You are on page 1of 347

Practitioners guide to

finite element modelling


of reinforced concrete
structures
State-of-art report prepared by
Task Group 4.4

June 2008

Subject to priorities defined by the Technical Council and the Presidium, the results of fibs work in
Commissions and Task Groups are published in a continuously numbered series of technical publications
called 'Bulletins'. The following categories are used:
category
Technical Report
State-of-Art Report
Manual, Guide (to good practice)
or Recommendation
Model Code

minimum approval procedure required prior to publication


approved by a Task Group and the Chairpersons of the Commission
approved by a Commission
approved by the Technical Council of fib
approved by the General Assembly of fib

Any publication not having met the above requirements will be clearly identified as preliminary draft.
This Bulletin N 45 was approved as an fib State-of-art report by Commission 4 in April 2004.
This report was drafted by Task Group 4.4, Computer based modelling and design, in Commission 4, Modelling
of structural behaviour and design:

Koichi Maekawa 10, 1, 3, 5, 6 (Univ. of Tokyo, Japan, Co-Convener), Frank Vecchio 1, 3, 5, 6, 10 (Univ. of
Toronto, Canada, Co-convener)

Stephen Foster 2, 6, 3, 4, 5, 8 (Univ. of New South Wales, Australia, Editor)


Oguzhan Bayrak 4, 6 (Univ. of Texas at Austin, USA), Evan Bentz 3, 4 (University of Toronto, Canada),
Johan Blaauwendraad 9 (Delft Univ. of Technology, The Netherlands), Jan Cervenka 5 (Cervenka
5, 6, 10
Consulting, Czech Republic), Vladimir Cervenka
(Cervenka Consulting, Czech Republic), Tetsuya
Ishida 6 (Univ. of Tokyo, Japan), Milan Jirasek 6 (Czech Technical Univ. in Prague, Czech Republic),
Walter Kaufmann 2 (dsp Ingenieure & Planer AG, Switzerland), Johann Kollegger 5 (Technische Univ.
7, 8
7
Wien, Austria), Daniel Kuchma
(Univ. of Illinois, USA), Ho Jung Lee (SC Solutions, Inc., USA),
3
Giuseppe Mancini (Politecnico Torino, Italy), Giorgio Monti (Sapienza Universit di Roma, Italy) 4, 6,
Josko O!bolt 5, 6 (Univ. Stuttgart, Germany), Clemens Preisinger 5 (Technische Univ. Wien, Austria),
Enrico Spacone 4 (Univ. of Chieti-Pescara, Italy), Tjen Tjhin 8 (Buckland and Taylor Ltd. Bridge
Engineering, USA)
1, 2, 3 ...

Chapter number for which this member was the main preparing author

1, 2, 3 ...

Chapter number for which this member provided contributions

Full address details of Task Group members may be found in the fib Directory or through the online services on
fib's website, www.fib-international.org.
Cover image: FE modelling of high strength squat shear walls (image courtesy of S. Foster)
fdration internationale du bton (fib), 2008
Although the International Federation for Structural Concrete fib - fderation internationale du bton - does its
best to ensure that any information given is accurate, no liability or responsibility of any kind (including liability
for negligence) is accepted in this respect by the organisation, its members, servants or agents.
All rights reserved. No part of this publication may be reproduced, modified, translated, stored in a retrieval
system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or
otherwise, without prior written permission.
First published in 2008 by the International Federation for Structural Concrete (fib)
Postal address: Case Postale 88, CH-1015 Lausanne, Switzerland
Street address: Federal Institute of Technology Lausanne - EPFL, Section Gnie Civil
Tel +41 21 693 2747 Fax +41 21 693 6245
fib@epfl.ch www.fib-international.org
ISSN 1562-3610
ISBN 978-2-88394-085-7
Printed by Sprint-Digital-Druck, Stuttgart

Preface
In September 2000, fib Commission 4 established a task group (TG4.4) with the objective of
preparing a report for guiding design engineers in the safe use of computer-based analysis
procedures for design of reinforced concrete structures. The working party that first met in
Berlin 2001 brought together a group of highly regarded researchers from Europe, the
Americas, East Asia and Australasia with the objective of producing a document for use by
engineers with some background in numerical modelling. In the six years since work started
on this report, advanced models have continued to be developed; however, this report is not
about picking one model over another but, rather, how designers can use existing and future
models as tools in design practice, in benchmarking of their models against established and
reliable test data and in selecting an appropriate safety factor as well as recognising various
pitfalls.
Non-linear computer analysis methods have seen remarkable advancement in the last halfcentury with much research activity in the manner of constitutive modelling of reinforced
concrete behaviour and in the development of sophisticated analysis algorithms. These
advancements are well documented in various state-of-the-art reports and remain the subject
of intensive research today. Linear and non-linear analysis methods, combined with plasticity
design processes, and with local detailing methods such as strut-and-tie modelling, can form
the basis of design of new, complex, structures that are not easily dimensioned using other
rational design methods.
The state-of-the-art in non-linear finite element analysis of reinforced concrete has progressed
to the point where such procedures are close to being practical, every-day tools for design
office engineers. No longer solely within the domain of researchers, they are finding use in
various applications; many relating to our aging infrastructure. Non-linear computer analysis
procedures can be used to provide reliable assessments of the strength and integrity of
damaged or deteriorated structures, or of structures built to previous codes, standards or
practices deemed to be deficient today. They can serve as valuable tools in assessing the
expected behaviour from retrofitted structures, or in investigating and rationally selecting
amongst various repair alternatives. Non-linear finite element analysis procedures are also
proving particularly valuable in forensic analyses. In the near future, they will likely form the
main engine in computer-based automated design software, although in a form likely invisible
to the user.
This report provides an overview of concepts and techniques relating to computer-based
modelling of structural concrete. It attempts to provide a diverse and balanced portrayal of the
current technical knowledge, recognizing that there are often competing and conflicting
viewpoints. The report is written primarily for the benefit of the practicing engineer, rather
than as a state-of-the-art for researchers, concentrating more on practical application and less
on subtleties in constitutive modelling. To the members of the working party, our sincere
thanks for the extensive and voluntary work undertaken over an extended period to get this
report competed.
Stephen Foster, Editor, chair of fib Commission 4
Koichi Maekawa, co-convener of TG 4.4
Frank Vecchio, deputy chair of fib Commission 4 and co-covener of TG 4.4
14 November 2007

