datalogger
A computer-controlled datalogger (Peekel Autolog 2005) was available to record
automatically all measurements at a constant time interval.

Measurement offorces
The forces were measured by means of calibrated load cells (Fig. 3.1.2a). These load cells,
with a range of 2000 kN, consist of dumbbell-shaped cylinders upon which four strain
gauges in a full Wheatstone-bridge configuration are mounted. The load cells were
calibrated regularly taking the range of load in a test into account.

MetJSUrement of azial dejorma.tions

Axial deformations were measured by means of strain gauges and LVDTs. The strain
gauges were glued to the four sides of the specimens as shown in Fig. 3.2.1. Their length
was 30, 60 and 120 mm (TML, PL-30/60/120-11, maximum s.train 2%) for the specimen
heights of 50, 100 and 200 mm, respectively. Further, at each side of the specimen an
LVDT (HBM, W10-K, range 10mm) was attached to the loadi~ platens. Fig. 3.2.1
shows how the LVDTs were attached to the steel brushes and th}! teflon platens. The
separate readings of the 4 strain gauges and 4 LVDTs were rec~ded for information on
the uniformity of deformations of a specimen during a test. Two extra LVDTs were
always present to be used for the test control {Fig. 3.2.1b}.

40
 Strain
gauge

LVDT
Strain
gauge
LVDT
LVDT -LVDT test control
LVDT •
•
LVDT
test control
Strain
LVDT gouge
(a) (b)

Fig. 3.2.1 Positions of strain gauges and LVDTs

Defomw.tior&s of a steel '6nsh

The LVDT measurements include deformations of part of the two loading platens and
have to be corrected to find the deformations of the concrete specimen. Therefore the
loading platens were taken out of the loading apparatus to measure their deformations
included in the LVDT measurements. Fig. 3.2.2 shows four different measurements of
brush deformations. First an initial setting is found, which depends on the accuracy of
preparation of the brush and the loading history in tests. It was found that the brush
structure is not very rigid and that use in a test resulted in some play in the clamping
system. Tests 2, 3 and 4 in Fig. 3.2.2 were carried out in sequence. Test 3 was carried out
on the same brush directly after test 2. A much smaller initial setting is found due to the
axial loading in test 2. Test 4 was carried out after the same brush was used once in a
compression test on concrete. This resulted again in a significant increase of the initial
setting. During the test series on concrete it was observed that the clamping system
became gradually weaker. Therefore the brushes had to be repaired regularly, which is a
laborious task. The second part of the curve in Fig. 3.2.2 is linear and not dependent on
the loading history of the brush. It represents the elastic properties of the brush and

41
unloading was found to coincide with loading for this part. Regression analysis showed a
good correlation between the different curves for this second part, which could be used for
the correction of the deformations measured by the LVDTs in the compression tests on
concrete after the initial setting had been eliminated.

-200

-150
1].
6 -100
~
l!
~ -50

oL-----L-----~----~----~----~----~
0 -200 -400 -600 -800 -1000 -1200

Force [kN)

Fig. 3.2.2 Deformations of a short brush

Deformo.tions of s tejlon pl&ten

Fig. 3.2.3 shows the results for 5 tests on te:Oon platens up to a stress of -100 N/mm2,
which is -1000 kN. In Fig. 3.2.3, the initial setting has been eliminated to make
comparison of the curves easier. The elimination of the initial setting will be1 discussed in
the next paragraph. Both loading and unloading are nonlinear and hysteresis is present.
The loading speed in these tests was chosen equal to the loading speed found in the
concrete tests. This is important because the deformations of the teflon platens probably
incorporate some creep. In one of the tests the te:Oon platen was preloaded with 50 kN for
5 minutes, because this occurred often during the preparation of a test. This influences
only the initial setting, but not the further deformations. The influence of a significant
nonflatness of the specimen was tested in two tests by placing a piece of brass foil
(50x50x0.025 mm) under the te:Oon platen during the test. The nonflatness of the test
specimen caused some concern, because it was of the same order pf magnitude as the
thickness of the sheet of te:Oon. Again, a significant influence on ~he initial setting was
found, but no appreciable effects in the further deformations. M can be seen in Fig. 3.2.3,
the curves show only a slight scatter after the initial setting has been eliminated.
Regression analysis was carried out to describe the necessary deformation corrections
using a first and a third-order polynomial for loading and unloading, respectively (drawn

42
lines in Fig. 3.2.3). Additional tests up to -450 kN were carried out to study the influence .
of the peak load on the unloading behaviour of the te:Oon platens. Fig. 3.2.4 shows that a
satisfactory prediction of unloading is given when the hysteresis is reduced proportionally
to the maximum load.

-40

1! -30
.6
-20

J -10

-200 -400 -600 -BOO -1000 -1200

Force [kN]

Fig. 3.2.3 Deformations of a te:Oon platen up to -1000 kN

-25.------------------------------------,

-20

1! -15

1 -10

-5 Predicted from -1000 kN tests

0~------~----~------~------~------~
0 -200 -300 -400 -500

Force [kN]

Fig 3.2.4 Deformations of a te:Oon platen up to -450 kN

Oorrecf.itm of iaiticl 1etting

The LVDT measurements had to be corrected not only for the deformations of the loading
platen, but also for extra deformations due to the setting of the loading platen against the

43
specimen (see also Schickert [1980], Van Mier [1984]). This setting is due to nonflatness of
the specimen surface and nonparallelism of loading platen surface and specimen surface.
To carry out the correction, a parabola was used, fitted to the force-deformation curve
between the point with the maximum tangent modulus and the point for a stress of -30
N/mm2. This is illustrated in Fig. 3.2.5.

Force oooo
•• 0
0 0

./ a=-30N/mm2

/E
"( t.max

.l Parabola
Deformation

Fig. 3.2.5 Correction ofinitial setting found in the LVDT measurements

It is clear that the initial deformations determined after correction of the initial setting
and by subtracting the loading-platen deformations from the total deformations are not
the most accurate. It is assumed that the strain-gauge ,measurements result! in more
accurate pre-peak deformations. The errors introduced by the correction procedure will
become relatively smaller when the deformations increase, that is during softening.

t
Strain gauges

~----------l

I I
Specimen

Fig. 3.2.6 Clip gauge attached to a spe~en

Measurement of lateral deformations

The lateral deformations were measured by means of clip gauges. An example is shown in

44
Fig. 3.2.6. The clip gauges were clamped along aJ1 four sides of a. specimen. Often two or
three clip gauges were used at one side at different heights of the specimen to investigate
the uniformity of deformations of the specimen. The deformations of the clip gauges were
measured by means of two strain gauges in a half Wheatstone-bridge configuration glued
to the thin steel strip in the middle of the clip gauge. Ca.libration of the clip gauges
showed that there is a slightly nonlinear relation between the displacement of the two
contact points and the bending of the steel strip. This was taken into account. The clip
gauges were used over a range of 6, 12 and 24 mm for the 50, 100 and 200-mm-wide
specimens, respectively.

3.3 PREPARATION OF SPECIMENS

Co.sting and hardening

The concrete mix proportions are given in Table. 3.3.1.

Portland c 340 kg/m3

Vater 165 kg/m3 170 kg/m3
Gravel/Sand 1879 kg/m3 1828 kg/m3
0.0 - 0.25mm 8 1. 8 7.
0.25- 0.5 mm 12 7. 10 7.
0.5 - 1.0 mm 12 7. 13 7.
1.0 - 2.0 mm 10 7. 16 7.
2.0 - 4.0 mm 14 7. 23 7.
4.0 - 8.0 mm 20 7. 30 7.
8.0 -16.0 mm

Table 3.3.1 Concrete mix proportions

The concrete was cast in large moulds on a vibrating table (5000Hz). For the concrete
with the sma.llest maximum aggregate size (8 mm), it was found necessary to use in

45
addition an extra vibration needle with a higher frequency (15000 Hz) to obtain good
compaction. Subsequently, the concrete in the .moulds was placed in the laboratory for
three days, keeping it wet and covered with plastic to prevent it from drying out. After
demoulding, the concrete blocks were immersed in water.

&iwir&g
At an age of 28 days, the blocks were taken out of the water to be sawn to the required
specimen size plus 3 mm. At least 1.5 times the maximum aggregate size was removed
from the outer layers of the blocks to obtain specimens which are as homogeneous as
possible. Due to a wall effect, the outer layers of cast concrete have a different structure,
which influences the mechanical behaviour (Schickert [1980], Van Mier [1984]).

Directior& of castir&g
The direction of casting was marked on the specimens in order to be able to apply the
loading in tests parallel to the direction of casting. Concrete is an anisotropic material
due to the process of casting. Significant differences in test results can be found for
loading parallel and perpendicular to the direction of casting (Hughes and Ash [1970],
Van Mier [1984]).

Grindi.r&g
Then the specimens were ground to their required size. Special attention was paid to the
flatness and parallelism of the loading surfaces. This is necessary for uniform loading. In
the first series the deviation of the ideal parallel surface was always found to be less than
:1:151-'lJl. For the second series a grinding apparatus with a higher stiffness was used, which
resulted in a maximum deviation of :1:51-'lJl. For the specimens used in the triaxial tests,
special care was also given to the orthogonality of the surfaces. It was found that the
deviation from orthogonality was always less than 0.01 degrees. This could be neutralized
by the free movement in the horizontal direction of all the loading axes {Van Mier [1984]
and the loosening of the hinges in the loading frames, when placing the loading platens
against the specimen.

Storage
After grinding, the specimens of the first series were stored in se;\Jed plastic bags with
some water added. The last half day before testing, the specim~'s had to be in the open
air to be prepared for testing. The second series was stored in a/case with a constant
relative humidity of lOO% and taken out of this case only two hours before testing. This

46
was done in order to keep the influence of drying shrinkage as slight as possible. The age
of concrete at testing varied from 4 to 6 months. At that age, standard compression tests
on 150 mm cubes, carried out according to the Dutch codes, resulted in compressive
strengths of 51.9 N/mm2 for concrete 1 and 56.6 N/mm2 for concrete 2.

3.4 CRACK DETECTION TECHNIQUE

In the present investigation both external and internal crack patterns were recorded at
the end of a test to be used in the analysis of softening. The external crack patterns were
recorded by simply tracing the macrocrack pattern on the specimen surface visible with
the naked eye. Internal crack patterns were recorded using infrared photography of slices
of concrete specimens impregnated with a fluorescent epoxy resin. The development of
this technique is described extensively by Goudswaard & Vonk [1989]. Recently this
technique also has been applied successfully for the detection of cracks in uniaxial tensile
tests (Van Mier [1991b]).

Methods of crtJCk tktection

Many different methods have been developed to record crack patterns and crack growth
in concrete (Wittmann (ed.} [1983,1986], Reinhardt (1989], Shah [1989], Mindess [1991]).
Three approaches can be distinguished:
1. continuous surface measurements during a test
2. continuous internal measurements during a test
3. recording internal crack patterns after a test.

Continuous recording of crac& .formo.t;ion

Highly sophisticated methods are available for recording surface displacements to detect
crack formation. Examples of these methods are:
1. a photo-elastic technique (Lierse & Ringkamp (1986], Van Mier & Nooru-
Mohamed [1990])
2. interferometry technique (Jacquot & Rastogi [1983], Ansari [1986], Maji & Shah
[1988], Cedolin et al. [1987])
3. stereophotogrammetry (Torrenti et al. [1989]).
However, concrete crack formation is a three-dimensional process and surface cracks do
not always correctly reflect the internal process of cracking (see for instance Bascoul et al.
[1987] and Van Mier [199lb]). The last fact is caused by the heterogeneity of the specimen

47
as a result of the casting and hardening process, by internal shrinkage stresses, by the
actual discontinuity of the specimen surface and by nonuniform loading. Therefore
methods have been developed to record continuously the internal crack process. Examples
are:
1. measurements of ultrasonic-pulse velocity (Alexander&; Blight [1986], Chhuy et
al. [1986,1989], Berthaud [1987])
2. acoustic emission measurements (Diederichs [1983], Maji &; Shah [1987], Izumi et
al. [1986], Chhuy et al. [1986,1989])
3. infrared vibrothermography (Luong [1986]).
These methods give information about the cracking process which is significantly less
detailed than the information acquired by the methods using surface measurements.
Notwithstanding, they have proved to give valuable information for tensile cracking.
However, compressive failure is a more complex three-dimensional process and it will not
be easy to extract the required information of the process of cracking by these techniques.

Recording of crack patterns after a test

The most detailed information about internal crack patterns can be acquired by sawing
specimens open and recording the crack patterns directly. This means that the continuity
of the monitoring process in time and in place is lost. Good examples of this technique are
found in the investigations at CorneD University to establish a relation between the
crack-formation process and the stress-strain curve. There internal crack patterns were
analyzed by means of a dyeing technique and an x-ray technique of slices of concrete
after a test (Hsu et al. [1963], Slate&; Olefski [1963], Slate [1983]). For intensively
cracked specimens this technique cannot be applied, because the process of sawing would
damage the specimen totally. Recently, a petrographic technique inspecting thin slices of
concrete impregnated with a fluorescent epoxy resin has been used to analyze the
structure and microcracking of concrete (Knab et al. [1986], Soers & Meyskens [1987}).
The fluorescent epoxy resin is impregnated in the cracks to stabilize the concrete during
cutting and polishing and makes the cracks and pores visible under ultra violet light. This
technique was only used for small sections.

It was decided to try to inspect the internal crack patterns by means of impregnating the
specimens with a fluorescent epoxy resin because 1
1. the epoxy resin stabilizes the intensively cracked specimer,
2. ultraviolet photography can give detailed pictures of tho/crack patterns and
3. both impregnation technique and photographic technique are very simple.

48
Impregnation technique
At the end of a test the specimen was unloaded and carefully taken out of the loading
apparatus. After a few weeks of drying in the laboratory, the specimen was placed in a
plexiglas mould. Subsequently, the mould was filled with a low-viscous (110mpas)
fluorescent (fluorol green, BASF) epoxy resin (EP-IS, CIBA Geigy). Under vacuum (6.5
mbar) the air was removed from the cracks in the concrete and replaced by epoxy resin.
After 30 minutes, atmospheric pressure was restored in order to let the epoxy resin
penetrate even deeper into the cracks. The pot life of the resin was about 50 minutes. The
epoxy resin easily filled all the macrocracks of cubes of 200 mm. After 8 hours of
hardening, the epoxy resin reached a strength of approximately 60 N/mm2. Subsequently,
the specimen was carefully sawn into slices to be able to record the internal crack
patterns. The introduction of additional cracks by the sawing had no influence on the
recording of the cracks, because only cracks which were filled with epoxy resin before the
sawing started are recorded.

mtraviolet photography
The crack patterns were recorded by means of ultraviolet photography (Roleiflex SL66
technical camera and a Kodak professional plus-x-pan film, black-white, 6x6 cm, 21
DIN). The use of a yellow-green filter resulted in the best contrast between concrete and
cracks.

This method works very well for specimens with continuous patterns of macrocracks. It
can also be extended to detect isolated microcracks by the techniques developed at
Cornell University (Slate [1983], Goudswaard & Vonk [1989]).

3.5 SUMMARY OF TESTS

In this research two main test series have been carried out:
1. boundary-condition tests
2. size-effect tests
The details of these tests are summarized in Tables 3.5.1 and 3.5.2. The test results have
already been reported in Vonk [1989], Vonk et al. [1989] and Vonk et al. [1990].

49
 Influence of boundary condi ·
Loading Size Specimen Number
Triaxial Platen Height Vidth of Tests
Uniaxial Long Brush 100 100 2
BL3 Triaxial Long Brush 100 100 3
BS1 Uniaxial Short Brush 100 100 2
BS3 Triaxial Short Brush 100 100 3
D1 Uniaxial Dry Platen 100 100 2
D3 Triaxial Dry Platen 100 100 2
HT1 Uniaxial Hinged Tefl. Pl. 100 100 2

T1 Uniaxial Teflon Platen 100 100 2

T3 Uniaxial Teflon Platen 100 100

Table 3.5.1

Test series 2 : Influence of size and shape

Code Uniaxial Loading Size Specimen Number
Test Triaxial Platen Height Vidth of Tests
n Platen 200 200 4

200 100 4

200 50 4

IL Teflon Platen 100 200 4

D Uniaxia.l Teflon Platen 100 100 4

IS Uniaxial Teflon Platen 100 50 4
SL Uniaxial Teflon Platen 50 200 4
SI Uniaxial Teflon PLaten 50 100 4

ss Uniaxial Teflon Platen 50 50 4

Table 3.5.2

50
 CHAPTER4
MODELLING OF SOFTENING

4.1 INTRODUCTION

A short review of literature concerning modelling of nonlinear concrete behaviour will be

given in this chapter, with special attention to softening. This review can not be
complete, as a great deal of literature on this subject has been published during the last
decade. It has been pursued to select the most significant developments that form a
background to the problem of modelling compressive softening of concrete.

Contefltl ofthitl chapter

The development of finite element analysis has been a great stimulus to the formulation
of nonlinear material models. Since this method was initially fully directed to the
description of continuum behaviour, a large number of continuum models has been
formulated. These models will be discussed in Chapter 4.2. As softening is due to crack
formation, many researchers have directed their efforts to formulating models describing
initiation and growth of individual cracks. Chapter 4.3 on fracture mechanics deals with
this subject. The ways in which the results of fracture mechanics are implemented in
finite element analysis are discussed in Chapter 4.4. Most of the crack models deal only
with tensile loading and often give a poor description of the behaviour under shear
loading. This last aspect, essential to modelling of softening under compressive loading, is
treated in Chapter 4.5. Problems encountered by calculating structural behaviour by
means of finite element analysis using continuum models for softening are discussed in
Chapter 4.6. Most researchers agree that the classical continuum formulation should be
enriched by a characteristic length based on the heterogeneous structure of concrete to be
able to describe concrete softening adequately. Several methods which are being
investigated at present are given a short review. A problem in modelling concrete
behaviour lies in the fact that the causal mechanism is not yet fully understood. Therefore
researchers try to investigate the mechanism of concrete softening by modelling its
heterogeneous structure to simulate the failure process. This subject is treated in Chapter
4.7.

51
4.2 CONTINUUM MODELS

Three main groups of continuum models used for the description of concrete behaviour
can be distinguished.
1. Nonlinear elastic models
2. Elastoplastic models
3. Damage models
Several other approaches have been applied, but these have not become so widespread and
are not discussed here.

Nonlinear eklstic models

Initially, only linear-elastic material properties were used in finite element analysis. The
next logical step was to formulate the stresses as nonlinear functions of the strains ( Chen
& Saleeb [1982]). However, the one-to--one relation between stress and strain in elastic
models (Fig. 4.2.1a) does not agree with concrete behaviour. Concrete shows path-
dependent behaviour and a difference in behaviour between loading and unloading. This
limitation was overcome by formulating incremental elastic moduli as functions of the
stress-and'""'5train history and treating loading different from unloading. It seems that by
this approach the nonlinear-elastic models are capable of describing most features of
concrete behaviour under any kind of triaxialloading. However, it has been shown that
this approach creates the danger of formulating a model violating thermodynamic laws by
generating energy in a closed deformation cycle (Chen & Saleeb [1982]). Basing the model
on energy-density functions eliminates this problem, but complicates the model and
limits its possibilities. Examples of elastic models for concrete can be found in the work of
Kupfer & Gerstle [1973], Cedolin et al. [1977], Ottosen [1979], Kotsovos [1980,1984],
Gerstle [1981a,1981b], Cedolin & Mulas [1984], Stankowski & Gerstle [1985] and Ahmad
et al. [1986].

a a a

t
(a) (b) (c)

Fig. 4.2.1 Examples of (a) nonlinear elastic, (b) elasto plastic and (c) damage modelling

52
~m.od.eliJ
Nonlinear beha.viour in elasto-plastic models is assumed to be due to plastic deformations
in the material (Fig. 4.2.1b). The theory of plasticity was first developed for metals, but
proved to be so powerful that it could also be applied to concrete (Chen &: Ban [1988])
and geomaterials (Chen &: Baladi [1985]). n provides a flexible and reliable basis for the
description of nonllnear material behaviour including path-<lependent behaviour,
unloading and softening. One or more state variables, formulated as a function of the
plastic deformations, govern the stress-strain behaviour. Examples of elasto-plastic
concrete models are found in the work of Chen &: Chen [1975], Chen &: Sehnobridge
[1981], Fardis et al. [1983], Vermeer &: De Borst (1984], Chen &: Buyukozturk [1985], Lin
et al. [1987] and Pramono &: Willam [1989].

IJu.mage m.ol.el4J
Nonllnear concrete behaviour is due to the formation of cracks, which weakens the
material. Damage models simulate this weakening by reducing the elastic moduli as a
function of one or more damage parameters, which represent the state of damage in the
material due to crack formation. This means that unmodified unloading is always directed
to the zero-stress zero-strain state according to these reduced elastic modnll (Fig.
4.2.1c). Dougill (1976] formulated the basis for damage modelling of concrete behaviour in
his work concerning the "progressively fracturing solid". This theory, which fits well
within the framework of thermodynamics {Kra.jcinovic &: Fonseka. [1981]), is a sound basis
for the formulation of complex concrete models. Examples of damage models can be found
in the work of Loland [1980], Fonseka &: Krajcinovic (1981], Mazars [1984,1986], Ortiz
[1985] and Pijaudier-Cabot &: Bazant [1987]. Because nonllnear concrete behaviour seems
to be a combination of damage and plastic slip, some researchers have even accepted the
challenge to combine plasticity and damage models (Ba.zant &: Kim [1979], Hsieh et al.
(1982], Ban&: Chen (1986]). These models are very complex. It has not yet been shown
that they result in a more accurate prediction of concrete behaviour.

Commmt on contiRVum '11Wtlel6

It seems that all three classes of models are basically able to describe any nonllnear
triaxial continuum behaviour under monotonicalloading including softening. Unloading
or cyclic loading can also be covered. However, the complexity of the model would be
enormous when all these features are covered and no such models are to be found in
literature.

The class of continuum models presented above has two important drawbacks:

53
1. Rotation of aus of principal stress and strain
A severe limitation is that the models are restricted to the principal stress or strain space.
This also applies to the tests to which these models are fitted. They do not take into
or
account a rotation of the axes principal stress strain, which can lead to significant
or
errors (Bazant (1983]), especially for softening when the orientation damage changes
the material from isotropic to highly anisotropic. This is no imaginary problem, because
nonlinear structural behaviour causes these changes in principal stress or strain direction
easily (see for example Rots (1988]). The correct approach is to link the state or damage
with its orientation to the material and to calculate the response or the material to some
or
kind of loading out this state of damage. This is done in the concept of the microplane
model proposed by Bazant et al. ( Bazant & Gambarova [1984], Bazant &; Oh [1986]).
or
This model takes into account the orientation damage on the mechanical response or
or
the material by integrating the influence of the state damage of a finite number of
planes with a uniformly--distributed orientation. The application of this model in finite
or
element analysis, has been limited because the significant demand for calculation time.
Recently, Ozbolt et al. (Ozbolt &; Bazant [1991], Ozbolt &; Eligehausen [1991]) and Carol
&; Prat [1991] have developed new microplane models and applied these for, structural
calculations. It should be noted that discrete crack models, the fixed smeared crack model
and the fixed smeared multidirectional crack model have also the correct fixation to the
material (see Chapter 4.4).

!. Localization of deformations
For this research the most severe limitation of the continuum models is found in the !act
that they do not take into account the localization of deformations caused by strain
softening. Some of the continuum models mentioned in this chapter have been fitted
directly to test results including softening, thus neglecting the phenomenon of
localization. This means that predicted deformations will not be correct for any specimen
or different size than the one used in the test or broadly speaking for any other concrete
structure. Localization or deformations due to softening is incompatible with the classical
continuum approach used up to now in modelling or interpreting test results. New ways
or
have to be found to take the localization deformations into account.

4.3 FRACTURE MECHANICS

Linear elastic fratture mechanil;$ (LEFM}

Because softening of concrete is due to crack formation, many researchers have directed

54
their efforts to the formulation of models for crack initiation and propagation. Fracture
mechanics was first developed for perfectly brittle linear-elastic materials. Griffith [1920]
was the ~rst to formulate a criterion for crack propagation. He stated that crack
propagation takes place when the energy released by the material during crack extension
is equal to that energy needed to create the new crack surface. However, elastic
calculations show that the stresses around a crack tip go to infinity (Fig. 4.3.1a), which
means that nonlinear effects will be present.

ctniform (1
I I
-IJ1.ttJ.tfJJ.ttJ1.tf_l -4. t tJ 1.ttJ.1.tt J t.tt J. +-
I
Macro crack
-~JT~L~e
I -+ ·
cr~~kfJl.[[:
· -1
I _::---:::;;
i. zones
t
,:::----
tk.mg.i
I

. I I . Wl•th m1crocrac
.
I i i
: I~CJt;p= V) -+I:
I -+ ·
1
r lll.r ll. Trll r r~ 1· t-
I
(a)
.. .. (1
(b)

Fig. 4.3.1 (a) Stresses around a crack tip in an elastic medium

and (b) microcracked zone in front of a crack tip

Nonlimar fracture mechanic& {NLFM}

Extensive research has shown that concrete does not act according to LEFM, because
microcracking occurs in the area in front of the crack tip (Fig.4.3.1b ). This means that
the energy needed for crack propagation is greater than the surface energy of the
macrocrack and that concrete is more ductile than LEFM predicts. To overcome the
limitations of LEFM, a number of nonlinear fracture models has been developed. The
cohesive-{:rack model has become the most popular for concrete, because it can predict
both crack initiation and crack growth and is easy to incorporate in finite element
programs (Wittmann (ed.) [1983,1986], Shah & Swartz (eds.) [1989], Shah et al. (eds.)
[1989], Van Mier et al. (eds.) [1991a]). This type of model was first developed by Dugdale
and Barenblatt for cracks in steel, which have a nonlinear zone around the crack tip due
to yielding of the material (see Broek [1986]). In a cohesive crack model the nonlinear
zone in front of the crack tip is modelled as a part of the crack, which is not stress free
but contains cohesive forces. Hillerborg et al. [1976] were the first to apply a
cohesive-{:rack model for the description of concrete cracking. Their "fictitious crack
model", which models a crack as a line crack, assumes that there is a microcracked zone
in front of a stress free macrocrack. This s~alled crack-process zone is assumed to
soften gradually as a function of the deformation of the microcracked zone (Fig. 4.3.2).

55
The model predicts microcrack formation, which initializes softening, when the tensile
strength ft is reached. A stress-free macrocrack is formed when the opening of the
fictitious crack is equal to Wmax· These two parameters, together with the shape of the
softening curve and the surface underneath it, the fracture energy Gr, are assumed to be
material properties .
. I
a

Wmox

I
crac~ Micro cracks
~
Fictitious crack
w
Wmax

Fig. 4.3.2 Fictitious crack model (Hillerborg et al. [1976])

4.4 CRACKING IN FINITE ELEMENT ANALYSIS

In finite element analysis two main crack concepts are used.

1. the discrete-crack concept
2. the smeared-erack concept

Discrete-crack concept
When the discrete-crack concept is used, each crack is modelled separately and is taken
into account as a discontinuity in the finite element mesh. Three important approaches
ca.n be distinguished.

1. Crack path known in advance

When the crack path is known in advance, it is easily incorporated in the initial mesh of
continuum elements as a. series of interface elements, which represents the potential crack
path (Fig. 4.4.la). The crack path ca.n be known in advance from tests, from preliminary
numerical analysis or estimated by trial and error (Rots [1988], Hillerborg &: Rots [1989]).
The need to know the crack path in advance is of course a severe limitation of this
approach.

56
2. Discon.tin.uum
When the crack path is not known in advance, it is possible to model a continuum
separated by a continuous mesh of interface elements, which represents all potential crack
paths (Fig 4.4.1b). Then it is better termed a discontinuum than a continuum. The mesh
of interfaces can be fully random, but can also be modelled taking into account the actual
internal structure of the material. In rock mechanics it is an accepted method to model
joints and discontinuities in this way. This method is used also in the Distinct Element
Method (Cundall [1911], Cundall & Strack [1979], Lorig & Cundall [1987]), which is
applied to the micromechanical simulations in the present investigation. Another
application is found in the analysis of the mechanical behaviour of masonry structures
(Rots (1991]). A drawback of this approach can be that the exact configuration of the
mesh is not known or can not be established.

9. Remeshin.g
When it is desired to keep crack propagation independent of the initially chosen finite
element mesh, a technique of remeshing can be applied (Fig. 4.4.1c, Ingraffea and Saouma
[1985]). Based on calculated continuum stresses, a prediction can be made for the
extension of the crack, which determines the new configuration of the mesh incorporating
the crack as a discontinuity. For implicit methods this means that the set of equations
governing the problem must be reformulated and solved again completely from the start.
It is clear that this takes a lot of computer time

(a) (b) (c)

Fig. 4.4.1 Examples of discrete modelling of cracks: (a) crack path known in
advance (Rots [1988]), (b) discontinuum (Lorig & Cundall [1987]), (c) remeshing
(Ingraffea & Saouma [1985])

57
Smeared crack concept
The smeared-crack concept was developed to have perfect freedom of crack initiation and
growth and to overcome the problems of incorporating discontinuities in the finite
element mesh (Bazant [1986], De Borst [1986], Rots [1988]). This concept soon became
more popular than the discrete-crack concept adopted earlier, because it fits in better
with the nature of the finite element method, which was directed initially only to the
description of continuum behaviour.

f1 f1

- -------
------ -
f1 ~ack
e.nox

Fig. 4.4.2 Smeared-crack concept

In the smeared-crack concept, the behaviour of the cracked concrete is formulated as

stress-strain behaviour of the finite element containing the cracks. This is illustrated in
Fig. 4.4.2. The total strain of the element is divided into a concrete strain and a crack
strain:

f = fconcrete + fcrack ( 4.4.1)

Crack formation is described according to the fictitious-crack model. Microcracking

starts when the stress in the continuum reaches the tensile strength ft. Then strain
softening is modelled as a decreasing function of the crack strain. The material is fully
fractured when the strain reaches fmax· A stress-free macrocrack has been created.

To obtain mesh-independent results (Bazant [1986]), the fracture energy Gr must be

independent of the element size h. For the area g under the stress-strain diagram this
means that

Gt
g=- (4.4.2)
h

58
When the shape of the softening curve is kept constant, Emax is related to the maximum
crack opening Wmax of the fictitious crack model by

Wmax
Emax=--- (4.4.3)
h

The orientation of the crack is not fixed in advance. As soon as the tensile strength is
reached in any direction in the material, cracking is initiated perpendicular to the
direction in which the tensile strength is reached. Then the constitutive behaviour of the
element is changed from isotropic to orthotropic by adding the crack softening strain to
the continuum strain perpendicular to the crack direction. The discussed approach has
been used mainly for simple tensile failure. A more general model, which covers softening
due to triaxial tensile and compressive loading, was recently proposed by Pramono &
Willam [1989] . For cracks with an inclination to the finite element boundaries, the value
of h is not so obvious as in Fig. 4.4.2. When a crack meanders through a finite element
mesh, it is even harder to determine this value (Rots [1988]).

Cmck-6arul theory
Bazant et al. (Bazant & Oh [1983], Bazant [1986]} describe crack formation by means of a
crack band of a certain width, because a concrete crack is no line crack and the
microcracked zone in front of the crack tip also covers some space. In finite element
analysis the crack-band model is practically treated as a smeared crack. Then the exact
width of the crack pattern seems not to be of great importance as long as it fits inside the
finite element. Determination of the width by fitting the model to test results showed
that the results of calculations with the model are not very sensitive to the exact value of
the width.

4.5 CRACK BEHAVIOUR UNDER SHEAR LOADING

LEFM prediction of shear cracking

Tensile cracking is the most important and perhaps the sole cause of concrete softening.
Still, it is difficult to model shear softening or compressive softening with tensile models.
Perhaps this can be done when the heterogeneity of concrete is modelled down to a very
low scale level. By means of LEFM it can be shown that tensile cracking can take place in
heterogeneous materials under shear loading or compressive loading (Di Tomasso [1984],
Van Mier [1984}, Horii [1986]). Even under triaxial compression tensile stresses can be

59
found around a void or crack {El-Rahman & Shrive [1984]). However, LEFM has not
brought forward a practical model for the description of shear cracking or compressive
cracking.

Mohr-Coulomb type models

Up to now, the most practical models have been the Mohr-Coulomb type models, which
limit and control the shear stress at a plane as a function of the normal stress on that
plane {Fig. 4.5.1a). Many of the continuum model~ given in Chapter 4.2 feature this
limitation too. Though they are formulated in principal-stress space, they 'actually limit
the shear stress on certain planes. How a shear band in a specimen loaded in compression
is predicted by the use of a Mohr-Coulomb continuum model is illustrativ~ {Fig. 4.5.1b,
De Borst [1986]).

(1

(a) {b)

Fig. 4.5.1 (a) Mohr-(Joulomb criterion and {b) shear band predicted With a Mohr-
Coulomb continuum criterion formulated in principal stress space (De IBorst [1986])

When the shear plane is known, it is also possible to use a Mohr-Coulomb type of model
for the description of interface behaviour. Simple interface models of this type have been
used by Roelfstra et al. (Roelfstra & Sadouki [1986], Roelfstra [1989]) and Lorig &
Cundall [1987]. In these models a tension cut--off criterion is added to the shear-failure
criterion. Recently, a more complex model for combinations of tensile and shear loading
including softening has been proposed by Stankowski [1990].

Shear loading of smeared cracks

As the orientation of principle stress can change after the initiation of cracks, it is also
important to model the behaviour under shear loading of a crack created due to tensile
loading. It was already mentioned in Chapter 4.2 that continuum models are often unable
to take this rotation of principal stresses into account correctly. The same is true for the

60
smeared-crack models. The greater part of the smeared crack models has the same
problem due to a poor description of the behaviour under shear loading. Most often the
shear modulus G of the finite element containing the crack is set to zero or multiplied by
a constant reduction factor /3 (/3<1), which is called the shear retention factor. Rots [1988]
showed that even low values for /3 (=0.05) can result in a significantly over-stiff reaction
of structures which fail due to tensile crack formation (Fig. 4.5.2). The nonlinea.r
behaviour of a structure causes a rotation of principal stresses, which can rise to
significant values because no softening is modelled for shear loading of the cracks. As
alternatives to the fixed crack concept the rotating crack concept (Cope et. al. [1980],
Willam et al. [1987]) and the fixed multidirectional crack concept (De Borst [1986], Rots
[1988]) have been proposed. In the rotating crack concept the crack eo-rotates with the
rotation of principal stresses, which means that the shear stresses are automatically kept
zero. This is similar to the use of continuum models formulated in principal stress or
strain space. ~he fixed multidirectional crack concept allows new cracks to be formed at
an inclination to the first crack. This results implicitly in a kind of shear softening. The
concept of this model is close to the microplane model proposed by Bazant et al. (Bazant
& Gambarova [1984], Bazant & Oh [1985]).

F,
(kN)
10

o~~~~~--~--~--~--~-+
0.0 0.1 0.2 0.3 0.4 0.5 0.6
CMOD(mm)
Fig. 4.5.2 Influence of crack model and shear retention factor /3 on load-deformation
behaviour of a specimen (Rots [1988])

Aggregate interlock models

It has long been recognized that the shear resistance of tensile cracks, which have opened
and are subsequently subjected to a shear deformation, still can have a significant value.
This is due to the irregular shape of the crack faces, caused by the fact that the cracks
have to curve around the aggregates in the concrete. Under shear loading, the irregularly

61
shaped crack faces make contact and cause the shear resistance. This phenomenon is
called aggregate interlock. Adequate models have been proposed to describe this
behaviour {Walraven (1980], Bazant and Gambarova [1985]) A complete model for crack
behaviour should of course cover both the behaviour of a crack under tensile and shear
loading during initiation and after opening.

4.6 STRAIN SOFTENING IN FINITE ELEMENT ANALYSIS

Problems tDith ltrt&m aoftening

Recently it was discovered that it is very difficult to incorporate softening correctly in the
classical concept of the finite element method by means of strain softening. The finite
element method was entirely conceived and developed to the discretization of continuum
behaviour on the assumption of displacement continuity. The displacement discontinuity
due to crack formation, which is described by softening, seems to be incompatible with
this method. It has been shown that the use of strain softening leads to
1. dependency of calculation results on the size of the finite elements (Bazant
[1986], De Borst [1986]),
2. dependency of calculation results on the orientation and shape of the finite
elements (Rots [1988]),
3. dependency of calculation results on the type of finite element (Steinmann &
Willam [1991]) and
4. spurious local kinematic modes, which lead to instabilities hampering the solution
procedure for the total.stmcture (De Borst [1988]).

Characteristic ltmgtA
Keeping the fracture energy independent of the size of the finite element by making strain
softening dependent on the size of the finite element only solves the problem of mesh
dependency in a few simple cases. A more general method is needed. Simple dimensional
analysis shows that a parameter representing a length is needed to change strain-
softening in a local fracture behaviour. Therefore, researchers have turned to enrichment
of the classical continuum formulation with a. characteristic length, which is used to
control the localization of strain softening in a certain area. This characteristic length is
usually assumed to be related to the heterogeneity of concrete. It is a common assumption
that the heterogeneity dominated by the large aggregates controls the spread of
microcracking and macrocracking and thus the area which shows softening. Most
important from a mathematical point of view however, is that enriching the governing

62
field equations for the continuum with terms incorporating a characteristic length,
prevents the system of equations from losing its elliptic character.

Se1JertJl approaches
Three main approaches are being explored at the moment. Ba.zant and eo-workers use the
nonlocal approach in which averaging procedures for strain and material state variables
over a finite domain are used (Bazant et. al. (1984], Pijaudier-Ca.bot & Ba.zant [1987),
Bazant & Ozbolt [1990], Ozbolt & Bazant [1991], Ozbolt & Eligehausen [1991]). The size
of the finite domain controls the area in which strain-softening is loca.lized. It is assumed
that this size is a material parameter which is set by the heterogeneous structure of
concrete. Others try to controlloca.liza.tion of deformations by using formulations for the
calculations of stress involving gradients of strain or inelastic state variables (Schreyer &
Chen [1986], Trianta.fyllidis & Aifantis [1986], La.sry & Belytschko [1988], Schreyer
[1989]). The nonlocal approach and the gradient approach are closely related (De Borst
[1991]). A different approach is proposed by MO.hlhaus [1986] and De Borst [1990,1991],
who use the Cosserat continuum to control the localization of strain softening. A Cosserat
continuum includes rot.ational degrees of freedom in the governing field equations. The
classical continuum theory contains only translational degrees of freedom. When the
rotational degrees of freedom are coupled to a certain characteristic length, the area in
which softening localizes can be controlled. However, it was recently found that this
approach is not able to control pure tensile softening, because in that case the rotational
degrees offreedom are not activated (De Borst [1991]) ..

Comment
The "nonlocal 11 approach and the gradient approach can technically control strain
localization under general loading conditions. What approach is most effective is not yet
clear. How well concrete behaviour is described by these methods under all loading
conditions is not known. The methods are very new and a thorough verification has not
yet been carried out. They are a logical extension from an existing method which has
proved its effectivity. The characteristic length introduces some notion of heterogeneity in
the homogeneous continuum. How this characteristic length is related to the
heterogeneity of concrete, whether it depends on the state of loading, what it really
represents and how it can be identified or measured is not yet known. Clearly, there is a.
gap between material science and computational mechanics.

63
4. 7 SIMULATION OF HETEROGENEITY

Localization of strain softening in a volume determined by a characteristic length is a

feature characteristic for continuum mechanics. Actually, discrete cracks grow in a
heterogeneous material causing softening on a macroscopic level. Models taking the
heterogeneity into account to predict softening due to crack growth will be discussed in
this chapter. This approach has not yet resulted in models fit to be used in the
calculations of complex concrete structures. They are research tools for investigating the
underlying mechanism of softening.

Stochastic models
Many features of concrete behaviour can be described by stochastic models using only
simple assumptions for its internal structure (Mihashi and Izumi [1977], Mihashi [1983]).
Assuming that cracking is a stochastic process leads to realistic stress-strain curves for
concrete. Dougill. [1971] showed this for a one-dimensional model in which he assumed
that an external uniform stress results in a nonuniform normally-distributed stress state
inside the concrete, which causes a progressive failure. Later, Burt and Dougill [1977]
formulated a random two-dimensional truss model simulating the influence of spatial
heterogeneity on softening. No link was made to the actual structure of concrete.
Qualitatively, stochastic truss models can describe the process of macrosc?pic softening
and localization of deformations. For this purpose it seems not to matter whether an
internal distribution of stiffness or strength is used or if the topology of the mechanical
model is related to the concrete structure (Schorn [1986], Schorn & Rode [1989],
Schlangen and Van Mier [1991a]).

M~chanical models

When more realistic predictions of failure under all kinds of multiaxial loading are
required, it is necessary to model the microstructure of concrete more realistically for use
in mechanical calculations. Such models are called micromechanical models. The most
important characteristic of the heterogeneity of concrete is its two-phase structure of
aggregates in a hardened cement paste matrix. Hardened cement paste itself is also
heterogeneous, but on a lower level. The lesser importance of its heterogeneity can be
deduced from its brittle failure behaviour. As shown in Chapter 2, the role of the
aggregates in concrete softening is very important.

LEFMmotlel
Using LEFM for a computer-simulated composite structure of concrete, Zaitsev and
Wittmann (Zaitsev & Wittmann [1981], Zaitsev [1983]) showed that realistic crack

64
formation can be predicted for both tensile and compressive loading taking into account
the composite structure of concrete An example for uniaxial compression is shown in Fig.
4.5.1.

Fig. 4.5.1 Two stages of crack formation in uniaxial compressive loading (Zaitsev [1983])

Truss models
Two truss models based on a particle structure were proposed by Zubelewicz & Bazant
[1987] and Bazant-et al. [1990] (Fig. 4.5.2). The idea of both models is based on the
distinct element method formulated by Cundall (Cundall [1971], Cundall & Strack
[1979]). This method models the behaviour of granular materials and discontinuous rock
masses by means of rigid elements exchanging forces at contact points which take all the
deformations. Most important is that crack formation is also localized in these contact
points. Recently, a lattice model, simulating the concrete structure in more detail, was
presented by Schlangen & Van Mier [1991b] (Fig. 4.5.3). For tensile loading these models
have proven to be able to realistically describe
1. crack patterns,
2. softening as a result of crack growth,
3. localization of deformations and
4. size effect in peak stress and softening.
In the models of Zubelewicz & Bazant and Schlangen & Van Mier failure on the
rnicrolevel is modelled as perfectly brittle. However, as shown in Fig. 4.5.3b, this results
in a highly irregular macroscopic stress-deformation relation. Probably a significant
scatter in results for different simulations is also found, although this is not reported. In
the model of Bazant et. al. [1990] softening is modelled at the microlevel. It gives
approximately the same results as the models with brittle failure, but less disturbed by
stochastic influences.

65
(a) {b)

Fig. 4.5.2 (a) Particle model of concrete and {b) predicted tensile cracking
{Zubelewicz & Bazant [1987])

a (N/mm 2 ) . - - - - - - - - - - - - - - ,
3.0

2.0

1.0

0+--T--~~--~~~~~~
0 20 40 60 80 1 00 1 20 1 40
(a) w( JLI11) {b)

Fig. 4.5 .3 (a) Tensile stress-deformation relation and {b) crack pattern predicted by
the model of Schlangen & Van Mier [1991b)

Finite element models

The above-mentioned models take into account the internal length scale and internal
structure set by the large aggregates in concrete. Mechanical interactions are modelled in
a simple way. More detailed models using finite element analysis to describe the
mechanical behaviour have been proposed by Roelfstra et al. {Roelfstra and Sadouki
[1986), Roelfstra [1989]) and by Willam, Stankowski et al. {Willam et al . [1989),
Stankowski [1990]). The model proposed by Roelfstra et al., named 11 Numerical
Concrete 11 , describes concrete very realistically by means of a two-phase material
comprising large aggregates in a mortar matrix. It uses a detailed description of the
composite structure of concrete {Fig. 4.5.4). It has been used to analyze mechanical and
physical behaviour of concrete. Only a limited number of tensile failure simulations have
been carried out with the model. The model of Willam et al. is similar, but simpler due to
its more schematic representation of the concrete structure {Fig. 4.5.5). In this model a

66
mixture is used of discrete cracking in the interfaces between aggregates and mortar and
continuum softening in the mortar. Simulations of both tensile and compressive softening
have been carried out and look quite realistic. Localized shear bands are predicted for
compressive softening. However, the predicted size effect for compression does not agree
with the test results of Van Mier [1984).

(a) (b)

Fig 4.5.4 (a) Finite element mesh of "Numerical concrete" and (b) prediction of cracking
due to shrinkage and tensile loading (Roelfstra [1989))

(a) ··"'1,~
...~~~-'::t;-;~~~-;,_,;';;;
...:--'-'"~
_~1.20!-'--'"~
-1.50
AXIAL S1lWN (11:-3)
(b)

Fig. 4.5.5 (a) Two-phase model of concrete and (b) predicted shear failure
in uniaxial compression (Willam et al. [1989))

67
 CHAPTERS

MICROMECHANICAL MODEL

5.1 INTRODUCTION

Analysis of test resulU

In Chapter 2 it was shown that the process of compressive softening is complex and not
yet well understood. Due to localization of cracking and deformations, softening is the
result of a structural behaviour of the specimen which is tested. Due to this structural
behaviour it is not possible to fit a simple model directly to the measured specimen
behaviour. First, assumptions have to be made for the structural process inside the
specimen. This means that the continuum behaviour and the local behaviour must be
known. For the local behaviour it is important to know where and when it is initiated.
These are facts that are unknown for compressive softening. Highly sophisticated methods
have been developed to record crack formation and deformations in concrete tests. These
were briefly summarized in Chapter 3.4. They have brought forward valuable information
about concrete softening. But it is not easy to extract the essential information from a
compression test. The process of cracking takes place inside the specimen and cannot be
easily followed in detail.

Micromuhanical modelling
It has recently been shown that micromechanical models can be a powerful tool for
analyzing softening of concrete. This was discussed in Chapter 4.7. These models enable
us to investigate the influence of the heterogeneous composite structure of concrete on its
behaviour. A great advantage is that the process inside the material can be monitored
continuously in time and space. It is also easy in such a model to control the boundary
conditions and investigate their influence. These aspects make a micromechanical model
suitable for investigating the structural behaviour of concrete which determines
compressive softening.

COfl.tentll
The philosophy behind the micromechanical model will be discussed in Chapter 5.2.
Chapter 5.3 gives the most important characteristics of the computer program UDEC,
which is used for the numerical simulations. A new model for interface behaviour,

68
representing crack formation, is then discussed in Chapter 5.4. Chapter 5.5 will show the
ability of UDEC to simulate softening and localization of deformations.

5.2 IDEA BEHIND THE MICROMECHANICAL MODEL

TVJO-pho.se model
The first and most logical step in modelling the heterogeneity of concrete is modelling it
as a two-phase material existing of aggregates in a hardened cement-paste matrix. The
aggregates act as the most important stress concentrators and define for normal concrete
an area where no cracking is found. Hsu et al. [1963] conclude from their research that
1. cracking in normal concrete starts at the interfaces between coarse aggregates
and mortar because the bond between these components is the weakest link in
concrete and that
2. continuous crack patterns are created by the growth of cracks through the mortar
forming bridges between the bond cracks.
This means that the aggregates determine the shape and spacing of the cracks and that
the interfaces between the aggregates and mortar play a very important role in the
crack-formation process.

1Ji.8crm -crack approach

To be able to describe the above mentioned process, it was decided to model crack
formation by means of a discrete-crack approach as done in the Distinct Element Method
{Cundall [1971], Cundall and Strack [1979], Lorig and Cundall [1987]). In this way the
problems of the smeared--crack approach are avoided. It was expected that the
heterogeneity of the model would determine localization of cracking and deformations.
The material between the discrete cracks, represented by continuum elements, is not
allowed to crack in the model.

Interface pattern
A random mesh generator was built to generate a continuous pattern of interface
representing all potential crack paths in concrete. Based on the conclusions of the research
of Hsu et al [1963], the pattern exists of two sets of interfaces, that is
1. one set of interfaces representing all potential bond cracks along the boundaries
of the aggregates and
2. one set of interfaces representing all potential mortar cracks connecting all
potential bond cracks in a logical and systematic manner.

69
The basic structure of the interface pattern is shown in Fig. 5.2.la. The aggregates are
schematized as regular hexagons. Only the coarse aggregates are modelled, because it is
beyond computational means to model all aggregates. This means that the matrix around
the aggregates represents mortar.

--bond interface
--- mortar interface

(a) (b)

Fig. 5.2.1 (a) Basic interface pattern and (b) randomly disturbed interface pattern

Simulation heterogeneity
The heterogeneity of concrete is simulated further by
1. giving different elastic properties to the aggregates and the mortar,
2. giving different strength to the potential bond cracks and mortar cracks and
3. disturbing the regular interface pattern in a stochastic manner.
As a result, the heterogeneous structure will determine the interfaces which will be
activated during the failure process and those which will not.

Random mesh generator

The mesh generator allows the regular pattern of Fig. 5.2.la to be disturbed in a random
manner. The result is shown in Fig. 5.2.lb. The size of the aggregates is varied between
the maximum aggregate size and approximately one third of this value. A random
displacement of the aggregates away from their original position can be specified. The size
and shape of the mortar layers between the aggregates can be varied. This results in a
change of shape of the aggregates from a regular hexagon to an irregular polygon with six
or less corners. Measures have been taken to prevent the formation of very small
elements, which means that aggregates can disappear or that the mortar layer can
disappear and the aggregates will touch each other. A part, having the size and shape of
the specimen, is cut out of the mesh at a randomly chosen position and angle of orien-
tation. More details about the mesh-generation procedure can be found in Appendix A.

70
FiRite element afltllym
A fimte element analysis was preferred to a simulation with a truss model because in
finite element analysis the formation and behaviour of a crack can be modelled more
naturally by means of an interface between the continuum elements. The truss models
have been found to give good results for the simulation of tensile crack formation, but it
was doubted whether phenomena like shear failure, friction and aggregate interlock could
be easily simulated with such a model.

Choice ojUDEC
To carry out the finite element analysis a choice was made in favour of the computer code
UDEC (Universal Distinct Element Code, Itasca Consulting Group [19891), as
1. it has great capabilities in splitting up structures into continuum parts separated
by interfaces, thus crea~ing a discontinuum,
2. it is especially programmed to handle discrete <:!&eking and mechanical
interactions of continuum elements after complete separation and
3. it takes into account the influence of large displacement& on the mechanical
interactions. This is very important because in the micromechanical simulations
displacement& of the size of the finite elements are found.

New CORBtituti?Je model

Initially it was hoped that the interface behaviour could be described by means of brittle
fracture and friction using a stochastic distribution for the characteristic properties. It
appeared that this was not suitable for the model presented here. A highly brittle and
irregular failure was found for specimens loaded in compression. Probably, this will not be
found when the heterogeneous structure of concrete is modelled down to a smaller scale
level. This was not done for the micromechanical model, because this would have resulted
in extreme calculation times. Because UDEC did not provide an interface model with
softening for combinations of tensile and shear loading, a new constitutive model
providing this feature was developed and implemented in UDEC.

5.3 UDEC

The computer program UDEC (Universal Distinct Element Code, ltasca Consulting
Group [1989]) was written especially for the examination of mechanical problems of
fractured media. UDEC has been applied extensively for modelling problems in rock
mechanics. It is based on the Distinct Element Method developed by Cundall [1971] and

71
is formulated quite differently from the better-known Finite Element Method. The most
characteristic features of UDEC are:
1. the division of structures into blocks and interfaces,
2. the allowance of large displacements and
3. the explicit dynamic solution procedure.

Contact Block

corner

Fig. 5.3.1 Structure consisting of 2 blocks, 8 zones, 1 interface and 3 contacts

Block system
Basically UDEC distinguishes 3 kinds of structural elements:
1. blocks
2. zones
3. interfaces
They are shown in Fig. 5.3.1. A structure is split up in a number of blocks by creating a
continuous pattern of interfaces through the str.ucture. Blocks are the structural elements,
which represent the continuum behaviour of the structure. They are not supposed to show
localized cracking, but can have any linear or nonlinear continuum behaviour. Blocks may
have any polygonal shape. For the calculation of stresses and strains, the blocks are
divided into finite elements, which are called zones in UDEC. UDEC only uses triangular
constant strain elements. The basic idea behind the Distinct Element Method is that
nonlinear behaviour and tensile or shear failure in particular is mainly found in a limited
number of weak planes. In UDEC these weak planes are modelled as interfaces between
the blocks. Interfaces have their own constitutive behaviour, which is thought to be
lumped in a number of contacts between the blocks. This is also shown in Fig. 5.3.1. The
constitutive laws of the contacts are actually force-displacement relations, but can also
be formulated as stress-displacement relations. Then the length of the interface
represented by the contact is taken into account.

72
Large displacements
One of the most important properties of UDEC is that it can take into account the
influence of large deformations on the interaction between the blocks. Blocks are allowed
to slip, separate, lose contact and displace over considerable distances. Contact points and
contact lengths are continuously updated. UDEC even tracks new contacts which did not
previously exist. Corners of blocks are rounded as is shown in Fig. 5.3.1. The use of
rounded corners has the advantage that small overlaps of two corners do not lock up the
total mechanical system. The use of rounded corners is justified by the fact that such
corners are easily crushed under such circumstances.

,
,---- .....
F F
/
~F ' ' \V
I
I 'I \ \I
, 1----7>
I 1 f m
I I m
\
, I •
' '~ /

---~--
!.t.w
.t.~r- _ _ _ t~
3 K K
_I_ _I_ -
'
t=O t =.t.t

Fig. 5.3.2 Illustration of the explicit calculation procedure in UDEC

Solution procedure
UDEC uses an explicit time-marching'solution procedure. In this explicit approach for
each time step the equations of motion for each separate element are solved neglecting the
boundary conditions for the displacements. This is illustrated in Fig. 5.3.2. At time t=O,
rigid block 1 with mass m is loaded with force F. Application of the law of motion gives
an acceleration a=F /m. This results in a displacement increment !::. w at the end of time
increment !::..t. Block 1 is connected to block 2 by 2 elastic springs with stiffness K. This
means that the displacement !::..w results in 2 reaction forces !::..F = K!::..w. In the next time
increment !::..t, the law of motion for block 1 is solved again taking the 2 reaction forces
!::.F into account. This is also done for block 2. In this way information travels through
the structure. The basic calculation cycle is shown in Fig. 5.3.3. The law of motion and
the constitutive laws are used alternating for each element for each time step. This
procedure is not restricted to rigid blocks. The application of the constitutive laws can
also include the application of constitutive laws for the continuum elements.

73
 (for each eleaent for each tiae step) Displaceaents

Fig. 5.3.3 Basic calculation cycle in UDEC

Time step
For the explicit formulation used in UDEC, numerical stability and convergence to the
exact solution are found only when a. small enough time step is chosen. An algorithm is
applied in UDEC which calculates a time step based on the speed of propagation of
information through the mechanical system. The time step is chosen sufficiently small, so
that information can not travel further than one structural element during a time step.

Quuisto.tic problems
UDEC also can solve quasistatic problems by following the nonlinear behaviour with
small increments. Then, the time step is still used, but merely as a means of
incrementation to reach the final solution. The kinetic energy is removed from the
mechanical system by adding an artificial damping. UDEC uses an adaptive method
dependent on the change in kinetic energy in the system that guarantees a critical
damping. This method was developed by Cundall [1982], who showed it to be effective in
several cases of structural failure.

Ezplicit method tlem&S implicit method

For many problems, for instance linear-elastic problems, the implicit method of solving
problems in most finite element methods is much more effective than the explicit method
used in UDEC. This is due to the small time step needed in UDEC to obtain numerical
stability. The explicit method of UDEC becomes competitive when
1. the problem is highly nonlinear,
2. the constitutive behaviour is path dependent and
3. there are changes in connectivity of the elements.

74
The small calculation increments in the explicit method used in UDEC make it easy to
follow highly nonlinear or path-dependent behaviour. In the implicit method this can
lead to a dramatic increase in calculation time, because of the increase in increments and
iterations needed to find a satisfactory convergence to the exact solution. This also
happens in the implicit method when the physical or geometrical changes make it
necessary to rebuild and invert the total structural matrix.

5.4 CONSTITUTIVE MODEL DESCRIBING INTERFACE BEHAVIOUR

A new constitutive model was formulated because no interface models were available in
UDEC describing failure due ~o general combinations of tensile and shear loading,
including softening for both (Vonk [1990]). The model.was formulated with the intention
of giving a simple description of the essentials of crack behaviour at microlevel to be used
in the micromechanical simulation&. The model was kept relatively simple for the reasons
that
1. many details modelled at the microlevel will probably hardly be recognized on
the macrolevel due to an averaging process. It was thought better to make the
model more complicated when the results of the simulations required it.
2. many details of crack behaviour at microlevel are not well known or not
known at all.
3. it reduces calculation time.

~
,
....
............
__ _.x:::.'llllo.o~~;

-.... ........ .... __

~1
....

--
c

a,

Fig. 5.4.1 Initial failure surface

75
Failure surface
The model is formulated linearly throughout. The initial failure surface is shown in Fig.
x
5.4.1. Tensile failure is governed by a tensile strength ft and a reduction angle to take
into account the presence of a shear stress. Shear failure is governed by a Mohr-Coulomb
criterion applying a cohesion c and a friction angle 'P· No failure is modelled for pure
compressive loading.

Eltud.ic tleformatitm.s
Until the failure surface is reached, the behaviour is assumed to be linearly elastic,
depending on the two different moduli of elasticity K 1 and K2 for normal and shear
deformations, respectively. No coupling between the two directions is modelled in the
elastic range. In the micromecha.nical simulations the moduli of elasticity are chosen very
high to keep the interface deformations negligible until failure starts.

--- --- -- _ ....

....

---- _ __ _
.... .....

Fig. 5.4.2 Shrinkage of the failure surface due to softening

Softming
Mter the failure criteria are met, linear softening is modelled both for tensile and shear
loading by reducing the tensile strength and the cohesion (Fig. 5.4.2). A plasticity
formulation is used, which means that the deformations during softening are irreversible
and unloading takes place according to the moduli of elasticity. The two kinds of
softening are coupled, assuming that both tensile strength and cohesion decrease at the
same time and at the same rate, depending on the damage to the interface. This results in
an isotropic shrinkage of the failure surface, which is shown in Fig. 5.4.2. Residual tensile
strength and residual cohesion are formulated as linearly decreasing functions of the
parameter K representing the state of damage. It varies from 0 in the undamaged state to
1 in the fully damaged state. It is assumed that K increases linearly depending on the

76
plastic deformations of the interface. The rate of increase differs between normal and
shear deformations. Pure tensile softening ends after a plastic tensile deformation w 1111ax,
whereas pure shear softening ends after a plastic shear deformation w2max· This is shown
in Fig. 5.4.2. Simultaneous softening due to tensile and shear deformations is treated as a
linear combination. Completed softening sets the tensile strength and the cohesion to 0,
which means that a pure friction model is left. During shear softening or frictional sliding
it is assumed that the interface opens with a dilatancy angle '1/J.

Formulation of the pltuticity model

In the plasticity formulation of the interface behaviour the incremental deformation
vector Aw is decomposed into an elastic and a plastic component, which gives

Aw= Awe+ AwP (5.4.1)

The elastic incremental deformation vector Awe is related to the incremental elastic
stress vector A ae by

Aae= K Awe (5.4.2)

where

(5.4.3)

Starting from an initial stress state trJ (Fig. 5.4.3), first an elastic trial stress state ot is
calculated by

ot = trJ + K Aw (5.4.3)

Yielding is detected by testing the trial stress state ut in the yield function f ( u,"), which
depends on the loading history by the damage parameter "· Yielding is found when

f ( ot,"O) >0 (5.4.4)

This means that the trial stress state falls outside the yield surface for the present damage
state "o (Fig. 5.4.3). The yield surfaces used in the model described here are

ft = u1 + u2 tgx- (1-") ft = o (5.4.5)

77
 ..... _
__ __
~'--- .}::::.
-- ---

Fig. 5.4.3 Correction of elastic trial stress states

9t=O a,

Fig. 5.4.4 Plastic potential surface with plastic strain vectors

for tensile failure and

f2 = Ut tg!p + U2- (1-~t) C = 0 (5.4.6)

for shear failure. A stress state is not allowed to fall outside the yield surface, but has to
stay on the yield surface after it is reached. This is achieved by satisfying the consistency
condition

df= 0 (5.4.7)

during yielding. The plastic deformations are determined using a nonassociated flow rule

78
(Fig. 5.4.4), which reads

8g
.:lwP =A- (5.4.8)
00:

Here, g is a plastic potential function. In this model the plastic potential functions are

(5.4.9)

for tensile failure and

g2 = O't tg.,P + 0'2- (1-") c = 0 (5.4.10)

for shear failure. The unknown multiplier Ain equatio~ 5.4.8 can be calculated by
satisfying the consistency condition in equation 5.4. 7 during yielding. As a result the final
damage "' and the final stress state ut can be calculated.

De Borst [1986] describes a rigorous method which puts the trial stress state directly back
on the yield surface. Using this method, it is not necessary to determine the intersection
point of the stress vector with the yield surface and further satisfy the consistency
condition. An expansion of the yield function f ( u,") in a Taylor series using only linear
terms leads to the direct relation

f(ut,"O) 8g
u' = ot- ---------...---------__,....---- K - (5.4.11)
!.!_ [!!:__] T ~ +
0" OWP 00'
[!..:]
00'
TK og
00'
ou

The method gives an exact solution when a model is formulated linearly throughout.

Combined softening occurs when the trial stress state falls inside the shaded area of Fig.
5.4.3. Then the same method as for regular softening is used by decomposing the plastic
deformation vector into two components perpendicular to the two plastic potential
functions (Fig. 5.4.4) and treating the case as a linear combination of both types of
softening (Vonk [1990]).

The linear softening for both tension and shear, as described previously, is represented by
the linear softening rule

79
 E~w 1 Pt E~w 2 P
I'>=
Wtmax
+ W:~max-
< 1.0 (5.4.12)

A superscript t is used for the plastic incremental normal deformation ~w 1P to indicate

that only the component due to tensile softening is taken into account and not the normal
component due to the dilatant behaviour of the interface during shear softening.

5.5 SIMULATION OF SOFTENING WITH UDEC

In UDEC a deformation controlled simulation can be carried out by

1. applying the required deformation in large steps.
2. applying a constant velocity.
The last method was chosen because the material behaviour is path depen~ent. Also, it
was found that the first method causes significant dynamic effects resulting in undesired
damage. The application of a constant velocity means that a perfect state of equilibrium
is virtually never found. UDEC gives an indication of the dynamic effects by continuously
monitoring the maximum unbalance force in the mechanical system. During the micro-
mechanical simulations this unbalance force was always very small compa~ed to the total
force in the system. Only incidentally was it more significant.

(a)

(b)
(c) (d)

Fig. 5.5.1 Simple structures used for simulation of softening with UDEC

Tensile •oftenin!J
Four calculations were carried out to investigate the ability of UDEC to simulate
softening and localization of deformations. In these calculations the applied boundary

80
velocity, the time step, the elastic stiffness, the fracture energy and the size of the
specimen were chosen in the same order of magnitude as in the micromechanical
simulations. A simulation of tensile softening was carried out for the specimen in Fig.
5.5.1a consisting of two blocks separated by one interface. The result is shown in Fig.
5.5.2. It shows the dynamic character of UDEC. Information is transmitted through the
mechanical system by means of stress or deformation waves. The automatically applied
damping in UDEC keeps the dynamic energy in the system low. It can be reduced further
by decreasing the velocity of the boundaries. Due to the dynamic effects, the total energy
dissipation was found to be 2% higher than the theoretical value for the fracture energy.

o~--~--._--~--~--~
~~--~--------~~--~
0.000 0.005 0..0 10 0.015 Q.020 ().()25 0.002 0.003 0.004 0.005 0.005
(a) Oef<:rmaticn Cotml (b)

Fig. 5.5.2 (a) Result of a simulation of tensile softening with UDEC

and (b) detail around peak stress

Locolization in tension
The ability to handle localization of deformations was shown by the second simulation of
tensile failure of the specimen with two interfaces in series (Fig. 5.5.1b). Only the upper
interface fractured, while the lower interface remained almost undamaged. The fact that
the upper interface fractured was to be expected, because the stress waves are initiated at
the upper boundary of the specimen. The simulation showed that UDEC is sufficiently
selective to make only the upper interface fracture. The fracture energy was 1% higher
than the theoretical value. This was due to a decrease of the dynamic effect. The size of
the stress waves halved with respect to those in the first simulation. In the calculations
with the micromechanical model the dynamic effects were also found to be much smaller.
The simulation of the first tensile test in this paragraph is a critical case due to its
homogeneity and symmetry.

She&r MJjtening
Two simulations of shear softening due to unia.xial compressive loading were carried out
too. The specimens are shown in Figs. 5.5.1c+d. The simulations gave qualitatively the

81
same results as those for tensile softening. The fracture energy for shear loading was found
to be approximately 1% higher than the theoretical value and localization of deformations
was found in the upper interface. This shows that the dynamic approach of UDEC
automatically takes care of localization of deformations without the necessity to carry out
special analysis like the bifurcation analysis required in implicit finite element programs
(De Borst [1986,1989b]). It is hard to prove that localization of deformations is always ·
predicted correctly by UDEC. The simulations presented in this thesis always show
realistic results. When no perfect localization is found, but the cracking is more diffuse,
this can be explained by the heterogeneity of the material.
-50 -50
- clt=2.906 - v=-0.2mmlsee

-40 x10"'seo -40 -- v=-OAmnlsec

I -30
clt=1.453
x10"'seo I -30
- - v=-D.!lrrmlsee

I !..
-20 -20
~
J -10
~ -10

oL---~--~--~----~~ 0
o.o -0.1 -02 -o.3 -o.4 -o.s 0.0 -0.1 -02 - o.:a -0.4 -0.5

(a) Deformation [mm) Deformation [mml (b)

-50

- max=3.0rrm

I -30

Fig. 5.5.3 Influence of (a) time step, I -20

~
(b) boundary velocity and (c) maximum
~
size of finite elements on simulations -10

with the micromechanical model

-0.1 -02 -o.:a. -0.4 -0.5

Deforrnat•cn [~TYT~l (c)

Softening simulated.fllitA the micromech.tl.nit!al mod.el

To determine the influence of some UDEC calculation parameters on the micromechanical
simulations, a sensitivity study was carried out for the simulation of a uniaxial
compression test. Fig. 5.5.3a shows that the influence of halving the time step, which was
automatically determined by UDEC, has a negligible influence. Varying the boundary
velocity does not influence the pre-peak behaviour, as shown in Fig. 5.5.3b. Some

82
influence is found for the post-peak behaviour when the boundary velocity is increased
from -0.4 mm/sec to -0.8 mm/sec. The influence of an increase from -0.2 mm/sec to
-0.4 mm/sec is smaller. The slight differences in stress-deformation behaviour are a
result of slight differences in the crack formation process. Fig 5.5.3c shows that the
simulation& are quite sensitive to the maximum size of the finite elements or zones in
UDEC. A reduction in maximum size from 7.0 to 5.0 mm has a significant influence on
the stress-deformation behaviour of the specimen; The influence of a reduction from 5.0
to 3.0 mm is less so.
-50
Single precision

-40 Dolble precision

;;-
E
! -30
$
~ ' \

iii .,-20 \
c: \
'
~
\

-10 '

-0.1 -0.2 -0.3 -0.4 -0.5

Deformation [mm]

Fig. 5.5.4 Influence of numerical precision of UDEC calculations on the average

stress-deformation relation of 4 simulation& of a uniaxial compression test

Choice of UDEC parameter~~

It was decided to carry out the micromechanical simulation& with a time step
~t=2.906x10·6 sec, a boundary velocity v=-0.4 mm/sec and a maximum finite element
size of 5.0 mm. For the pre-peak behaviour a velocity of -0.8 mm/sec was regularly used
to shorten the calculation time. It was not possible to choose smaller values because of the
limitations in computer time. A complete simulation of a compressive softening test
required 320000 to 400000 calculation steps ("time" steps). This took 4 days CPU time on
an Alliant FX2600 minicomputer. Because of the large number of calculation steps, it was
feared that the cumulation of numerical errors would significantly influence the
calculation results. Therefore UDEC was changed to make the calculations in double
precision. Fig. 5.5.4 shows the influence of the numerical precision on the average
stress-deformation relation of 4 simulations.

83
 CHAPTER6

PROCESS OF SOFTENING

6.1 INTRODUCTION

The concrete softening process will be analyzed in this chapter. First of all, some
experimental results will be discussed in Chapter 6.2. The chapter will concentrate mainly
on uniaxial compressive loading, but the results of a triaxial compression test will also be
presented for comparison. Results of micromechanical simulations will then be given.
Chapter 6.3 discusses simulations of some tensile tests. These simulations have been
carried out to show that the micromechanical model can also describe this type of failure.
The process of compressive softening is then discussed in Chapter 6.4 in a detailed
investigation of one simulation of a uniaxial compressive test. The numerical results are
compared with experimental results. The conclusions are drawn in Chapter 6.5.

The micromechanical simulations were carried out with one set of parameters. In view of
the difficulty of calibrating the micromechanical parameters, a sensitivity study will be
carried out in Chapter 7 in which the results of the micromechanical simulations will be
investigated with regard to variations of the material parameters. There it will be shown
that the results of the model are relatively insensitive to the variation of most material
parameters.

6.2 DISCUSSION OF A COMPRESSION TEST

In this chapter results of compressive experiments will be discussed, which are

representative of tests on cubical specimens of 100 mm whose loading platens only
introduce a slight restraints to the lateral deformations. These experiments htclude tests
with short brushes, long brushes and teflon platens. Mainly the results of uniaxial tests
will be discussed. For comparison, the results of a triaxial test will be summarized.

Stress-dejormatioa relo.tion
Fig. 6.2.1 shows the relation between the axial stress and the axial deformations measured
by means of LVDTs and strain gauges for test BSl-1 with short steel brushes. In the

84
pre-pea.lt region, the overall deformations measured by the LVDTs are in good agreement
with the surface deformations measured by the strain gauges. At peak stress, the surface
deformations start to lag behind the overall deformations and eventually even decrease.
This points to localization of deformations, which develops gradually, indicating a process
of progressive macrocrack growth.

-40

I -30 - - LVOTs

I -20
- - - stratn
gauges

J -10

-1 -2 -3 -4 -5 -6 -7
Nominal strain [o/oo]

Fig. 6.2.1 Relation between the axial stress and the axial deformations measured
by means of LVDTs and strain gauges for test BS1-1

-50

;;"'

! -30
--left

- - - back
!
IJ)

-20 --- --- -- na"~t

J -10
· front

-4 -5
Nominal strain [o/oo)

Fig. 6.2.2 Separate strain-gauge readings for test BS1-l

85
Fig. 6.2.2 sho,ws the 4 separate strain-gauge readings. The strain gauges were glued to the
4 sides of the specimen as shown in Fig. 3.2.1. Deformations are quite uniform up to peak
stress. At peak stress the strain-gauge readings deviate. The pattern becomes chaotic.
Some strain gauges show further deformation, others a decrease in deformation. This can
be explained by a random process of macrocrack growth, which causes a redistribution of
forces and the localization of deformations. When the macrocracks start to grow, it is just
a matter of chance whether a strain gauge is glued to a piece of concrete which is
unloaded or on a piece of concrete which is further loaded.

Lateral deformations
The average lateral deformations as a function of the average axial deformations are
shown in Fig. 6.2.3. The first part of the curve is governed by the Poisson ratio, which
results in a decrease of the volume of the specimen. Then a gradual change to a dilatant
behaviour is found. Macrocracks are formed and start to shear off, thus causing the
dilatant behaviour. After the gradual change, a more or less constant dilatant behaviour
is found, which indicates that the process of macrocrack formation has stopped and the
final crack pattern has been developed.

2.5

E' 2.0
.s
!5
1.5
j
~ 1.0
iii
lii
o;
...J 0.5

0.0 L-......::;;;.;a;;;==:::::::::::i...__--1._ __,___ __,'--_....J__ ___j

0.0 - o. 1 - 0.2 - 0.3 - 0.4 - 0.5 0.6 - 0.7

Axial deformation [mm)

Fig. 6.2.3 Relation between axial and lateral deformations for test BSl-1

The separate clip-gauge measurements of the lateral deformations are shown in Fig.
6.2.4. Three clip gauges were used over the height of each side of the specimen. As in the
case of the strain gauges, a significant scatter in the measurements is found after peak
stress has been passed. The unrestrained lateral boundary conditions, combined with the

86
stochastic crack process, cause significant scatter in the lateral deformations. An
important part of this scatter is due to the fact that the specimen fails from the bottom to
the top, as is shown in Fig. 6.2.5. Such indications of nonuniform failure were found for
most tests. The specimens showed failure from top to bottom, left to right and front to
back. No dominant shear failure in one direction was found. The failure mechanisms were
always more or less axial symmetric.

2.5

2.0

1.5

1.0

0.5

0 .0 -===---'----...J....__ _J....__
L - -........ __jL____ _j

0.0 -0.1 -02 -0.3 -0.4 -0.5 -0.6 -0.7

Axial deformation [mm)

Fig. 6.2.4 Separate measurements of lateral deformations for test BS1-1

3 .0

2.5 Bottom

Ic 2.0
Middle
T op
0
~
~Q)
1.5
"0
lij 1.0
lii
AI
...J
0 .5

0.0 '------""'"""-===---....l....--...J...._- - J...._-___j_ __ j

0.0 - 0. 1 - 0 .2 - 0.3 - 0.4 - 0.5 - 0 .6 - 0 .7

Axial deformation [mm)

Fig. 6.2.5 Distribution of the lateral deformations over the height of the
specimen in test BS1- 1

87
From the available measurements it could not be concluded that the specimens failed
from the outside to the inside. This failure mode has been reported by some researchers
(Van Mier [1984,1986ab), Shah and Sankar [1987)). Sometimes it is used to explain
softening of a specimen. It is thought to be due to a triaxial stress state in a specimen,
caused by the restraint of lateral deformations by the loaded boundaries. Then softening
is the result of a decrease in size of a more or less intact core. This mechanism certainly
plays an important role in tests with a high lateral boundary restraint (Hudson et al.
[1971)), but it is not clear whether it also plays an important role in tests with a much
smaller boundary restraint (Van Mier [1984,1986ab), Shah & Sankar [1987)). In these
last-named cases, the phenomenon of crack propagation from the outside to the inside
can also have been caused by nonuniform shrinkage stresses in the specimens.

(b)

Fig. 6.2.6 (a) Surface crack patterns for test T1-1 (£=-{l.0064) and (b) conical pieces
of a specimen which has failed in compression

Cra.cl: patterru
Surface cracks recorded after a test with teflon platens are shown in Fig. 6.2.6a. The
derived crack patterns are irregular and often seem to be discontinuous. It is not easy to
recognize that the specimen is split up into conical pieces. This is more clearly seen when
a heavily fractured specimen is pulled to pieces by hand. Some of the pieces are shown in
Fig. 6.2.6b. The formation of these pieces was earlier reported by Stroeven [1973) and van
Mier [1984).

A horizontal and a vertical cross-section showing crack patterns recorded by ultraviolet

photography after a test are given in Fig. 6.2.7. The amount of cracking found by means
of this technique is surprisingly great. The photographs in this thesis show this only

88
partially, because a significant amount of information is lost due to the presentation
technique used. Much more detail is revealed when the crack patterns are investigated by
direct inspection of the specimens under ultraviolet light or by inspection of enlargements
of p.nt>lGJtl'al~ru! ~or~~m

{b)

Fig. 6.2.7 Internal crack patterns recorded by ultraviolet photography:

(a) horizontal cross section {bottom, test Tl-1, E=-D.0064) and {b) vertical cross-
section {middle, test Tl-2, E=-D.0069)

The horizontal cross-section shows the large number of pieces into which the specimen is
split up. Many bond cracks are found, whereas the number of aggregate cracks is small.
The cross-section of the conical pieces varies from the size of one aggregate to
approximately 3 times the maximum aggregate size. Inside the conical pieces, identified
by the continuous wide-open cracks, secondary less-open cracks are often found. It is
difficult to determine whether they are continuous or not.