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

iii

Contents
1

Introduction
1.1 Preamble
1.2 Notation
1.3 Sample applications
(1.3.1 Kimberley!Clark warehouse 1.3.2 Sleipner A offshore platform
1.3.3 Frame corner 1.3.4 Base slabs in LNG storage tank)
The question of accuracy (1.4.1 Reasons for caution)

1.4
1.5 Challenges remaining
1.6 Objectives
1.7 Scope of report
1.8 References
Design using linear stress analysis
2.1 Introduction
2.2 Membrane structures

1
1
2
2
20
27
29
30
30
33
33
34

(2.2.1 Notation 2.2.2 General 2.2.3 Reinforcement in one direction


2.2.4 Isotropically reinforced panels 2.2.5 The general solution
2.2.6 Some comments on the angle ! 2.2.7 The design concrete
compression strength, fcd. 2.2.8 Example Design of a reinforced concrete
squat shear wall)

2.3

Slabs and shells

52

(2.3.1 General 2.3.2 Stress resultants 2.3.3 Equilibrium, stress


transformation and boundary conditions for slabs 2.3.4 Normal moment
yield criterion for slabs 2.3.5 Sandwich model for the dimensioning of shell
elements 2.3.6 Dimensioning of slab and shell elements in design practice
2.3.7 Example 1 2.3.8 Example 2)

2.4

3D solid modelling

70

(2.4.1 Introduction 2.4.2 Background 2.4.3 Application to reinforced


concrete 2.4.4 Reinforcement dimensioning for 3D stresses ! example 1
2.4.5 Reinforcement dimensioning for 3D stresses ! example 2)

2.5 References
Essential nonlinear modelling concepts
3.1 Introduction
3.2 Nonlinear concrete behaviour

78
83
83
84

(3.2.1 Concrete in compression 3.2.2 Concrete in tension


3.2.3 Modelling of tension stiffening 3.2.4 Modelling of concrete cracks
3.2.5 Modelling of reinforcement)

3.3

Nonlinear concrete modelling framework

98

(3.3.1 Elasticity 3.3.2 Plasticity 3.3.3 Damage 3.3.4 Mixed models


3.3.5 Discrete modelling frameworks)

3.4

Solution methods

102

(3.4.1 Newton!Raphson method 3.4.2 Modified Newton!Raphson


method)

3.5
3.6
3.7
3.8
3.9
4

iv

Precision of nonlinear concrete FE analyses


Safety and reliability
Statistical analyses
Concluding remarks
References

Analysis and design of frame structures using non!linear models


4.1 Introduction
4.2 Notation

104
105
114
115
115
121
121
122

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

4.3

Nonlinear models of frame elements

123

(4.3.1 Lumped versus distributed plasticity 4.3.2 Distributed models


4.3.3 Section models: fibre elements vs. strut!and!tie 4.3.4 Modelling of
shear 4.3.5 Modelling Bond Slip in Beams 4.3.6 Analysis of a section)

4.4

Interpretation of results

148

(4.4.1 Localisation problems 4.4.2 Physical characteristics of localised


failure in concrete 4.4.3 Regularisation techniques for force!based frame
elements 4.4.4 Practical considerations)

4.5
5

References

Analysis and design of surface and solid structures using non!linear models
5.1 Introduction
5.2 Notation
5.3 2D Structures with in!plane loading
5.4 Plate and shell structures (5.4.1 Layered elements)
5.5 Three dimensional solid structures

160
165
165
165
166
170
173

(5.5.1 Introduction 5.5.2 Models based on non!linear elasticity


5.5.3 Fracture!plasticity modelling 5.5.4 Microplane model
5.5.5 Examples of the application of 3D FE modeling)

5.6 References
Advanced modelling and analysis concepts
6.1 Introduction
6.2 Constitutive frameworks

190
195
195
195

(6.2.1 Non!linear elasticity 6.2.2 Plasticity 6.2.3 Continuum damage


mechanics 6.2.4 Smeared crack models 6.2.5 Microplane models)

6.3

Solution strategies

214

(6.3.1 Introduction 6.3.2 Newton!Raphson method 6.3.3 Modified


Newton!Raphson method 6.3.4 Incremental displacement method
6.3.5 The constant arc length method 6.3.6 Line searches
6.3.7 Convergence criteria 6.3.8 Load!displacement incrementation)

6.4

Other issues

223

(6.4.1 Post peak response of compression elements 6.4.2 Effects of ageing


and distress in concrete 6.4.3 Effects of ageing and distress in reinforcing
steel 6.4.4 Second order effects)

6.5 References
Benchmark tests and validation procedures
7.1 Introduction
7.2 Calibration and validation of NLFEA models

227
233
233
234

(7.2.1 Overview of model calibration and validation process 7.2.2 Level 1:


model calibration with material properties 7.2.3 Level 2: validation and
calibration with systematically arranged elementlevel benchmark tests
7.2.4 Level 3: validation and calibration at structural level)

7.3
7.4

Selection of global safety factor


Other issues in the use and validation of NLFEA programs

239
241

(7.4.1 Problem definition and model selection 7.4.2 Working within the
domain of the programs capability)

7.5

Case 1: Design of a shear wall with openings

244

(7.5.1 Objective 7.5.2 Level 1 calibration 7.5.3 Level 2 and 3


validation 7.5.4 Evaluation of global safety)

7.6

Case study II: design of simply supported deep beam

250

(7.6.1 Objective 7.6.2 Calibration and validation of NLFEAP!1


7.6.3 Calibration and validation of NLFEAP!2 7.6.4 Analysis of deep
beam)
fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