Failure mechanism
The vertical cracks in the vertical cross section of Fig. 6.2. 7b are most visible and
dominate the picture of the crack pattern. The inclined cracks have often opened less than
the vertical cracks and are less visible. Parts of the inclined cracks which carry load are
closed and do not show up in the picture. This makes identification of the failure
mechanism difficult. The continuous macrocrack pattern determining the failure
mechanism is schematically illustrated in Fig. 6.2.8. It has been determined after close
scrutiny of the crack pattern in Fig. 6.2. 7b. Detection of this pattern is also hindered by
the spread in cracking. This spread is probably for an important part due to the shearing
of the irregular shaped cracks. This causes anonuniform stress transfer over the length of
the cracks, which introduces complex stress patterns in the concrete around them and

89
results in the formation of secondary cracks. In particular, ma.ny vertical splitting cracks
a.re found because their formation ta.kes less energy tha.n the formation of inclined shea.r
cracks. The phenomenon of formation of secondary cracks due to shearing of cracks ha.s
also been reported by Bieniawski [1967] a.nd Va.n Mier et al. ,[1991].

i\\
I
\\ I
I
I \ 1 I
I \ '\ I
I \ I , /
I \I ' \/
I V ' Y
1 I ' I
4 ( ~\ {
I\ ( I , 1\
1\ I 1 1 '11\
I\ I 111\
I \ I I I I \
I \ I I 11 \
~\ I \ I I I I \
.J \ ! 1 I! '
Fig. 6.2.8 Schematic failure mechanism of Fig. 6.2. 7b

The inclined cracks and not the vertical cracks a.re essential to the localization of
deformations. Investigation of the cra.ck pattern in Fig. 6.2. 7b shows that a. limited
number of inclined cracks is found on a. vertical line, that is the direction of compressive
loading. The number of cracks is small, but not always equal to one. This indicates a
localization of deformations, but not a. perfect one, which would mea.n one inclined crack
found in the direction of compressive loading.

Tri4Zi4l test
In test series 1, the boundary condition tests, a series of tria.xial tests wa.s also carried out.
In these tests, 100-mm cubic specimens were loaded axially in deformation control. In
this direction the maximum compressive stress wa.s applied. In the two lateral directions,
stresses were applied equal to 30% a.nd 5% of the axial stress in the first direction. In this
chapter the results of the tests with the short brushes are presented. These results are
representative of the triaxial tests with loading platens which cause a slight restraint to
the lateral deformations of the specimen.

Result,
The average results of the uniaxial a.nd triaxial experiments with the short brushes a.re
given in Fig. 6.2.9. The strength doubles and the ductility increases dramatically due to
the fact that the specimen is confined by two compressive stresses, the smallest of which

90
is only 5% of the largest compressive stress. The cracking process is slowed down
considerably by the confining stresses. The state of loading causes a two-dimensional
failure mode. The specimen is compressed in the direction of the largest stress and
expands in the direction of the smallest confining stress (Fig. 6.2.9b ). In the direction of
the intermediate stress the deformations remain relatively small. The surface-crack
patterns in Fig. 6.2.10a show the two-dimensional failure mode. The specimen shears off
due to the formation of a pronounced shear-crack pattern in the plane of the maximum
and the minimum stress applied. Fig. 6.2.10b shows almost the same internal crack
pattern as that found for the same type of test with teflon platens (test T3-2}. Some

-100r----------------------------------------,
.... ---- ....
....
....

' \

' \
....
,a a '

~·
m-:::-
./

.CD 0.3a
-20

. o- ~~'-·~~~~~~--~----------~------~
. ·p.b. ' ;~_ :t 't: 0.5 . -1 .0 - 1.5 - 2 .0
(a)
Deformation· [mm]

3.0 r-------------------------------------,/,-----,
/
/
I
2.5 /
/
/
I
2.0 I
I
/

1.5

1.0

0.5

0 .0 L.--o:::::c...::..::..__. L __ _ _ _ _ _ _ _..:L__ _ _ _ _ _ _ ___i__ _ _ _ _ _ _ __ j

0.0 -0.5 -1.0 - 1.5 -2.0
(b) A x ial deformation [mm]

Fig. 6.2.9 Comparison of the uniaxial and triaxial tests with short brushes

91
secondary cracks are found around the continuous shear cracks, but they have not been
well developed due to the presence of the confining lateral stress. The average angle of the
shear cracks to the direction of the maximum applied stress is approximately 20°.

Fig. 6.2.10 shows that shear failure should be taken into account in the description of
compressive failure. It has to be determined whether or not this failure mode loses
importance when a compressive stress state changes from triaxial to uniaxial. This should
be investigated in relation to a failure mode caused by the formation of splitting cracks.
This will be done both in the present chapter and in Chapter 7 with the micromechanical
model, which is able to simulate the formation of both splitting cracks and shear cracks.

{b)

Fig. 6.2.10 (a) Surface crack patterns for test BS3-2 (a=-31 N/mm2, E=-0.020)
and internal crack pattern for test T3-2 ( a=-18 N/mm2, E=-0.019)

6.3 SIMULATIONS OF TENSILE TESTS

Specimen and boundary conditions

In these numerical simulations specimens of 50x100 mm were loaded in uniaxial tension
subjecting the upper boundary to a constant velocity, while the lower boundary remained
at rest. No rotation of the upper and lower boundaries was allowed. The lateral
displacements of all the boundaries were kept unrestrained. Plane-5tress conditions were
assumed. The input for the mesh generator was chosen in such a way that a coarse
aggregate (4-16 mm) content of approximately 31% was simulated, as was present in the
concrete used in the boundary-condition tests. The composite structures and the interface

92
meshes used in the simulations are shown in Fig. 6.3.1. An example of the finite element
mesh used for the calculation of continuum stresses and strains is given in Fig. 6.3.2. The
maximum length of the sides of the finite elements was 5 mm.

(1) (2)

(3) (4)

Fig 6.3.1 Meshes used in the simulations of the tensile tests

Fig. 6.3.2 Finite element mesh of simulation 1 (see Fig. 6.3.1)

Material properties
The material properties used in the simulations of both tensile and compressive softening
tests are shown in Table 6.3.3. They were chosen so that the results obtained are close to
test results. Only very limited information is available in literature to determine their
values. There was no attempt to fit the results of the simulations more closely to test
results by adapting the material parameters, because this would have taken a considerable
amount of time and would probably not have resulted in a better insight into the process
of failure.

93
For these simulations it is interesting to note that the heterogeneity of the simulated
concrete is modelled by means of
1. different moduli of elasticity for mortar {Em=25000 N/mm2) and aggregate
{Ea=70000 N/mm2).
2. a tensile bond strength {ftb=3N/mm2, Cb=10N/mm2), which is 50% of the mortar
strength {ftm=6N /mm2, Cm=20N /mm2).

Continuum Interface
Aggregate lortar Bond lortar
E=70000 N/mm2 E=25000 N/mm2 ft=3.0 N/mm2 ft=6.0 N/mm2
c=10.0 N/mm2 c=20.0 N/mm2
v=0.2
K1=5x105 N/mm3 tgVJ=0.3
K2=5x105 N/mm3 tg,=O.O
Wtmax=0.02 IDDI tgx=O.O
li2max=1.00 mm

Table 6.3.3 Material parameters used in the simulations of tensile tests

The moduli of elasticity for the interfaces were chosen very high to make their elastic
deformations as small as possible. These deformations were approximately 1% of the total
elastic deformations of the specimen. The moduli could not be increased further, because
this would have had a negative influence on the stability of the UDEC calculations and
consequently on the calculation time.

ToW. stress-deformation relation

Fig 6.3.4 shows the stress-deformation relations found for the different meshes in Fig.
6.3.1. They compare well with test results. The pre-peak behaviour of the different
meshes is almost the same. Near the peak stress differences occur, which grow in the
softening regime. In the pre-peak regime the stress-deformation relation is determined by
the average response of the total volume of the specimen. Due to the localization of the
damage process, a gradually smaller part of the random internal structure of the specimen
determines its behaviour. This makes the stochastic variation in the heterogeneous
structure more apparent during softening. This aspect is magnified by the fact that the
model is two-dimensional. In a two-dimensional model, an aggregate of the maximum
size occupies a much larger portion of the cross section than in a three-dimensional
model. This means that this characteristic element of the heterogeneity of concrete has a
much greater influence on the specimen behaviour in a two-dimensional model than in a
three-dimensional model.

94
 - - - Sim. 1

Sim. 2

Sim. 3

- - Sim. 4

- - Average

OL-----~--~---L----~~~~~

0.00 0.02 0.04 0.06 0.08

Deformation [mm]

Fig. 6.3.4 Stress-deformation relations for 4 simulations of tensile tests

on the specimens shown in Fig. 6.3.1

Peak stress
Peak stress is found to be 3.7 N/mm2. The work of Riisch & Hilsdorf [1963] and Hordijk
[1989] shows that this is a realistic value for the type of concrete used in the present
research. The value obtained for peak stress shows it to be more strongly influenced by
the bond strength (3.0 N/mm2) than by the mortar strength (6.0 N/mm2). This can be
explained by the brittleness of tensile failure of the interfaces. As a consequence, the
weakest component has the greatest influence on peak stress.

Post-peak beluJviour
In the post-peak regime the well-known softening curve with the long tail is found. The
long tail is remarkable because it was not introduced in the material parameters.
Apparently, it is a result of the heterogeneous composite structure. The material
parameters were chosen so that localization in a straight-line crack parallel to the loaded
boundaries would have resulted in the end of the softening process at a deformation of
0.02 mm. In Fig. 6.3.5, the final crack patterns for the different composite structures are
shown. The cracks are tortuous and show overlaps. Comparison of the stress-deformation
diagrams and the macrocrack patterns reveals that the post- peak stress is greater when
the crack pattern is more tortuous and shows more overlaps. When no crack overlaps are
found, as in simulation 1, no tail is found in the stress-deformation relation. When the
crack overlaps grow in size and number, a more significant tail is found. The tail is most
pronounced for simulation 4, which shows two of the largest crack overlaps.

95
(1) (2)

(3) (4)

Fig. 6.3.5 Final crack patterns (interfaces with an opening>0.02 mm) for the 4
simulations of tensile tests

Softening and heterogeneity

The heterogeneity of the material has a significant influence on the fracture energy. When
it is assumed that a crack through the specimen consists of 50% bond cracks
(Gr=30Jfm2) and 50% mortar cracks (Gr=60J/m2), which is close to the results of the
simulations, the tensile-fracture energy for a straight-line crack can be estimated to be
45 Jfm2. By integrating the stress-deformation curves in Fig. 6.3.4, fracture energies of
54 up to 98 Jfm2 are found. The last value is lower than the actual value because the
integration could only be carried out up to 0.08 mm, where the simulation was
terminated. The calculated values show that modelling the heterogeneity of the material
increases the fracture energy substantially. This indicates that concrete behaviour can be
described realistically by brittle behaviour at microlevel, when the heterogeneity is
modelled down to a sufficiently low scale level. The results of the brittle truss model of
Zubelewicz [1987] and the brittle lattice model of Schlangen & Van Mier [1991ab] also
point in this direction.

Gr0111f.h of damage
Fig. 6.3.6 shows the growth of damage in simulation 4. The damage is represented by the
damage parameter K. of the constitutive model (see Chapter 5.4). It varies from 0 in the
undamaged state to 1 in the fully damaged, which means fractured, state. The thickness
of the lines in Fig. 6.3.6 indicates the value of K.. The thinnest lines indicate a K. value of
between 0.0 and 0.2 and the thickest lines indicate a value of K. of 1.0.

96
Due to the heterogeneity, damage is already found in the bond interfaces for small
deformations {Fig. 6.3.6a). Gradually the damage zones grow along the aggregate
boundaries, but stay distributed {Fig. 6.3.6b ). The higher strength of the mortar prevents
the damage zones in the bond interfaces from connecting through the mortar. At peak
stress the overall damage has grown to such a level that the damage zones connect
through the mortar and become continuous. There the localization of damage and
deformations starts. The state of damage just past peak stress is shown in Fig. 6.3.6c. The
final damage pattern is shown in Fig. 6.3.6d. The heterogeneity of the material causes
some spread in damage. Forks and overlaps are found in the crack pattern. This makes
the actual length of the crack pattern much greater than the width of the specimen, which
explains the increase of the fracture energy for the specimen due to the heterogeneity.
Modelling the material on a lower level will increase the length of the cracks further and
introduce more forks and overlaps on a lower scale level. Making the material more
heterogeneous by increasing the difference in bond and mortar strength will widen the
zone of damage, thus also making the material more ductile.

(a) (b)

(c) {d)

Fig. 6.3.6 Damage found in simulation 4 for a deformation of

(a) 2.9 pm, {b) 5.8 pm, (c) 11.6/1-m and {d) 86.7 pm

Crack groVJth
The process of cracking can be followed more easily, when only the stress-free cracks
{~t=1) are taken into account {Fig. 6.3.7). Due to the heterogeneity of the material the
macrocracks are initiated at several places in the specimen (Fig. 6.3. 7a+ b). These cracks
grow in each others direction, but also avoid each other. Simha et al. [1986] showed that
two approaching cracks in an elastic medium will always avoid each other. After the

97
crack tips have passed each other (Fig. 6.3.7c), a bending mechanism is created, which
makes one crack-tip bend in the direction of the other crack (Fig. 6.3.7d). This
mechanism, illustrated in Fig. 6.3.8, is very ductile and causes the long tail in the
stress-deformation. relation.

(a) (b)

(c) (d)

Fig. 6.3. 7 Macrocrack patterns in simulation 4 at a deformation of

(a) 21.7 JC!Il, (b) 28.9 JC!Il, (c) 57.8 JCII1 and (d) 86.7 JCIIl

Fig. 6.3.8 Failure mechanism for simulation 4

Comparison with. tests of Van Mier {1991}

The results of these simulations correspond well to recent test results of Van Mier
[1991b], who investigated macrocrack patterns in uniaxial tensile tests with the
impregnation technique developed in this research. His tests show that only few
continuous macrocracks are found at the beginning of softening. He reasons that the first
steep drop in the softening diagram is due to isolated debonding of coarse aggregates as
described here by the micromechanical model. His tests show further that, even at a
deformation of 50 JC!Il, the macrocrack does not go completely through the specimen. At a

98
deformation of 100 Jml, the macrocrack has passed completely through the specimen, but
is still discontinuous, due to overlaps between the macrocracks. Fig. 6.3.9 shows shows a
crack overlap and a crack fork found by Van Mier [1991]. Van Mier finds a strong
correlation between the stress transfer in the tail and the maximum aggregate size of the
concrete. His experiments show that the maximum size of the bridges is determined by
the maximum size of the aggregates and thus determine the stress level in the tail of the
stress-deformation curve. This observation is supported by the micromechanical
calculations, which show that the stress transferred in the tail depends on the size of the
bridges, which is determined by the internal structure of concrete set by the large
aggregates.

Fig. 6.3.9 Crack overlap and crack fork in concrete loaded in tension (Van Mier [1991b])

6.4 SIMULATION OF A COMPRESSION TEST

Specimen and boundary conditions

A specimen of 100x100 mm is loaded in uniaxial compression giving the upper boundary a
constant downward velocity, while the lower boundary is given a zero velocity. The
lateral displacements of the boundaries are kept unrestrained. No rotation of the loaded
boundaries is allowed. Plane stress conditions are assumed. The input for the mesh
generator was chosen so that the type of concrete used in test series 1 (boundary-

99
condition tests) was simulated. The input was the same as that used for the simulations
of the tensile tests. The composite structure with the mesh of interfaces is shown in Fig.
6.4.la and the finite-€lement mesh is shown in Fig. 6.4.lb. This simulation will be
referred to further as simulation Tl.

(a) (b)

Fig. 6.4.1 (a) Composite structure with interfaces and (b) finite-€lement mesh used in the
simulation of a uniaxial compression test (simulation Tl)

Material properties
The material properties are shown in Table 6.4.2 and Fig. 6.4.3. They are the same as
those used in the simulations of the tensile tests in chapter 6.3. After some trials, these
parameters were found to give reasonable results. The sensitivity of the results of the
model with respect to their value will be investigated in Chapter 7.

Continuum Interface
Aggregate lortar Bond lortar
E=70000 N/mm2 E=25000 N/mm2 ft=3.0 Nfmm2 ft=6.0 N/mm2
c=10.0 N/mm2 c=20.0 N/mm2
v=0.2
K,=5x105 N/mm3 tgp=0.3
K2 =5x105 N/mm3 tg;=o.o
litmax=0.02 mm tgx=O.O
li2max=1.00 mm

Table 6.4.2 Material parameters used in the simulation Tl of compression test

100
 a
-... ......
......
...... ......
r---
tg
-q1=;.3r.........
...
c,.,=20N/m~
......
... _'I"
I
I
c,.,=20N/mm2

c., =1DN/mm2 t - - - -
c., =10N/m~ ""'-I
I I ftm=6N/mm2
I I
I I f1b=3N/mm2
a, w
ftb=3N/mm 2 ftm=6N/mm2 w1mox= 0.02 mm w2mox=1.0mm

(a) (b)

Fig. 6.4.3 Parameters for interface behaviour of Table 6.4.2

StrW~-tkjormation relation
Fig. 6.4.4 shows the axial stress-deformation relation and Fig. 6.4.5 shows the relation
between the axial and the lateral deformation. In both figures, the results of tests with
teflon platens on 100 mm cubes are shown for comparison. The tests (T and MM, see
section 3.5) with teflon platens have been chosen because those platens cause the least
restraint to the lateral deformations of a specimen during softening.

Qualitatively, the predicted behaviour corresponds well to the response found in

experiments. The most important difference, besides slighter differences in pre--peak

-50

,.-- .....
-40
.... ''

~
~
~
Q)
-30
"' \ \
\
' \

\
',,
\

' \
Simulation
T1
Tests T
.t; \ \
1/)

\
\

lij -20 \
Tests lvl'v1
c:
E ' \
\

~
- 10 "" '
' -.......::

- 0. 1 - 0.2 - 0.3 - 0.4 - 0.5

Deformation [mm]

Fig. 6.4.4 Axial stress-deformation curve for simulation Tl

101
 2.5

E 2.0 I
.E. I
c Simulation
0 11
T1
~ 1.5
;'/
~ 11
Tests T
.EIll
0 1.0
)' Tests Mv1
11
tii
ffi
n; tl
I'
.J 0.5
//
/-'
/

0.0 L - - - = " = = - - ' - - - - ' - - - - - L . . __ __j

0.0 - 0. 1 - 0.2 - 0.3 - 0.4 - 0.5

Axial Deformation [mm]

Fig. 6.4.5 Relation between axial and lateral deformation for simulation Tl

behaviour and peak stress, is that the predicted softening behaviour is more brittle than
what is found in tests. This will be discussed further in Chapter 7. Another striking detail
is the irregular shape of the softening branch. Like the scatter in the simulations of tensile
tests, this is caused by the localization of damage and deformations in the two-
dimensional model. This means that the stability of only a small part of the total
specimen determines a significant part of its average behaviour. When an aggregate of the
maximum size is torn loose, this has a significant influence on the behaviour of the
specimen. This can have a brittle character, as the steep drops of the stress-deformation
curve in Fig. 6.4.4 show.

Failure mode
Fig. 6.4.6 shows the predicted failure mode. Fig. 6.4.6a gives the interfaces with the
largest normal or shear deformations. Only deformations larger than 0.1 mm are given.
The thickness of the lines indicates the value of the deformations. The displacements of
the corners of the finite elements are given as vectors. Fig. 6.4.6b gives the stress-free
interfaces.

The specimen is split up into a number of pieces whlch shear off. Localization of
deformations is found. Each vertical line, the direction of the compressive loading, crosses
approximately one inclined shear crack. The number of cracks crossed is sometimes larger
than one due to the heterogeneity of the material, whlch causes some spread in the

102
 I
I
' ,,
\
/jl\_
! '-. . . .
' \
II I 1\ '
1/ \
\
I I
I ')· (

I\
/
'I'
f·
I
I I
I

\ I
I
I

'
I

\ I
' \_ \
I 1\
I /. I
I
\ I
) 'I
I
I
\I
,; L),
(a) I \I \I (b)

Fig. 6.4.6 Failure mechanism for simulation Tl: (a) interfaces with the largest normal
and shear displacements (>0.1 mm) combined with displacement vectors of the finite
elements and (b) stress- free interfaces

cracking. Additional cracks are found, which do not seem to be an essential part of the
failure mechanism. They were created before the final failure mechanism was formed or
were caused by stresses introduced by sliding of the crack faces.

Comparing the failure mode in Fig. 6.4.6 to the crack pattern for test BS1-1 in Fig.
6.2. 7b shows that the predicted number of cracks is significantly smaller than found in the
test. This results in the specimen being split up in a smaller number of pieces. It can be
that either the lateral tensile stresses due to the heterogeneity in the micromechanical
model are too low or that the assumed tensile fracture energy is too high. It must be
mentioned that the damage to the specimen is more distributed than is shown in Fig.
6.4.6a, but that this has not resulted in the formation of more stress-free macrocracks.
The reason can also be that the model is two-dimensional. The heterogeneity and crack
growth in the third dimension are not taken into account in the micromechanical model.
It is not unlikely that the formation of the crack pattern in the third dimension will cause
an extra amount of cracking, due to the fact that cracks are initiated and arrested at
more places in the third dimension in the specimen.

Fig. 6.4.7 shows the finite-element mesh in its actual deformed state. It can be seen that
the crack opening or shearing has assumed the size of the finite elements. This shows the
necessity of incorporating geometric nonlinearities into this kind of simulation.

103
(a) (b)

Fig. 6.4.7 (a) Finite element mesh in its actual deformed state
and (b) upper-right quarter of the mesh (u=-3.1N/mm2, w=-G.56mm)

Gr01Jith, of damage
The growth of damage in the interfaces is shown in Fig. 6.4.8. The thickness of the lines
indicates the magnitude of the damage parameter K of the interface model. Damage is
already found for a low stress level (Fig. 6.4.8a). The damage is found in the bond
interfaces, more or less parallel to the direction of compressive loading. It is initiated by
the combination of splitting and shear stresses caused by the composite structure of the
concrete. During further deformation the damage zones grow along the boundaries of the
aggregates. At 75% of peak stress (Fig. 6.4.8b) the damage zones start to extend through
the mortar, connecting the damage zones around the aggregates. The damage is still
distributed. At peak stress the growth of damage accelerates (Fig. 6.4.8c) The localization
of damage starts from this moment on. The final damage pattern is shown in Fig. 6.4.8d

Failure process
All stages in the failure process that were investigated are shown in Fig. 6.4.9. The failure
process, including localization of deformations, is illustrated best by the deformations of
the interfaces and velocities of the finite elements. These are shown in Fig. 6.4.10. Both
normal and shear deformations are shown, but the larger of the two is visible for each
interface. The thickness of the lines is proportional to the size of the deformations. It
should be noted that the scale of deformations is different for the different pictures in Fig.
6.4.10. Their size can only be compared within one picture. To simplify the discussion,
groups of interfaces, which deform significantly more than the other interfaces, will be

104
(a) (b)

(c) (d)

Fig. 6.4.8 Growth of damage: (a) w=-{).047 mm, (b) w=-{).093 mm,
(c) w=-{).0186 mm (peak stress),(d) w=-{).558 mm
-00 ,---------------------------. 3.0

--40
I 2.~

6
I -30 1 2.0 10

I -8
i
1.5

.! 1.0

f 0.~

1 2 3
0.0
-0.1 -02 -0.3 -0.-4 -M -0.6 0 .0 -0.1 -02 -03 -0.~ -0.5 -0.6

Axial dwformaticrl 1"""1

(a) (b)

Fig. 6.4.9 Stages in simulation Tl, which have been investigated

105
called cracks. They are actually fictitious cracks, which include microcracking and
stress-free macrocracks (see Chapter 4.3).

At stage 2, at 75% of peak stress, a number of small cracks is found which are isolated
and do not seem to influence each other (see Fig. 6.4.10). Yet, some cracks have started
to grow through the mortar and combinations of two bond cracks and one mortar crack
can be found. This corresponds closely to the observations of Hsu et al. [1963]. At stage 3,
at 90% of peak stress, the length of the cracks has increased considerably. They have
started to influence each other, but have not yet formed a continuous pattern. The cracks

(2) (3)

(4A) (4A)

Fig. 6.4.10 Interface deformations and incremental displacements of finite element

corners at several stages of simulation Tl. The numbers of the stages in this figure
correspond with those in Fig. 6.4.9.

106
 ·· ..·
.. -:.
·.-·. .... .
~·

(4B) : ~ ~· . .. ' .
.. · ·.·

--~--
. .----
- .... .- -....
. .. . . . .. : ': :~::_: -~:~
·.- ·.·: · . "_\

(5A) (5A)

· -: ·: .

· -·
. . .·"' · .
·_. _. : - ': "': ·
--- -
-----
- :"" '
(6) (6)
.· ..

Fig. 6.4.10 continued

107
are arrested by zones of higher strength, i.e. mortar and aggregate, or by cracks in other
directions. As in the simulations of tensile tests, crack overlaps are found. Overlapping
cracks seem to be quite stable, probably because it takes a considerable amount of energy
for these cracks to connect by means of a curved crack path.

At peak stress, some of the larger cracks connect in the upper-middle part of the
specimen and form a continuous pattern. The result is shown at stage 4A, just past peak
stress. This is the beginning of the growth of the final crack pattern in which the
localization of deformations takes place. The first steep drop in the stress-deformation
curve is due to the failure of the upper-right corner of the specimen, which is shown at
stage 4B. This corner is split off in a brittle mode. Further crack growth is mainly
determined by the two triangular pieces in the upper part of the specimen, which are
pushed downward. This causes the growth of two splitting cracks. The second steep drop
in the stress-deformation curve is due to the failure of the bottom-left corner of the
specimen, when the left splitting crack reaches this area. The result is shown at stage 5A.
The crack pattern is completed when the right splitting crack reaches the bottom of the
specimen at stage 6. This causes the last drop in the stress-deformation curve. The final
part of the softening curve is not due to crack growth, but to a decrease in resistance of
the contacts (shear softening) and a decrease in the number of contacts due to opening of
the crack pat tern.

Figure 6.4.9b shows that the lateral deformations accelerate during crack growth. The
ratio of increase of axial and lateral deformations becomes constant after the crack
formation has stopped and the failure mechanism has been completed.

Localization of deformations
It was shown above that localization of cracking and deformations is found in a uniaxial
compression test as the result of a continuous crack-growth process. Fig. 6.4.11 shows
what this means for the deformations of the pieces of concrete between the macrocracks.
Soon after the peak is passed, the deformations of these pieces decrease. The results
compare quite well with the measurements of strain gauges in real tests (Fig. 6.2.2). Only
in tests, some parts of the specimen show larger deformations after peak stress has been
passed. This can be due to the restraint of the lateral deformations by the loading platens,
which slows down crack growth in the specimen. No lateral boundary restraint has been
modelled in the simulations in this chapter. The subject of lateral boundary restraint will
be treated in Chapter 9.

108
 -50

"EE
~
-40

(/
..
\
'\ , ) I
,

"'
"'
QJ

~
-30 2
' :
. .i "i'
I
/'
I
\
11
0
- 20 i / (\
iii
c I /. I V
E ' ,,t..· 1'
""
~
-10
0
' '
0 L_--'------'"---'----'---'------_J
0 -1 -2 -3 -4 -5 -6

Nominal strain [o/oo]

Fig. 6.4.11 Overall and continuum deformations for simulation Tl

6.5 CONCLUSIONS

The main characteristics of the new micromechanical model are that

1. it is a two-phase model existing of large aggregates in a mortar matrix, with
different moduli of elasticity for the two phases.
2. the bond strength between aggregate and mortar is lower than the mortar
strength.
3. all damage is localized in a mesh of interfaces, which is determined by the
two-phase structure.
In this chapter it has been shown that a model based on these assumptions gives a
realistic description of the failure process of concrete loaded in tension and in
compression.

Tensile test
According to the model, tensile failure starts with microcracking of the interfaces between
aggregate and mortar. This Inicrocracking is distributed. First, the cracks grow along the
interface of mortar and aggregate. The cracking stays distributed, because the mortar has
a higher strength. When the microcracks penetrate right through the mortar, the total
specimen soon becomes unstable. Deformations and macrocracks localize in a zone with a
width of approximately the maximum aggregate size. Due to the heterogeneity, the

109
macrocracks grow through this zone starting from several places. When crack-tips miss
each other, overlaps are formed, which make further crack growth a very ductile process.
The micromechanical model shows that the tail found in the stress-deformation relation
is a result of these crack overlaps. This agrees closely with the test results of Van Mier
[1991] . It has been assumed that the tail in the stress-deformation relation is caused by
friction between the two crack faces (Hordijk [1991], Duda [1991]) . This is not confirmed
by the micromechanical model. Also, the crack-detection experiments of Van Mier
indicate that friction between the crack faces is not the cause of the tail. His pictures of
crack patterns show that crack faces in tensile cracks very soon lose all contact, so that
friction cannot be present any longer.

The results of simulations with the micromechanical model show that the fracture energy
of a material increases when it is more heterogeneous. It is likely that macroscopic
softening of concrete can be simulated well by modelling concrete as perfectly brittle at
the microlevel, when its heterogeneity is modelled to a lower scale level than was done for
the present model.

Compres8ion test
The pre-peak crack growth in a compression test is described by the micromechanical
model in the same way as by Hsu et al. [1963]. Microcracking starts at the interfaces
between aggregate and mortar, which make a small angle with the direction of
compressive loading. The microcracks grow along the interfaces, but stay distributed. At
75% of peak stress, the microcracking connects through the mortar. From this moment
onwards, cracking is accelerated and cracks grow significantly in size. At peak stress, the
cracks have grown to such a size that the specimen starts softening. The growth of the
final macrocrack pattern starts around peak stress. The first softening is mainly due to
macrocrack growth. Further softening is due the opening of the final crack pattern, which
causes a reduction of effective contacts and a decrease of crack resistance due to shearing.
The simulations show that localization of deformations is found in compressive softening.
It starts at peak stress when the growth of the final crack pattern starts.

llO
 CHAPTER 7

PARAMETER STUDY

FOR THE MICRO MECHANICAL MODEL

7.1 INTRODUCTION

In this chapter the sensitivity of the micromechanical simulations of compressive

softening will be investigated with respect to a number of variables. In Chapter 6, the
results of simulations with the micromechanical model with a, quite arbitrarily chosen,
set of parameters were presented. In this chapter the question will be addressed whether
and how much a variation in model parameters influences the results of simulations with
the micromechanical model and whether this affects the conclusions of Chapter 6.

Contents
The micromechanical model is a stochastic model, due to the random generation of the
composite structure. The influence of the stochastic variation in the composite structure
will be investigated and discussed in Chapter 7.2. A variation of the material parameters
will be investigated in Chapter 7.3. The influence of two other causes of softening will
then be investigated. In general, macroscopic softening is due to progressive failure caused
by a variation at microlevel of
1. stiffness
2. strength
3. initial stresses, deformations or damage.
Other models use a stochastic variation in strength as the main source of softening
.(Schlangen & Van Mier [1991a], Rossi & Wu [1991]). It was therefore interesting to see
how significantly the simulations were influenced by a stochastic variation in the strength
parameters of bond and mortar. This question is pursued in Chapter 7.4. It is known that
initial damage is present in concrete due to initial stresses and deformations caused by the
casting and hardening process (Hsu et al. [1963]). The sensitivity with regard to the
initial stress or deformation state in the composite model is investigated in Chapter 7.5
by assuming the presence of autogeneous shrinkage of the mortar. The influence of drying
shrinkage was not investigated, because its influence depends largely on the dimensions of
the structure and is not considered to be a fundamental cause of concrete softening.
Finally, conclusions are drawn in Chapter 7.6.

111
7.2 INFLUENCE OF RANDOM COMPOSITE STRUCTURE

To investigate the influence of the random generation of the composite structure, three
more simulations were carried out with meshes generated with the same set of parameters
used for simulation Tl in Chapter 6.4. Actually, only one new mesh was generated for
simulation M2. The other two simulation&, Ml and M3, were carried out by loading the
meshes used for simulations T1 and M2 in the lateral direction. The results of the 4
simulations are summarized in Figs. 7.2.1 to 7.2.3. inclusive.
-50.------------------------------,

-40 T1
N'

~ M1
~
-30
~
.b
M2

"' -20 M3
~
~ -10
- - Average

Deformation [rrm]

Fig. 7.2.1 Stress-deformation curves for 4 different composite structures

2.5

1 2.0 /
/
,
- - - T1
8 /

~
/
/
/
M1
I
.E 1.5 /

-8 I
/
/
h M2
iii / ~
!i~ 1.0
/
/ /
M3

2!. - ·-Average
Ill
~
0 .5
>
<(

0 .0 L---"""""~=-'----'---'---__.J
0.0 -0.1 -o.2 -o.3 -0.4 -o.5
Axial deformation [rrm]

Fig. 7.2.2 Relation between axial and lateral deformation for 4 different simulations

112
Streu-defonnation rel4tion
The nominal stress-deformation relations in Fig. 7.2.1 agree very well up to peak stress.
Post peak, the differences in the composite structures result in significant differences in
softening. These have to be taken into account when the results of micromechanical
simulations for different composite structures are compared. Fig. 7.2.2 shows that the
lateral deformations are more sensitive to the stochastic variation in the composite
structure. This is due to the fact that the lateral boundary conditions are unconstrained
in a uniaxial compression test. The larger spread in the lateral deformations is also
observed in laboratory tests.

(Tl) (Ml)

(M2) (M3)

Fig. 7.2.3 Failure modes found for 4 different simulations

Crack patterns
Due to the heterogeneous material structure and the unconstrained lateral boundary
conditions, a significant variation in crack patterns is found in Fig. 7.2.3. All the

113
specimens are split up into pieces which shear off, but shape and size of the pieces differ
significantly. The process of crack formation leading to these patterns was found to be
qualitatively the same. This is also reflected by the relatively slight differences in the
axial stress-deformation relation.

Approximately vertical splitting cracks and more inclined shear cracks join in a quite
random crack pattern. The angle of the combined cracks is not characteristic for the
failure mode of uniaxially loaded concrete. For shear failure of a continuum with the same
friction angle as the interfaces (tgcp=0.3), minimizing the failure load leads to an
optimum angle for a shear plane of 37o to the direction of compressive loading. When the
angle is measured of a number of more or less straight, combined cracks, angles are found
between 20o and 30o. This can be explained by the fact that splitting cracks join the shear
cracks in the combined cracks, which makes these combined cracks steeper. A spread in
cracking is found where the combined cracks cannot develop easily. This can be where the
combined cracks reach the loaded boundaries or where the combined cracks are arrested.
A good example can be found in the crack pattern of simulation M3 (Fig. 7.2.3). In the
middle of this specimen, a spread in cracking is found due to the fact that the shear crack
was arrested there.

7.3 INFLUENCE OF MATERIAL PARAMETERS

Table 7.3.1 gives a summary of the variation of material parameters used to study the
sensitivity of the micromechanical model with respect to their values. First, the
parameter values are given for the default case simulation Tl (see also Table 6.3.3). Next
the modified parameter values are given, which were substituted in the default case to
test their influence on the results. One parameter is substituted each time, except in the
case of the strength parameters. The values for the cohesion Cb and Cm are replaced in
pairs, keeping the ratio between bond and mortar strength constant at 0.5 (variation code
C). The same is done for the values of the tensile strength ftb and ftm (variation code FTr.
The ratio between bond and mortar strength (variation code RBM) is varied from 0.25 to
1.0 for both cohesion and tensile strength at the same time. The values for the mortar
strength are kept constant, while the values for the bond strength are varied.

The codes for the variations given in this table will be used further on in the discussion in
this chapter. A number added to this code indicates the substitution of value 1, 2 or 3 for
the parameters which are varied. The results are presented in Fig. 7.3.2 by means of the

114
nominal stress-deformation relations combined with the crack patterns and the relations
between axial and lateral deformations. They are also denoted by the codes for the
variations ofthe specific parameters. Table 7.3.1 and Fig. 7.3.2 are organized in
alphabetical order with respect to the code of the variation.

Default case T1 Variation

Parueter Value Code Value 1 Value 2 Value 3

Cb 10 c 5 15 N/u2
Cm 20 10 30 N/m2
tgx 0.0 en 0.15 0.3 -
Em 25000 El 10000 20000 30000 N/m2
tg, 0.3 FI 0.2 0.4 -
ftb 3.0 FT 2.0 4.0 Nfu2
ftm 6.0 4.0 8.0 N/u2
tg; 0.0 PSI 0.15 0.3 -

fb 3.0 Ul 1.5 4.5 6.0 N/m2

Cb 10 5 15 20 N/mm2
Wtmax 0.02 V11 0.01 0.03 JIJD

Wamax 1.0 V21 0.5 2.0 1x106 JIJD

Table 7.3.1 Summary of parameter variation

Nominal atress-tle/O'I'f1UI.tioa relation

In general, the parameter study shows that, in a qualitative sense, the axial stress-
deformation relation is not influenced significantly by a reasonable variation of the values
of the material parameters. Now, taking into account the stochastic scatter in the results,
some more conclusions will be drawn.

Petit stress
The most important change in the nominal stress-deformation relation is found, when
peak stress changes significantly. The peak stress is mainly determined by the cohesion
(Fig. 7.3.2-C), the friction angle (Fig. 7.3.2-FI) and the ratio between bond and mortar
strength (Fig. 7.3.2-RBM). When peak stress changes, the deformation at peak stress
also changes. It seems that this change in deformation is mainly due to an elastic shift

115
 -?or-------------------~ ao.--------------------.

I
-60
.·/~ I 4.0
cb= 5Nhml2
cm•15Nhml2

/[\) I
-60 cb=10Nhml2

-40 3.0 cm=20Nfmn2

I -30
1 2.0
cb=15Nhml2
cm•2QN1mml!
!!

§ 1.0
1
0.0 L---"""':s!.:..:..:__..__ __.__ __J
-0.1 -02 -0.3 -OA -0.5 0.0 -0.1 -0.2 -0.3 -OA -0.5
(C)
Deformation [mm] Axial deformation {mm]

-7o.-------------------~ ao.-----------------.
-60 I tgx=O.OO

Ii .,
4.0
tgx=0.15
tgx=0.30

"'.·-
··~
2.0 .. ...
"
.·"
1
<(
1.0
.··~·;,·'
. r
.: I

o.o~---~~-~--~--~
-0.1 -0.2 -0.3 -0.4 -0.5 0.0 -0.1 -0.2 -0.3 -0.4 -0.5
(CHI)
Oeforrnation [mm] Axial deformation [mm]

-?or-----------------~ 5.0 .-----------__.;..-----,

Em=10000Nhml2
Em=20000Nhml2
Em=25000Nirrm2
Em"'30000N/mm2

i 2.0
//
j& ./.'·9"
...
1.0

o.oL-----~~~~~__._
~
(;" I
../.."/
._ __ ..,' - J

_ __J
0.0 ·-0.1 -0.2 -0.3 -0.4 -0.6
(EM)
Deformation [nm] Axial deformation [mm]

Fig. 7.3.2 Results of a variation of the cohesion (C), the reduction 'factor for
tensile strength (CHI) and the modulus of elasticity of mortar (EM)

116
 -7or---------------------~ 6.0 .---------------------~

I
-~
-60 - - . tQ9>=0.2

I 4.0
tg<P=0.3

I
-50

-40

-30
.,
/1\
,
/ : ...

.
iI 3.0

2.0
..... tg<P=0.4

i -20

-10 ~----,_ j 1.0

... ...
0 '----~---'-----'-----'------' 0.0 '----~'"""':::::L..'-----'----"----...J
0.0 -0.1 -0.2 -0.3 -0.4 -0.5 0.0 -0.1 -0.2 -0.3 -0.4 -0.5
{FI)
Deformation [mm) A><ial deformation [mm)

-70.---------------------~ 6.0 .-----------------------,

2
-60 111> =2Nfmm
I
2 /
ftm=4Nfmm
I
2
111> =3Nfmm
/
3.0 ftm=6Nfmm2 I
I
fib =4Nfmm2
I
2.0 ftm=8Nimm2

1.0 ...
-10

0 ._____.____._____,.____...____....J
0.0 1.....---==-----'-----'------'
0.0 -0.1 -0.2 -0.3 -0.4 -0.5 0.0 -0.1 -0.2 -0.3 -OA -0.6
(FT)
Deformation [mm] Axial deformation [rrrn)

-70 .---------------------.., 5.0 . - - - - - - - - - - - - - - - - - - - - - - ,

-60 I tg>p=().OO
,· /

I
4.0
tg>p=0.15

I
1.'
-50 tg>p=0.30 ,.,.I."
3.0 t".,·
-40 I ..

!., -30
2.0
1,:

j -20
1.0
-10

0 '-----'-----'-----"-----'--............J o.o'----~~~----'~---'------'
0.0 -0.1 -o.2 -0.3 -OA -0.5 0.0 -0.1 -0.2 -0.3 -OA -0.5
(PSI)
Deformation [mm] Axial deformation [mm)

Fig. 7.3.2 Results of a variation of the friction (FI), the tensile strength (FT)
and the dilatancy (PSI)

117
 ao~--------------------~

- - . 111> ~ 1.f!Ninm:2

4.0
c., = 5Nfmm2 I

- - - 111> =3.0Nimm2 ,/
cb ~10Nflnm2 r

...•. 111> =4.5Nfmm2 /

cb •15Nfmm2 /

2.0 - - 111> •6.0Nhnm 2 I

t
~ -20
c., =20Nflnm2/
I
1.0
-10

0.0 L---""""~=:........____._____:
-0.1 -0.2 -0.3 -OA -0.6 o.o -0.1 -0.2 -0.3 -OA -o.e;
(RBM) ...;.ial defol+nation [rrm)
Deformation [mrn)

-70r---------------------~ ao~----------------------,

I w1'"""=0.01mn

I -00
4.0

3.0
w1..,.=0.02nm
w1max=0.03nm
I
I
I

"'
I 2.0
I
I
I

I 1.0
... ...
o.oL--~~~~~~--~--~
-0.1 -0.2 -0.3 -OA -O.e; 0.0 -0.1 -0.2 -0.3 -OA -05
(WlM)
Defannation (nm) Axial deformation (IMI)

-70r---------------------~ ao~--------------------~

w-"=05nm
w-=1.onm
w_=:i!.onm
6
w-=1x10 mm

-10
j 1.0

0.0 L._----===----'-----'---~
-0.1 -0.2 -o.3 -OA -o.e; 0:0 -0.1 -0.2 -0.3 -OA -o.e;
(W2M)
Deformation [mrnJ Axial defcrmation [mrn]

Fig. 7.3.2 Results of a. variation of the ratio between bond and mortar strength (RBM),
the maximum crack opening (WlM) and the maximum crack shear displacement (W2M)

118
caused by the change in stress level at the peak. The inelastic part of the deformations at
peak stress seems not to be influenced significantly by these parameters. As a consequence
of the shift of peak stress and deformations at peak stress, the brittleness of the post-peak
behaviour changes because the deformation due to which the softening ends is unaffected
by a change in the above mentioned parameters.

EM of sojl.ening
The maximum shear displacement w211ax for the interface behaviour is the parameter that
determines the end of compressive softening (Fig. 7.3.2-W2M). What is interesting, is
that an increase in maximum shear displacement to infinity does not make the softening
disappear. This indicates that compressive softening in the model is not only due to shear
softening of the interfaces, but also to a decrease in the number of contacts caused by a
gradual opening of the cracks ..

Tensile Jro,cture energy

The figures for the influence of the tensile strength (Fig. 7.3.2-FT) and the maximum
crack opening (Fig. 7.3.2-lM) indicate that an increase of the tensile fracture energy has
a positive influence on the fracture energy of the specimen under compressive loading.
This influence is less than that of the parameters mentioned above. It modifies the
compressive softening of the specimen, because it influences the opening of the cracks. It
can also prevent the formation and opening of splitting cracks. This causes the formation
of more regularly shaped shear cracks. The number of effective contacts in these regularly
shaped shear cracks does not decrease very fast during sliding. This means that the rate of
softening decreases when tbe shape of the crack becomes more regular.

Heterogeneity
A change in the modulus of elasticity of the mortar causes an elastic shift in the total
stress-deformation relation (Fig. 7.3.2-EM). The failure mode is not changed. When the
modulus of elasticity of the mortar decreases, slightly more damage is found in the
pre-peak regime due to the larger stresses caused by the composite structure. The
influence of the increase of these stresses is limited. The damage caused by these stresses,
corresponding to the formation of splitting cracks, seems a stable process which does not
influence the final failure process significantly. Also, the influence of the ratio between
bond and mortar strength is less than expected (Fig. 7.3.2-RBM). The shape of the
stress-deformation curve changes only slightly.

The slight influence of the difference in moduli of elasticity and strength suggests that the
heterogeneity introduced by the configuration of the mesh of interfaces determined by the

119
composite structure plays a more important role in the simulation of uniaxial compression
than the other factors. Actually, the acceptance of cracking in the interface elements
alone is the most significant factor determining the heterogeneity in the model.

Ltltertll deformation and crack pattern

A general trend, which seems to be present, is that the lateral deformations increase when
more vertical splitting cracks are present. This also makes the combined cracks steeper
due to the fact that more splitting cracks are combined with inclined shear cracks. More
splitting cracks are found when the tensile fracture energy decreases due to a decrease of
the tensile strength (Fig. 7.3.2-FT) or the maximum crack opening (Fig. 7.3.2-WlM).
The same is found for an increase of the cohesion, which means a relative decrease of the
tensile strength (Fig. 7.3.2-C). More splitting cracks are found too, when the bond
strength decreases (Fig. 7.3.2-RBM).

Two reasons can be given why the formation of a larger number of splitting cracks causes
larger lateral deformations. Firstly, the combined cracks become steeper when more
splitting cracks join in these cracks. This is illustrated in Fig. 7.3.3. In consequence, more
cracks are needed to span the width of the specimen. This results in larger lateral
deformations for the same axial deformation, because each combined crack adds to the
lateral deformations. Secondly, the specimen deforms more easily in the lateral direction,
when more splitting cracks are present.

I
lW
I
'-V

wtg a wtga wtga

~- ~- -~

Fig. 7.3.3 Increase of lateral deformations, when more splitting cracks are present

Finally it should be mentioned that an increase o!JP.e dilatancy increases the lateral
deformations (Fig. 7.3.2-PSI). This causes faster opening of the cracks during sliding.
Also the spread in the cracking increases slightly, due to the additional stresses caused by
the dilata.ncy.

120
7.41NFLUENCE OF VARIATION IN STRENGTH

To test the sensitivity of the micromechanical model to a random variation of strength

parameters, 6 sets of strength parameters were introduced in the simulations. For both
bond and mortar interfaces two extra sets of material parameters were defined with a
tensile strength and a cohesion being 50% higher or lower than the original values. No
further sets of parameters were introduced because UDEC limits their number. Moreover,
it was not thought to be essential for the simulation results to have more sets of material
parameters. The sets of parameters were randomly distributed over the mesh of
interfaces, taking into account the difference between bond and mortar interfaces.

Rti8fllta
The results are shown in Fig. 7.4.1. It shows that the variation in strength does not
essentially change the behaviour predicted by the micromechanical model for a unia.xial
compression test. The peak stress is reduced slightly and some more splitting cracks are
found.

Models for concrete behaviour with brittle failure and no heterogeneity at the microlevel
need a variation in strength to be able to describe softening. The influence of the
variation in strength on the predicted behaviour by the micromechanical model is limited,
because variation in strength, significant heterogeneity and softening were already present
on the microlevel. This indicates that the source of softening in the micromechanical
model was only slightly extended by adding a random variation in bond and mortar
strength.

-50 2.5

-40 I 2.0
- V!JP.: 0%

I -30
.,-(2]
.5
~
. - - V!JP.:50%

!., i 1.5

!
1
-20

-10

0
~
/\ ... ....

<(
,
i 1.0

0.5
~
,._

0.0
0.0 -0.1 -0.2 -0.3 -0.4 -0.5 0.0 -0.1 -0.2 -o.3 -o.-4 -0.5
(a) O..formation lrrm! Axial deformation !irml (b)

Fig. 7.4.1 Influence of random variation of strength

121
7.5 INFLUENCE 01' AUTOGENEOUS SHRINKAGE

Values up to 1.5x1o-a have been reported for autogeneous shrinkage of hardened cement
n
paste (RlLEM [1981]). is not reasonable to prescribe such a value for the shrinkage of
the mortar in the micromechanical model. The autogeneous shrinkage of mortar is less
because the small aggregates and sand in the mortar restrain the shrinkage. :Further, it
should be realized that the modulus of elasticity increases constantly during the process of
autogeneous shrinkage. This means that using the final, and thus highest, value for the
modulus of elasticity in the calculations would overestimate the stresses due to the
autogeneous shrinkage. Therefore, two lower levels of autogeneous shrinkage were chosen
to test their in11.uence on the simulations, a low level of lxl0-4 and a high level of 4xl0-4.
These strains are equivalent to elastic mortar stresses of 2.5 N/mm2, which is of the same
order of magnitude as the t;ensile bond strength, and 10 N/mm2, which is greater than the
tensile strength of both bond and mortar.

Rualt6
The low value ~f the shrinkage resulted only in slight damage at places of highest stress
concentration. The high value resulted in a continuous damage pattern throughout the
specimen. Because this damage was highly distributed, the average damage (damage
parameter ")of the interfaces was only 2%. This means that only 2% of the fracture
energy had been absorbed. This explains why the simulation results in Fig. 7.5.1 are not
very sensitive to the initial autogeneous shrinkage of the mortar.

-50 2.5

I 2.0
- c.-ox1o""'
<~: 8 = 1x1o""'
I -30
.6

I 1.5
..... «a=4X10""'

!..
f
j
-20
i 1.0

-10
i 0.5

0.0
-0.1 -02 -03 -0.4 -0.5 0.0 -0.1 -0.2 -03 -0.4 - 0.!>
(a) Deformation {mn)
(b)
· Axial deformation {mn)

Fig. 7.5.1 Influence of autogeneous shrinkage

122
7.6 DISCUSSION AND CONCLUSIONS

Se~ of the model

It has been shown in this chapter, that a relatively large variation in parameters does not
significantly change the failure process as described by the model in a qualitative sense.

Stochastic charGCter of the model

The softening predicted by the model shows a significant stochastic scatter. The scatter is
negligible for the pre-peak behaviour. When softening is found, the specimen behaviour is
determined by the behaviour of only a small part of the specimen, that is where the
macrocracks localize. This is the reason why the stochastic variation in the microstructure
becomes more apparent when the specimen undergoes softening. The larger scatter during
softening is also found in real tests. The scatter in the results of the micromechanical
model is more significant than in real experiments, because the model is two-dimensional.
This is caused by the fact that a two-dimensional model is much more sensitive to the
failure of a part of a certain size than a three-dimensional model, because in a
two-dimensional model a much larger percentage of the cross section is occupied by this
part than in a three-dimensional model.

It should be realized that other parameters, like the accuracy in preparation of the
specimen and application of the load, also influence the scatter of results in real tests.
These factors have not been taken into account in the micromechanical simulations.

lnjl'IUJU:e of material parameters

The variation of the material parameters did show that peak stress of the compression
test simulated with the micromechanical model is mainly determined by the cohesion and
the friction. The completion of softening depends on the maximum shear displacement in
shear softening. The parameters determining the tensile fracture energy modify
compressive failure, but have less influence than the parameters determining shear failure.
On the other hand, the value of the tensile fracture energy significantly influences the
lateral deformation of the specimen and the crack pattern.

Crack propagation and localization of deformations change with the variation in

parameters, but they are present in all simulations. The conclusions in Chapter 6.5
concerning compressive softening are not influenced by the variation in material
parameters. Even in simulation WlMl, where the tensile fracture energy of simulation Tl
is halved, shear failure is decisive in compressive failure. Localization of deformations is
still present, but some spread in cracking, and consequently in deformations, is found. On

123
a vertical line more than one inclined crack is often encountered. The crack pattern found
for this simulation compares very well with test results (see Fig.6.2. 7b ).

I
I ' ,/·~ \ I
~~.\ ;'f/, 1\i '\ ..
\ ., I 1-, I
I :; ,:
\71 I I
I J~ I
I If I
I \ ') ( ) I I
I \
I
I
\
\
I I • '
\
. '\;'\ \ I
' \• I\J
\
\\1\,---- 1
1
' \ I / I I I
11 !1)11 I
I q/
' I
I I 1_\ \! I
I I I I I \ I
(a) ,\ I ;-' '
/
I
I
I

Fig. 7.6.1 Comparison of crack patterns found in (a) test T1-2 and (b) simulation W1M1

Influence of interface muh

The insensitivity of the process of compressive softening with respect to the variation in
material parameters points to the very important role the mesh of interfaces plays in the
failure process. In fact, the most significant heterogeneity in the model is the acceptance
of cracking only in the interfaces and not in the continuum between the interfaces. This
basic property of the model is justified by the observations of compressive crack formation
by Hsu et al. [1963). It reflects the important role the aggregates in normal-iltrength
gravel concrete play in the crack-formation process.

Softening at the microlevel

The softening modelled at the microlevel makes the micromechanical model relatively
insensitive to variations in strength and variations in initial stress, deformation or
damage at the microlevel. It can be reasoned that these variations have already been
taken into account by modelling softening at the microlevel. The variations only extend
the heterogeneity of the model slightly and have a relatively slight influence on the results
of the model.

Attempts to introduce brittle failure at the microlevel in the micromechanical model

resulted in very brittle compressive failure and a significant scatter in results. Therefore
softening was introduced at the microlevel. It is assumed by several researchers that it
should be possible to model softening of concrete with brittle behaviour at the microlevel.

124
In Chapter 6, it was shown that post-peak ductility increases when the heterogeneity of a
material increases. This also suggests that concrete softening can be modelled by brittle
behaviour at the microlevel if the heterogeneity is modelled down to a very low scale
level.

The micromechanical model describes the heterogeneity of concrete in two dimensions

down to the level of the largest aggregates. On this scale level, cracks are modelled by a
straight interface. Failure of each of these interfaces represents a complex three-
dimensional process of crack growth. This means that it would be incorrect to model
failure of these interfaces as brittle, as long as the geometrical heterogeneity is not
modelled in finer detail and in three dimensions. One should realize that a two-
dimensional heterogeneous model is actually homogeneous in the third dimension.
Modelling softening realistica.Uy by a two-;:limensional model with brittle failure at the
microlevel can probably only qe done by making the geometrical heterogeneity more
extreme than it is, so as to compensate for the missing heterogeneity in the third
dimension.

The above indicates that it is necessary to translate three-dimensional properties, but

also nonlinear processes, in to two-;:limensional ones in order to obtain a quantitatively
correct description of three-dimensional processes by a two-;iimensional model. Whether
this is possible without restrictions is a question which cannot yet be answered.

Comparison of simulation renlts 'll1ith ezpe.rimef&t41 renlts

Comparing the average results of the 4 simulations in Chapter 7.2 to the experimental ·
results with teflon platens, it can be concluded that
1. the stress-deformation relation of the simulations compares well with the
experimental results (Fig. 7.6.2a),
2. the lateral deformations of the simulations are smaller than found in the experiments
(Fig. 7.6.2b) and
3. the number of splitting cracks and the spread in cracking in the simulations are less
than those found in the experiments (compare Fig. 7.6.1a with Fig. 7.2.3)

It is not easy to get a better fit of the model results with the experimental results. This is
a complex process of trial and error, which has to be carried out for all kinds of tests like
uniaxial tensile and compression tests, but also for multiaxial tests. For compressive
softening it should be noted that the restraint to lateral deformations of the teflon platens
has not yet been taken into account. This shows that it is a laborious task to fit the model
results to test results and do so for a.ll features of concrete failure with the same degree of

125
accuracy. It should also be borne in mind that a two-dimensional model has its
limitations in simulation of three-dimensional processes and that all types and features of
concrete failure can possibly not be described with the same accuracy by one model. This
aspect calls for closer investigation in the future. In the present thesis, optimization of the
model parameters will not be pursued further. It has been shown that the model is able to
provide a good qualitative insight into the process of concrete failure and this will be used
further in the thesis.
-50r-------------------------,

-40

I -30
Tests T

J"' --Tests MM

~ -20 Average of
strn.llattons

~ -10

oL---~----~----~--~--~
0.0 - 0. 1 - 0.2 - 0.3 - 0.4 0.5

(a) Deformation [mm]

3.0 . - - - - - - - - - - - - - - - - - - - - ,

2.5 I
I,
I,
2.0 lt - - - Tests T
I'
1.5
1/ Tests MM
/I
!/ Average of
/;
1.0 stmulattons
I!
h
0.5

0.0 L__.....,==:.::;_ __.__ _!..J._ __j

0.0 -0.1 - 0.3 - 0.4 - 0.5

(b) Axial deformation [mm)

Fig. 7.6.2 Comparison of average simulation results and results of tests

with teflon platens

126
 CHAPTERS

INFLUENCE OF NONUNIFORM DEFORMATIONS

8.1 INTRODUCTION

Nonv.niform deformation in tensile test6

In many material property experiments the objective is to create a uniform state of stress
and strain in a specimen in order to analyze its behaviour by the theory of the
homogeneous continuum. When a specimen is large enough, this method can give accurate
results for concrete before pea~ stress is reached. It has become standard practice in these
tests on concrete specimens to use hinged loading plat~ns because they facilitate uniform
loading in the pre-peak loading regime.

Early tensile tests on concrete showed that the application of a hinged loading platen
results in highly nonuniform deformations over the cross-section of a specimen and in
crack growth from one side of the specimen to the other as soon as soon as the peak stress
is passed (for instance, see Riisch & Hilsdorf [1963]). Apparently, softening not only
causes a non uniform distribution of deformations parallel to the axis of loading, which is
generally denoted as localization of deformations, but also causes a nonuniform
distribution of deformations perpendicular to the axis of loading.

To achieve more uniform deformations perpendicular to the axis of loading, hinges were
no longer applied, but the specimen ends were effectively connected to the loading
apparatus. However, significant nonuniform deformations were still found just after peak
stress was passed (Reinhardt [1984], Cornelissen et al. [1986]). The cause and the effect on
the stress-deformation relation of the nonuniformity of deformations were not yet
understood. Van Mier [1986b] was the first to show that the nonuniform deformation
state in a tensile test depends on the slope of the softening branch of the stress-
deformation relation. Hordijk et al. (Hordijk et al. [1987], Hordijk & Reinhardt [1989])
showed how the nonuniform deformations in the softening regime influence the
stress-deformation measurements and explained the irregular shape of the softening
curves found in many uniaxial tensile tests (Reinhardt [1984], Willam et al. [1985],
Cornelissen et al. [1986], Van Mier [1984,1986], Hordijk et al. [1987]).

127
Nonuniform tlefornr.ationtl in compreuion tut6
There is no reason why the phenomenon of nonuniform deformations during softening
should be an exclusive property of uniaxial tensile tests. It will be found for other types of
uniform loading and other materials where softening is present. As localization of
deformations parallel to the axis of loading, a nonuniform distribution of deformations
perpendicular to the axis of loading is a characteristic result of softening.

The compression experiments presented in the present thesis were planned and carried
out when the above-mentioned nonuniform deformations in tensile tests were discussed.
Up to that time, a lack of uniformity of deformations in compressive softening tests had
not been addressed. Neither had the choice between a hinged and a fixed loading platen
for compressive softening tests been discussed explicitly in literature. Hinged loading
platens have been used in compressive softening tests without special attention having
been paid to the nonuniformity of deformations (Kotsovos [1983], Van Mier [1984],
Torrenti et al. [1989]). It was decided to investigate this subject as part oftest series 1,
which was dedicated to the boundary conditions. Identical tests were carried out with
completely fixed and with hinged loading platens. In these tests the rotation of the
loading platens was measured with 4 LVDTs at the corners (see Chapter 3.2). The
rotations were also measured for the fixed loading platens because it was uncertain
whether the stiffness of the loading apparatus was sufficiently high to guarantee uniform
deformations during softening. This turned out not to be the case. Unfortunately, it was
also found in test series 2, the siz~ffect tests, that the nonuniformity of deformations
played an important role in some of the tests. It will be shown in this chapter that it is
very difficult to keep the nonuniformity of deformations negligible in siz~ffect tests for
compressive softening.

Contents
In Chapter 8.2, some experimental observations of nonuniform deformations during
softening are discussed. Then in Chapter 8.3, an explanation for the loss of uniform
deformations will be given and the parameters which govern the nonuniformity of
deformations will be shown. In Chapter 8.4 a simple numerical model will be used to
investigate the subject further and show how the stress-deformation relation measured
for a specimen is influenced by the nonuniformity of deformations. In Chapter 8.5 the
results of simulations with the model will be compared with test results. The influence of
nonuniform loading will be examined in Chapter 8.6. Conclusions will be drawn in
Chapter 8.7.

128
8.2 OBSERVATIONS IN TESTS

Fi:ud tefton plii.UM

Although the platens were firmly fixed to the loading apparatus, it was found in the
boundary-condition tests with teflon platens that significant nonuniform deformations
occurred during softening. This is shown for test T1-l in Fig. 8.2.1. The nonuniformity of
deformations grows slowly in the pre-peak regime. As soon as the peak stress has been
passed, nonuniformity grows very rapidly. When most of the softening has taken place,
uniformity of deformations is more or less restored. The observations in these experiments
agree well with those reported by Hordijk et al. [1987] for tensile tests.

-50

.:-..... -........

I<'
-40
'"\'
! ·. '
' -_,
- left-front
corner

' -
\
-30 \ left-back
'\
m \ corner
.1::; \
(/)
\
I \ - n~t~t-back

iil -20 I I corner

1
I
\
I ri<;;-,t-front
\ \ corner
-10 ·.\ \
··:,," ........
·.":-

0'-----'-----'----'------l
0.0 -0.2 -0.4 -0.6 -0.8

Deformation [rrrn]

Fig. 8.2.1 Deformations of the 4 corners of the specimen in test T1-1

Although the loading platens were firmly fixed to the loading apparatus, the deformations
of the specimens were highly nonuniform due to the rotation of the loading platens. This
points to an insufficient bending stiffness of the loading apparatus. The loading apparatus
was designed as a stiff loading apparatus taking into account only the axial stiffness and
not the bending stiffness in particular (Van Mier [1984]}. The results of these tests point
out that this aspect should also be taken into account when designing a loading
apparatus. It puts forward the question of how and how much the test results were
influenced by the rotations of the loading platens.

It is possible that significant rotations were also present in the tests with the steel
brushes. In these tests it was very difficult to connect measuring devices to the loading
platens close to the specimen so as to get reliable information about the uniformity of

129
deformations of the specimen. For the tests with the dry platens only very little rotation
of the loading platens was observed.

Hinged. tejlon platens

The deformations of the 4 corners of the specimen in test HTl-1 with a hinged teflon
platen a.t:e shown in Fig. 8.2.2a. As for test Tl-1, the nonuniformity of deformations
grows significantly after peak stress is passed. It is so extreme that only one side of the
specimen continues to deform further, while the other side shows a decrease in
deformation. At that side the loading platen even loses contact with the specimen.
Strain-gauge readings indicated that even significant elongations were found in that part
of the specimen which indicate the presence of tensile stresses. Fig. 8.2.2b shows that only
little more than half of the specimen failed, while the other half showed no <apparent
damage. A comparison of the nomina.l-t~tress-deformation curves for the tests with the
fixed and hinged te:flon platens is given in Fig. 8.2.3. It is found that the measured rate of
softening is increased by the application of a hinge. This agrees with the observation of
the partial failure of the specimen, which should result in a lower fracture energy of the
specimen. The influence is less than expected from the partial failure mode observed in
Fig. 8.2.2b.

-50.-----~----------------~

-40
"'E - - - right-back

! -30
corner
- - - right-front
"' corner
1 - - left-front
"' -20 corner
~ left-back
~ -10
corner

0 '-------"-------'-------'------'
0.5 0.0 -0.5 -1.0 1.5
(a) Deformation [mm]
(b)

Fig. 8.2.2 (a) Deformations of the 4 corners of the specimen in test HT1-1
and (b) surface crack patterns for test HT1-1

The straight lines in the stress-deformation curves for the tests with the hinged loading
platens in Fig. 8.2.3 indicate that the tests were unstable during those parts of softening.
Almost no measurements were obtained in those parts, because the deformations there

130
took only a second. This instability was caused by a snap back in the deformation whose
the measurement signal was used for the test control. The LVDTs for test control could
not be attached to the loading platen near the specimen, because there the LVDTs were
attached to measure the rotation of the loading platen. Therefore the LVDTs were
attached at the other side of the hinge. It appeared that the deformations of the hinge
were so significant that, during the brittle softening of the specimen, a snap back was
found in the LVDT signals used for the test control. This means that the actual behaviour
of the specimens with the hinged loading platens was more brittle than shown in the part
with the straight lines in Fig. 8.2.3.

-50

-40

I -30
\
'
\

\
',' ,<E--- Fixed

g:ll
'\
'\
fJ) '~
-20
'\
~ ',,,,,,
~ -10
~
0~--~--~----~--~----~--~

0.0 -0.1 0.2 - 0.3 - 0.4 - 0.5 - 0.6

Nominal deformation [mm)

Fig. 8.2.3 Comparison of tests with fixed and hinged teflon platens

Influence of 'll1itlth. of specimen

In the size-effect tests, the height and width of the specimens were varied. It was found
that the 200-mm-wide specimens showed significant nonuniform deformations. An
example is given in Fig. 8.2.4. The influence of the nonuniformity of deformations on the
nominal-stress-deformation relation for these tests is shown in Fig. 8.2.5. In this figure
the results of the tests on the 200-m.m-eubes are compared to the tests with the same
height but with a specimen width of 100 mm, which yielded almost uniform deformations.
It shows that the nonuniformity of deformations particularly influences the end of the
softening curve where a plateau in the stress-deformation relation is found, which often
ends abruptly. Cracking was found to propagate through the specimen. The abrupt end of

131
the plateau was found when the crack growth extended over the width of the specimen.
At that moment the nonuniformity of deformations reduced significantly. These
observations show that the test results of the 200-mm-wide specimens could not be used
in the comparison of the size-effect tests.

-50.-----~----------------------------~

Front
-40 •, \. Right

I1/)
(J)
-30
\'·.
\ ·..
\ ·..
\
Back
Left

~ \
-20 \
~ I
\
~ -10
I
I
\
'- ...... _
-....... ....................... '•,
') .·
0
---- ..::::::

0.0 -0.5 - 1.0 -2.0 -2.5 -3.0 -3.5

Deformation [mm]

Fig. 8.2.4 Deformation of the 4 sides of specimen LLl (200x200x200 mm)

-50

-40
~
~
-30
"'
~"'
"' -20
~
~ -10

0
0.0 0.5 - 1.0 - 1.5 -2.0 -2.5
Nominal deformation [mm)

Fig. 8.2.5 Comparison of stress-deformation curves for tests on specimens of

200x100xl00 mm (LM) and 200x200x200 mm (LL)

132
8.3 CAUSE OF NONUNIFORM DEFORMATIONS

Nonuniform loading
A specimen is never loaded uniformly in a test. Three of the most important factors
causing nonuniform loading in a compression test are:
L eccentricity of the load
2. nonparallelism of loading-platen surface and specimen surface
3. nonflatness of loading-platen surface and specimen surface.
It is assumed that the last two factors can represent the total effect of initially imperfect
contact between the loading platen and the specimen. Both surfaces can have an infinite
variety of shapes and relative positions to each other when the test starts. In the following
calculations, the nonparallelism will be used as the factor representing the component of
the imperfect contact which results in moment.

Moment caused by a nonuniform reaction of the specimen due to its heterogeneity is not
taken into account explicitly. It is hard to estimate this property of a specimen. In an
ideal test specimen it is not present. In this research, special attention was given to
preparing specimens which were as homogeneous as possible (see Chapter 3.3).

Antllytical model
A simple model for the analysis of a uniaxial compression test is shown in Fig. 8.3.la. A
specimen of dimensions h x b x d is loaded by means of rigid loading platens. The upper
loading platen moves downward in deformation control. It is placed on the specimen with
eccentricity e and initially makes an angle cp0 with the specimen surface. The directions of
e and cp0 have been chosen so that both result in a moment on the specimen in the same
direction. The resistance of the loading apparatus to a rotation of the loading platen is
represented by a rotational spring with stiffness C connected to the upper loading platen.
It is assumed in the calculations that the nominal stress-deformation relation of a
uniformly-deformed specimen is representative of the behaviour of any part of the
specimen (Fig. 8.3.1b). This assumption of homogeneity of the specimen is, strictly
speaking, not correct for softening concrete, but it will be shown in Chapter 8.5 that it
results in a good description of test results.

In the calculations it is assumed that complete contact has been established between
loading platen and specimen surface. The initial rotation of the loading platen is taken
into account by introducing a moment Ctp0 on the specimen. The angle cp in the
calculations is the angle of both loading platen and specimen surface to the initial position

133
of the specimen surface. The displacement w is the average displacement of the surface of
the specimen and, consequently, the average deformation of the specimen.

w
b
(a) (b)

Fig. 8.3.1 (a) Model of a unia.xial compression test and (b) stress-deformation relation of
a uniformly-deformed specimen

Ctlltultl.tion of rotation
Now, it is assumed that the specimen is loaded uniformly in deformation control up to
deformation w and stress u. The rotation rp can then be calculated from the equilibrium of
moments of point A (Fig. 8.3.1), which reads

Meccentricity + Minitial angle= Mapparatus + Mspecimen

or

d b3 d (j
-dbue+Crp0 =Crp+- {-)rp {8.3.1)
12 dw

For the rotation this yields

134
 -dbue + Ctpo
tp= (8.3.2)
db3 du
c + (-)
12 dw

This equation gives a good approximation for low values of tp and tp0 , because physical
and geometrical nonlinearities then play a minor role. In that case the use of the
derivative do'/dw provides a reasonable approximation of the stress~eformation relation
around the point (w,u) in Fig. 8.3.1b.

stiffness

r- ---- ---~

If" - Capparatus+ .9.JiS

12 du dw I
L------
"'S)'stem -
___ !..1

Deformation

Fig. 8.3.2. Evolution of the rotational stiffness of the system of specimen

and loading platen during a softening test

Limitation of rotation
Equation 8.3.2 shows that the rotation tp can be kept low by keeping the moment due to
nonuniform loading ( = numerator of equation 8.3.2) low or by making the rotational
stiffness of the total system of loading apparatus and specimen (= denominator of
equation 8.3.2) high. It is now essential in a softening test that the slope of the stress-
deformation relation du/dw continuously decreases in the pre-peak and at the beginning
of the post-peak regime. It becomes even negative in softening. This means that the
counteracting moment of the specimen against a rotation of the loading platen

d b3 du
Mspecimen = (-) lP (8.3.3)
12 dw

continuously decreases and changes sign when the peak stress is passed. Then, it no longer
opposes the rotation of the loading platen, but in fact stimulates it. During softening, the
rotation of the loading platen is only opposed by the rotational stiffness of the loading

135
apparatus C. The evolution of the rotational stiffness of the total system of specimen and
loading apparatus during a test is illustrated in Fig. 8.3.2.

This figure clearly shows how the rotational resistance of the system decreases and
reaches a minimum where the lowest negative value of the derivative du/dw is found.
When this resistance decreases so far that it becomes negative, no stable solution for
equation 8.3.2 exists. This means that the rotation of the loading platen in equation 8.3.2
can only be limited when

db3 du
C+- (-)>0 (8.3.4)
12 dw

Given a loading apparatus with a stiffness C, this means that the stability criterion in
equation 8.3.4 is violated when the derivative du/dw becomes smaller than a critical
negative value

du 12C
(-)critical = - - - (8.3.5)
dw db3

When the steepest slope of the softening curve represented by (du/ dw )critical is known,
rotational instability will not be present when the criterion

db3 du
C > - - (-)critical = Ccritical {8.3.6)
12 dw

is satisfied. A similar criterion has been derived by Hassanzadeh et al. [1987].

In reality the rotation 1fJ will not increase to infinity because physical and geometrical
nonlinear effects will result in forces opposing the rotation.

Some researchers suggest that satisfying the stability criterion of equation 8.3.4 will
prevent significant nonuniform deformations in a softening test {Hassanzadeh et al.
[1987], Hillerborg [1989], Hordijk [1991]). Equation 8.3.2 shows that this will not be so.
That equation still predicts a significant rotation of the loading platen when the stability
criterion is just satisfied, which leaves the numerator of equation 8.3.2. quite small.
Whether this results in a significant influence on the stress-deformation relation
measured in a test will be investigated in Chapter 8 paragraphs 5 and 6. ''

136
BifurctJ.tion point
A bifurcation point is reached when the numerator of equation 8.3.2 becomes zero while
the load is centric. This, for instance, is equal to the buckling of a centrically loaded
column. Up to the bifurcation point the rotation rp is zero. At the bifurcation point the
rotation of the loading platen is no longer opposed by the total system of loading
apparatus and specimen and is undetermined. Equilibrium can be found for values of rp
differing from zero.

Rotation of loading platen

Equation 8.3.2 shows that the rotation rp of the loading platen increases when
1. the eccentricity e and the initial rotation rp0 of the loading platen increase. This
means that the rotation rp depends on the accuracy of testing.
2. the rotational stiffness of the loading apparatus C decreases.
3. the cross section b x d of the specimen increases. The rotation rp is quite sensitive
to this change in cross section. This is expressed by equation 8.3.4, which shows
that the critical stiffness for the loading apparatus increases 16 times when the
dimensions of the cross section double.
4. the combination of u and du/dw, which are coupled, becomes more critical. For a
constant value of du/dw this means when the compressive stress u increases. For a.
constant u this means when du/dw decreases. Generally speaking, this
combination becomes more critical for
a. a more brittle type of concrete. Examples are high-strength concrete and
lightweight concrete.
b. a specimen of greater height. This is caused by localization of deformations which
makes such a. higher specimen more brittle.

8.4 SIMULATIONS WITH A SIMPLE NUMERICAL MODEL

In this chapter the behaviour observed in the uniaxial compression tests will be
investigated and explained by means of a simple numerical model. The model allows the
influence of nonuniform deformations on the measured stress-deformation behaviour of a
specimen to be investigated.

Other sim'llllltions
Simple models for analysing the influence of nonuniform deformations on tensile tests
have also been used by Hordijk &: Reinhardt [1989] and Hillerborg [1989]. The model

137
presented here differs from these other models in its ability to capture nonuniform loading
and take into account a more complex stress-deformation behaviour including unloading
and reloading different from loading and the loss of contact between loading platen and
specimen. Finite element analyses have been used to analyse tensile tests by Rots et al.
(1987], Zhou [1988] and Rots & De Borst [1989]. However, it is not easy to use finite
element analysis in a parameter study, because of its significant demand on computer
time.

Namerit:Gl motlel
The model used for the analysis is shown in Fig. 8.4.1a. It is similar to that used in
Chapter 8.3. The difference is that the specimen is represented by n springs, each of which
represents an nth part of the specimen. All springs have the same nonlinear stress-
deformation relation (Fig. 8.4.1b). For each spring the initial gap w0 between loading
platen and specimen is taken into account. After contact is established between spring
and loading platen, the stress-deformation curve comes into effect. Unloading and
reloading of the springs is modelled different from loading. It is assumed to be linear
elastic. The contact between loading platen and spring is lost as soon as the stress
becomes zero. After the contact is again established, reloading takes place until the
nonlinear loading curve is reached.

< n springs ) w
(a) b (b)

Fig. 8.4.1 (a) Model of a uniaxial test and (b) stress-deformation relation of the springs

138
Calculatitm procetlfltl'fl
When the eccentricity e and the initial rotation tp0 of the loading platen are not zero, the
calculations are straightforward. For each calculation step, the average displacement w of
the loading platen is incrernented, while the rotation tp is kept constant. Then the
resulting moment on the loading platen is calculated. Subsequently, tp is adapted in an
iterative procedure until equilibrium of moments is approximated with a prescribed
accuracy. Finally, the resulting force on the loading platen is calculated and divided by
the specimen surface to get the nominal stress.

Bifurcation tmtJlyBis
In the case of initially uniform loading, a bifurcation analysis has to be carried out at each
calculation step until a rotation is found. Due to the fact that no moment acts on the
specimen, a straightforward calculation, as described above, only results in uniform states
of deformation. A non uniform state of deformations ca:n only be found when the stability
criterion in equation 8.3.4. is violated. Therefore, at each calculation step, the loading
platen is given a very slight rotation tp and the counteracting moment delivered by the
total structure of loading apparatus and specimen is calculated. As soon as the moment
due to the rotation tp is found to change sign and thns starts to stimulate the rotation
instead of counteracting it, this is the indication that the bifurcation point has been
passed. Subsequently, the deformation state with the slight rotation is taken as the
starting point for the above-mentioned iterative procedure. The moment now stimulating
the rotation guarantees that convergence to the nonuniform deformation state is found. In
this way, the bifurcation point is allocated and passed and the desired solution with
nonuniform deformations is obtained.