7.7
7.8
7.9
8

Summary and future trends in model validation


Future trends in model validation
References

Strut!and!tie modelling
8.1 Introduction
8.2 Notation
8.3 Overview of the STM
8.4
8.5

(8.3.1 Strut!and!tie models 8.3.2 Components of strut!and!tie models


8.3.3 Admissible strut!and!tie models)
STM design steps (8.4.1 Complications in STM design)

Some considerations in using the STM

260
261
263
265
265
266
267
270
271

(8.5.1 Rules in defining D!regions 8.5.2 Two! and three!dimensional


D!regions 8.5.3 Capacity of struts 8.5.4 Uniqueness of strut!and!tie
models 8.5.5 Strain incompatibility of struts and ties 8.5.6 Tension
stiffening in ties 8.5.7 Influence of tie anchorages 8.5.8 Size, geometry,
and strength of nodal zones 8.5.9 Load redistribution and ductility
requirements)

8.6
8.7

Computer!based STM
Modelling aspects using computer!based STM

279
280

(8.7.1 Identifying strut!and!tie models 8.7.2 Refining strut!and!tie


models 8.7.3 Other considerations 8.7.4 Static indeterminacy of
strut!and!tie models 8.7.5 Procedures to solve statically indeterminate
strut!and!tie models 8.7.6 Dimensioning nodal regions)

8.8

Design example using computer!based tools

298

(8.8.1 Problem statement 8.8.2 Solution)

8.9 References
Special purpose design methods for surface structures
9.1 Introduction
9.2 Notation
9.3 Design of slabs and shear walls: perfect plastic approach

303
307
307
307
309

(9.3.1 Slabs subjected to bending loads 9.3.2 Ultimate load determination


9.3.3 Failure mode determination 9.3.4 Material optimization
9.3.5 Plates subjected to in!plane loads)

9.4

Design of slabs using the reinforcement field approach

318

(9.4.1 Linear yield conditions for element nodal forces 9.4.2 Material
optimisation through stress redistribution 9.4.3 Slab subjected to bending
loads 9.4.4 Dimensioning procedure)

9.5

Design of shear!walls: the stringer!panel approach

321

(9.5.1 Linear!elastic version 9.5.2 Non!linear version


9.5.3 A three!step design procedure 9.5.4 Example)

9.6 References
10 Concluding remarks
10.1 Introduction
10.2 Structural performance based design in practice
10.3 Benefits of non!linear modelling and analyses
10.4 Code provisions
10.5 Specification of design loads
10.6 Maintenance
10.7 References

vi

329
331
331
331
333
335
335
336
337

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

Introduction

1.1

Preamble

Computer-based analysis procedures for reinforced concrete structures, and finite element
analysis procedures in particular, have seen tremendous advancement in the last half-century.
Much research activity has occurred in the manner of constitutive modelling of reinforced
concrete behaviour and in the development of sophisticated analysis algorithms. These
advancements are well documented in various state-of-the-art reports, and still remain the
subject of intensive research.
Occurring at the same time, and no less significant, has been the accelerated development of
computing technology and hardware. Data compiled by Bentz (2006), shown in Figure 1.1,
provide a clear measure of the exponential growth in computing power in recent years. Shown
is the time required to conduct a nonlinear shear analysis of a prestressed T-beam using a
layered beam element algorithm. It is seen from the graph that, in 25 years, computing speed
has increased by five orders of magnitude. Analyses that required several days of CPU time
on supercomputers two decades ago run in minutes on personal desktop computers today. The
advent of powerful and relatively inexpensive computers has greatly expanded the size and
complexity of problems that can be analysed, and has greatly reduced the computer time
required for their solution.
Computer Performance
100

Pentium 4

15 sec

Pentium III

Estimated SPECint95

10

Pentium Pro

3 min

Pentium

30 min

0.1

5 hours

80486
80386
80286

0.01

2 days

0.001

20 days

8088

8080

0.0001

7 months

1970

1975

CFT
published

1980

MCFT
published

1985

1990

Kobe
Earthquake

1995

DSFM
published

2000

2005

Year of CPU introduction

Figure 1.1: Increase in computing power in recent years (Bentz, 2006).

The state-of-the-art in nonlinear finite element analysis (NLFEA) of reinforced concrete has
thus progressed to the point where such procedures are close to being practical, every-day
tools for design office engineers. No longer solely within the domain of researchers, they are
finding use in various applications; many relating to our aging infrastructure. NLFEA
procedures can be used to provide reliable assessments of the strength and integrity of
damaged or deteriorated structures, or of structures built to previous codes, standards or
fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
1!

practices deemed to be deficient today. They can serve as valuable tools in assessing the
expected behaviour from retrofitted structures, or in investigating and rationally selecting
amongst various repair alternatives. In situations that have not turned out well, NLFEA
procedures are finding applications to forensic analyses and litigations that follow. In the near
future, they will likely form the main engine in computer-based automated design software,
although in a form likely invisible to the user.

1.2

Notation

Agm

maximum aggregate size

As
D
!"#
!$#

cross section area of rebar


rebar diameter
compressive strength of concrete cylinder at 28 days
tensile strength of concrete

fy

yield strength of reinforcement

Mu

sectional moment capacity of beam (hand calculated)

Pu

ultimate load capacity of beam (finite element analysis)

Vu1

sectional shear capacity of beam (Simple Method of CSA A23.3)

Vu2

sectional shear capacity of beam (General Method of CSA A23.3)

!u

midspan deflection at ultimate load (finite element analysis)

"o
#
$
%

concrete strain at peak compressive stress


shear strain
reinforcement ratio
shear stress

1.3

Sample applications

The failure of two reinforced concrete structures is recounted below; one involving a
warehouse structure and the other an offshore platform base-structure. The structures were
subsequently analysed using nonlinear finite element analysis procedures, taking into account
relevant second-order behaviour models. The analyses were found to provide an accurate
assessment of the load capacities and failure modes observed; in addition, they provided
meaningful insights into the underlying behaviour mechanisms and factors leading to the
failures. Hence, these two sample applications serve to show that nonlinear analysis
techniques can be useful everyday tools for design office applications, particularly in forensic
work. As well, they provide evidence that errors made in the design of modern structures can
be potentially more catastrophic, and that advanced assessment techniques will assume
increased importance as a result.
Also discussed below are two additional examples in which nonlinear finite element analysis
procedures were used to aid in the design of structures. The first involves a prestressed
concrete frame; the second, the base slab in an underground liquid natural gas (LNG) storage
tank. Again, both examples serve to illustrate the usefulness of advanced analysis procedures
in solving difficult design problems.