Injlueace of softening
It will now be investigated what the influence of softening is on the nonuniformity of
deformations and the nominal stress-deformation relation in a compression test. For this
purpose, uniform loading will be assumed. This is done just to investigate the influence of
softening without disturbing the picture by the influence of nonuniform loading. The
influence of nonuniform loading will be shown in Chapter 8 paragraphs 5 and 6.

"Uniform" lotul.ing 'With a .{iutllotul.ing platen

For this analysis the stress-deformation relation of test T1-1 with teflon platens will be
used to describe the specimen behaviour under uniform loading. For a description of
unloading, a modulus of elasticity K=400N/mm3 (K=E/h, E=40000 N/mm2, h=lOOmm)
is used which is approximately equal to the initial modulus of elasticity. A minimum
slope of -136 N/mm' is found for the stress-deformation relation of T1-L According to

139
equa,tion 8.3.6 this results in a critical spring stiffness of approximately l.lx109
Nmm/rad. In the ease of uniform loading, non uniform deformations will only be found
when the rotational stiffness of the loading apparatus is less than the critical stiffness of
equation 8.3.6.

-50.--------------------------------.

C=1.0Ccr
-40
C=0.8Ccr

I -30
C="0.5Ccr
C=0.3Ccr

i
"@ -20
· - C=O

~ -10

oL-----~----~------~----~--~~
0.0 -0.2 -0.4 0.6 -0.8 - 1.0
(a) Nominal deformation [mm]

0.04 .------------------/.,....--------------,

I
I - - C="O.BCcr
0.03
I - - C="0.5Ccr

!
I
· · ···· C=0.3Ccr

5 0.02 I
I --C=O
:0
.l!l
{i.
I
0.01
1
/ .. •····
1.. ... ·;,...-
0.0 -0.2
/

~·.:::;
0.00 L..__ _ _ _....w..:_.:..<;....; --
·;

.... .............. '----···~ ............ .

_ _--'------'-"'-'---==----..:..:...:."'-'-'J

-0.4 0.6 -0.8 -1.0

(b) Nominal deformation [mm]

Fig. 8.4.2 Influence of rotational stiffness C of loading apparatus on (a) the

stress-deformation relation for test Tl-1 and (b) the rotation of the loading platen

The influence on the nominal stress-deformation relation predicted by the model is shown
in Fig. 8.4.2a for several values of the spring stiffness. The rotations of the loading platen
are given in Fig. 8.4.2b. It is clearly shown that the rotations due to a decrease in the

140
rotational stiffness of the loading apparatus can significantly influence the nominal
stress-deformation relation. For low values of the spring stiffness, the "bump" in the
stress-deformation relation, which has been found in many tensile tests, is clearly visible.
For 30% of the critical spring stiffness, a vertical drop is found in the stress-deformation
relation and the rotation of the loading platen. Here the calculation jumps from one stable
state to another. The continuously stable snap back behaviour can not be followed
because the calculation is deformation controlled. The same phenomenon is observed in
deformation controlled tests (see Fig. 8.2.5) .

Discussion of a test at 5l1Yo of the critical rotational stiffness

The observed behaviour will now be discussed for the simulation with a spring stiffness of
50% of the critical value. It is illustrated in Fig. 8.4.3. In this figure the nominal stress-
deformation curve found for nonuniform deformation is compared to that found for
uniform deformation. Also, the stress distributions over the cross-section of the specimen
are given. The rotation of the loading platen is shown in the upper part of the figure.

Deformation ritj-1t side

Deformatoon left sode

-50

-40

~
~ -30
tll
tll
(])
e.tll
iii -20
c
E
z0
-10

0
0 .0 -0.2 -0.4 -0.6 -0.8 - 1.0

Nominal deform;~tion [mm]

Fig. 8.4.3 Simulation of a test with a rotational stiffness of 50% of the critical value

141
The deformations are uniform until the stress-deformation state with the critical
derivative is reached. Then, the loading platen starts to rotate. The rotations stay
limited, because of the nonlinear concrete behaviour. At the right-hand side, the
specimen starts unloading according to the initial modulus of elasticity. At the left-hand
side the specimen reaches a state of softening in which the derivative of the
stress-deformation relation is smaller. This activates a stabilizing moment. Due to the
fact that the specimen is softening at the left side and unloading at the right side, the
average softening of the specimen is accelerated by the rotation of the loading platen. The
nonuniformity of deformations of the specimen then grows constantly building up moment
in the specimen. During this process the softening of the specimen slows down until the
moment delivered by the specimen cannot increase any longer. Then the rotation of the
loading platen decreases and the specimen is totally in the softening regime. This
accelerates the softening again. Due to the significant rotation, loading platen and
specimen even lose contact at the left side. Finally the deformations become uniform
again.

These calculation results show that the response of the specimen is more complex than
was assumed in the simple model in chapter 8.3, which did not take physical nonlinear
behaviour into account. The simple model predicts the bifurcation point correctly, but
overestimates the nonuniformity of deformations because the forces opposing the rotation
caused by the physical nonlinearity are not taken into account.

• Uniform" loading with a hinged loading platen

Fig. 8.4.2 also shows the simulation results for a test with a hinged loading platen (C=O).
The rotation starts at peak stress, as predicted by the stability criterion. It is extreme
and has a significant influence on the nominal stress-deformation relation. The process is
illustrated in Fig. 8.4.4. Initially the same behaviour is observed as in the tests with a low
spring stiffness. The nonuniformity of deformations is more extreme because the rotation
of the loading platen is not opposed by a moment generated by the loading apparatus. As
a result, the softening is not slowed down. The stress distribution becomes more and more
nonuniform. A little bit more than half of the specimen fails, while the other half is
unloaded and stays undamaged. A significant part of the specimen loses contact with the
loading platen. This explains why the test with the hinged loading platen is so brittle and
why a much lower fracture energy is found than for the tests with fixed loading platens.

Some comment on other simulations

Hillerborg [1989] concluded from similar calculations carried out with a simple analytical
model, that the application of a hinge does not have a significant influence on the nominal

142
stress-deformation relation of a tensile test. A parameter study carried out with the
model presented here, showed that this was caused by the way Hillerborg modelled
unloading. For the sake of simplicity of the model he assumed that concrete was infinitely
stiff in the case of unloading. The results of the simulations are significantly influenced
when such an extreme assumption is made. The rotations decrease when the unloading
stiffness is increased and at the same time the influence on the nominal stress-
deformation relation decreases. This also explains why Hordijk & Reinhardt [1989] found
a more significant influence of the application of a hinged loading platen. In their model,
unloading was modelled according to the loading curve. Finit~lement calculations of
Rots & De Borst [1989] show almost no difference in tensile tests with or without a hinge.
This is probably due to the slenderness of the specimen (h x b = 250 x 50 mm). In their
simulations of a test with a hinge, a significant "bump" is found in the softening branch.
This can only be explained by.the creation of moment opposing the nonuniform rotations.
This has to be a result of geometrically nonlinear behaviour, caused by the extreme
nonuniform deformations of the slender specimen. This shows that a test with a hinged
loading platen is a complex test, which is sensitive to many parameters, which are not
easily taken into account.

Deformat ion right side

Deformation left side

-50r----------------------------------------.

- 40
"E
!
(})
-30
(})

..,
~

lii -20
t:
E
0
z
- 10

0
0.0 -0.2 -0.4 -0.6 -0.8 - 1.0

No m1nal deformation [mm]

Fig. 8.4.4 Simulation of a test with a hinged loading platen

143
8.5 COMPARISON OF TESTS AND SIMULATIONS

Test TJ-1
To be able to simulate test T1-1, the rotational stiffness C of the loading apparatus with
the teflon platens was measured first. This was found to be approximately 1.35x109
Nmm/rad, which was higher than the critical value for test T1-1, 1.1x1Q9 Nmm/rad.
This means that uniform loading would not result in nonuniform deformations for this
test. Therefore, a simulation was carried out giving the loading platen an eccentricity
-50r---------------------------~

Uniform
-40 '' Nonuniform

I -30

-10

O'-----'----...J----"-------L----__.J
(a) 0.0 -0.2 -0.4 -0.6 -0.8 -1.0

Nominal deformation [mm]

-5o.----------------------------.
Calculat>on

-0.2 -0.4 -0.6 -0.8

(b)
Deformation [mm]

Fig. 8.5.1 (a) Influence of nonuniformity of deformations on the stress-deformation

relation of test T1-1 and (b) comparison of measured and calculated deformations

144
e=0.5 mm and an initial angle with the specimen surface !p0 =0.0005 radial {0.05 mm over
100 mm). These values were assumed to be realistic for test T1-1.

Fig. 8.5.1 shows. the results of the simulation compared with the test results. The
stress-deformation relations have both been corrected for the initial contact effect due to
the initial angle between loading platen and specimen surface. According to the model,
the nominal stress-deformation relation was scarcely affected by the nonuniformity of the
deformations (Fig. 8.5.1a). Fig. 8.5.1b shows that the nonuniformity of deformations was
simulated quite well.

It should be mentioned that in test T1-1 rotations of the loading platen were found in
two directions. However, the rotations in one direction dominated those in the other
direction. Therefore it was thought that a. two-dimensional simulation could approximate
the three-dimensional behaviour well. Actually, there ~ill be a nonlinear interaction
between the response in the two directions.

-40
"E
!
<I)
-30
IJ)
Q)
.1;:
IJ)

-20
I~
-10

o~--~--~~--~--~----~--~
0.0 -0.1 -0.2 -0.3 -0.4 0.5 -0.6

Nominal deformation [rrrn]

Fig. 8.5.2 Comparison of a simulation and test results

for tests with a hinged loading platen

lnjluetu:e of a hinge
The tests with a hinged teflon platen were simulated using the stress-deformation
relation of the tests with the fixed teflon platens as an input. Just as in the simulation of
test Tl-1, an eccentricity e=0.5 mm was assumed to be present. Fig. 8.5.2 shows a
comparison of the results of the simulation with test results. It is found that the influence

145
of the hinge on the stress-deformation relation is less than predicted by the model. Also,
the rotation of the loading platen was found to be significantly less than predicted.

The model seems to overestimate the effect of a. hinged loading platen. This is possibly
because it neglects to take into account that compressive softening of concrete is a three- .
dimensional process of crack formation. In the model no interrelation between the stress-
deformation behaviour of the neighbouring parts of the specimen is taken into account.
This deficiency, of course, becomes most apparent in a test with a hinged loading platen,
which shows extreme gradients of deformation.

PerjoJWU~,J&Ce of the hinge

There should be a question mark at the performance of the hinge: It functions perfectly
when it is not loaded, but needs moment to rotate when it is loaded. This is caused by the
friction in the spherical seating (Fig. 8.5.3). Making a rough estimate, it can be assumed
that the moment needed to make the hinge rotate, is in the order of a R F. Assuming
that a=0.02 (teflon coating) and R=50mm, it is found that the load F may have an
eccentricity of 1 mm before the hinge rotates. This means that the moment is of the same
order of magnitude as that due to eccentricity of the load. This eccentricity determines
the initiation of unstable rotation of the loading platen, but also the further rotation.
Theoretically it can even prevent completely rotational instability.

Fig. 8.5.3 Friction in a hinged loading platen

InflueJ&Ce of the fl1i4th. of a specimen

The tests on specimens with a width of 200 mm (tests LL) were simulated by using the
stress-deformation relation found in the tests on specimens of the same height but with a
width of 100mm (tests LM). These last tests showed more or less uniform deformations
during softening. The loading platen was given an eccentricity e=0.5 mm and an initial

146
rotation 'J'o=0.0002 radial (0.02 mm over 200 mm). The initial rotation was chosen less
than in the simulation of test Tl-1 because in test series 2 the flatness of the specimens
was significantly better than in test series 1 due to the use of a stiffer grinding apparatus.

For the tests on the 200-mm-wide specimens, the loading apparatus was equipped with
an other set of loading platens than that used for test T1-1. This made the rotational
stiffness of the loading apparatus increase from C=1.35x109 Nmmjrad to C=1.60x109
-50

Calculation
-40 Tests
1E
~ -30
Ill

!
iii -20
c
E
~
-10
;:----
I ' \
~ -~ \
0
(a) 0.0 -0.5 -1.0 - 1.5 -2.0 -2.5

Nominal deformation [mm]

0.020
__ ,
Calculation
Tests
0.015 '
'I
"0 I
\ll
..:. I

..,c
.0 0.010
...
.l!l
fl.
0.005

0.000
0.0 -0.5 - 1.0 -1.5
(b)
Nominal deformation [mm]

Fig. 8.5.4 Comparison of simulation and experimental results for tests on specimens of
200x200x200 mm: (a) nominal stress-deformation relation, (b) rotation of loading platen

147
Nmmfrad. Based on the stress-deformation relation for the tests on the 100-mm-wide
specimens, the critical rotational stiffness of the loading apparatus was calculated to be
C=13.4x109 Nmm/rad. Hence, the rotational stiffness of the loading apparatus was ouly
12% of the critical value.

A comparison of the results of the simulation and test results are shown in Fig. 8.5.4. The
model predicts the stress-deformation relation and the rotation of the loading platen well.
Analysis of equation 8.3.6 for the critical rotational stiffness of the loading apparatus
shows that the critical stiffness changes dramatically when the size of a specimen is
changed. When all the dimensions of a specimen are doubled, the critical stiffness for the
specimen increases 24 = 16 times. Actually, this will be more because the derivative
du/dw will.also be influenced by the change in height of the specimen (see Fig. 2.3.2). A
minimum ratio of sizes in a test series investigating a size effect is 4, which indicates that
the largest specimen is more than 256 times as sensitive to nonuniform loading as the
smallest specimen. This indicates that one should always pay attention to this
phenomenon when carrying out size-effect tests.

8.6 SIMULATIONS OF NONUNIFORM LOADING

Here,· the influence of nonuniform loading on the nominal stress-deformation relation of a

specimen will be investigated. This will be done for test T1-1, which satisfied the
stability criterion of equation 8.3.6. In test T1-1, the rotational stiffness was 23% higher
than the critical value. Test simulations with hinged loading platens are also given for
comparison.

Eccentricity of load
The results of simulations with an eccentricity of load are given in Figs. 8.6.1 and 8.6.2. It
is found that the test with the fixed loading platen is not very sensitive to an eccentricity
of load. The test with a hinged loading platen is much more sensitive. This is due to
possibility of free rotation of the loading platen in this test. In Chapter 8.4 it was shown
that the rotation of the loading platen significantly influences post-peak behaviour. The
simulations in this Chapter reveal that the rotations already become important before
peak stress is reached and have a significant influence on peak stress.

Fig. 8.6.3 shows the reduction of peak stress due to an eccentricity of load. The results of
the simulations are compared with test results of Ruseh [1955] and Newman &:
Sigvaldason (1965]. These researchers carried out tests on specimens of the same width,

148
 -50
e=Omm
-40 - e=imm

1
~
e=2mm
e=3mm
())
-30
()) e=4mm
QJ
l:;
()) e=5mm
-20
~
~ -10

0
0.0 -0.2 - 0.4 -0.6 - 0.8 - 1.0

Nominal deformation [rrmJ

Fig. 8.6.1 Influence of the eccentricity of load on a test with fixed loading platens

-50

e=Omm
-40 e=1mm
1
~
e=2mm
e=3mm
-30
m e=4mm
~
u; e=5mm
6i -20

1 -10

0
0.0 -02 -0.4 -0.6 -0.8 - 1.0

Nominal deformation [mm]

Fig. 8.6.2 Influence of the eccentricity of load on a test with a hinged loading platen

but of a lower strength. In general, the conclusion that tests with hinged loading platens
are more sensitive to an eccentricity of load than tests with fixed loading platens is
confirmed by these test results. But more factors play an important role, for instance the
shape and size of the specimen, the shape of the stress-deformation curve and the
rotational stiffness of the loading apparatus. Some remarks can be made concerning these
influences.

149
 1.00 1.00~---J~---------,

I I
0.95

t
5 0.95

.
E
.. fj, Calculated
> ~'
ia: 0.90 0 Calculated (low strJ .!
rl
0.90

0 Newman & Sigvaldason

0.85 '---'----'---~'---~'------' 0.85 '----"'---~'-----"'-----"'------'

0 2 3 4 0 2 3 4
(a) Eccentricity [......,] Eee<Ontrieity [mm!
(b)

Fig. 8.6.3 Influence on peak stress of a.n eccentricity ofload for (a) a t.est with fixed
loading platens and {b) a test with a hinged loading platen

The tests of Newma.n & Sigvaldason with fixed loading platens, given in Fig. 8.6.3a, show
less sensitivity to a.n eccentricity of the load than the simulations. Because they used a
lower-strength concrete, a series of simulations with a lower strength concrete was also
carried out. The stress-deformation relation of this lower-strength concrete was derived
from the stress-deformation curve of test Tl-1 by multiplying the values for the stress
by a factor 0. 75. Fig. 8.6.3a shows that the influence of the eccentricity on peak stress
decreases significantly when this is done. In general, the simulations have shown that the
influence of a.n eccentricity of load is mostly determined by the difference between the
present rotational stiffness and the critical rotational stiffness. In the simulations of test
Tl-1, the rotational stiffness of the loading apparatus was only 23% higher than the
critical value. In the simulations with the lower-strength concrete, the rotational stiffness
of the loading apparatus was 64% higher than the critical value. Increasing the rotational
stiffness to infinity would reduce the influence of the eccentricity to zero, as rotation is
then prevented.

In Fig. 8.6.3b, the influence of a.n eccentricity of load on peak stress is shown for a test
with a hinged loading platen. The model appears to overestimate the effect of the
eccentricity on peak stress, just as it overestimates the effect on the post-peak behaviour.
It was found that changing the total stress-deformation relation as described above to
simulate the lower-strength concrete did not affect the results. The only influencing
factors for a test with hinged loading platens are the shape of the stress-deformation

150
curve, which was not changed above, and the size of the specimen. One final remark
should be made. In this research it was found that it is very difficult to position a
specimen on the loading platen with an estimated accuracy of less than 0.5 mm in two
directions. Therefore, it can be questioned how accurate the zero eccentricity in the above
presented test results was. Possibly, part of the difference between simulation and test
results ~an be explained by the fact that peak stress for a zero eccentricity has been
underestimated because the eccentricity was not zero.

-wr-------------------------~----.

- q:..,=<>.OOO
-40 - - - q:..,=0.001

I -30
'l0 0 =0.002
'l0 0=0.003
I

i
o.95

I -20
'l".,=0.004 0.90

J -10 o.ao '-----~---....J

0.000 0.001 0.002 0.003 Q.004

Initial rotation [rad]

-0.2 -0.4 -0.6 -0.6 -1.0

(a) (b)

Fig. 8.6.4 Influence of an initial angle between loading platen and specimen surface on
(a) the nominal stress strain relation of test Tl-1 and (b) peak stress

Initial angle of the loading pltlten

The sensitivity of test Tl-1 with fixed loading platens to an initial angle between loading
platen and specimen surface is shown in Fig 8.6.4. All curves have been corrected for the
initial contact effect due to the angle between loading platen and specimen surface. It is
clear that an initial angle between loading platen and specimen surface can have a
significant influence on the nominal stress-deformation relation and peak stress. One
should be aware of this problem, but experience in testing has shown that it is not too
difficult to solve. The use of a stiff grinding apparatus can result in a sufficient flatness of
the specimen surface and plan;>arallelism of upper and lower surface. Further, it is
advantageous to place the loading platen upon the specimen using a hinge which is
loosened. Then, the loading platen is guided to the most favourable position with respect
to the specimen surface. Subsequently, the hinge must be effectively locked. Another
method was used by Van Mier (1984], who used a stiff epoxyresin to fill the gaps between
loading platen and specimen surface. This can only be done, when steel brushes are

151
applied. For other loading platens this will result in a significant restraint of the lateral
boundary displacement& with, in turn, a significant infiuence on the test results.

8.7 CONCLUSIONS

Jajlue'fll!e of softtm,iRg
In this chapter it has been shown that nonuniform deformations in a test specimen are
stimulated by softening a.nd that this nonuniformity of deformations can have a
significant effect on the nominal stress-deformation relation of the specimen.

Like localization of deformations, the stimulation of non uniform deformations belongs to

the structural behaviour of a softening test. It can be looked upon as anot}J.er example of
the tendency of softening to localize in the smallest volume possible. By a rotation of the
loading platen, softening takes place only in part of the specimen, which results in the
most brittle failure mode. U is also an example of the preference of nature for non-
symmetric failure. It points once more at the danger of assuming symmetry in the
analysis of structural failure.

The stimulation of non uniform deformations by softening is a gradual process, caused by

the gradual weakening of the specimen. The stability criterion of equation 8.3.4 gives the
minimum value for the rotational stiffness of the loading apparatus for which uniform
deformations are found in a uniformly loaded specimen. However, satisfyiqg this criterion
is no guarantee that the deformations will be more-or-less uniform, as is suggested by
several researchers (Hassanzadeh et al. [1987], Hillerborg [1989], HordiJ'k [1991]). Equation
8.3.2 shows that when the stability criterion is just met and the denominator of the
equation is small, significant nonuniform deformations can still be found because of
nonuniform loading. When it is desired to keep the deformations as uniform as possible, a
significantly higher value than the critical value for the rotational stiffness in equation
8.3.6 is required.

Simple numerical model

A simple numerical model has been presented in this chapter to investigate the above-
mentioned phenomena. It has been shown that the model provides a good description of
the behaviour observed in tests well.

152
Nonuniftnm loading
It has been shown that a. test with fixed loading platens is relatively insensitive to a.n
eccentricity of load. When a hinged loading platen is used, a high sensitivity is found. A
test with fixed loading platens is more sensitive to an initial angle between loading platen
and specimen surface. Experience in compressive testing has shown that this angle can be
kept sufficiently narrow, when care is taken in preparation and loading of the specimen.

Hinged loading plo.ten

A test with a. hinged loading platen is an extreme test, as the deformations in this test are
highly nonuniform and physical and geometrical nonlinearities play an important role in
its behaviour. The measured response is due to complex structural behaviour significantly
influenced by many parameters. The softening found in such a test is more brittle than
that found in a test with fixed loading platens. It is a test which is difficult to analyse and
interpret.

Shear crt~ck in uniaiol eompf'U8i.on

The significant rotation of the loading platen in a test with a. hinge probably stimulates
the formation of one large shear crack in a. slender specimen. It is initiated at one side and
grows gradually to the other side of the specimen when the loading platen rotates. This is
clearly shown in the test by Torrenti et al. [1989] (see Fig. 2.6.3). Tests by Newman &
Sigvaldason [1965] have shown that one pronounced shear crack is found when a hinge is
used, but that a more complex and distributed crack pattern is found when the hinge is
deactivated. In cubic specimens one continuous shear crack cannot cross the width of the
specimen and only half the specimen fails. This is sometimes observed too in slender
specimens, where the boundary restraint to lateral deformations is slight. The inclination
of a shear crack depends highly on the restraint to lateral deformations by the loading
platens (Kotsovos [1983]).

153
 CHAPTER9
INFLUENCE OF LATERAL BOUNDARY RESTRAINT

9.1 INTRODUCTION

Injlaeru:B of 4 multio.:rial stress~

Compressive failure of concrete is highly sensitive to lateral confinement and thus to the
lateral boundary conditions. When concrete fails in compression, a failure mechanism is
created which enables it to shear off in the lateral direction. This means that lateral
expansion must be allowed and is of vital importance in finding compressive failure.
Softening does not take place when there is no possibility of lateral expansion (Bazant et
al. [1987]). A lateral confining stress slows down compressive failure. This is most extreme
for stress states near uniaxial compression. This is shown by the test results of Jamet et
al. [1984] (Fig. 2.5.3). In their tests, a small confining stress of 3 N/mm2 makes peak
stress increase by 50%. In the same test, the post-peak behaviour also becomes more
ductile and a. residual stress level of 20 N/mm2 is found. This shows that compressive
failure for low confining stresses is highly sensitive to lateral stresses in the order of
magnitude of 1 N/mm2. It makes clear that compressive failure can never be accurately
predicted when the lateral boundary conditions are not taken into account.

Influence of lotuling platens

It is well known that the frictional restraint of lateral deformations of a concrete specimen
by the loading platens influences peak stress in a uniaxia.l compression test (for instance,
see Schickert [1980]). In his tests with normal steel platens, this results in .a 20 to 25%
higher uniaxial compressive strength for a cubic specimen than for a slender specimen
with a height-width ratio of 2.5 or more. It is assumed that the stress state in in the
middle section of such a. slender specimen is not essentially influenced by the lateral
boundary restraint and fails when the stress in this section reaches the uniaxial strength.

Tests of Kotsot10s
The test results of Kotsovos [1983] (Fig. 9.1.1) show that softening of concrete loaded in
uniaxial compression is more sensitive to the lateral boundary restraint than peak stress.
ffis tests on slender cylinders (250x100 mm) show that softening is still influenced by the
lateral boundary restraint when peak stress is no longer influenced. His point of view that

154
compressive softening is modelled best by a. complete and immediate loss of load-carrying
capacity, is extreme, but illustrates the role that boundary conditions ca.n play in this
phenomenon. Observing a crack growth from the unia.xially-stressed centre of the
specimen to the triaxially-stressed end zones, Kotsovos postulates that softening is due to
crack growth in the specimen caused by a nonuniform stress state and is not a material
property but purely a structural property.

F/Fpeok F/Fpeok
1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2

00 0.5 1.0 1.5 2.0 2.5 0.5 1.0 1.5

wjwpook w/wpeak

(a) (b)

Fig. 9.1.1 Influence oflateral restraint on softening of (a) a low-strength and (b) a
normal-strength concrete: (1) active restraint, (2) plain steel platen, (3) MGA pad used
before, (4) brush platen, (5) rubber layer, (6) MGA pa.d not used before (Kotsovos [1983])

Discus8Wn oftut results of KotsOflos

Kotsovos' extreme point of view seems not to be justified by the test results for the
low-strength concrete (Fig. 9.1.1a). In these tests, even for the lowest boundary restraint,
substantial ductility, i.e. softening, is found that it is not likely to disappear completely
when the boundary restraint is reduced further. It is hard to estimate the lateral
boundary restraint in Kotsovos' tests. This can be illustrated for his tests with loading
platens with a rubber layer. Kotsovos reports that splitting stresses were present in the
concrete specimen due to the expansion of the rubber layers. This caused a reduction in
peak stress. For the softening regime Kotsovos states that the rubber will have restrained
the lateral expansion of the concrete, due to the fact that the lateral expansion of the
concrete during softening increases considerably while lateral deformations of the rubber
decrease. Frictional slip and frictional stresses are not mentioned or taken into account by
Kotsovos.

Two observations have to be made concerning the interpretation of Kotsovos' tests. First,
one should be aware that localization of deformations is also found for compressive

155
softening. This means that softening can disappear for a specimen with a height of 250
mm, as used by Kotsovos, but will probably not disappear for shorter specimens. The
importance of softening in the behaviour of a real structure cannot be simply estimated,
because it will depend on the composition, geometry, dimensions and boundary conditions
of the structure. To be able to describe softening in a structure of any size it will be
necessary to find a type of compressive softening model like the fictitious-crack model for
tensile softening, which is independent of the size of the structure. Second, Kotsovos used
hinged loading platens in his tests. It was shown in Chapter 8 that this resuil.ts in a
complex structural behaviour and makes the specimen behaviour more brittle. To what
extent the structural behaviour was influenced by the different loading platens is hard to
say. The failure modes of the specimens show a significant variation, which is due to the
combined influence of lateral restraint and rotation of the loading platen.

Testa with steel brushes

The first tests at Eindhoven University of Technology were carried out by Van Mier
[1984], who used steel brushes. At that time steel brushes were then believed to have the
least possible influence on compressive failure of concrete. The test results of Kotsovos
put forward the question, how much the results of the tests with the brushes had been
influenced by lateral restraint. This was also questioned by Van Mier, who suggested
carrying out further research. Therefore it was decided to carry out comparative tests
with brushes with a. different rod length.

Testa with tejlon platef&B

At the same time it was realized that the lateral restraint by brushes increases when the
lateral deformations of the specimen increase. Due to the bending of the brush rods, the
shear stresses induced in the specimen are proportional to the lateral deformations, which
increase significantly during softening. It has been shown by several researchers that steel
brushes have a negligible influence on the concrete behaviour up to peak stress (Kotsovos
[1983], Van Mier [1984]). However, during softening, there must be a. time when the
continuously increasing shear stresses remain no longer negligible. In order to investigate
this, it was decided to carry out also tests with frictional platens. Frictional platens
induce greater shear stresses in a specimen around peak stress than. steel brushes, but
become more favourable during softening. As the frictional stresses are related to the
axial stress, they decrease during softening. Thus there will be a. moment when the
frictional platens influence softening less than steel brushes. Therefore tefl.on platens with
very low frictional properties were developed (see Chapter 3).

156
Test& with dry pl4tefts
Tests were also carried out with platens without special friction-reducing measures for
reasons of comparison. These platens will henceforth be called dry platens. The only
difference with platens used in standard compression tests is that the dry platens had
been polished because they had been used for the teflon platens too.

Contents of this chapter

The results of the boundary condition tests will be discussed in Chapter 9.2. Then these
tests will be simulated with the micromecha.nical model. The results will be discussed in
Chapter 9.3. Conclusions will be drawn in Chapter 9.4.

9.2 TEST RESULTS

The most important test results are given in Figs. 9.2.1 and 9.2.2. To support the
interpretation of the test results, Fig. 9.2.3 gives the estimated values for the average
shear stresses induced in the specimen by the loading platens. The frictional stresses
induced by the teflon platens and the dry platens have been calculated, interpolating or
extrapolating the results of the tests carried out to measure the frictional properties of
these platens (see Chapter 3.1 and Vonk [1989]). The shear stresses induced by the steel
brushes have been calculated, taking the average value for a rod with two ends fixed and
-60

-50
Dry platen
lV
E
E
-40
~
m
g
1/)
-30

iii
c -20
·~ Short brush
z Long brush
-10
Teflon platen

-0.2 - 0.4 0.6 -0.8 - 1.0

Nominal deformation [mm!

Fig. 9.2.1 Axial stress-deformation relations for tests with different loading platens

157
 5r--------------------------------------,
Dry platen
Short brush
Long brush
Teflon platen

oL-----~~--~------L-----~----_J
0.0 -02 -0.4 -0.6 -0.8 - 1.0

Axial deformation lrrml

Fig. 9.2.2 Relations between axial and lateral deformation ·

for tests with different loading platens

7r--------------------------------------,
i' 6
E
~ 5

Short brush

/
/
,. ,.. .r ..- _ _ Long brush

~~~:..,..-:"".~.--:-:~
o~----~~~~~--~~~----~~--~
Teflon platen
0.0 -0.2 -0.4 -0.6 -0.8 - 1.0

Axial deformation [rnrn]

Fig. 9.2.3 Average shear stresses induced in the concrete specimens

by the different loading platens

a rod with one end fixed and the other hinged. This average value was taken, because the
end condition is not known exactly and is difficult to measure (see chapter 3). In the
calculations geometric second-order effects were taken into account (see Linse [1978]).

158
(a) (b)

Fig. 9.2.4 External crack patterns

(a) for test Dl-2 (w=-1.10 mm,
u=-27.6 N/mm2) and internal crack
patterns (middle sections) for test
Dl-2 (b) and Dl-1 (c) (w=-1.10 mm,
u=-30.1 N/mm2)

(c)

High bounda'll restraint

Due to the high boundary restraint, the results of the tests with the dry platens differ
significantly from the test results with the other loading platens. Peak stress is higher,
post-peak behaviour is significantly more ductile and the lateral deformations are smaller
for this type of test than for the others with a lower boundary restraint. Due to the
frictional restraint, significant confining stresses are built up in the specimen near the
loading platen. Fig. 9.2.3 shows that the average shear stress induced in the concrete by
the dry platens is more than 6 N/mm2, which means that the average confining stress in
the middle vertical cross-section of the specimen is also more than 6 N/mm 2. This
prevents crack formation in this part of the specimen, as is shown in Fig. 9.2.4. Cracking
starts from the more uniaxially stressed sides of the specimen, which split off. This is also

159
clearly illustrated in Fig. 9.2.5. This process results in the well-known hour-glass failure
mode (see Fi~ ·- 9.2.4b). Softening in this test is caused by a decreasing load-carrying
capacity of a more ?I less intact core of the specimen, which reduces gradually in size.
Compressive failureJn this tes.t is dominated by a nonuniform triaxial stress state, which
causes a typical·~t'ructurifb~haviour. ·J~ ~v.es the impression that concrete loaded in
uniaxial compression ·is ductile, what is not always true.

~60

-50
/
--
I
/'
-~ I
I I
I
-40
~ I I
I I

"'"'
~
-30
I I
I
~
~ -20
~ LVDTs
Strain gauges
-10

0
0 -2 -4 -6 -8 -10

Nominal strain [o/oo]

Fig. 9.2.5 Axial stress-strain relation measured by LVDTs and strain gauges
for test Dl-1

L0111 bovndi&ry restraint

Due to~the lo~ bound8ff restraint, the tests with the brushes and the teflon platens show
a lower p~ ··stresii, a more bri~tle post-peak behaviour and a larger lateral deformation
than the tests with the dry platens: Due to the reduction of the boundary restraint, peak
stress reduces by 12% and the deformation at peak stress decreases from -{).34 mm
(3.4 °/ 00 ) to 0.23 mm (2.3 °/ 00 ) for the brushes and 0.20 mm(2.0 °/ 00 ) for the teflon
platens. The reduction of peak stress is less than what is often reported by other
researchers. This can be due to the fact that the steel platens had been polished and had a
lower frictional restraint than the loading platens normally used for compressive tests.
However, it should be noted that this reduction of strength is by no means a constant, but
depends significantly on the type of concrete. The compressive strength of a slender
specimen is found to be 10 to 27% lower than the cube strength (Schickert [1980],
Kotsovos [1983], FIP /CEB [1990]).

160
The axial stress-deformation curves for the brushes and the teflon platens show no
significant differences until halfway along the softening branch, where curves start to
diverge. Fig. 9.2.3 shows that then the shear stresses induced in the specimen by the
brushes become greater than those induced by the teflon platens. This causes the
softening of the specimens loaded with the brushes to slow down. This is not found for the
tests with the teflon platens due to the decrease in shear stresses found for these, tests.
The smaller lateral boundary restraint by the teflon platens is also reflected in the lateral
deformations in Fig. 9.2.2, which are greater at the end of the tests for the teflon platens
than for the brushes.

For all tests with brushes and teflon platens, qualitatively the same crack patterns were
found at the end of the test. Characteristic examples of crack patterns found in these tests
were shown in Figs. 6.2.6 and 6.2.7. They show a random configuration of splitting and
shear cracks, which reveals no clear influence of the boundary restraint as found for the
tests with the dry platens. It seems that the higher shear stresses induced by the teflon
platens around peak stress did not have a significant influence on the configuration of the
crack pattern. The divergence of the softening curves is probably due to the fact that the
already been formed cracks cannot open and shear off easily due to the restraining shear
stresses.

Triazial tests
Fig. 9.2.6 shows the influence of the brushes and teflon platens on the results of the
tria.xial boundary-condition tests. In these tests, two confining stresses, 30% and 5% of
the largest stress, were present in the lateral directions. This stress state resulted in a
two-dimensional failure mode, with a shear failure in the plane of the highest and the
lowest stress (see Fig. 6.2.10). Fig. 9.2.7 gives a comparison of the smallest confining
stress and the average shear stress induced in the specimen by the loading platens in the
same direction. It shows that during softening the average shear stress induced by the
brushes is of the same order of magnitude as the lowest confining stress, which is very
important for the failure of the specimen. The influence of the shear stresses is found in a
reduction of the softening for the tests with the steel brushes. The tests with the teflon
platens show significantly more softening. It seems that the shear stresses induced by the
teflon platens slow down pre-peak deformations and cause an increase in peak stress.

These test results show that one should bear in mind that loading platens with a low
boundary restraint can still have an influence on tria.xial compression tests. Teflon platens
will have the greatest influence for high tria.xial stress states around peak stress, while
brushes will have the most influence in the softening regime for large lateral deformations.

161
 -120
Short brush
~ -100 Long brush
~ '
'' Teflon platen
.... -so
UJ
' \
\
-~ \
-60 er
''
~MOo
UJ
UJ ~
''
i -40 '' ...... _
'
iii
' ....__
~ -20 0.3 0"

0
0 -5 -10 -15 -20 -25

Nominal strain axis 1 [o/oo]

Fig. 9.2.6 Influence of different loading platens on triaxial test results

6.-------------------------------------~

0.050"
Short brush
Long brush
Teflon platen

oU-~--~-~-~~~=-~~_L_______ L_ _ _ _ _ __ L_ _ _ _ _ __J
0 -5 -20 -25

Nominal strain axis 1 [o/oo]

Fig. 9.2.7 Comparison of the smallest confining stress with the average shear stress
induced in the specimen in the same direction by the loading platens

The triaxial test results presented here showed that differences in test results were less
significant owing to the wide scatter in triaxial test results (see Vonk [1989]). This wide
scatter in triaxial tests is generally found in literature (for instance, see Gerstle et al.
[1978,1980], Newman [1979], Nojiri [1984]). It is probably due to the difficulty ofloading a

162
specimen uniformly in three directions. Furthermore, it can be due to the triaxial
constraint of the specimen, which can cause that one large aggregate hinders shearing of a
crack.

9.3 MICROMECHANICAL SIMULATIONS

To investigate the influence of the lateral boundary restraint on compressive softening

further, a number of simulations with the micromechanical model were carried out. Both
the influence of a frictional restraint and a restraint due to the bending of brush rods were
simulated.

Frictional restraint
To be able to simulate the frictional restraint, two loading platens were modelled together
with the specimen. The two loading platens were assumed to be infinitely stiff and were
not allowed to rotate or move in the lateral direction. The load was applied in
deformation control, giving the upper loading platen a downward velocity. The frictional
restraint is modelled by prescribing a coefficient of friction tgrp for the interface between
loading platen and specimen. The elastic normal and shear stiffness of the interface was
2x105 N/mms. The high value for the shear stiffness means that the shear deformations
were negligible until the maximum frictional stress was reached. Then, plastic slip was
assumed to take place. In the simulations, plane-stress conditions were assumed. For
these simulations the same material properties were used as for simulation T1 (see Table
6.4.2).

Simulation results
The results of the simulations are given in Figs. 9.3.1 to 9.3.3 inclusive. The coefficient of
friction tgrp was varied from 0 to oo. The coefficient of friction tg<p=oo means that all
lateral deformations along the loaded boundaries were prevented completely. The axial
stress-deformation relations in Fig. 9.3.1 show that compressive softening is highly
sensitive to the frictional restraint along the loaded boundaries. A coefficient of friction of
0.01 to 0.05 already has a significant influence. This is also reflected in the lateral
deformations in Fig. 9.3.2, which are slowed down considerably by the presence of friction
along the loaded boundaries.

For a low frictional restraint the crack pattern does not change in a qualitative sense.
This is shown in Fig. 9.3.3. In the case of a low boundary restraint opening and shearing

163
 -60

-50
_____ ____ ---... ... t~OO
/..:=--------........ ---
" /. --
!
m
-40
/~. :..:~ :.:..;.
.· \ f~· ........ .
'1
1..
. • ....
tg~0.20

·. tg~0.05
Gl
.b -30 \::---....:......_ .
Ul

1\\
' .... \~ ' ......
~ tg~0.02
c
.E -20 '
~ '
-10

0
0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6

Deformation [mm]

Fig. 9.3.1 Axial stress-deformation relations for the micromechanical simulations of the
influence of the frictional boundary restraint

2.5 r----------------------:r--..

2.0
I t~O.OO
-~ 1.5

I 1.0
t~0.01

t~0.02

0.5

-0.1 -0.2 -0.3 -0.4 0,5 -0.6

Axial Deformation (mm]

Fig. 9.3.2 Relations between axial and lateral deformation for the micromechanical
simulations of the influence of the frictional bound,ary restraint

of the cracks are delayed. This results in a slower reduction in t~e number• of contact
I
points in the crack pattern. An example of how the lateral boU}idary restraint influences
the number and length of contacts in the crack pattern is shown in Fig. 9.3.4. This figure
shows the stress-free interfaces for the simulations with coefficients of friction of 0.01 and
0.05. Clearly, the specimen loaded with a loading platen with a coefficient of friction of

164
0.05 shows fewer stress-free interfaces for the same deformation. Besides, a higher
boundary restraint results in more contact points, each having a higher shear resistance
due to the presence of a higher confining stress.

~.7'\) \
I' /

\) f \
I /\
I

' \.
'
\
,,
I
1/
'

(a)
I ___ _
_L _____tL
'/ ~ I

(b)

1-7
1',

'· \

''\1
I

I
I
/1
/' I
' I

\
.'
r-~
'''· ,,
\
I

'1'\ I
I \,,I
,,/

\
,/

I I
I ) >' / \
\/'"'
' I
I
1'\j"'- ' \

\ \

,,/
(
I '.

\
"tLI ___)_ I
/
I
/

I /
/ "
,/
\ \ '" 'I
/
/'/"

__ _ J
\ I '

(c) (d)

Fig. 9.3.3 Evolution of the crack pattern for an increase of the frictional boundary
restraint: (a) tgc,o=O.OO (w=--0.465 mm, u=-5.6 N/mm2), (b) tg!p=0.02 (w=--0.422 mm,
u=-21.6 N/mm2), (c) tgc,o=0.05 (w=--0.422 mm, u=-30.4 N/mm2) and (d) tgc,o=oo
(w=--0.464 mm, u=-50.2 N/mm2)

Fig. 9.3.3 shows that, when the frictional restraint increases, the crack pattern changes
gradually to the well-known hourglass shape. The crack pattern in the specimen can no
longer have a random shape governed by the heterogeneous material structure, but is
forced into a pattern determined by the nonhomogeneous state of stress in the specimen.
Due to the significant confining stress near the loading platen, this part of the specimen
stays free of cracks, while the more uniaxially stressed sides of the specimen are forced to
split off. To illustrate the failure mode, Fig. 9.3.5 shows the principal stresses in the
specimen at the end of the simulation with the coefficient of friction tgc,o=oo. It is shown

165
clearly how the uncracked core of the specimen carries almost a.ll the loa.d. The creation of
this fa.ilnre mode takes considerably more energy than the failure mode with the zero
boundary restraint. In the last case the crack pattern ca.n search for the weakest links in
the specimen to combine with them in a fa.ilnre mode. In the first case the stress state
determines what interfaces have to fail, regardless of the energy it takes to cause their
failure. For an increase of the coefficient of friction from 0 to oo, pea.k stress increases by
23% and the deformation at peak stress increases from -o.185 mm (-1.850/ 00) to -o.335
mm (-3.35 of 00 ), which is in good agreement with test results.

-,- 7
'\ \ 7:
' I I /
, 'I ) \
/.r I' ( I

I\ I '
I f
I
\ ' I
\ I
'
\ / -(\ I
J

\ I
I
,}'• I
I\!'
I/'
I

II ', I I
_I ) I I
(a) ' (b)

Fig. 9.3.4 Stress-free interfaces for simulations with (a) tgrp=0.01 (o=-14.4 N/mm2)
and {b) tgrp=0.05 (o=-30.4 N/mm2) at a deformation w=-o.422 mm

Fig. 9.3.5 Principal stresses at the end of the simulat,on with tgrp=oo
(u=-49.8 N/mm2, w=-o.526 mm) J

Influence of steelhrushu /
The lateral restraint by the steel brushes ha.s been modelled by giving the elastic shear
stiffness in the interface between loading platen and specimen the appropriate,value (see
Table 3.1.3). In Chapter 3.1 it was shown that the shear stiffness of the long brush varies

166
between k=0.70 N/mms and k=2.80 N/mm3 and that of the short brush between k=2.00
N/mm3 and k=7.99 N/mms. Simulations have been carried out for all these values. In
order to simplify the simulations, geometrical nonlinearity like shortening of the brush
rods due to bending, rotation of the rod top, which touches the concrete, and the influence
of buckling have not been taken into account (see Linse [1978], Winkler [1985]). These
phenomena can be taken into account by UDEC, but would increase the calculation time
significantly and would probably not change the results of the simulations very much.

Discussi.or& of results
Figs. 9.3.6 to 9.3.8 inclusive show the results of the simulations of the tests with the steel
brushes. The influence of the steel brushes on compressive softening grows gradually
during softening. Peak stress is almost uninfluenced by the presence of the brushes
because the shear stresses are still very low around peak stress. Post-peak the softening is
slowed down more and more due to the growing shear stresses. This agrees well with the
test results. Fig. 9.3.7 shows that the lateral deformations also become smaller and
smaller when the lateral shear stresses increase. These differences in softening are not due
to differences in the crack pattern. The crack patterns found for k=0.70 N/mms and
k=7.99 N/m.m3 in Fig. 9.3.8 do not differ significantly from the crack pattern found for
simulation Tl with k=O N/m.m3, which is shown in Fig. 9.3.3a. As for the tests with the
low frictional restraint, the softening is mainly influenced due to the fact that opening
and sliding of the already existing crack pattern is slowed down by the restraining shear
stresses.

-50
k=O.OON/mm3
- k=0.70N/mm3
-40

I
k=2.00Nimm3
· ..... k=2.81N/mm3
-30 "\. _ - · k=7.99Nimm3
~
.!:; ..... . '""--
"""'-"·i\'·
tiJ

iii -20
-..:.;..·-:...:.,. ....
.................
J -10

-0.1 -0.2 -0.3 -0.4 -o.5 -0.6

Deformation [mm]

Fig. 6.3.6 Axial stress-deformation relations for the micromechanical simulations of the
tests with steel brushes

167
 2.5
k=O.OO Nlmm'
k=0.70 Ntmm•
"E 2.0
s k=2.00 N/mm'
c
k=2.8 1 N/mm•

j 1.5
k=7.99 Nlmm2

t!! 1.0

~
j 0.5

0.0
0.0 -0.1 -0.2 -0.4 0.5 -0.6

Axial Deformation [mm)

Fig. 9.3. 7 Relations between axial and lateral deformation for the micromechanical
simulations of the tests with steel brushes

. . .-r----,-,--- ------,~----7/l.,._....v
1 ~ \ I I', ~ \ I
1 I ~~ I
I I \,/
I '/ '!! \ I/ I

I
I
~

I
,.'"

I
.I
\
Ift

•
I
I
11
. \
I
I
I\.
/
I
I
I
I

'
r
\

. I /\ \ I /\
I ~
'-r,, I '\
A /' \ I

v,,,
I
';~·,/ I\!

(a)
.... ' '-
1
_u\
I '\
(,___ \ _
(b)

Fig. 9.3.8 Crack patterns found in the simulations of the tests with the steel brushes for
(a) k=0.70 Nfmm3 and (b) k=7.99 N/mm3

The above agrees well with what was found in Chapter 6. There it was shown that the
first part of softening is due to the formation of the crack pattern. When the restraining
shear stresses are low around peak stress, the process of crack grpwth does not really
I
change, but is slowed down. When the restraining shear stressesjbecome significant in the
last part of the softening branch, as is found for the brushes, o~y the behaviour of the
already formed cracks can be influenced.

168
9.4 CONCLUSIONS

Influence of lateral boundary restraint

It has been shown in this chapter that uniaxial compressive failure is a complex
three-dimensional process. This process is very sensitive to a restraint of the lateral
deformations along the lateral boundaries, but also to a lateral restraint along the loaded
boundaries. The failure process can not be accurately predicted, if the lateral boundary
conditions are not taken into account.

lHgh and low 6oun44'll restraint

For tests on a cubic specimen of normal strength concrete, both peak stress and ductility
depend significantly on the type of loading platen used in a test. The use of a dry steel
platen with a high lateral restraint results in ductile post-peak behaviour, while the use
of a teflon platen with a low lateral restraint results in fairly brittle post-peak behaviour.
The standard compression test, when a cubic specimen is loaded with a normal steel
platen, suggests that uniaxial compressive failure is fairly ductile. This is not really true.
One should always be aware that the ductility strongly depends on the lateral boundary
restraint.

Steel broshu 'l/er8U8 tejlon plate.m

The application of steel brushes and teflon platens results in a low boundary restraint in
uniaxial compression tests. It has been shown that steel brushes cause the lowest
boundary restraint up to and including peak stress, while teflon platens restrain lateral
deformations the least as softening proceeds. Tests and micromechanical simulations
indicate that it will be very difficult, if not impossible, to find unrestrained compressive
softening in a test. Unrestrained compressive softening is convenient for analysis and
simulations of test results.

It has been shown that the restraining effect of steel brushes has to be taken into account
when oompressive softening is investigated. This will influence the conclusions drawn
from the results of tests on specimens of different height by Van Mier [1984], which will
be discussed in Chapter 10. There it will also be shown that the influence of teflon platens
for this kind of tests can't be neglected either.

Micromechll:nicallimvlations
The case of zero boundary restraint is only of relative importance. A correct model should
be able to describe oompressive softening under the complete range of boundary
conditions. It has been shown that the micromechanical model is in a good position to do

169
this. The heterogeneous structure in combination with the continuous mesh of interfaces
is able to describe the distributed microcracking as well as the local macrocrack growth
under influence of the boundary conditions. U seems that all the aspects needed to
describe the three-dimensional compressive softening behaviour are basically present. A
limitation is that the micromechanical model is still a two-dimensional model describing
a three-dimensional process.

170
 CHAPTERlO
INFLUENCE OF SIZE

10.1 INTRODUCTION

Size effect in tensile softeni.n.g

It is now well known that localization of cracking and deformations takes place in tensile
softening. This means that the theory of the homogeneous continuum is no longer
applicable and that the behaviour of the material becomes dependent on the size of the
specimen. This was shown in Fig. 2.3.2. As the size of the specimen increases, the
softening response of the specimen becomes more brittle.

Size effect in comprusive softeni.n.g

For a long time it was not recognized by many researchers that localization of
deformations and a size effect are present in compressive softening. This is probably due
to the fact that nnia.xial compression tests on cubic specimens with and without a.
negligible boundary restraint show no clear localization of cracking and deformations. On
the other hand, clear examples of local failure can be found for unia.xial compression tests
on slender specimens loaded with a. hinged loading platen and also in the case of a nUI'nber
of multiaxial compression tests.

The tests on specimens of different height by Van Mier (1984] gave the first indication
that there is a size effect in compressive softening just as there is in tensile softening (see
Fig. 2.3.3). Van Mier used steel brushes to load the specimens which were assumed to
have a negligible influence on the compressive failure of concrete.

Two other test series to investigate the size effect in compressive softening should be
mentioned here. Since concrete and rock qualitatively show the same failure behaviour, it
is also interesting to look at test results for rock specimens. Hudson et al [1972] varied
size and shape of a number of rock specimens to investigate the size effect in compressive
softening. The tests showed a. significant size effect, but it was attributed entirely to the
varying influence of the frictional restraint along the loaded boundaries. Recently Rokugo
et al. [1992] carried out a. series of unia.xial compression tests on specimens of different
height. Their test results indicate that there is a size effect due to localization of cracking

171
and deformations, but the interpretation of these test results is difficult because they are
disturbed significantly by the size effect caused by the frictional boundary restraint.

In Chapter 9 it was shown that the steel brushes used by Van Mier slow softening down.
As this effect of the brushes varies with the height of the specimen, it means that the
post-peak softening curves will not be equal if this effect is eliminated. To investigate the
size effect further, it was decided to carry out tests with teflon platens on specimens of
different height. The teflon platens were used because it was shown that they slow
compressive softening down less than steel brushes. The width of the specimens was
varied to get an indication of the influence of the lateral boundary restraint, which is also
still present when teflon platens are used.

The results of the size-effect tests will be presented and discussed in Chapter 10.2.
Subsequently, the results of micromechanical simulations of compressive failure of
specimens with a different height will be given in Chapter 10.3. A summary of the results
and conclusions will be given in Chapter 10.4.

10.2 TEST RESULTS

Tmsetup .
The different specimen geometries used in the size-effect tests are shown in Fig. 10.2.1.
Three different heights, 50, 100 and 200 mm, and two different widths, 50 and 100 mm,
were used. The tests on the 200-mm-wide specimens, mentioned in Chapter 3.5, are not
discussed here, as these tests suffered from significant nonuniform deformations, as was
discussed in Chapter 8. All tests were carried out with teflon platens. A constant strain
rate of 1x1o-s 1/sec was applied in all the tests irrespective of the height of the specimen.
Further details about the testing technique can be found in Chapter 3.

Extensive measurements of deformations were carried out for all tests. The axial
deformations were measured by means of 1 L VDT and 1 strain gauge at each side of the
specimen. The lateral deformations were measured by means of l, 2 or 3 clip gauges
attached to each side of the specimens of a height of 50, 100 and)200 mm, respectively.
The measurements showed that the rotations of the loading pl~ens for the tests on the
100-mm-wide specimens were always less than in test T1-1 (see Fig. 8.2.1). The
rotations for the tests on the 50~mm-wide specimens showed significantly smaller
rotations of the loading platens, as predicted by equation 8.3.2. In view of the influence of

172
the rotation in test T1-1 on its stress-deformation relation (see Fig. 8.5.1) it is assumed
that the rotations of the loading platens in the size-effect tests did not significantly
influence the stress-strain relations.

A:r:ial stress-strain cun~es

The nominal stress-strain curves for the different specimen geometries are shown in Fig.
10.2.2. This figure confirms that there is a size effect in compressive softening. The
/ /

Specimen codes:
200
LM LS L = Large = 200 mm
M =Medium= 100 mm
S = Small = 50 mm

100 ILD MM
[] MS

sol~ SM ITl ss
~00 ~
50 50
100

Fig. 10.2.1 Specimen geometries used in the size-effect tests

-50

I~
-40

~ -30
"'
~
·- -20
~
-10

-2 -4 -6 -a -10 -12 -14

Nominal strain [o/oo]

Fig. 10.2.2 Nominal stress-strain relations for the different specimen geometries

173
post-peak stress is less for a certain strain, the greater the height of the specimen. The
differences a.re statistically significant. The separate test results are shown in Fig. 10.2.3.
It seems that the width of the specimen has also a slight influence on the stress-strain
relation, but that this in:fluence is less significant. It is also found that the influence of the
specimen geometry on the stress-strain relation extends into the pre-peak regime. This
results in an increase in peak stress as the height or the width of the specimen decreases.

Fig. 10.2.4 shows the dependence of the fracture energy per unit of volume on specimen
geometry. The fracture energies have been calculated by integrating the stress-strain
curves to a softening stress of -2.9 Nfmm'J. Integrating the curves to a certain strain or

-60 -60
w=1001rm

II -50

-40 -
w= 5()tm
11=200mn
11=1001rm I -50

-40
- - I1=200mn
- 11=1001rm

~
-30
I -30

! -20 -20

~ -10

0
I -10

0
0 -2 -4 -6 -8 -10 -12 -14 0 -2 -4 -6 -e -10 -12 -14

Nominal strain [o/oo] Nominal strain [o/oo)

(a) (b)

Fig. 10.2.3 Nominal stress-strain relations for all tests

0.30

0 Width= 50nm
1'0' 0.25 ~ (J Width= 1 OOmm

!>.
0.20
~\
''
''
'
0

'i--,
~
0.15

Ill ------~
0.10
~ 8
0
1\1
U: 0.05

0.00
0 50 100 150 200" 250

Specimen heii;tlt [rrm]

Fig. 10.2.4 Fracture energies per unit of volume for all the specimens

174
deformation would have incorporated either the continuum theory or the fracture
mechanics theory in the method of analysis. It seems that the softening stress is a more
neutral indicator of a certain softening level. Estimating the energy in the tail of the
stress--fitrain relation was not attempted. This can not be done with any degree of
accuracy. Theoretically the tail will extend to a strain of 1 (=1000 °/ 00 ).

Lateral de/Of'TIUI.tiofltl
The lateral deformations of the specimens with a height-width ratio ~ 1 showed a more or
less uniform lateral expansion. The scatter in lateral deformations was significant, as was
already shown in Fig. 6.2.4 in test BS1-1. The relations between axial and lateral strains
did not differ significantly for the different specimen types.
8.0 r-------------------~

7.0
E' 6.0
Bottom
!.
<:
Q 5.0
-;;;
E
0 4.0
Qj
"0 Middle
3.0
lii
~
-;;; 2.0
..J

1.0

0.0
0.0 -0.1 - 02 - 0.3 - 0.4 - 0.5 - 0.6 - 0.7

Axial deformation [mm)

Fig. 10.2.5 Average lateral deformations in test LM3

for the bottom, middle and top of the specimen

The measurements of the lateral deformations of all the slender specimens with a
height-width ratio ~ 2 showed a localization of deformations. An example is given in Fig.
10.2.5 for test LM3. The bottom of the specimen shows a significant lateral expansion
during softening, while the top of the specimen stays almost passive and even shows a
decrease in deformations. This indicates that it is mainly the bottom of the specimen that
fails, while the top of the specimen stays more or less intact.

Crack pattenu
Examples of internal crack patterns, recorded with ultraviolet photography after the
tests, are shown in Figs. 10.2.6 and 10.2.7. There is a clear difference between the crack

175
(a)

(b) (c)

Fig. 10.2.6 Internal crack patterns (middle sections)

for specimens with a width of 50 mm:
(a) test SS4 (u=-3.56 N/mm2, t:=-12.35 ofoo)
(b) test MS3 (u=-2.76 N/mm2, t:=-9.62 ofoo)
(c) test LS2 (u=-2.57 N/mm2, f=-5.71 °/ 00 )

176
(a)

(b) (c)

Fig. 10.2. 7 Internal crack patterns (middle sections)

for specimens with a width of 100 mm:
(a) test SM3 (a=-2.76 N/mm2, f=-9.62 ofoo)
(b) test MM4 (a=-1.87 N/mm2, f=-9.63 °/ 00 )
(c) test 1M3 (a=-3.52 N/mm2, f=--0 .13 ofoo)

177
patterns for specimens with a height-width ratio S 1 and a height-width ratio ~ 2. The
specimens with a height-width ratioS 1 show crack patterns, distributed over the whole
specimen, while specimens with a height-width ratio ~ 2 show parts which are free of
macrocracks. In other words, these tests show localization of cracking. This agrees with
the localization of lateral deformations for test LM3 shown in Fig. 10.2.5.

Furthermore it is found that the crack patterns are qualitatively the same as that already
discussed in Chapter 6.2. The concrete is split up into many small pieces which shear off.
The more or less vertical cracks are distributed and have a spacing which is
approximately the size of the largest fraction of the aggregates (4--8 mm). Localization of
cracking and deformations in the direction of compressive loading is not easy to detect.
For the specimens with a height of 50 mm it can be found that the number of inclined
shear cracks crossed by a vertical line is limited, but is often greater than one. This
indicates some localization. However, for the higher specimens, the crack patterns become
more complex and the number of inclined shear cracks crossed by a vertical line is often
larger than 1 or 2. A parallel shows up with the shear bands found in the case of granular
materials.
-55 , - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - .

a
&--- -o Width= 50rrm

- 50 <>- - -o Width= 1OOrrm

]
6
£ -45
0
5i
bi
-40

-35L---~----~----L---~-----L----~--~

D D D 0
Fig. 10.2.8 Values of peak stress for all tests

Discwsion of size effect in peak strus

Fig. 10.2.2 indicates that peak stress increases as the height or the width of the specimen
decreases. All values for peak stress are given in Fig. 10.2.8. It shows that the scatter in
test results is significant, compared to the influence of height or width. However, multiple
linear regression analysis (Montgomery [1983]) shows that the influence of height and

178
width on peak stress is statistically significant (error level < 0.1 %). Therefore some
possible canses will be briefiy discnssed here.

1. Frictional boundary restraint

First of all, the influence of the frictional boundary restraint by the toefion platens has to
be taken into account when an increase in peak stress is found for a decrease of specimen
height. There is no assurance that the friction by the tenon platens is so slight, that it
causes no differences in peak stress for different specimen heights. Knowing that the
coefficient of friction of the tenon platens is many times smaller than that of untreated
·steel platens, the effect of the height for the 50-mm-wide specimens is much more
significant than what would be expected from tests on specimens of different height
loaded with untreated steel platens (Hudson et al. [1972], Schickert [1980], Rokugo et al
[1990]). Against the influence of the frictional restraint are the facts that an increase of
15% is found for specimens of the same shape when the size is halved-and that an increase
of 8% is found for specimens of the same height when the width of the specimen is halved.
In the first case an indifference to a change in size would be expected, while in the second
case no increase but a decrease would be expected if the frictional restraint were to play
an important role.

2. J:Jrying of specimens
Drying of a specimen causes an increase in compressive strength. For a similar type of
concrete, Bonzel & Kadlecek (1970] found an increase in strength of approximately 6% for
cubic specimens of 150 mm after 24 hours of drying. After a few days a maximum increase
of 12% in strength was found. In the present tests the drying time was significantly
shorter ( ~ 2 hours}, but the size of the specimens was smaller (50 and 100 mm). Probably,
the drying has had some influence, but it is not likely that the differences in peak stress
(up to 22%) have been cansed only by the differences in strength increase caused by
drying.

9. Fracture mechanics
A third phenomenon which should be considered is the size effect on peak stress predicted
by fracture mechanics when a structure fails due to crack propagation (Ba.zant [1984,-
1986], Di Tomasso [1989]). The qualitative dependence of peak stress on a characteristic
size of the structure is shown in Fig. 10.2.9. It has been observed in many types of
concrete structures (Ba.zant [1986]). It is found that peak stress decreases with an increase
of the size of the structure. Nonlinear fracture mechanics predict a. gradual change from
the conventional size-independent strength criterion for small structures to the
size-dependent strength criterion predicted by linear fracture mechanics, which applies to

179
large concrete structures. Qualitatively, it is the same size effect as found in softening,
which extends to the pre-peak region on the assumption that stable crack propagation
can take place before peak stress is reached. Without a doubt, stable crack growth takes
place in a uniaxial compression test before peak stress is reached, which means that peak
stress must also be dependent on the size of the specimen. However, the theory of
nonlinear fracture mechanics does not yet give a quantitative prediction of the influence
of size on peak stress for uniaxial compressive testing.

........
Ill
Ill
I!
1i Plasticity

-5
tt
l '
.-----=---~ . . . -- Linear fracture
' , , !<('mechanics
c
c: 1 ',~::,..

-
E
0
c: Nonllnear fracture
mechanics
C)
0
....1

log (size)

Fig.10.2.9 Size effect for nominal peak stress

~ oflize t./fed ita softtming

Fig. 10.2.2 shows that a similar size effect on softening is found for the present tests with
the teflon platens as was found by Van Mier [1984). A comparison of the post-peak
curves ofthe specimens with a width of 100 mm with the post-peak curves found by Van
Mier for the specimens of the same width is given in Fig. 10.2.10. To facilitate comparison
of results the stresses have been normalized by dividing them by peak stress.
Corresponding to the findings in Chapter 9, the tests with the steel brushes carried out by
Van Mier show a more ductile post-peak behaviour than the present tests with teflon
platens. Fig. 10.2.10 shows that the difference is larger as the height of the specimen
decreases.

Assuming that localization of deformations takes place at peak stress, analogous to tensile
failure, Van Mier compared the post-peak stress-deformation curves for the different
specimen heights and found a striking resemblance (see Fig. 2.3.3). The poshpeak curves
for the present tests are shown in Fig. 10.2.11. These curves were determined by

180
subtracting not just the deformation at peak stress, but by subtracting the inelastic
pre-peak deformations as shown in Fig. 10.2.12a. This was done to take into account
elastic unloading of the continuum due to softening.

1.2 . - - - - - - - - - - - - - - - - - - - - - ,

Brushes (Van Mier)

1.0
Teflon platens
I 0.8
"'"'Gl
~ 0.6
QJ
>
~
0.4
~
0.2
~-
'~
--- --- ----
0.0
0 -2 -4 -6 -8 10

Post-peak strain [o/oo]

Fig. 10.2.10 Comparison of softening curves found by Van Mier [1984]

with the present test results
-50.--------------------.
h/b=200/1 OOrrm
-40 h/b=200/ 50rrm
~ h/b= 100/1 OOrrm
~ -30 hlb= 100/ 50rrm
:f!Gl hlb= 50/1 OOmm
.b h/b= 50/ 50rrm
t1'J
-20

) -10

o~--~--~--~---~--~--~
0.0 -0.2 -0.4 -0.6 -0.8 - 1.0 1.2

Post-peak inelastic displacement [mm]

Fig. 10.2.11 Post-peak stress-displacement curves for all specimen geometries

Fig. 10.2.11 shows that the post-peak curves are not equal, as found by Van Mier, due to
the fact that the influence of the increasing boundary restraint by the steel brushes has
been eliminated. Here it is found that the post-peak resistance for a certain deformation

181
increases as the height of the specimen increases. This indicates that the post-peak
resistance is not only a result of a local process, but also depends on a diffuse process,
which makes that the height, and thus the volume, of the specimen has an influence. The
same is observed when the estimated local fracture energy per unit of area is compared for
all the different specimens (Fig. 10.2.13). The local fracture energy has been estimated by
integrating the post-peak curves of Fig. 10.2.11 to a softening stress of -2.9 N/mm2, as is
shown in Fig. 10.2.12b. Iflocalization of cracking and deformations were perfect, the
fracture energy would not depend on the height of the specimen as is found in Fig.
10.2.13. This indicates that compressive fracture is a mixture of a local and a. diffuse
process. A possible division into local and volumetric components is shown in Fig. 10.2.13.

a unloading a
continuum
Continuum
damage
energy

t,W
Continuum Local inelastic
inelastic deformation
deformation
(a) (b)

Fig. 10.2.12 Determination of continuum and local components of

deformations and energies
30
o width= 50mm
10 25 o width= 1 OOmm

j 20
0
0

r
Q)
15
1
Continuum fracture energy
a
0
10
----------- -1-------
e
IL
5 Local fracture energy

0
J
0 50 100 150 200 250

Specimen hei~t [mm)

Fig. 10.2.13 Post-peak fracture energies for all different specimen geometries

182
Frictitn&tU tiiOf'li:
It is not easy to estimate the influence of the lateral boundary restraint on the fracture
energy. What can be estimated is the external work by the friction between loading
platen and specimen. The frictional stresses can be estimated by multiplying the nominal
stress by the coefficient of friction. The displacements needed to calculate the work can be
estimated from the lateral deformations of the specimen. It was assumed that the lateral
displacement& showed a linear variation over the cross-section. The results are shown in
Fig.10.2.14. It shows that the external frictional work should not be neglected, not even
for the teflon platens with their relatively low coefficient of friction. For the specimens
with a height of 50 mm and a width of 100 mm, the frictional work is in the order of 10%
of the local fracture energy. Although the coefficient of friction is small, the external
frictional work is significant because the lateral deformations are significant during
softening. For all the other specimens, the external frictional work is relatively less. The
external frictional work has to be subtracted form the fracture energies estimated
previously to find the actual fracture energy of the specimen. It should be noted that this
does not eliminate the influence of the frictional stresses on the fracture energy of the
specimen itself. This influence is still difficult to estimate.

1.5 . - - - - - - - - - - - - - - - - - - - - - ,

0 width= 50mm
Cl
1:( Cl width= 100mm
''
Cl ' ' ....
,o
EJ',_
Cl
i
g .. , -- -~

--g...--
0
0

0.0 L -_ _ --~. _ _ __..L.._ _ _

--- ----&
..~..,_ _ _ _..___ __

0 50 100 150 200 250

Specimen height [mm)

Fig. 10.2.14 External frictional work for all specimen geometries

The external frictional work around peak stress was significantly less than at the end of
softening because the lateral deformation at peak stress were relatively slight. It was

183
found that the frictional work at peak stress was in the order of magnitude of 0.5% of the
energy found by integrating the stress-strain curve up to peak stress.

lajluence of the qecimea volume oa 10fte:n.iag

How can the influence of the specimen volume on the post-peak behaviour be explained?
It seems .that the diffuse cracking should be taken into account. Especially, the specimens
with a height-width ratio S 1 show a significant amount of almost vertical cracks
distributed over the entire volume. The more slender specimens with a height-width ratio
~ 2 also show a significant number of vertical cracks, but they are often not distributed
over the entire volume. The many vertical cracks are probably a result of the forces
introduced into the specimen by the shearing of the concrete pieces.

The average distance between the vertical cracks is approximately equal to the size of the