!
1 Introduction !

1.3.1

Kimberley-Clark warehouse

Details of warehouse failure


The Kimberley-Clark building was built in 1944, in Niagara Falls, Canada, in accordance
with then-current building codes. The building was a simple four-storey structure with
basement, having plan dimensions of approximately 38 x 36 m (see Figure 1.2). The structural
system employed was primarily a reinforced concrete flat slab system with six bays in each
direction. The centre-to-centre column spacing was approximately 6.25 m in the N-S
direction, and 5.85 m in the E-W direction. Exterior columns were rectangular with haunches,
while interior columns were circular with capitals. Column and capital diameters decreased
with increasing elevation; the columns supporting the third floor were 450 mm in diameter
with 1.5 m diameter capitals. The floor-to-floor height ranged from 3.35 m to 3.65 m. Exterior
walls were constructed of brick masonry, and stair/elevator shafts were located at various
points around the perimeter of the structure.

Figure 1.2: Floor plan of Kimberley-Clark warehouse.

The floor slabs were typically 200 mm thick. At the third floor level, the slab was thickened
by 150 mm around the perimeter, over a width of 1.3 m. The floor slabs were reinforced with
No.4 (13 mm dia.) and No. 5 (16 mm dia.) deformed bars. The reinforcement details were
consistent with a column-strip/middle-strip design method, as shown in Figure 1.3. Similar
reinforcement patterns and amounts were used in both directions.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
3!

Figure 1.33: Reinforcem


ment details for third storey floor
f
slab.

The enggineering plans


p
for thhe structuree specified service livve loads off 12.0 kPa for the
basemennt and firstt floor, 6.0 kPa for thhe remainin
ng floors, and
a 1.9 kPaa for the ro
oof. The
specifieed concrete strength waas 20 MPa, and the allowable tenssile stress inn the reinfo
orcement
was 1388 MPa. Thee actual com
mpressive sttrength of th
he concretee, determineed from corres taken
in 19788, had an average
a
vallue of 37.22 MPa. Fro
om couponss taken at tthe same tiime, the
reinforccing steel was
w found to have a yield strength of 442 MPa and tensile streength of
710 MP
Pa, with a modulus
m
of elasticity
e
of 210,000 MPa.
The thirrd floor of the warehoouse becam
me the site for the storrage of druums of nick
kel pellet
beginninng in mid-D
December 1977. The drums, on wooden paallets, weree transporteed to the
third flooor by a freiight elevatoor, and then moved to the
t storage location
l
by forklift trucck. Each
drum weighed
w
2255 kg, and there were eiight drums to
t a pallet. The palletss measured 915 mm
square. In general, the pallets were stackked side by side, two high,
h
giving a live floorr load of
about 43.1 kPa. Cooincidentallly, an alternnate-bay loaading patterrn evolved as one bay was left
free to facilitate foork-truck acccess. Anotther perhap
ps relevant detail is that the floorrs below

!
1 Introduction !

housed a paper-prroducts mannufacturingg plant. Wiith such opperations, thhe dangers of dust
explosioons and firee are an everr-present cooncern.
On Januuary 4, 19788, a large seection of thhe third floo
or collapsed; see Figuree 1.4. The collapsed
c
area enccompassed approximattely 16 bayss of the stru
ucture, withh columns, sslabs, and drums
d
of
nickel falling
f
throuugh to the basement
b
leevel. At abo
out the sam
me time of thhe collapsee, a large
explosioon occurredd and fire engulfed thee structure. The fire coontinued forr 48 hours before
b
it
could bee controlledd, and two men
m died in the inciden
nt. Indicatedd in Figure 11.4 is the nu
umber of
pallets of
o nickel peellet, per bayy, determineed to be preesent at the time
t
of colllapse.

F
Figure
1.4: Looading and co
ollapse patternn.

The conntention off the paper manufacturring firm was


w that thee stored nicckel overloaaded the
floor, resulting
r
inn its collappse and coonsequently
y sparking an explossion and fire.
fi
The
propriettors of the nickel opeeration arguued that an explosion and fire inn the plant was the
primaryy cause of collapse;
c
thaat is, the inttense heat from
f
the firre instantly weakened the
t floor
above and
a led to itss failure.
Analysees of warehouse structture
To obtaain an estim
mate of the load-bearing capacity
y of the slaab at the thhird floor leevel, the
structurre was anallysed usingg nonlinear finite elem
ment analyssis proceduures. In add
dition to
materiall and geoometric nonnlinearities, the analy
yses considdered the membrane action,
temperaature degraadation of material
m
strrengths, tim
me-related effects andd other infl
fluencing
factors. The analysis approacch was to model
m
portions of the third floor slab, and columns
above and
a below, using
u
plane frame stripps. The mosst critically loaded porttion of the structure
s
was assumed to be along Coluumn Line 4.. In the mod
delling, fourr bays extennding from Column
Lines A through E
E, one bay wide,
w
were considered.
c
To take intto account tthe influencce of the

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
5!