largest aggregate fraction (4- 8 mm). When it is assumed that the vertical cracks go
completely through the specimen (area= b x h) in two directions, the energy needed for
the formation of these vertical cracks can be estimated by

b 8 b 2h
E=2 b h Gr =- - - Gr {10.2.1a)
0. 75dmax 3 dmax

or

8 h
E/b2=---Gr (10.2.1b)
3 dm a x

Here, dmax is the maximum aggregate size. The tensile fracture energy per unit of area Gr
can be estimated as 0.12 Nmm/mm2 for a normal-strength gravel concrete with a
water-cement ratio of 0.5 and a maximum aggregate size of 8 mm (Hordijk [1991]). Then
equation 10.2.1b results in fracture energies per unit area of 2, 4 and 8 Nmm/mm2 for
specimens with a height of 50, 100 and 200 mm, respectively. Comparing these fracture
energies to the values found in Fig. 10.2.13, it shows that this phenomenon can not be
neglected and can explain the dependence of the fracture energy on the volume of the
specimen. The spread, which is also found in the more inclined cracks, will make the
contribution of the volume to the fracture energy more significant, because more energy is
needed to create these cracks than the vertical splitting cracks.

184
10.3 MICROMECHANICAL SIMULATIONS

In addition to the tests, a limited number of simulations was carried out to investigate
the size effect in compressive softening and to test the capabilities of the micromechanica.l
model. Three tests on specimens with a width of 50 mm and a height of 50, 100 and 200
mm were simulated. The tests on specimens with a width of 100 mm were not simulated
because of the limitations in computer time.

Details of the m.it:rtm&echanicalsimulatior&tl

The parameters for the mesh generation were chosen in such a way that the procedure
resulted in a large aggregate content (2-8 mm) of 0.375. This was equal to the large
aggregate content in the concrete used for the size-effect tests. The stochastic parameters
to disturb the interface mesh were kept equal to those used earlier to generate the meshes
for the simulations in Chapters 6, 7 and 9.

The material parameters used in the simulations were approximately equal to those used
in the previous simulations. They are summarized in Table 10.3.1. The moduli of
elasticity were increased slightly, because in the previous simulations it was found that
the actual modulus of elasticity of the concrete was underestimated. Also the values for
the cohesion of bond and mortar were slightly increased, because the tests on the 50 mm
wide specimens showed a higher peak stress than the boundary condition tests.
Furthermore, a small dilatancy of 0.15 was introduced in the interfaces.

Continuum Interface

Aggregate Bond lortar

E=75000 N/•2 lf/u2 ft=3.0 Nfo2 ft=6.0 N/u2
c=ll.O Nfu2 c=22.0 N/u2
v=0.2
1 1=5x10S NfuS tgp=0.3
12=5x105 N/uS tgt-=0.15
Wtmax=0.02 IIII tgx=O.O
Wamax=LOO IIII

Table 10.3.1 Material parameters used for the simulations of the size-effect tests

All specimens were loaded by moving the upper loading platens downwards in
deformation control. Upper and lower loading platens were not allowed to rotate or move
in the lateral direction. The properties of the tefl.on platens were simulated as closely as

185
possible with the interface model. Figure 10.3.2 shows the interface parameters and the fit
of the model parameters to the results of the friction tests on tefion platens (see also
Chapter 3.1). The reduction of friction during the first sliding was modelled by
introducing softening for the first 0.115 mm of sliding.

1.0 , - - - - - - - - - - - - - - ,

0.8 CJ 1" peak

(test)
[;. w=0.5rrm
Interface paraaeters 0.6 (test)
CJ
0 w=1.5rrm
It · 5x10S lf/DS (test)
l2 = 200 lf/mma - - - 1" peak
ft = 0.0 lf/mm2 (model)
c = 0.39 N/mm2 - - w>0.115mn
tgp = 0.0115 (model)
tg; = o.o
tgx = o.o
Wtmax = 0.0 1111
W2max = 0.115 1111

(a) Normal stress IN!rrvn"l (b)

Fig. 10.3.2. (a) Interface properties to simulate the frictional properties of the tefion
platens and (b) comparison of the interface properties and the results of tests
on tefion platens

Results of simvlo.tions
The calculated nominal stress-strain curves are given in Fig. 10.3.3. A significant size
effect for softening is predicted by the micromechanical simulations. On comparing the
post-peak stress-deformation curves in Fig. 10.3.4., which are corrected for the inelastic
pre-peak deformations, it is found that the micromechanical model, with the present set
of material parameters, predicts an almost perfect localization after peak stress has been
passed. This does not agree with the test results, which showed an influence of the
continuum on the fracture process.

Fig. 10.3.5 shows the distribution of damage, the damage parameter x. of the interface
model, for the different specimens at the end of the simulations. The stress-free interfaces
are shown in Fig. 10.3.6. All specimens fail due to the formation of continuous patterns of
inclined shear cracks, which are incidentally connected by more vertical splitting cracks.

186
 -50

w= 50mm
-40 h= 50mm
"E h=100mm
i -30
h=200mm
m
<ll
~
tii -20
c:
E
~ '\
-10 \

.... .....
0
0.000 -0.002 -0.004 -0.006 -0.008 -0.010

Nominal strain [-)

Fig. 10.3.3 Stress-.strain curves found for the micromechanical simulations

-60,-----------------------------------~

w= 50mm
h= 50mm

I~
-30
h=100mm
h=200mm

b
tJ)

~ -20

~ -10

OL-----L-----L-----L-----L-----L---~
0.0 -0.1 -02 -0.3 -0.4 -o.5 -0.6

Local deformation [mm)

Fig. 10.3.4 Post-;>eak stress-deformation curves, which are corrected for the pre-peak
inelastic deformations

Localization of cracking and deformations is most clearly found for the 200-mm-high
specimen, which fails due to the formation of one crack band. The bottom half of the
specimen shears off to the left, which allows the axial deformation of the specimen during
softening. The forces introduced into the parts of the specimen which shear off, cause
secondary cracks to be formed and the crack band to be widened. The same can be found
in the specimens with a. height of 50 and 100 mm, but it can not be recognized as easily,

187
(a)

(b) (c)

Fig. 10.3.5 Distribution of damage (damage parameter K. of the constitutive

model) for the different specimens at the end of the simulations:
(a) h =50 mm, o=-2.99 N/mm2, E=-11.4 ofoo
(b) h = 100 mm, a=-3.28 N/mm2, E=-4.5 o/ 00 ,
(c) h = 200 mm, o=-5.95 N/mm2, c=-2.1 0/ 00

188
 ·,- ·----------·
\ )..,
I l .
I . ' I
b J I • }
I I
, 1
'·· I r: I
I

(a) I
I
/
'I f\
-:_. C. _)- '{:..:_
\\ /
~·
I ~·.
1:.
{
1 I
\
\
I
I

I '·
\ f \

I
'\
~'\ ·. ,,
I
\
I
;,
' \
',

'.L\\
!?\
f) \
\
\ /

\
', ,•I•
I I
', , I
---- _j {__ ~->

(b) (c)

Fig. 10.3.6 Stress-free interfaces for the different specimens

189
because the shear cracks zigzag through the specimen and are surrounded by the
secondary cracks.

There is a tendency in the simulations to a more distributed crack formation, but it is less
than what is observed in the test results. The 5G-mm-high specimen shows the most
significant spread in cracking. This is reflected by the slightly more ductile post-peak
behaviour. In this simulation, the left part of the shear crack pattern locked up during
softening, which resulted in the increase of the post-peak toughness observed in Fig.
10.3.2 and the splitting cracks in Figs. 10.3.5 and 10.3.6. This was due to the fact that
new contacts were created as a result of the shearing of the cracks and that a piece of
mortar was tom loose and got stuck in the shearing crack. This caused a redistribution of
forces, which resulted in additional cracking in the left part of the specimen. It shows that
taking geometrical nonlinearity into account is an essential aspect of this type of
calculation. It shows further that it will be necessary to increase the geometrical
heterogeneity in the model to find a cracking and failure process which comes closer to
the test results. A more significant heterogeneity will stimulate the formation of
additional cracks and will make the influence of the volume of a specimen on the failure
process more significant. It is not unlikely that a part of the necessary increase in the
heterogeneity can be found in extending the model to the third dimension. Also an
increase of the coefficient of friction and the dila.tancy can help too. It was shown by
Hobbs [1990) that the size of a shear band in a heterogeneous material increases when the
friction or the dilatancy are increased.

10.4 CONCLUSIONS

The test results presented in this chapter have affirmed those of Van Mier [1984) that
there is a size effect in compressive softening. It has been shown that the lateral boundary
restraint by the loading platens plays an important role in the size effect in compressive
softening and that it should always be taken into account. It was found that steel brushes
as well as te:O.on platens can have a significant influence on size effect tests. The use of the
steel brushes by Van Mier led him to discover that the post-peak stress-deformation
curves were equal. Based on this result, he concluded that there is a similar size effect in
compressive softening as there is in tensile softening. In Chapter 9 it was shown that
te:O.on platens slow down softening less than steel brushes do. The use of these te:O.on
platens in the size-effect tests showed that localization of cracking and deformations is
present in compressive softening, but that it is not equal to those phenomena for tensile

190
softening and that they should be described by a more complex model. It is indicated that
compressive softening is not only determined by a local process, but also by a diffuse
process, which results in the influence o£ the volume o£ a specimen on the fracture energy.
The influence of the volume can be explained from the spread in cracking observed in the
compression tests.

The micromechanical model is very capable of predicting localization of cracking and

deformations resulting in the size effect. However, in its present state and with the
present material parameters it overestimates the localization and size effect in
·compressive softening. A tendency to the creation of more distributed cracking is
observed, but this is not enough to let the simulation results compare well to the test
results. A solution can be found in an increasing of the heterogeneity of the model.

The mixture o£ local and volumetric components in the compressive fracture process
indicates that crack-band, shear-band or nonlocal models can be applicable. However,
the tests and micromechanical simulations show that the characteristic size present in
these models will not be a simple constant, but a complex variable influenced by a
number of parameters.

191
 CHAPTERll

RETROSPECTIVE VIEW

Subject of research
In this research, softening of concrete specimens loaded in uniaxial compression has been
investigated by analysis and modelling of the structural behaviour of these specimens.
This has been done for a normal-strength gravel concrete. Sofiening is a decrease in
mechanical resistance due to a continuous increase of deformation forced upon a specimen
or structure. It is caused by a crack-growth process. The specimen is split up in a number
of pieces which shear off. Softening is the combined. response of these pieces and the
cracks separating them and has therefore to be analysed as a structural property.
Continuum mechanics no longer applies, but fracture mechanics govern the behaviour.
The structural behaviour is determined by the crack formation process, which depends on
the heterogeneous structure, the dimensions and the boundary conditions of the concrete
specimen.

Recortli.ng of crack patterns

Extensive surface measurements during a test can give valuable information about the
softening process, but this information is always indirect. It does not reveal details of the
process of crack growth which causes sofiening. The test specimen is treated as a black
box. Therefore a technique was developed to record in detail the macrocrac~ patterns in
cross-sections of specimens after a test. All continuous macrocrack patterns! are filled
with a fluorescent epoxy resin by vacuum impregnation. The fluorescent epoxy resin
stabilizes the fractured specimen during the cutting process and makes it possible to
record the internal crack patterns by means of ultraviolet photography. This technique
has proved to be simple and effective. It can be extended easily to the detection of
discontinuous rnicrocrack patterns by combining it with the techniques developed by Hsu
et al. [1963].

New micromechanical model

A new rnicromechanical model has been developed to investigate the structural behaviour
of softening concrete. It describes concrete as a two-phase material consisting of large
aggregates in a mortar matrix. Its mechanical behaviour is analysed by means of the
computer program UDEC, which is based on the Distinct Element Method (Cundall
[1979]). Most characteristic of the model is its continuous mesh of interface elements of

192
which the geometry depends on the aggregates. It describes all potential crack paths. No
cracking is allowed in the continuum elements. A new constitutive law has been
developed to describe the failure of the interfaces under combinations of tensile and shear
loading. The heterogeneity of concrete is further modelled by different moduli of elasticity
for aggregate and mortar and different strengths for interfaces between aggregate and
mortar and interfaces in the mortar. Crack formation in the model is determined by the
heterogeneous structure, the boundary conditions and the dimensions of the concrete
structure.

Cra.d: jo'ITIUfl.ion and failure me.ch.anism.

The test results show that compressive failure is caused by a gradual process of crack
growth, which becomes unstable and localizes at peak stress. The concrete is split up in
pieces which shear off. A significant increase in lateral expansion is characteristic and
essential to the formation of this failure mechanism.

Internal crack patterns recorded after a test show that the large aggregates have a
significant influence on the crack formation. In normal-strength gravel concrete, the
cracks grow around the aggregates, because the aggregates have a considerable strength
while the bond between the aggregates and the mortar is relatively weak. Many splitting
cracks are found, which often have a spacing in the order of magnitude of the largest
aggregate fraction.

Not the splitting cracks are most important for compressive failure, but the shear cracks,
because they are essential to the failure mechanism. Many splitting cracks are
prominently visible, but only modify compressive failure, as their formation takes a
relatively smaller amount of energy than that of shear cracks. The shear cracks are often
not well visible and hard to distinguish, because their length is relatively small, they are
surrounded by many other cracks and because they do often not open far. This explains
why the splitting cracks are often mentioned to he the cause of compressive failure.
However, there are examples of uniaxial and multiaxial compressive tests which show
failure due to the formation of shear cracks only. Simulations with the micromechanical
model have shown that the assumption of shear failure being the cause of compressive
softening results in a realistic description of compressive failure.

Nonumfonn tle/ofm~JtiDns
Highly nonuniform deformations have been found in the softening regime of uniaxial
compression tests due to rotation of the loading platens. In these tests the loading platens
were fixed effectively to the loading apparatus. It was observed that the nonuniformity of

193
deformations could have a significant effect on the nominal stress-deformation relation
registered in the test. The phenomenon is similar to that found in uniaxial tensile tests
(Hordijk et al. [1987]).

It has been shown that softening stimulates the above-mentioned nonuniform

deformations. A criterion has been derived, which gives an indication of the tendency of a
test to suffer from nonuniform deformations. It shows that the bending stiffness of the
loading apparatus is of importance in a softening test. Satisfying the criterion will not
prevent any nonuniformity of deformations, as suggested by other researchers. This will
depend further on the nonuniformity of loading.

A simple numerical model has been proposed to calculate the rotation of a loading platen
in a compression test and its effect on the nominal stress-deformation relation. It takes
into account the dimensions of the specimen, the constitutive behaviour of concrete, the
bending stiffness of the loading apparatus and the nonuniformity of loading. Good
agreement was found between the results of the numerical simulations and test results.

It should be noted that even when no rotation of the loading platen is found, this does not
mean that the deformations inside a specimen are uniformly distributed. Tests and
micromechanical simulations have shown that crack propagation always takes place,
which results in a nonuniform distribution of deformations in the specimen. The
nonuniformity of deformations found in compressive softening tests is one more exam:rle
that failure is often asymmetric and this should always be taken into account when
modelling failure.

Fized or hinged loading platen?

It has been shown that the use of a hinged loading platen in a compressive softening test
results in an extreme rotation of this platen and consequently in a highly nonuniform
distribution of deformations in the specimen. The test is very complex and difficult to
analyse. It results in a more brittle post-peak behaviour than a test with fixed loading
platens on an identical specimen. Peak stress in such a test is more sensitive to an
eccentricity of load. The formation of the characteristic pronounced shear crack found in
tests on slender specimens is probably the result of the use of a hinged loading platen.
Softening tests can be better carried out with fixed rather than with hinged loading
platens, because the first-mentioned tests are less sensitive to many parameters and can
be analysed more easily. Researchers, interpreting test results mentioned in literature
should be aware that data concerning peak stress and softening have often been
determined in tests with hinged loading platens.

194
Lateral boundary restraint
It has been shown that compressive failure of concrete is a three-dimensional process,
which is highly sensitive to a restraint of the lateral deformations. Lateral expansion is
essential to the formation of the failure mechanism. Compressive failure cannot be
predicted accurately when the lateral boundary conditions are not taken into account
explicitly. For a cubic specimen loaded in unia:x:ial compression, the post-peak behaviour
can vary, depending on the lateral boundary restraint by the loading platens, from rather
ductile to fairly brittle for a normal-strength concrete. One should always be aware of the
influence of the lateral boundary restraint in unia:x:ial compression tests. Even loading
platens like teflon platens and steel brushes can have a significant influence on softening
in such tests. It seems impossible to develop a loading platen with a zero boundary
restraint. This is of relative importance. A correct model should be able to describe
compressive failure for the complete range of boundary conditions.

Influence of size
The present research has confirmed that there is a size effect in softening of concrete
loaded in uniaxial compression. This was earlier found by Van Mier [1984). The
post-peak behaviour becomes more brittle when the specimen size increases. The fracture
energy increases not in proportion to the size of the specimen due to localization of
cracking. Analysis of test results shows that the size effect in compressive softening is
more complex than in tensile softening. Localization of cracking is present in compressive
softening, but it has a more diffuse character than in tensile softening. Test results of Van
Mier [1984] suggested similarity, but it has been shown in the present research that this
was caused by the lateral boundary restraint by the steel brnshes used by Van Mier in his
tests.

Ruults of micromeclw.nicalsimulations
The micromeehanical model provides a realistic description of uniaxial compressive
failure. Both distributed microcracking in the pre-peak regime and more localized
macrocrack growth in the post-peak regime are described well by the model. According
to the model, a gradual change is found: microcracks grow in size until the total specimen
becomes unstable and the final macrocrack pattern starts to develop. Just past peak
stress, softening is mainly due to the growth of the final macrocrack pattern. Further
softening is mainly due to a decreasing resistance of the macrocracks, which open and
shear off.

A parameter study has shown that compressive failure as described by the model is
mainly dependent on the formation of shear cracks. A variation in the tensile fracture

195
energy only modifies compressive softening. It has been shown that the crack patterns
fonnd in tests, with the significant amonnt of splitting cracks, can be accurately
reproduced by the micromechanical model, but that shear failure then still determines
compressive failure. This supports the postulate that shear failure is the determining
factor in compressive failure and not instability caused by the formation of a dense
pattern of splitting cracks.

The tests with a varying lateral bonnda.ry restraint have shown that the micromechanical
model is perfectly able to describe the structural behaviour of compression tests under
varying boundary conditions. They confirm that unia.xial compressive softening is
extremely sensitive to the slightest lateral boundary restraint.

The size effect in compressive softening is described by the micromecha.nical model, but is
overestimated. The localization of cracking is stronger than is observed in tests. The
spread in cracks should be increased. A possible solution is an increase of the geometrical
heterogeneity. That can be done by modelling the concrete structure down to a smaller
scale level, but also by an extension of the model in the third dimension. One should
always realize that a two-dimensional heterogeneous model is homogeneous in the third
dimension.

Tensile softening is also described realistically by the micromechanical model. The results
compare well with test results of Van Mier (1991]. The model shows that the tail of the
tensile stress-deformation relation is due to the formation of overlaps betw~n cracks,
which is caused by the heterogeneous structure of concrete. It is fonnd that ~he fracture
energy increases when the heterogeneity increases. This implies that softening of concrete
can probably be described realistically by brittle behaviour of the microstructure when
the microstructure is modelled down to a smaller scale level.

Future research
The present research has shown that our knowledge of compressive failure is still limited
and that a significant amonnt of work has to be done before a model can be proposed
which describes concrete failure nnder general loading conditions. The micromecha.nical
model has been shown to be promising in this respect, but it is a complex research tool,
which is limited at the moment by the calculation time involved. Its approach, based on
the heterogeneous structure of concrete and fracture mechanics, seems adequate and it
can be used for further research and serve as a basis for the formulation of more simple
models. Further research should be directed to

196
1. Different loading conditions
The extension of the investigations to the complete range of multiaxial loading conditions
will improve our knowledge further and can be used to develop the micromechanical
model. This has to be done with special attention to crack formation, localization of
deformations and the resulting size effect. Also, it will be necessary to investigate failure
in the case of gradients of stress or deformation.

2. Different types of concrete

The type of concrete has a significant influence on the failure of concrete. Therefore the
present research has to be extended to other types of concrete. This is especially
important in the case of more brittle types of concrete like high-i!trength concrete and
lightweight concrete. Investigating other types of concrete will teach us more about
failure of concrete in general and what aspects are of importance. The influence of the
type of concrete can be investigated well with the micromechanical model.

3. Time-dependent effects
The present research has not taken into account the effects of loading rate, creep or
drying out on the failure of concrete. They certainly have a significant influence on the
failure of concrete and have to be investigated. Micromechanical models can be useful
research tools in these investigations.

Concluding remtri
A standard compression test on a 150-mm...-cubic specimen, as used in the Netherlands,
gives the impression that concrete loaded in compression is ductile. This is not true. One
should always be aware of the dependence of the ductility on the lateral boundary
restraint and the size of the structure. Post-peak ductility decreases when the lateral
boundary restraint decreases or the size of the structure increases. It has been shown that
the question "How brittle is concrete?" can not be answered. This can only be done for a
structure with known composition, dimensions and boundary conditions. This has to be
investigated further.

197
APPENDIX A: MESH GENERATION PROCEDURE

As an input for the mesh generator, the following variables can be specified:
1. height and width of the specimen
2. mean distance between the aggregate centres
3. standard deviation of displacement of the centre of the aggregate away from its
original position
4. mean value and standard deviation of the thickness of the mortar layer between the
aggregates
5. standard deviation of rotation of the two sides of the mortar layer
6. minimum distance between two corners of the aggregates

bond Interface
---mortar interface

Fig. Al Basic composite structure and mesh of interfaces

Centre
aggregate t
~Sides of
aggregates

Fig. A2 Procedure for mesh generation

The basic structure of the mesh of interfaces is shown in Fig. Al. The mesh generator
starts to build a. regular pattern of equilateral triangles as shown in Fig A2. The corners
of the mesh are the centres of the aggregates. The distance between the centres is

A.l
specified in the input for the mesh generator. Subsequently, the centres are given a
random displacement away from their original position. The direction is chosen at random
and the distance is normally distributed with a mean value of zero and a specified
standard deviation. Then each aggregate is given a relative size parameter between 0.5
and 1.5. This means that the generated aggregates represent, roughly estimated, the
fraction larger than one third of the maximum aggregate size. Each aggregate has 6
neighbours, unless it is situated on the edge of the mesh.

For each pair of neighbouring aggregates the following routine is carried out. The
thickness of the mortar layer between the aggregates is calculated on the the basis of
mean value and standard deviation specified in the input. The resulting distance between
the two centres of the aggregates is distributed over the two aggregates in proportion to
their size parameters. First the sides of the mortar layers between the aggregates are
assumed to be perpendicular to the line connecting the two aggregate centres. Next a
rotation of the sides is calculated from a mean value zero and the specified standard
deviation. After this routine has been carried out for all neighbouring aggregates, all the
sides of the aggregates are known and their corners can be calculated.

A final inspection of the mesh is carried out to find out if corrections are nec~ssary.
Sometimes, corners have to be deleted because the random generation process places them
outside the aggregate. Then the hexagon reduces to a polygon with fewer corners. Corners
of aggregates that penetrate into other aggregates are drawn back until they •fall outside
the other aggregate. Corners that are closer than a specified minimum distance are put
together, because extremely small and irregularly shaped elements must be avoided in the
UDEC calculations. Finally, a part, having the size and shape of the specimen is cut out
of the mesh at a randomly chosen position and angle of orientation.

The input for the mesh generator, which has been used to simulate the concrete for the
boundary condition tests, was:

Mean distance between aggregate centers 16.0 mm

Standard deviation of displacement of aggregate centers 3.0 mm
Mean thickness of mortar interlayers 6.3 mm
Standard deviation of thickness of mortar interlayers 1.2mm
Standard deviation of size parameters of aggregates 0.3
Standard deviation of rotation of sides of mortar
interlayers 0.40
Minimum distance between two corners of the aggregates 2.0mm

A.2
 REFERENCES

ACI committee 446 [1990]

"Fracture mechanics of concrete: concepts, models and determination of material
properties"
Concrete International, 12 (12), 67-70
Ahmad, S.H. & Shah, S.P. [1982]
"Complete tria.x:ial stress-fitrain curves for concrete"
Journal of the Structural Division, ASCE, 108 (ST4), 728-742

Ahmad, S.H., Shah, S.P. & Khaloo, A.R.J1986]

"Orthotropic model of concrete for tria.x:i stresses"
Journal of Structural Engineering, ASCE, 112 {1), 165-181
Alexander, K.M., Wardlaw, J. & Gilbert, D.J. U968l
11
Aggrega.te--eement bond, cement paste strength and the strength of concrete"
in Proceedings of the International Conference on the Structure of Concrete, London,
September 1965, Cement & Concrete Association, session 3, 59-81
Alexander, M.G. & Blight, G.E. [1986]
"The use of small and large beams for evaluating concrete fracture characteristics"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F .H. Wittma.nn, Elsevier
Science Publishers B.V., Amsterdam, 323-332
Andenaes, E., Gerstle, K. &; Ko, H.Y. [1977)
"Response of mortar and concrete to biaxial compression"
Journal of the Engineering Mechanics Division, ASCE, 103 {EM4), 515-526

Ansari, F. [1986]
11
Analysis of micr()-{!racked zone in concrete"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F.H. Wittmann, Elsevier
Science Publishers B.V., Amsterdam, 229-240
Bartenev, G.M. &; Lavrentev, V.V. [1981]
11
Friction and Wear of Polymers"
Tribology series 6, eds. L.H. Lee & K.C. Ludema, Elsevier, Amsterdam
Bascoul, A., Kharcbi, F. & Maso, J.C. [1989]
"Concerning the measurement of the fracture energy of a micr()-{!Oncrete according to the
crack growth in a three points bending test on notched beams"
in "Fracture of Concrete and Rock", eds. S.P. Shah & S.E. Swartz, Springer-Verlag, New
York, 396-408

Bazant, Z.P.J1976J
"Instability, uctility and size effect in strain-fioftening concrete"
Journal of the Engineering Mechanics Division, ASCE, 102 (EM2), 331-344

Bazant, Z.P. [1983]

"Comment on orthotropic models for concrete and geoma.terials"
Journal of Engineering Mechanics, ASCE, 109 {3), 849-865

R.l
Bazant, Z.P. [1984]
"Size effect in blunt fracture: concrete, rock, metal"
Journal of Engineering Mechanics, ASCE, 110 ( 4), 518-535

Bazant, Z.P. [1986]

"Mechanics ol distributed crackin:g"
Applied Mechanics Review, 39 (5), 675-705
Bazant, Z.P. & Gambarova, P.G. (1984:)
"Crack shear in concrete: crack band nncroplane model"
Journal of Structural Engineering, ASCE, 110 (9), 2015-2035

Bazant, Z.P. & Kim, S.S. [1979]

"Plastic-fracturing theory for concrete"
Journal of the Engineering Mechanics Division, ASCE, 105 (EM3), 407-428

Bazant, Z.P. & Oh, B.H. [1983)

"Crack band theory for fracture of concrete"
Materials and Structures, RILEM, 16, 155-177

Bazant, Z.P. & Oh, B.H. [1985J

"Microplane model for progresstve fracture of concrete and rock"
Journal of Engineering Mechanics, ASCE, 111 (4), 559-582
Bazant, Z.P. & Ozbolt, J. [1990] .
"Nonlocal microplane modcl: tensile and compression fractures and triaxial damage"
in Proceedings of Second International Conference on Computer Aided Analysis and Design
of Concrete Structures, eds. N. Bicanic & H.A. Mang, Pineridge Press, Swansea, 809-829
Bazant, Z.P. & Ffeiffer, P.A. [1986]
"Shear fracture tests of concrete"
Materials and Structures, RILEM, 19 (110), 111-121

Bazant, Z.P., Belytschko, T.B. & C~, T.P. (1984]

"Continuum theory for strain-fioftening
Journal of Engineering Mechanics, ASCE, 110 (12), 1666-1692
Bazant, Z.P., Bishop, F.C. & Chang, T.P. [19861
"Confined compression tests on cement paste ana concrete up to 300 ksi"
ACI Journal, 83 (48), 553-560
Bazant, Z.P., Tabbara, M.R., Kazemi, M.T. & Pijaudier-Cabot, G. ~1990]
"Random particle modcl for fracture of aggregate or fiber composites
Journal of Engineering Mechanics, ASCE, 116 (8), 1686-1705

Bazant, Z.P., Xi, Y. &; Reid, S.G. [1991]

"Statistical size effect in quasi-brittle structures: I. Is Weibull theory applicable?, II.
Nonlocal theory"
Journal of Engineering Mechanics", 117 (11), pp 2609-2640

Bertacchi, P. & BelloUi, R. [1972]

"Experimental research on deformation and failure of concrete under triaxialloads"
in Proceedings of the RILEM Symposium on the Deformation and the Rupture of Solids
Subjected to Multiaxial Stresses, paper 1/3, 37-52

R.2
Berthaud, Y. [1987]
"An ultrasonic testmg method: an aid for material characterization"
in preprints of Proceedings of SEM-RILEM International Conference on Fracture of
Concrete and Rock, Houston, Texas, eds. S.P. Shah & S.E. Swartz, 644-654
Bieniawski, Z.T. [19671
"Mechanism of brittle fracture of rock, part I, II and ill"
International Journal of Rock and Mining Science, 4, 395-430
Bonzel, J. & Kadlecek, V. [1970]
"Einfluss der Nachbehandlung und des Feuchtigkeitszustands auf die Zugfestigkeit des
Betons"
Betontechnische Berichte, 99-132
Broek, D. (1986)
"Elementary Engineering Fracture Mechanics"
Martinus Nijhoff Publishers, Dordrecht, The Netherlands
Burt, N.J. & Dougill, J.W. [1977]
"Progressive failure in a model heterogeneous medium"
Journal of the Engineering Mechanics Division, ASCE, 103 (EM3), 365-376
Carol, I. & Prat, P.C. [1991]
"Smeared analysis of concrete fracture using a microplane-based multicrack model with
static constraint"
in "Fracture Processes in Concrete, Rock and Ceramics", eds. J.G.M. Van Mier, J.G. Rots
& A. Ba.kker, RILEM Proceedings 13, E. & F.N. Spon, London, 619-628
Carpinteri, A. [1986]
"Mechanical Damage and Crack Growth in Concrete"
Martinus Nijhoff Publishers, Dordrecht, The Netherlands
Carrasquillo, R.L., Nilson, A.H. & Slate F.O. [1981a]
"Properties of high strength concrete subject to short-term loads"
ACI Journal, 78, 171-178
Ca.rrasqnillo, R.L., Slate, F.O. & Nilson A.H. [1981b}
"Microcracking and behavior of high strength concrete subject to short-term loading"
ACI Journal, 78, 179-186
CEB-FIP [19911
"CEB-FIP Mod'el Code 1990"
Final draft, Bulletin d'information No. 203, 204 and 205, CEB, Lausanne
Cedolin, L. & Mulas, M.G. [1984]
"Biaxial stress-strain relation for concrete"
Journal of Engineering Mechanics, ASCE, 110 (2), 187-206
Cedolin, L., Crutzen, Y.R.J. & Dei-Poll, S. (1977]
"Triaxial stress-strain relationship for concrete"
Journal of the Engineering Mechanics Division, ASCE, 103 (EM3), 423-439
Cedolin, L., Dei Poli, S. & lori, I. [1987]
"Tensile behavior of concrete"
Journal of Engineering Mechanics, 113 (3), 431-449

R.3
Chen, A.C.T. &; Chen, W.F. (1975]
"Constitutive relations for concrete11
Journal of the Engineering Mechanics Division, ASCE, 101, 465-481
Chen, E.Y.T. &; Schnobrich, W.C. [1981]
"Material modeling of plain concrete"
in Proceedings of IABSE Colloquium on Advanced Mechanics of Reinforced Concrete,
Delft, IABSE, ZUrich, 31-52
Chen, W.F. &; Baladi, G.Y. U985J
"Soil Plasticity: Theory and lmplementation"
Elsevier, Amsterdam
Chen, E.S. &; Bu;yukozturk, 0. [1985]
"Constitutive model for concrete in cyclic compression"
Journal of Engineering Mechanics, ASCE, 111 (6), 797-814
Chen, W.F. &; Ran, D.J. f1988]
"Plasticity for Structural ":Engineers"
Springer-Verlag, New York
Chen, W.F. &; Saleeb, A.F [1982] .
"Constitutive Equations for Engineering Materials- Vol.l: Elasticity and Modeling"
John Wiley and Sons, New York
Chhuy, S., Canna.rd, G., Robert, J.L. &; Acker, P. [1986] .
"Experimental investigations into the damage of cement concrete with natural aggregates"
in "Brittle Matrix Composites 1", eds. A.M. Brandt and I.H. Ma.rshall, Elsevier Applied
Science, London, 341-354
Chhuy, S., Canna.rd, G., Robert, J.L. &; Rossi, P. [19891
"Experimental investigations on the damage of Portlana cement concrete made with
natural aggregates"
Materials and Structures, RILEM, 22, 98-106
Comelissen, H.A.W., Hordijk, D.A. &; Reinha.rdt, H.W. (1986)
"Experiments and theory for the application of fracture mechanics to normal and
lightweight concrete"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F .H. Wittmann, Elsevier
Science Publishers, Amsterdam, 565-575
Cope, R.J., Rao, P.V., Cla.rk, L.A. &; Norris, P. f1980]
"Modelling of reinforced concrete behaviour for ffnite element analysis of bridge slabs"
in "Numerica.l Methods for Nonlinea.r Problems", eds. C. Taylor et al., Pineridge Press,
Swansea, 457-470
Cundall, P.A. [1971] .
"A computer model for simulating progressive large scale movements in blocky rock
systems"
in Proceedings of the Symposium of the International Society of Rock Mechanics, Nancy,
France, 1 (11-8)
Cundall, P.A. [1982] .
"Adaptive density scaling for time-explicit calculations"
in Proceedings of the 4th International Conference on Numerical Methods in
Geomechanics, Edmonton, 23-26

RA
Cundall, P.A. [1989]
"Numerical experiments on localization in frictional materials"
Ingenieur-Archiv, 59, 148-159
Cundall, P.A. & Sirack, O.D.L. [1979]
"A discrete numerical model for granular assemblies"
Geotechnique, 29 (1), 47-65
Darwin, D. & Attiogbe, E.K. (1983)
"Load induced cracks in cement paste"
in Proceedings of the Fourth Engineering Mechanics Division Speciality Conference, ASCE,
New York, 1051-1054
De Borst, R. r1986)
"Non-linear Analysis of Frictional Materials"
PhD Thesis, Delft University of Technology, The Netherlands
De Borst, R. [1989a]
"Analysis of spurious kinematic modes in finite element analysis of strain-softening solids"
in "Cracking and Damage", eds. J. Mazars & Z.P. Bazant, Elsevier Applied Science,
London,335-345
De Borst, R. [1989b]
"Numerical methods for bifurcation analysis in geomechanics"
Ingenieur-Archiv, 59, 160-174
De Borst, R. (1990]
"Simulation of localisation using Cosserat theory"
in Proceedings of the Second International Conference on Computer Aided Analysis and
Design of Concrete Structures, eds. N. Bicanic & H.A. Mang, Pineridge Press, Swansea,
931-944
De Borst, R. [19911
"Continuum modefs for discontinuous media"
in "Fracture Processes in Concrete, Rock and Ceramics", RILEM Proceedings 13, eds.
J.G.M. Van Mier, J.G. Rots & A. Bakker, E. & F.N. Spon, London, 601-618
Di Tomasso, A. [1984]
"Evaluation of concrete fracture, in Fracture mechanics of concrete"
in "Fracture Mechanics of Concrete", eds. A. Carpinteri & A.R. Ingraffea, Martinus
Nijhoff Publishers, The Hague, 31-65
Di Tomasso, A. (1989}
"Size effects and britt eness"
in "Fracture Mechanics of Concrete Structures", RILEM Report 3, ed. L. Elfgren,
Chapman & Hall, London, 191-207
Diederichs, U., Schneider, U. & Tenien, M. [1983)
"Formation and propagation of cracks and acoust1c emission"
in "Fracture Mechanics of Concrete", ed. F.H. Wittmann, Elsevier Science Publishers
B.V., Amsterdam, 157-206
Dorinson, A. & Ludema, K.C. [1985)
"Mechanics and Chemistry in Lubrication"
Tribology Series 9, Elsevier, Amsterdam

R.5
Dougill, J.W. [19711
"Further consideratiOn of a mathematical model for progressive fracture of a heterogeneous
material"
Magazine of Concrete Research, 23 {74), 5-10
Dougill. J.W. [1976)
"On stable progresSively fracturing solids"
Journal of Applied Mathematics and Physics, 27 (4), 423-437
Doda, H. [19911
"Grain mOdel fOr the determination of the stress-crack-width-relation"
in "Analysis of Concrete Structures by Fracture Mechanics", RILE M Proceedings 6, eds. L.
Elfgren & S.P. Shah, Chapman & Hall, London, 88-96
El-Rah.ma.n, M. &; Shrive, N.G. [1984]
"Crack initiation in multiaxial compressive stress fields"
in Proceedings of RILEM-CEB Symposium on Concrete under Multiaxial Conditions,
INSA, Toulouse, 220-229
Elfgren, L. (ed.) [1989]
"Fracture Mechanics of Concrete Structures"
RILEM Report 3, Chapman & Hall, London, 1989
Erdei, C.K. [1980]
"Finite element analysis and tests with a new load-transmitting medium to measure
compressive strength of brittle materials"
Materials and Structures, RILEM, 13, 83-90
Fardis, M.N., Alibe, B.&; Ta.ssoulas, J. [1983]
"Monotonic and cyclic constitutive law for concrete"
Journal of Engineering Mechanics, ASCE, 109 (EM2), 516-536
FIP /CEB [1990]
"High Strength Concrete- State of the Art Report"
FIP, Bulletin d'information 197, Chameleon Press, London,