column capitals and reduced clear spans, beam elements representing the drop panels were
used. The four stairwells, located roughly at the corners of the warehouse floor plan
(Figure 1.2), likely influenced the behaviour of the floor slab as will be discussed. To estimate
their lateral stiffness, the stairwells were considered to be cantilevered from the basement
floor. All constitutive modelling was based on the models of the Modified Compression Field
Theory (Vecchio and Collins, 1986). Vecchio and Collins (1990) provide complete details of
the modelling.
The nonlinear analyses indicated that the ultimate capacity of the floor in the vicinity of the
collapse was approximately 45.5 kPa (superimposed load), as shown in Figure 1.5a. Including
the dead load, the total capacity was 50.3 kPa. Note that the estimated floor capacity is only
slightly higher than the estimated floor loading at the time of collapse, but is 7.6 times the
value of the specified floor capacity.
The most significant difference between linear and nonlinear analyses, in this situation, lies in
the fact that the nonlinear analyses accounted for the net elongation that typically occurs in
reinforced concrete elements in flexure (see Figure 1.5b). Tensile strains on the cracked face
of an element are normally much larger than the compressive strains on the opposite face,
particularly if the reinforcement is yielding. Hence, the tendency is for a slab to elongate as it
flexes under the applied transverse loads. When the slab is prevented from elongating, as it is
in this case by the columns, stairwells, elevator shafts, and floors above and below, a
compressive thrust is induced in the slab. For reinforced concrete elements, axial compressive
forces generally serve to increase flexural and shear capacity. This behaviour, known for
many years, is commonly referred to as compression membrane action.
In the case of the Kimberley-Clark Warehouse, a linear elastic analysis of the third floor
subsystem predicts very little gain in the axial compressive force as the floor load is
increased. The flexural capacity at the midspan of the floor is exceeded at approximately
21.5 kPa live load. Allowing for partial moment redistribution, as permitted by ACI 318
(1983), the estimated ultimate floor capacity increases to 25.4 kPa. If complete moment
redistribution is considered, the calculated floor capacity increases to approximately 31.1 kPa.
This remains well below the observed capacity at failure. The nonlinear analyses, on the other
hand, show a much more pronounced increase in axial force with increasing floor load; see
Figure 1.5c. The proper consideration of membrane action, only possible through nonlinear
analyses, was instrumental in achieving an accurate estimate of the load capacity of this
structure.
Further analyses were conducted considering the effects of elevated temperatures if a fire
were present below the heavily loaded third floor slab. A temperature of 1000!C was assumed
to be acting instantaneously on the bottom surface of the slab, and standard uni-directional
heat flow analyses were used to compute the progression of the thermal gradient through the
slab. Reductions in materials strength and stiffness, arising from the elevated temperatures,
were considered for both concrete and steel. The analyses indicated that collapse would have
occurred within 2 to 3 minutes from the onset of the fire. Hence, it was concluded that the
collapse could as likely have been triggered by a fire below weakening the floor as by floor
overload.

!
1 Introduction !

Figurre 1.5: Resultss of analyses for column linne 3/4.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
7!

1.3.2

Sleipner A offshore platform

Details of platform failure


The Condeep offshore platforms, of which Sleipner A was one, serve in the hostile
environment of the North Sea in water depths of up to 300 m. The construction of a typical
Condeep platform starts in a large dry-dock area, and is moved to a sheltered fjord as the
construction progresses. In later stages, the base structure is floated out to sea, then partially
submerged and mated with the deck structure. After deck-mating, the completed structure is
towed to its offshore site and lowered to a final position on the sea floor. A critical factor in
the design of the base structure is the thickness of the cell walls. If the walls are too thin, they
may fail under the very high water pressures to which they are subjected during deck-mating.
If the walls are too thick, the structure will not float, or will not be hydrostatically stable
during the towing operations. The margin between the two is narrow.
The gravity base structure of the Sleipner A offshore platform, which was 110 m high,
consisted of a cluster of 24 buoyancy cells. Four of these cells extended upward to form shafts
connecting to the deck structure. Details are given in Figure 1.6. The exterior walls of the
cells were circular, with radii of 12 m, whereas the interior walls separating the cells were
straight. The intersection of the interior walls formed a small triangular void called a tricell.
There were a total of 32 such tricells. Because the tricells were open at the top, they filled
with water once the tops of the cells were submerged. Hence, the walls of the tricells had to
resist a substantial hydrostatic pressure.
On August 23, 1991, the gravity base structure for Sleipner A was slowly being submerged as
part of the deck-mating operation. The intention was to lower the structure until the base was
104 m below the ocean surface. However, when a depth of 99 m was reached, a loud
rumbling noise emanated from one of the drill shafts and water could be heard entering.
Within minutes, the structure began to sink in an uncontrollable manner. Moments after
disappearing below the surface, a series of implosions were felt as the buoyancy cells
collapsed. Evidence showed that the loss of the structure was attributed to the shear failure of
the wall of Tricell 23 adjacent to drill shaft D3. At failure, this 550 mm thick wall was
resisting a 65 m head of seawater, resulting in a pressure of about 655 kPa.
The reinforcement details in the tricell walls near the failure location, described in detail by
Collins et al. (1997), are summarized in Figure 1.7. In general, a grid of horizontal and
vertical bars was provided near each face of each wall of the tricell. For approximately the
bottom third of the height of the cell walls, 12 mm diameter stirrups were provided at a 170 x
170 mm spacing. In the middle-third height of the walls, the stirrup spacing was increased to
170 x 340 mm spacing. Just below the location of the failure, the stirrups were terminated and
the walls contained no shear reinforcement thereon up. Another detail of note is the provision
of a T-headed bar placed across the throat of the tricell joint. This 25 mm diameter bar was
about 1.0 m long and had steel plates welded on its ends to provide anchorage.
The failure of the Sleipner base structure, which involved a total economic loss of about $700
million USD, was probably the most expensive shear failure on record.

!
1 Introduction !

Figure 1.6: Details off Sleipner A co


oncrete base structure.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
9!

Figgure 1.7: Reinfforcement dettails for tricelll 23.

10

!
1 Introduction !

Analysees of platforrm base stru


ucture
To deveelop a betteer understannding of thee factors in
nfluencing thhe failure oof the tricell wall, a
series of
o three-dim
mensional nonlinear
n
fiinite element analysess were unddertaken. Th
he finite
elementt mesh usedd for the innvestigation is shown in
i Figure 1.8. Because of symmettry, only
one-sixtth of the trricell needed to be moodelled. Thee one elem
ment thick m
mesh contaiined 342
brick ellements (244 dof) and 338
3 wedge elements
e
(1
18 dof). All constitutivve modelling
g was in
accordaance with thhe Modifiedd Compressiion Field Th
heory. In the analyses, the horizon
ntal axial
force inn the tricell wall was heeld constantt at 5000 kN
N/m while the
t hydrosttatic pressurre on the
inner faace of the wall
w was inccreased until failure. In
n addition, a constant vvertical com
mpressive
stress of 7 MPa waas assumed acting in thhe walls. Full details of
o the finitee element modelling
m
are disccussed by Coollins et al. (1997).

F
Figure
1.8: Fiinite element model
m
of tricelll.

The anaalyses indiccated that thhe as-built structure


s
wo
ould sustainn a brittle, sshear-criticaal failure
when thhe applied water presssure on thhe inner facces of the tricell reacched 620 kP
Pa. This
correspoonded to a head of seaawater 62 m high, and agreed welll with the 665 m head acting
a
at
the critiical locationn at the tim
me of failuree (see Figuree 1.7b). Note that this analysis waas a first
run andd involved the
t input of details annd material properties as then knnown, and no
n finetuning of analysiis parameteers was invvolved. Hence, the abbility of N
NLFEA metthods to
properlyy represent behaviour in this situaation was excellent. It is also inteeresting to note
n
that
ACI 3118-95 proceedures leadd to a prediicted failurre pressure of 1200 kkPa while th
he 1994
AASHT
TO LRFD Specification
S
ns gives 4500 kPa; both significantlly in error.
The dessigners of thhe structuree were interrested in leaarning how the strengthh of the triccell wall
would have
h
changeed if the stirrrups, whichh had been terminated just below the failure location,
l
had beeen continuedd up the wall. They weere also keen
n to know how
h the lenggth of the T-headed
T
bar in thhe throat off the joint reegion influenced the faiilure load. To
T answer tthese questions, two
series of nonlinear finite elem
ment analysees were cond
ducted. In one
o set the w
walls were assumed
a
to contaain 12 mm diameter
d
stiirrups at 1700 x 340 mm
m spacing, while
w
for thee other set the
t walls
were asssumed to contain
c
no stirrups. Within
W
each set, the lenngth of thee T-headed bar was
varied from
f
zero (i.e., no bar)) to the maaximum posssible lengthh (1.5 m). T
The results of these
studies, summarizeed in Figuree 1.9, indicaate that the tricells couuld have resisted an ad
dditional

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
11 !

20 m off water heaad if either the stirrupss had been continued up


u the walll or if the T-headed
T
bars haad been aboout 500 mm
m longer. The structu
ure would have survivved in eith
her case.
Changes were madde accordinngly to the design of the replaceement struccture. On April
A
29,
1993, thhe new conncrete gravvity base strructure wass successfuully mated w
with the deeck, and
Sleipnerr A2 was reeady to be toowed to seaa. It remainss in service today.
t

Figgure 1.9: Influuence of stirruups and lengthh of T-headed bar on prediccted failure prressure for triccell.

1.3.3

F
Frame
corn
ner

Problem
m descriptioon
Frame corners
c
are a typical deesign probleem where reinforcemennt detailingg plays an im
mportant
role. Thhe region of
o a frame corner is subjected to
t a compleex stress sttate and caannot be
analysedd by beam theory. Connsequently, it is design
ned by emppirical rules which typiically do
not cover specific situations. The
T followiing analysiss was perfoormed withiin the project of the
newly built
b
Brixeen Universiity (structuural designeed by Berggmeister, B
Brixen). Th
he frame
structurre is a part of the lectture auditorrium buildiing and is visually exxposed on the
t back
elevatioon with a laarge glass facade
f
creaating a signiificant aesthhetic impreession. Therrefore, a
stringennt requiremeent on the crack
c
width limit was im
mposed. Thhe girder is ppre-stressed
d and the
cable annchors signnificantly innfluence thee stress statee in the fraame corner. The reinfo
orcement
detailingg was an im
mportant isssue in this project.
p
Thee model is reduced to a section in
ncluding
the sym
mmetrical haalf of the giirder with the
t floor slaab and partss of columnns; see Figu
ures 1.10
and 1.111. Numerical analysiis based onn the work
k of Cervennka (1999) was used for the
reinforccement detaailing optimiisation.
Numeriical model
The moodel developpment was based on thhe analysis of the wholle 5-storey frame, Figu
ure 1.10.
At firstt the question of model complexxity was co
onsidered: How
H
far doo we have to
t go to
capture the structurral behaviouur realisticaally? In this case two models
m
weree considered
d; a two12

!
1 Introduction !

dimensional model, based on a plane stress state, and a three dimensional model with a fully
3D stress state. Obviously, the choice of model has a strong impact on calculation time and
objectivity of results. While a 2D model is simpler and less time demanding, it cannot
appropriately capture the confinement effect in the anchoring region of the prestressing
cables. On the other hand, the 3D model is more realistic, but it requires more time for data
preparation and calculation. In this case it was decided to use the 2D model for global
information on the structural behaviour and for parametric studies relating to structural
optimisation. The 3D model was used for a detailed analysis of selected cases.
The numerical model reflected in a realistic way the geometry and reinforcement. It included
three types of reinforcing models: smeared reinforcement (stirrups and slab), bars (main
girder and column reinforcement), prestressing cables, see Figure 1.11.

Figure 1.10: Frame structure (left) and frame corner (right) dimensions (in m) and loading.

Figure 1.11: Reinforcing by ordinary bars (left) and pre-stressing cables (right).

Constitutive models used for concrete were of two types: For plane stress analysis the
damage-based model (Cervenka, 2002), and for 3D analysis the fracture-plastic model
(Cervenka and Cervenka, 1999) were used. Both models employ the same approach for
tensile cracks, which is based on smeared cracks, crack band and fracture energy. Hordijks
exponential function is applied for the crack opening law. Compressive response in plane
stress is based on orthotropic damage with the failure function according to Kupfer. In 3D
analysis, the plastic flow theory with a Menetre-Willam surface is utilized. The post-peak
softening is employed in both tension and compression responses.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
13 !