Fonseka, G.U. &; Krajcinovic, D. [1981]
"The continuous damage theory ol brittle materials -part 2: uniaxial and plane response
modes"
Journal of Applied Mechanics, ASME, 48, 816-824
Gerstle, K.H. [198la]
"Simple formulation of biaxial concrete behavior"
ACI Journal, 78 (5), 62-68
Gerstle, K.H. [1981b]
"Simple formulation of triaxial concrete behavior"
ACI Journal, 78 (34), 382-387
Gerstle, K.H., Linse, D.L., Bertacchi, P., Kotsovos, M.D., Ko, H.Y., Newman, J.B., Rossi,
P., Schickert, G., Taylor, M.A., Traina, L.A., Zimmerma.n, R.M., &; Bellotti, R. [1978]
"Strength of concrete under multiaxial stress states"
in "Concrete and Concrete Structures", Proceedings of the Douglas McHenry International
Symposium, SP 55-5, ACI, 103-131

R.6
Gerstle, K.H., Aschl, H., Bellotti, R., Bet1acchi, P., Kotsovos, M.D., Ko, H.Y., Linse, D.,
Newma.n, J.B., Rossi, P., Schickeri, G., Taylor, M.A., Traina, L.A., Win1der, H. &;
Zimmerman R.M. [1980]
"Behavior of concrete under multiaxial stress states"
Journal of the Engineering Mechanics Division, ASCE, 106 (EM6), 1383-1403
Glavind, M. &; Stang, H. (1991)
"Evaluation of the complete stress-strain curve for high-strength concrete"
in "Fracture Processes in Concrete, Rock and Ceramics", RILEM Proceedings 13, eds.
J.G.M. Van Mier, J.G. Rots & A. Bakker, E. & F.N. Spon, London, 749-760
Gopalaratnam, V.S. &; Shah, S.P. [1985]
"Softening response of plain concrete in direct tension"
ACI Journal, 82, 31o-323
Goudswaard, I.&; Vonk, R.A. [1989]
"A detection method for internal cracks in concrete specimens 11
Report TUE-BK0-89.12, Eindhoven University of Technology, The Netherlands (in
Dutch)
Griflith, A.A. [1920]
"The phenomena of rupture and flow in solids"
Philosophical Transactions, Royal Society of London, A 22, 163-198
Gri:ffith, A.A.J1924]
"The theory o rupture"
in Proceedings of the 1st International Congress on Applied Mechanics, Delft, 55-93
Ban, D.J. &; Chen, W.F. [19861
"Strain-space plasticity formulation for hardening-softening materials with elastoplastic
coupling"
International Journal of Solids and Structures, 22 (8), 935-950
Ha.ssa.nzadeh, M., Hillerborg, A.&; Zhou, F.P. (19871
"Tests of material properties in mixed mode I and If"
in Proceedings of the SEM-RILEM International Conference on Fracture of Concrete and
Rock, Houston, Texas, eds. S.P. Shah & S.E. Swartz, 353-369
Hillerborg, A. [1988]
"Rotational capacity of reinforced concrete beams"
Nordic Concrete Research, 7, Norske Betonforening, Oslo, 121-134
Hillerborg, A. fl989]
"Stability problems in fracture mechanics testing"
in "Fracture of Concrete and Rock: Recent Developments", Eds. S.P. Shah, S.E. Swartz &
B.Barr, Elsevier Applied Science, London, 369-378
Hillerborg, A. (19911
"Size dependency o{ the stress-strain curve in compression"
in "Analysis of Concrete Structures by Fracture Mechanics", RILEM Report 6, eds. L.
Elfgren & S.P. Shah; Chapman & Hall, London, 171-178
Hillerborg, A. &; Rots, J. (19891
"Crack concepts and numericaf modelling"
Chapter 5 in "Fracture Mechanics of Concrete Structures", ed. L. Elfgren, RILEM Report
3, Cha.pman & Hall, London, 128-146

R.7
Hillerborg, A., Modeer, M. &; Petersson, P .E. (1976]
"Analysis of crack formation and crack growth. in concrete by means of fracture mechanics
and finite elements"
Cement and Concrete Research, 6, 773-782
Bilsdorf, H. [1965]
"Die Bestimmung der Zweiachsiger Festigkeit des Betons"
Deutscher Ausschuss fiir Stahlbeton, 173, Berlin
Hirsch, T.J. [1962]
"Modulus of elasticity of concrete affected by elastic moduli of cement paste matrix and
aggregate""
ACI Journal, 59 (3), 427-452

Bobbs, B.E. &; Ord, A. [1989]

"Numerical simulation of shear band formation in a frictional-dilational material"
Ingenieur-Archiv, 59, 209-220
Hordijk, D.A. [1989] ·
"Deformation.....Controlled Uniaxial Tensile Tests on Concrete"
Report 25.5-89-15/VFA, Stevin Laboratory, Delft University of Technology, The
Netherlands
Hordijk, D.A. [1991]
"Local Approach to Fatigue of Concrete"
PhD thesis, Delft University of Technology, The Netherlands
Bordijk, D.A. & Reinha.rdt, H.W. [1989]
"Macro-structural effects in uniaxial tensile test on concrete"
In "Brittle Matrix Composites", eds. A.M. Brandt & I. H. Marshall, Elsevier Applied
Science, Amsterdam, 486-495
Bordijk, D.A. & Reinha.rdt, B.W. [1990]
"Fracture of concrete in uniaxial tensile experiments as influenced by curing conditions"
Engineering Fracture Mechanics, 35 (4/5), 819-826
Hordijk, D.A., Reinhardt, B.W. &; Comelissen, H.A.W. [1987]
"Fracture mechanics parameters of concrete from uniaxial tensile tests as influenced by
specimen length 11 ,
in preprints of Proceedings of the SEM-RlLEM International Conference on Fracture of
Concrete and Rock, Houston, USA, eds. S.P. Shah & S.E. Swartz, 138-149
Hordijk, D.A., Van Mier, J.G.M. & Reinha.rdt, H.W. (1989]
"Material properties 11
Chapter 4 in 11 Fracture Mechanics of Concrete Structures", RILE M Report 3, ed. L.
Elfgren, Cha.pman & Hall, London, 67-127
Horii, H. & Nemat-Nasser, S. [1985]
"Compression-induced microcrack growth in brittle solids: axial splitting and shear
failure"
Journal of Geophysical Research, 90 (B4), 3105--3125

Borii, B. & Nemat-Nasser, S. (1986]

"Brittle failure in compression: splitting, faulting and brittle-ductile transition"
Philosophical Transactions, Royal Society of London, A319, 337-374

R.8
Hsieh, S.S., Ting E.C. &; Chen, W.F'. (1982)
"A plastic-fracture model for concrete"
International Journal of Solids and Structures, 18 {3), 181-197

Hsu, T.T.C. &; Slate, F.O., (1963]

"Tensile bond strength between aggregate and cement paste or mortar"
ACI Journal, 465-486

Bsn, T.T.C., Slate, F.O., Siurman, G.M. &; Wmter, G. (1963)

"Microcracldng of plain concrete and the shape of the stress-l!train curve"
ACI Journal, 60 {14), 209-224

Hn, X.Z., Coiierell, B.&; Mai, Y.W. (1986]

"Computer simulation models of fracture in concrete"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F .H. Wittmann, Elsevier
Science Publishers B.V., Amsterdam, 91-100
Hudson, J.A., Brown, E.T. &; Fairhursi, C. [1972] .
"Shape of the complete stress-lltrain curve for rock"
in "Stability of Rock Slopes", Proceedings of the 13th Symposium on Rock Mechanics,
1971, University of lllinois, Urbana, ed. E.J. Cording, ASCE, 773-795
Hughes, B.P. &; Ash, J.E. (1970]
"Anisotropy and failure criteria for concrete"
Materials and Structures, RILEM, 3 {18), 371-374

Ingraffea, A.R. [1989]

"Fracture mechanics studies of cracking in dams"
in "Fracture Mechanics of Concrete Structures", ed. L. Elfgren, RILEM Report 3,
Chapman & Hall, London, 319-335
Ingraffea, A.R. &; Panthaki, M.J. (1985]
"Analysis of "shear fracture" tests of concrete beams"
in Proceedings of Seminar on Finite Element Analysis of Reinforced Concrete Structures,
Tokyo, 1, 71-91

lngraffea, A.R. &; Saonma, V. (1985]

"Numerical modelling of discrete crack propagation in reinforced and plain concrete"
in "Fracture Mechanics of Concrete", eds. G. Sih & A. Di Tomasso, Martinus Nijhoff
Publishers, Dordrecht, The Netherlands, 171-225

ltasca Consulting Group, Inc. [1989]

"UDEC (Universal Distinct Element Code), Volume 1, Software Summary"
Topical Report 006~1-T12, Minneapolis, Minnesota, USA

Iznmi, M., Mihashi, H.&; Nomura, N. [1986]

"Acoustic emission technique to evaluate fracture mechanics parameters of concrete"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F.H. Wittmann, Elsevier
Science Publishers B.V., Amsterdam, 259-268

Jacquot, P. &; Rastogi, P .K. [1983)

"Speckle metrology and holographiC interferometry applied to the study of cracks in
concrete"
in "Fracture Mechanics of Concrete", ed. F.H. Wittmann, Elsevier Science Publishers
B.V., Amsterdam, 113-155

R.9
Jamet, P. Millard, A.&; Nahas G. [1984)
"Triaxal behaviour of a microconcrete- complete stress-strain curves for confining
pressures ranging from 0 to 100 MPa"
in Proceedings of RILEM-CEB Symposium on Concrete under Multiaxial Conditions,
INSA Toulouse, session 4, experimental results, 1, 133-140
Knab, L.I., Jennings, H., Walker, H.N. &; Clifton, J.R. (1986)
"Techniques to observe the fracture zone in mortar and concrete"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F.H. Wittmann, Elsevier
Science Publishers B. V., Amsterdam, 241-247
Kobayashi, S. &: Koyanagi, W. (1972]
"Fracture criteria of cement paste, mortar and concrete subjected to multiaxial
compressive stresses"
in RILEM Symposium on the Deformation and the Rupture of Solids Subjected to
Multiaxial Stresses, paper 1/9, 131-148

Kotsovos, M.D.J1980l
11
A mathematic model of the deformational behavior of concrete under generalised stress,
based on fundamental material properties"
Materials and Structures, RILEM, 13, 289-298
Kotsovos, M.D. (1983]
"Effect of testing techniques on the post-ultimate behaviour of concrete in compression"
Materials and Structures, RILEM, 16 (91), 3-12
Kotsovos, M.D. [19841
"Concrete- a brittle t'Iaeturing material"
Materials and Structures, RILEM, 17, 107-114
Krajcinovic, D.&: Fonseka, G.U. r1981]
"The continuous damage theory ol brittle materials. Part 1: General theory"
Journal of Applied Mechanics, ASME, 48, 809-815
Krishnaswamy, K.T. [1968]
"Strength and microcra.cking of plain concrete under triaxial compression"
ACI Journal, 65 (64), 856-862

Kupfer, H.J1973l
"Das Verh ten des Betons unter mehrachsiger Kurzzeitbelastung unter besonderer
Beriicksichtigung der zweiachsiger Beanspruchung"
Deutscher Ausschuss ffir Stahlbeton, 229, Berlin
Kupfer, H.B. &: Gerstle, K.H. [1973)
"Behavior of concrete under biaxial stresses"
Journal of the Engineering Mechanics Division, ASCE, 99 (EM4), 852-866
Lasry, D. &: Belytsehko, T. [1988]
"Localization limiters in transient problems"
International Journal of Solids and Structures, 24, 581-597
Launay, P. &: Gaehon, B. [19711
"Strain and ultimate strength ol concrete under triaxial stress 11
in Proceedings of the First International Conference on Structural Mechanics in Reactor
Technology, Berlin, paper H1/3, 23-34

R.10
Lierse, J. &; Ringkamp, M.J1983l
"Investigations into cracke reinforced concrete structural elements with the aid of
photoelastic methods"
in "Fracture Mechanics of Concrete", ed. F.H. Wittmann, Elsevier Science Publishers
B.V., Amsterdam, 95-111
Lin, F.B., Bazant, Z.P., Chem, J.C. &; Marchertas, A.H. (1987)
"Concrete model with normality and sequential identification"
Computers & Structures, 26 (6), 1011-1025
Linse, D. r1978]
"Losung "'ersuchstechnischer Fragen bei der Ermittlung des Festigkeits und
Verformungsverhaltens von Beton unter dreiachsiger Belastung"
Deutscher Ausschuss fiir Stahlbeton, 292, Berlin
Liu, T.C.Y., Nilson, A.H. &; Slate, F.O. [1972)
"Stress-strain response and fracture of concrete in uniaxial and biaxial compression"
ACI Journal, 69 (31), 291-295
Loland, K.E. rt980]
"Continuous damage model for load-response estimation of concrete"
Cement and Concrete Research, 10, 395-402
Lorig, L.J. &; Cundall, P.A. [1989]
"Modeling of reinforced concrete using the distinct element method"
in "Fracture of Concrete and Rock", Proceedings of SEM-RILEM International
Conference, Houston, Texas, USA, 1987, eds. S.P. Shah & S.E. Swartz, Springer-Verlag,
New York, 276-287

Luong, M.P. [1986)

"Infrared vibrothermography of plain concrete"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F.H. Wittmann, Elsevier
Science Publishers B.V., Amsterdam, 249-257
Maji, A.K. &; Shah, S.P. [1988)
"Application of acoustic emission and laser holography to study microfracture in concrete"
in ACI special publication ACI-sP-112 "Nondestructive Testing", 83-109
Mazars, J. [1984]
"Description of multiaxial behavior of concrete with an elastic damaging model"
in Proceedings of RILEM-CEB Symposium on Concrete under Multiaxial Conditions,
INSA, session 6, Vol. 1, Toulouse, 190-200
Mazars, J. [1986]
"A model of a unilateral elastic damageable material and its application to concrete"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F.H. Wittmann, Elsevier
Science Publishers B.V., Amsterdam, 61-71
Mazars, J. &; Bazant, Z.P. (eds.) [1989]
"Cracking and Damage"
Proceedings of the France- U.S. Workshop on Strain Localization and Size Effect due to
Cracking and Damage, Cachan, France, 1988, Elsevier Applied Science, London
Mihashi, H. [1983)
"A stochastic theory for fracture of concrete"
in "Fracture Mechanics of Concrete", ed. F .H. Wittmann, Elsevier Science Publishers
B.V., Amsterdam, 301-339

R.ll
Mihashl H. k Izumi, M. [1977]
11
A Stochastic Theory for concrete fracture"
Cement and Concrete Research, 7, 411-422
Mills, L.L. k Zimmermann, R.M. [1970]
"Compressive strength of plain concrete under multiaxalloading conditions"
ACI Journal, 67, 802-807

s.
Mindess, [1991)
11
Fracture process zone detection"
in "Fracture Mechanics Test Methods for Concrete", RILEM Report 5, eds. S.P. Shah &
A. Carpinteri, 231-261
Mindess, S. k Shah, S.P. (eds.) (1988)
"Bonding in Cementitious Composites 11
Materials Research Society Symposium Proceedings, 114, MRS, Pittsburgh, Pennsylvania

Montgomery, D.C. (1983]

"Design and AnalySis of Experiments"
John Wiley & Sons, New York

Miihlhaus, B.BJ1986J
"Shear band an ysis m granular material by Cosserat theory"
in 11 Numerical Models in Geomechanics", eds. G.N. Pande & W.F. Van lmpe, Jackson and
Sons, R.edruth, 115-122
Nelissen, L.J.M. (1972]
"Biaxial testing of normal concrete"
Heron, 18 {1)
Newman, K. k Sigvaldason, O.T. (1965-1966)
"Testing machine and specimen characteristics and their effect on the mode of deformation,
failure and strength of materials 11
in Proceedings of the Institution of Mechanical Engineers, 180 (3A), 399-410
Newman, J.B. [1979]
11
Concrete under complex stress"
in 11 Developments in Concrete Technology -1 11 , ed. F.D. Lydon, Applied Science
Publishers LTD, London, 151-219

Nojiri, Y., Kotani, K. k Abe, Y. [19841

"Failure envelope of concrete subjected to multiaxial compressive stresses"
in Proceedings of RILEM-CEB Symposium on Concrete under Multiaxial Conditions,
INSA Toulouse, 141-148

Ortiz, M. [1985)
"A constitutive theory for the inelastic behavior of concrete"
Mechanics of Materials, 4, 67-93

Ottosen, N.S. [19791

"Constitutive mode for short-time loading of concrete 11
Journal of the Engineering Mechanics Division, ASCE, 105 (EM1), 127-141
Ozbolt, J. & Bazant, Z.P. r1991)
"Clclic microplane model for concrete" ·
in 'Fracture Processes in Concrete, Rock and Ceramics", RILEM Proceedings 13, eds.
J.G.M. Van Mier, J.G. Rots & A. Bakker, E. & F.N. Spon, London, 639-6M

R.12
OzboU, J. &; Eligehausm R. [1991)
"Analysis of reinforced concrete beams without shear reinforcement using non-local
microplane model"
in "Fracture Processes in Concrete, Rock and Ceramics", RILEM Proceedings 13, eds.
J.G.M. Van Mier, J.G. Rots & A. Bakker, E. & F.N. Spon, London, 919-930
Paterson, M.S. f1978]
"Experimental ftock Deformation- The Brittle Field"
Springer-Verlag, Berlin

Pijaudier-Cabot &; G. Bazant, Z.P. [1987)

"Nonlocal damage theory"
Journal of Engineering Mechanics, 113, 1512-1533
Pramono, E. &; Willam, K. [1989]
"Fracture energy-based plasticity formulation of plain concrete"
Journal of Engineering Mechanics, 115 (6), 1183-1204
Read, H.E. &; Begemier, G.A. [1984)
"Strain softening of rock, soil and concrete - a review article"
Mechanics of Materials, 3, 271-294
Rehm., G., Diem, P. &; Zimbelm.ann, R. [1977]
"Technische Moglichkeiten zur ErhOhung der Zugfestigkeit von Beton"
Deutscher Ausschuss fur Stahlbeton, 283, Berlin

Reinha.rdt, H.W. [19841

"Fracture mechanics oi an elastic softming material like concrete"
Heron, 29 ( 2)
Reinha.rdt, H.W. (1989)
"Various techniques for the assessment of the damage zone between two saw cuts"
in "Cracking and Damage", eds. J. Mazars & Z.P. Bazant, Elsevier Applied Science,
London, 3-14

RoeHstra, P .E. (1989}

"Simulation of strain localization processes with numerical concrete"
in "Cracking and Damage", eds. J. Maza.rs & Z.P. Bazant, Elsevier Applied Science,
London, 79-90
RoeHstra, P .E. &; Sadouki, B. (1986]
"Fracture process in numerical concrete"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F.H. Wittmann, Elsevier
Science Publishers B.V., Amsterdam, 105-116
RoeHstra, P.E., Sadouki H.&; Wittmann, F.H. [1985}
"Le beton numerique" ·
Materials and Structures, RILEM, 18 (ton 327-335
Rokugo, K. & Koyanagi, W. (1992]
"Role of compressive fracture energy of concrete on the failure behavior of reinforced
concrete beams"
in "Applications of Fracture Mechanics to Reinforced Concrete", ed. A. Carpinteri,
Elsevier Applied Science, London, 437-464

R.13
Robgo, K., Ohno, S. et Koyanagi1 W. (1986&]
"Automatical measuring system ot load:...rusplacement curves including post-failure region
of concrete specimens"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F.H. Wittmann, Elsevier
Science Publishers B.V., Amsterdam, 403-412
Rotugo, K., lwase, H. & Koyanagi, W. (1986b]
"Application of fracture energy of concrete in compression to section design of reinforced
concrete members" ·
in "Fracture Toughness and Fracture Energy of Concrete", ed. F .H. Wittmann, Elsevier
Science Publishers B.V., Amsterdam, 587-596
Rossi, P. k rucher, s. [1987]
"Stochastic modelling of concrete cracking"
in "Constitutive Laws for Engineering Materials -Theory and Applications", eds. Desai et
al., Elsevier Applied Science, London, 915-922
Rossi, P. & Wu, X. (19911
"Use of a stochastic model for material behaviour analysis and appraisal of structures"
in "Fracture Processes in Concrete, Rock and Ceramics", RILEM Proceedings 13, eds.
J.G.M. Van Mier, J.G. Rots and A. Bakker, E. & F.N. Spon, London, 931-940
Rots, J.G. (1988]
11
Computat1onal Modeling of Concrete Fracture"
PhD thesis, Delft University of Technology, The Netherlands
Rots, J.GJ1991]
"Numeric simulation of cracking in structural masonry 11
in "Computational Mechanics: Recent Developments in DIANA" eds. R. De Borst & D.G.
Roddeman, Heron 36 (2)
Rots, J.G. &; De Borst, R. [1989]
"Analysis of concrete fracture in "direct" tension"
International Journal of Solids and Structures, 25 (12), 1381-1394
Rots, J.G., Hordijk, D.A. &; De Borst, R. [1987]
11
Numerical simulation of concrete fracture in a 11 direct" tension"
in 11 Numerical Methods in Fracture Mechanics", eds. A.R. Luxmoore et al., Pineridge
Press, Swansea, 457-471
Riisch, H. (19551
"Versuche zur F'estigkeit der Biegedruckzone11
Deutscher Ausschuss fiir Stahlbeton, 120, Berlin
Riisch, H. &; Hilsdorf, H. (1963] ·
Verformungseigenschaften von Beton unter zentrischen Zugspannungen"
11

Bericht Nr. 44, Materialpriifungsamt fiir das Bauwesen der Technischen Hochschule
Miinchen
Schickert, G. [19801
"Schwellenwerte be1m Betondruckversuch"
Deutscher Ausschuss fiir. Stahlbeton, 312, Berlin
Schick:ert, G. &; WinJder H. [1977]
"Versuchsergebnisse zur Festigkeit und Verformung von Beton bei mehraxialer
Druckbeanspruchung 11
Deutscher Ausschuss filr Stahlbeton, 277, Berlin

R.14
Schlangen, E. & Van Mier, J.G.M. (199la]
"Boundary effects in mixed mode I and II fracture of concrete"
in "Fracture Processes in Concrete, Rock and Ceramics", RILEM Proceedings 13, eds.
J.G.M. Van Mier, J.G. Rots &: A. Bakker, E. &: F.N. Spon, London, 705-716
Schlangen, E. & Van Mier, J.G.M. [1991b]
"Experimental and Nnmerical Analysis of Micromechanisms of Fracture of Cementbased
Composites"
Report 25.5-91-1/VFC, Stevin Laboratory, Delft University of Technology, The
Netherlands
Schorn, H. (19861
''Numerical simulation of composite material as concrete"
in "Fracture Toughness and Fracture Energy of Concrete", ed. F.H. Wittmann, Elsevier
Science Publishers B. V., Amsterdam, 177-188
Schorn, H. & Rode, U. [1989]
"3-D modeling of process zone in concrete by numerical simulation"
in "Fracture of Concrete and Rock", eds. S.P. Shah & S.E. Swartz, Springer-Verlag, New
York, 220-228
Schreyer, H.L. rt989]
"Formulations tor nonlocal softening in a finite zone with anisotropic damage"
in "Cracking and Damage", eds. J. Mazars & Z.P. Bazant, Elsevier AppliooScience,
London,426-439
Schreyer, H.L. & Chen, Z. [19861
"One-dimensional softening with localization"
Journal of Applied Mechanics, ASME, 53, 971-979
Scrivener, K.L. & Pratt, P .L. [1987]
"The characterization and quantification of cement and concrete microstructures"
in Proceedings of 1st RILEM International Conference "From Materials Science to
Construction Materials Engineering", ed. J.C. Maso, Versailles, 61-68
Shah, S.P. (19891
"Experimental observation of cracking and damage"
in "Cracking and Damage'', eds. J. Ma.zars & Z.P. Ba.zant, Elsevier Applied SCience,
London, 15-29
Shah, S.P. & Sa.nbr, R. [1987]
"Internal cracking and strain softening response of concrete under uniaxial compression"
ACI Materials Journal, 84 (3), 20Q-212
Shah, S.P. & Swartz, S.E. (19891
"Fracture of Concrete and Rock 1
Springer-Verlag, New York
Shah, S.P., Swariz, S.E. & Barr, B. (eds.) (1989]
"Fracture of Concrete and Rock: Recent Developments"
Elsevier Applied Science, London
Slate, F.O. (1983]
"Microscopic observation of cracks in concrete, with emphasis on techniques developed and
used at Cornell University"
in "Fracture Mechanics of Concrete", ed. F .H. Wittma.nn, Elsevier Science Publishers
B.V., Amsterdam, 75--93

R.15
 Slate, F.O. k Hover, K.C.l1984]
"Microcracking in concrete
· in "Fracture Mechanics of Concrete", eds. A. Carpinteri &; A.R. Ingraffea, Martin us
Nijhoff Publishers, The Hague, 137-159
Slate, F.O. k Olefski, S. [1963)
"X-rays for study of internal structure and microcracking of concrete"
ACI Journal, 60 l31), 575-588
Soers, E .. k Meyskens, M. [1987]
"Concrete petrography - microscopic investigation of concrete"
Cement, 11, 16-22, The Netherlands (in Dutch)

Smith, S.S., Willam, K.J., Gerstle, K.B. k Sture, S. rt989]

"Concrete over the top, or: is there life after the peak'?"
ACI Materials Journal, 86 (M45), 491-497
Stankowski, T. [1990]
"Numerical Simulation of Progressive Failure in Particle Composites"
PhD Thesis, University of Colorado, Boulder, USA

Stankowski, T. k Gerstle, K.B. f1985]

"Simple formulation of concrete behavior under multiaxial load"
ACI Journal, 82 (20), 213-221
Steinmann, P. k Willam, K. [19911
"Finite elements for capturing localized failure"
Applied Mechanics, Springer-Verlag, 61, 259-275
Stroeven, P. [1973]
"Some Aspects of the Micromechanics of Concrete"
PhD thesis, Delft University of Technology, The Netherlands

Sture, S. k Desai, C.S. [1979]

"Fluid cushion truly tria:x:ial and multiaxial testing device"
Geotechnica.l Testing Journal, GTJODJ, 2 (1), 20-33

Taerwe, L. (1991)
"Brittleness versus ductility of high strength concrete"
Structural Engineering International, IABSE, 4, 40-45
Torrenti, J.M., Desrues, J., Acker, P. &; Boulay, C. U989]
"Application of stereophotogrammetry to the strain localization in concrete compression"
in "Cracking and Damage", eds. J. Mazars & Z.P. Bazant, Elsevier Applied Science,
London,30-41

Triantafyllidis, N. &; Aifantis, E.C. [1986]

"A gradient approach to localization of deformation: I. hyperelastic materials"
Journal of Elasticity, 16, 225-237
Van Mier, J.G.M. [1984]
"Strain-softening of Co~crete under Multiaxial Loading Conditions"
PhD thesis, Eindhoven University of Technology, The Netherlands
Van Mier, J.G.M. [1986a]
"Multiaxial strain-softening of concrete, part I: fracture, part II: load-histories"
Materials and Structures, RILE M, 19, ( 111 ), 179-200

R.16
Van Mier, J.G.M. [1986b]
"Fracture of Concrete under Complex Stress"
Heron, 31 (3)
Van Mier, J.G.M. [1991a)
"Mode I behaviour of concrete: influence of the rotational stiffness outside the crack-zone"
in "Analysis of Concrete Structures by Fracture Mechanics", eds. L. Elfgren & S.P. Shah,
RILEM Proceedings 6, Chapman & Ha.ll, London, 19-31
Van Mier, J.G.M. r1991b]
"Mode I fracture ol concrete: discontinuous crack growth and crack interface grain
bridging"
Cement & Concrete Research, 21 (1), 1--'15
Van Mier, J.G.M. &; Nooru-Mohamed, M.B. (1990)
"Geometrical and structural aspects of concrete fracture"
Engineering Fracture Mechanics, 35 (4/5), 617-628
Van Mier, J.G.M., Rots, J.G. &; Bakker, A. [1991a]
"Fracture Processes in Concrete, Rock an Ceramics"
RILEM Proceedings 13, E.~ F.N. Spon, London
Van Mier, J.G.M., Nooru-Mohamed, M.B. &; Timmers, G. [1991bl
"An Experimental Study of Shear Fracture and Aggregate Interloc'k in Cementbased
Composites"
Heron, 36 ( 4)
Vermeer, P.A.&; De Borst, R. U984]
"Non-associated Plasticity for ~oils, Concrete and Rock 11
Heron, 29 (3)

Vonk, R.A. r1989] ~

"Influence of boundary conditions on softening of concrete loaded in compression"
Report TUE-BK0-89.14, Eindhoven University of Technology, The Netherlands
Vonk, R.A. [1990)
"Constitutive Model describing Interface Behaviour with Tensile and Shear Softening"
Report TUE-BK0-90.12, Eindhoven University of Technology, The Netherlands
Vonk, R.A., Rutten, B.S., Van Mier, J.G.M. &; Fijneman, H.J. [1989)
"Influence of boundary conditions on softening of concrete loaded in compression"
in "Fracture of Concrete and Rock: Recent Developments", eds. S.P. Shah, S.E. Swartz &
B. Barr, Elsevier Applied Science, London, 711-720
Vonk, R.A., RuUen, B.S.1 Van Mier, J.G.M. &; Fijneman, B.J. [1990)
"Size effect in softening ot concrete loaded in compression"
in "Fracture Behaviour and Design of Materials and Structures", ed. D. Firra.o, EMAS,
United Kingdom, 767-772
Vonk, R.A., Rutten, B.S., Van Mier, J.G.M. &; Fijneman, H.J. (1991)
"Micromechanica.l simulation of concrete softening"
in "Fracture Processes in Concrete, Rock and Ceramics", eds. J.G.M. Van Mier, J.G. Rots
& A. Bakker, RILEM Proceedings 13, E. & F.N. Spon, London, 129-138
Walraven, J.C. [1980)
"Aggregate Interlock; a Theoretical and Experimental Analysis"
PhD thesis, Delft University of Technology, The Netherlands

R.17
w~. n. k KJauea, n. L1979)
"Die Sehalemissiouanalyae .
Beionwerk + Feriigteil-Technik, 45 (12), 7~116

Wi11am, K., Hnrlbu~, B• .tr; S~ure, S. [19861

"Experimental and. couti1;u~ive aspects ofeoncr~ tailure"
in Proceedings of Seminar on Finiie Element Analysis of Reinforced Concr~e S~ructures,
Tokyo, ASCE Special Publication, New York, 226-254 .
Willam, K., Pram.ooo, E. .tr; Sture, S. (1989)
"Fundamental issues of smeared crack models"
in "Fracture of Concr~e and Rock", eds. S.P. Shah & S.E. Swartz, Springer-Verlag, New
York, 142-157
Wdlam, K., Stankowski, T., RlJJleiiiOn, K• .tr; Sture, S. [1989)
"Simulation issues of dinributed and localized failure computatiou"
in "Cracking and Damage", eds. J. Mazars and Z.P. Bazant, Elsevier Applied Science,
London,363-378
WlSChen, G. (1918)
11
Aufnahme und Auswirkungen von Druckbeanspruchungen auf Beton"
B~ontechnische Berichte, 19, 31-56

Wittm.ann, F.H. (ed.) f1983)

"Fracture Mechanics ol Concr~e"
Developments in Civil Engineering 7, Elsevier Science Publishers B.V., Amsterdam
WiUm.ann, F.H. (ed.) (1986]
"Fracture Toupness and Frac~ure Energy of Concrete"
Developments m Civil Engineering 18, Elsevier Science Publishers B.V., Amsterdam
Wittm.ann, F.H. (1987]
"Structure of concrete and crack formation"
in "Fracture of Non-M~allic Materials", eds. K.p. Herrmann & L.H. Larssdn, Reidel,
Dordrecht, 309-340
Zaitsev, Y. (1983]
"Crack propagation in a composite material"
in "Fracture Mechanics of Concr~e", ed. F.H. Wittmann, Elsevier Science Publishers
B.V., Amsterdam, 251-299
Zaitsev, Y.B • .tr; Wittmann, F.H. (19811 .
"Simulation of crack propagation and failure of concr~e"
Materials and Structures, RILEM, 14 (83), 357-365
Zhou, F.P. (19881
"Some Aspects of Tensile Fracture Behaviour and Structural Response of Cementitious
Materials"
Report TVBM-1008, Division of Building Materials, Lund Institute of Technology
Zubelewicz, A. .tr; Bazant, Z.P. f1987]
"Interface element modeling of fracture in aggregate composites"
Journal of Engineering Mechanics, ASCE, 113, 1619-1630

R.18
SAMENVATTING

SOFTENING VAN BETON ONDER EEN DRUKBELASTING

Dit proefschrift behandelt een gecombineerd experimenteel-numeriek onderzoek naa.r

softening van beton onder een drukbelasting. Softening is de afname van de mechanische
weerstand van een proefstuk of een constructie ten gevolge van een continu toenemende
opgelegde vervorming. Het is het gevolg van scheurvorming. In dit onderzoek is softening
van beton geanalyseerd en gemodelleerd als het gedrag van een constructie, aangezien het
de respons is van een samenstel van componenten. Gedurende het bezwijkproces wordt
beton door middel van scheurvorming opgesplitst in een aantal delen, die samen met de
scheuren het gedrag bepaleJl. Het beton gaat over van een continuum in een
discontinuum, wat impliceert dat het gedrag niet langer is te beschrijven met een
continuum model. Dit onderzoek heeft zich daarom geconcentreerd op een beschrijving
van het bezwijkgedrag van beton onder een drukbelasting met behulp van de breuk-
mechanica. Bijzondere aa.ndacht is besteed aan de wijze waa.rop het scheurvormingsproces
wordt bepaald door de heterogeniteit, de afmetingen en de randvoorwaa.rden van het
beton. Met name zijn eenassige drukproeven uitgevoerd en geanalyseerd. Da.a.rbij is
bijzondere aandacht geschonken aan het meerdimensionale aspekt van het bezwijkgedrag,
dat voor een drukbelasting een essentiele rol speelt.

In hoofdstvk 2 wordt de achtergrond geschetst van het bier gepresenteerde onderzoek. Een
beknopt overzicht wordt gegeven van het scheurvormingsproces in beton onder een
drukbelasting. Bijzondere aandacht wordt besteed aan de lokalisatie van scheurvorming.
De invloed van diverse factoren, zoals de samenstelling van het beton en de spannings-
toestand, wordt besproken. Er wordt nader ingegaan op de invloed van de rand-
voorwa.a.rden in proeven, die vaak een onverwacht domina.nte rol spelen. Dit hee£t er toe
bijgedragen dat een aantal aspekten van het bezwijkgedrag van beton onder een
drukbelasting onduidelijk zijn.

In hoofdstuk 9 warden de toegepaste experimentele technieken beschreven. Speciale

aandacht gaat uit na.a.r de belemmering van de laterale vervormingen van het beton door
de belastingplaten. De in dit onderzoek ontwikkelde scheurdetectiemethode wordt
beschreven. De proe£stukken vol scheuren warden na een proef onder vacuiim
geimpregneerd met een fluorescerende epoxyha.rs. Dit maakt het mogelijk om na

S.l
verharding van de epoxyhars de proefstukken door te zagen en de inwendige
scheurpatronen met behulp van ultraviolet fotografie vast te leggen.

In hoofdst'Uk 4 wordt het modelleren van betongedrag besproken, waarbij speciale

aandacht wordt geschonken aan softening. De problematiek van de beschrijving van
softening met behulp van continuum modellen en de eindige elementen methode wordt
toegelicht. Daarnaast wordt de benadering via discrete scheurmodellen en de modellering
van de heterogeniteit van beton door middel van micromechanische modellen beschreven.

In hoofdst'Uk 5 wordt een nieuw micromechanisch model gemtroduceerd voor de analyse en

modellering van scheurvorming in beton. Het model beschrijft beton als een composiet
bestaande uit grote toeslagkorrels in een mortel matrix. Het mechanisch gedrag wordt
berekend met behulp van het computer programma UDEC. Karakteristiek voor het model
is het continue net van interface elementen tussen de continuum elementen, waarvan de
geometrie bepaald wordt door de composietstructuur van het beton en met name door de
grote toeslagkorrels. Deze bepalen in hoge mate de scheurvorming in beton.
Scheurvorming in het model is alleen toegestaan in de interface elementen. De
composietstructuur bepaalt welke interface elementen geactiveerd worden tijdens het
bezwijkproces. Een nieuw constitutief model voor de interface elementen is OI!.twikkeld en
geimplementeerd in UDEC om het gedrag van scheuren in beton adequaat te kunnen
beschrijven.

In hoofdstuk 6 wordt het bezwijkgedrag van beton geanalyseerd aan de hand van
proefresultaten en simulaties met het micromechanisch model. Het model blijkt de
proefresultaten goed te kunnen beschrijven. Het bezwijken van beton is het gevolg van een
geleidelijke scheurgroei, die op een gegeven moment instabiel wordt en lokaliseert. Dit
resulteert in softening. Bij het bezwijken onder een drukbelasting wordt het beton
opgesplitst in een aantal stukken die langs elkaar afschuiven. Voor de vorming van het
bezwijkmechanisme en de lokalisatie van vervormingen zijn de afschuifscheuren
maatgevend en niet de splijtscheuren. De splijtscheuren worden vaak als oorza.ak voor het
bezwijken genoemd aangezien zij dominant zichtbaar zijn. Dit onderzoek laat zien dat ze
het bezwijken slechts modificeren. De proefresultaten laten zien da.t, wanneer de
afschuifscheuren in aanmerking worden genomen, lokalisa.tie van scheurvorming aanwezig
is voor het bezwijken onder een drukbelasting, maar dat deze lokalisatie niet zo sterk is
als bij het bezwijken onder een trekbela.sting. Verder wordt in dit hoofdstuk getoond da.t
het micromecha.nisch model het bezwiJ'ken van beton onder een trekbelasting ~oed
beschrijft.

8.2
In hoofdstuk 7wordt een gevoeligheidsanalyse nitgevoerd voor het micromechanisch
model. Deze toont aan dat de conclusies van hoofdstuk 6, die mede gebaseerd zijn op de
resultaten van de simulaties met het model, kwalitatief niet sterk bemvloed worden door
een va.riatie van parameters in het model. Het model onderschat in het algemeen de
spreiding in het scheurpatroon bij het bezwijken onder een drukbelasting. Dit kan
mogelijk ondervangen worden door de geometrische heterogeniteit in het model te
vergroten. Hierbij moet een uitbreiding van het model van twee naar drie dimensies zeker
in overweging worden genomen.

In hoofdstuk 8 wordt aangetoond dat softening een niet-uniforme vervorming van een
proefstuk stimuleert. Bij een proef met een scharnierende belastingplaat resulteert dit in
een extreme rotatie van de belastingplaat. De proef blijkt gevoelig te zijn voor een groot
aantal moeilijk in te schatten factoren en is niet eenvoudig te analyseren. Dit maakt een
dergelijke proef ongeschikt voor systematisch onderzoek naar softening. Ook wanneer de
belastingplaat momentvast bevestigd is aan het belastingframe blijken significante
rota.ties op te kunnen treden. Deze kunnen een grote invloed hebben op het softening
proces. Een criterium is afgeleid waarmee de gevoeligheid van een proefopzet voor rotaties
van de belastingplaat kan worden afgeleid. Hiermee kan de benodigde rotatiestijfheid van
het bela.stingframe worden ingescha.t. Een eenvoudig numeriek model is ontwikkeld om
het fenomeen te beschrijven. Een goede overeenkomst tussen de voorspellingen van het
model en proefresultaten wordt gevonden.

In hoofdstuk 9 wordt de invloed van de belernmering van de laterale vervorrningen van

beton door diverse belastingplaten onderzocht. Softening van beton onder een
drukbelasting blijkt hiervoor uiterst gevoelig te zijn aangezien de laterale uitzetting van
het beton essentieel is voor het bezwijken. Afhankelijk van de gebruikte belastingplaat
varieert het bezwijken van betonkubussen met een ribbe van 100 mm van taai tot vrij
bros voor grindbeton van een normale sterkte. De invloed van de belemmering van de
laterale vervormingen mag nooit verwa.arloosd worden. Bezwijken van beton onder een
drukbelasting is het gevolg van een meerdimensionaal proces en moet als zoda.nig
gemodelleerd worden. Het micromecha.nisch model blijkt het gedra.g onder invloed van de
varierende randvoorwa.arden goed te kunnen beschrijven.

In hoofdst:uk 10 wordt de invloed van de afrnetingen van het beton op softening onder een
eenassige drukbelasting onderzocht. Door de lokalisa.tie van scheurvorrning neemt de
benodigde breukenergie niet evenredig toe met de afrnetingen van het proefstuk, wat
betekent dat het bezwijkgedrag brosser wordt bij een toename van de afrnetingen. Het
verband tussen afrnetingen en de benodigde breukenergie blijkt voor bezwijken onder een

S.3
drukbelasting complexer te zijn dan voor bezwijken onder een trekbelasting. Dit wordt
veroorza.akt door het feit da.t de lokalisatie van scheurvorming bij een drukbelasting niet
zo sterk is als bij een trekbelasting. Het scheurpa.troon vertoont een enigszins diffuus
karakter wa.a.rdoor ook het volume van het proefstuk invloed heeft op de benodigde
breukenergie. Hierbij spelen ook de splijtscheuren een rol. Het micromecha.nisch model is
goed in staa.t om de loka.lisa.tie van scheurvorming weer te geven, ma.a.r onderschat in de
huidige vorm het enigszins diffuse ka.ra.kter bij het bezwijken onder druk. Dit kan
mogelijk ondervangen worden door een vergroting van de geometrische heterogeniteit van
het model.

De standa.a.rd drukproef, zoals deze in Nederland is voorgeschreven, wekt de indruk da.t

het bezwijken van beton onder een eena.sssige drukbelasting een ta.a.i proces is. Dit is niet
juist. Men moet zich rea.liseren dat het ka.rakter van het bezwijken van beton onder een
drukbela.sting a.fhangt van de microstructuur, de a.fmetingen en de la.terale rand-
voorwa.a.rden van het beton. Het ka.ra.kter van bezwijken kan da.a.rom slechts bepa.ald
worden voor een totale constructie, wa.a.rvan de sa.menstelling, a.fmetingen en
randvoorwa.a.rden bekend zijn.

S.4
 STELLINGEN
behorende bij het proefschrift van R.A. Vonk
"SOFTENING OF CONCRETE LOADED IN COMPRESSION''

1. Softening is geen ma.teria.aleigenscha.p, ma.a.r een eigenscha.p van een constructie. Dit
impliceert da.t de vra.a.g "Hoe bros is het ma.teria.al beton?" niet beantwoord kan worden.
De vra.a.g omtrent brosheid kan a.lleen bea.ntwoord worden voor een constructie, waarvan
samenstelling, afmetingen en randvoorwaarden bekend zijn.
- Bieniawski, Z.I. [1967], "Mechanism of brittle failure of rock", lnt. J. Rock Mech. Min.
Sci., vol. 6, pp. 395--430
-Van Mier, J.G.M. [1984], "Strain-softening of concrete loaded in compression",
proefschrift, Technische Universiteit Eindhoven
- dit proefschrift

2. Het bezwijken van een beioncontructie onder een drukbela.sting is in essentie een
driedimensionaa.l proces, dat gedomineerd wordt door de mogelijkheid tot la.terale
uitzetting.

3. De afschuifscheuren, d.w.z. de scheuren onder een hook met de drukbela.sting, zijn

bepa.lend voor het bezwijken van een betonconstructie onder een drukbelasting. De
splijtscheuren, d.w.z. de scheuren parallel aan de drukbelasting, modificeren dit
bezwijkgedrag.

4. Bij het bezwijken van beton onder een drukbelasting treedt loka.lisatie van
scheurvorming op. Dit heeft a.ls consequentie dat het gedrag van een constructie brosser
wordt naarma.te de afmetingen toenemen. Dit scha.a.leffect is aa.nmerkelijk complexer voor
bezwijken onder een drukbelasting da.n voor bezwijken onder een trekbelasting.

5. Het schaa.leffect bij bezwijken van betonconstructies onder een drukbelasting maakt
verder onderzoek en aanpassing van .de normen noodzakelijk.

6. Een drukproef, waarbij de bela.sting wordt aangebracht met behulp van een
scharnierende plaat, is niet geschikt voor systematisch onderzoek naar softening.

7. Het micromechanisch model geba.seerd op de heterogene samenstelling van beton en

discrete scheurvorming, zoa.ls ontwikkeld in dit onderzoek, verschaft een goede basis voor
een realistische beschrijving van het bezwijkgedrag van betonconstructies onder zowel
trek- als drukbelastingen.
8. De in dit onderzoek ontwikkelde scheurdetectiemethode verschaft een eenvoudige en
effectieve techniek om continue macroscheurpatronen in betonproefstukken vast te leggen
en te onderzoeken.

9. Het is van groot belang dat onderzoekers zich realiseren dat twee-dimensionale
heterogene modellen homogeen zijn in de derde dimensie.

10. Tijdens fundamenteel wetenschappelijk onderzoek moet een onderzoeker leren

accepteren dat Achilles de schildpad nooit zal inhalen. Pas dan kan hij voldoening halen
uit de ongelijke strijd met de natuur en iedere terreinwinst op waarde schatten.

11. Het plannen van de ontwikkeling van zowel numerieke als ook experimentele
technieken binnen een 4-jarig promotieonderzoek is een riskante zaak en zou alleen
mogen gebeuren wanneer voor beide voldoende faciliteiten en personele ondersteuning
aanwezig zijn.

12. Aangezien details beslissend kunnen zijn voor het slagen van een wetenschappelijk
experimenteel onderzoek is het noodzakelijk dat een onderzoeker veel eenvoudig
tijdrovend werk begeleidt of zelf uitvoert.

13. De noodzaak tot publiceren bij wetenschappers heeft tot gevolg dat het aantal
publicaties toeneemt, maar tegeliJltertijd de gemiddelde originaliteit en kwaliteit afneemt.
Dit maakt de uitvoering van een effectieve literatuurstudie steeds moeilijlter.

14. Deelname aan een congres van een onderzoeker in de aanloopfase van zijn onderzoek
kan een grote positieve invloed hebben op de verdere ontwikkeling en moet daarom
gestimuleerd worden.

15. Vaak wordt bij congressen te veel gestreefd naar het scheppen van de mogelijkheid om
iedere onderzoeker zijn werk te laten presenteren, terwijl te weinig tijd wordt g~eserveerd
voor het houden van algemene, fundamentele discussies over de ontwikkelingen binnen het
vakgebied.

16. Het is onrechtvaardig en onacceptabel dat stakingsacties een derde niet in het conflict
betrokken partij duperen.

17. Al is de computer nog zo snel, de eisen van de onderzoeker achterhalen hem wel.