Serviceability analysis
The prime objective of the analysis was to determine the crack width under service load in the
frame corner. It was difficult to calculate these cracks with usual beam theory and therefore a
FE-based simulation was employed. Crack patterns resulting from the analysis are shown in
Figure 1.12 for both models. In the 2D model the maximum crack width in the frame corner
was 0.28, mm and in the 3D analysis it was 0.15 mm. Both were calculated assuming a crack
spacing of 100 mm. Both crack patterns are very similar in locations and directions. It is
interesting that the 3D model gives smaller crack width. This is explained by the capability of
the model to capture the variable crack width in the third direction and thus producing smaller
crack width near the surface, where the reinforcement is located. The distribution of the
principal compressive stress in the corner in 2D analysis is shown in Figure 1.13.

2D model

3D model

Figure 1.12: Crack patterns under service load.

Figure1.13: Response of frame corner.

14

!
1 Introduction !

The crack spacing must be also examined. The crack model is based on a fracture mechanics
approach, and the crack width within this frame is independent of the element size. The crack
spacing develops naturally as the strains localize. This works well for single discrete cracks.
However, in reinforced concrete with several parallel cracks of similar thickness, one main
crack need not develop due to the restraining effect of reinforcement. Then the crack width is
a product of crack strain and element size, while the latter represents the assumed crack
spacing. In the above case this spacing was 100 mm and represented a conservative estimate
of a real spacing. Thus, another crack width could be obtained by a simple recalculation for a
different spacing.
It should be realized that the above crack model includes the effects of strain softening in
cracks and also the so-called tension stiffening due to the contribution of uncracked and
partially cracked concrete regions, while considering the real geometrical arrangement of the
reinforcement. In the given case of design, the accuracy of the model cannot be easily
examined and we have to rely on a separate validation through known experimental data, for
which the tests of Hartl (1977) and Braams (1990) were used.
Ultimate analysis
After service load conditions were examined, the loading was further increased and the
ultimate failure load and corresponding failure mode were evaluated. The full analysis was
performed for two types of concrete post-peak behaviour in compression. The first type,
described as brittle, utilizes linear softening according to van Mier (1986), which can be
considered as a lower-bound behaviour. The second type was a perfect plasticity with
horizontal plateau after peak, which clearly overestimates ductility. Response for both models
and two types of compression softening can be seen from load-displacement diagrams in
Figure 1.14.

Figure 1.14: Load-displacement diagrams.

The brittle model in 2D shows a local instability at the load of 3.2 times the service load. At
this stage the compressive failure of the cracked concrete in the frame corner is initiated. Note
that this region is under a strong local effect of the prestressing tendon anchor. The corner
rotation due to brittle failure causes a temporary load decrease, then the load is again
increased until the bending capacity in the midspan is exhausted. The ultimate load of 63.8

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
15 !

kN/m only slightly exceeded the first peak load of 58.6 kN/m before the corner fails. The
redistribution of internal forces was also confirmed by changes in moment distribution in
girder.
Similar behaviour was observed in the case of the 3D model. However, in this case the local
instability was less pronounced and the load could be significantly increased to the maximum
75 kN/m after the partial compressive failure of the frame corner at 46.6 kN/m. The higher
ultimate load in the 3D model is explained by the better ability of the corner to resist moments
when lateral confinement is included. In spite of the partial brittle failure of concrete in the
frame corner, the structure showed relatively large ductility. The mid-span defection of 355
mm at failure in the 3D analysis was 14 times that at service load. The failure at the frame
corner can be observed in Figure 1.13b, where the localized compressive strains are indicated
near the tendon anchor.
In order to identify the source of limited ductility, both models were also analysed by models
with perfectly plastic behaviour in compression. In this case the response of the 2D and 3D
models are almost identical and the ultimate load is increased to 97 kN/m with very ductile
behaviour. It shows that if brittle failure of concrete in the frame corner is avoided, the
bending capacity can be fully utilized. However, the brittle response is considered more
realistic.
Interesting is also information about the computer time required for analyses as shown in
Table 1.1. Calculations were performed on a Pentium 4 (2.2 MHz). The loading was imposed
in load steps with a Newton-Raphson iteration scheme up to service load. Then the solution
method was changed to arc-length and loading continued to failure. The arc-length method
was capable of capturing the local instability. The convergence criterion was 1% error in
equilibrium, and was satisfied in all steps until the global failure.

Table 1.1: Frame corner analysis computer time.

Model
2D
3D

Number of finite
elements
830
11777

Load
steps
20
38

time
[minutes]
15
184

Conclusions
The nonlinear analysis has proven as an efficient tool for optimising reinforcement detailing.
The three-dimensional analysis was more helpful for assessment of crack widths in the frame
corner and of the failure mode. It better described the effects of confinement on concrete
behaviour. The two-dimensional analysis could be considered as a safe estimate and, in the
view of considerably lower computer time requirement, it was used for parameter study. The
2D and 3D analyses showed different results in case of brittle response. In case of ductile
response the results were almost identical.

16

!
1 Introduction !

1.3.4

Base slabs in LNG storage tank

Problem description
The nominal punching shear strength of RC circular slabs (3D) is larger than that of beam
shear (2D) because hoop reinforcement creates confining radial compression under flexure
and the shear span-to-depth ratio becomes substantially smaller compared with 2D beams and
columns. As the nominal punching shear strength is size-dependent, this factor leads to higher
cost performance with reduced hoop and transverse reinforcement when RC circular slabs of
large scale are designed in shear.
When a circular slab is supported with some deformational constraint, the hoop effect is
thought to be reduced and the shear capacity is probably diminished. On the contrary, the
effective shear span to depth ratio will be smaller and the capacity will be increased. Since the
design formulae are based upon the experiments of simply supported slabs, they cannot be
directly applied to more generic cases without any special discussion. For designing the world
largest underground LNG tank, a new structural type was adopted, i.e., the base slab and
sidewall shells are monolithically combined; see Figure 1.15. Here, nonlinear finite element
analysis was applied for estimating the punching shear capacity of the wall confined slabs,
because FEM can be freely applied to any boundary condition; see Figure 1.16.

Figure 1.15: Circular slab in punching shear and axi-symmetric 3D analysis (Akiyama et al., 1996; Maekawa
et al., 2003)

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
17 !

Figure 1.16: Conventional punching shear design formulas and FEM simulation (Kawabara, 2002).

Limit state analysis


First, a 3D nonlinear analysis was made of a large-scale conventional circular slab in shear
and checked by the available shear strength formulas. As the hydrostatic uniform force is
applied to the slab, a volume-control method or arc-length method are effective in computing
the softened failure states. After checking the validity of nonlinear analysis, a parametric
study was performed with different peripheral boundary conditions; see Figure 1.17. It was
found that the nominal punching shear strength does not drop but increases due to more
confined boundary conditions and hoop reinforcement effects expected in these cases. The
most current LNG storage tank under construction has been designed with the hoop effect and
successful cost savings were achieved. In this practical case, computer based modelling was
utilized to extend the applicability of design formula to more general boundary conditions.

Figure 1.17: Shear capacity of different boundary conditions (Kawabara, 2002).

18

!
1 Introduction !

Detailing
Another application of axisymmetric 3D nonlinear analysis is addressed to slab-to-wall joints
in combined flexure and shear. Linear finite element analysis was used for deciding
reinforcement ratio and orientation in the slab-wall connection. Its size was so large that no
way of arranging reinforcing bars was found, because available bar diameters are inescapably
much smaller than the size of the structure. Furthermore, the back-check by nonlinear finite
element analysis revealed that the connection may fail in shear before the failure of the slab
and the sidewall due to size effect of shear failure. In general, elastic analysis based
arrangement of reinforcement is safe and possible for construction when the size of structures
is small. But it is a different situation with large-scale structural design.
Thus, several other design details were considered and the nonlinear finite element analysis
was conducted for verification of its safety performance under static design loads
(underground water and soil pressure); see Figure 1.18. Practitioners in both design office and
construction site could have the modified dimensioning and steel arrangement which satisfies
the design requirement of safety and waterproofing serviceability.

Figure 1.18:- RC slab-sidewall connection: linear analysis based arrangement of rebars (left) and nonlinear
analysis aided design (right) (Nakano, 2002).

Discussion
In the four case studies presented, the use of advanced analytical techniques served to greatly
aid in the understanding of the factors and mechanisms influencing behaviour. With the
warehouse structure, the puzzle from an analysis perspective was not what event finally
triggered collapse but rather how could the floor sustain such high overloads in the first place.
In the case of the offshore structure, the challenge was to rigorously and accurately capture

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
19 !

the influence of local reinforcement details in an analysis of predicted response. In both cases,
as was shown, conventional analysis/design calculations were largely inadequate in providing
an accurate assessment of load capacity. Advanced nonlinear analysis procedures, on the
other hand, provided much improved modelling capability.
The four structures examined, while in themselves insufficient to form the basis for a
definitive statement, also point to another finding. That is, as design procedures become more
refined, and as economic and technologic pressures increase, the margin for error diminishes.
In the bulky concrete structures constructed a half-century ago, there was often much
redundancy in the structural system and a large margin of safety inherent in the design and
construction procedures. Hence, a structure such as the Kimberley-Clark warehouse was able
to withstand service floor loads almost an order of magnitude higher than design-specified.
On the other hand, with modern precision-engineered structures of today, based on more exact
methods of analysis and design, the margin for error is small and the penalty for
miscalculation is high. In such an environment, the value of sophisticated analysis procedures
is great provided they have been adequately verified and are properly applied.

1.4

The question of accuracy

Despite the increasing sophistication of NLFEA tools, users must be ever mindful of the
question of how accurately and reliably can they represent the behaviour of reinforced
concrete structures. In this regard, it is useful to examine the results of three prediction
competitions.
At the 1981 IABSE Symposium in Delft, a blind competition was organized involving four
panels tested in a comprehensive research program then underway at the University of
Toronto. The test panels were orthogonally reinforced, and subjected to uniform, proportional
and monotonically increasing stress conditions; seemingly a very simple problem to model
and analyse. The results of these panel tests were not disclosed prior to analysts submitting
their predictions of strength and load-deformation response. Approximately thirty entries
were received, many from the leading researchers in the field at the time. Shown in
Figure 1.19 is the range of responses for Panel C, one of the better predicted of the four
panels. The analysis results submitted showed a wide variation in predictions of the panels
shear strength, and an even wider divergence in computed load-deformation responses
(Collins, Vecchio and Mehlhorn, 1985). Clearly the collective ability to model nonlinear
behaviour of reinforced concrete, particularly in shear-critical conditions, was not well
advanced.
More recently, in 1995, the Nuclear Power Engineering Corporation of Japan (NUPEC)
staged a prediction competition involving a large-scale 3D shear wall subjected to dynamic
cyclic loading (NUPEC, 1996). The flanged shear wall exhibited highly nonlinear behaviour
before sustaining a sliding shear failure along the base of the web. One facet of the
competition called for estimates of the ultimate strength, and corresponding displacement, of
the wall as determined from static push-over analyses. Again, over thirty sets of predictions
were received; the results are summarized in Figure 1.20. The predictions of strength, as a
group, showed better correlation than was seen with the Toronto panels; however, the
deformation estimates still showed large scatter. Nevertheless, it could be concluded that the
ability of NLFEA to accurately capture the behaviour of reinforced concrete had measurably
advanced. It should be noted, however, that this was not a completely blind competition since
some of the test results had been disclosed to analysts prior to the competition.

20

!
1 Introduction !

Most recently, ASCE-ACI Committee 447 organized an informal competition centred on


results from a series of large-scale columns tested at the University of California at San
Diego. Many of the analyses undertaken pursuantly are documented in papers contained
within an ACI Special Publication (William and Tanake, 2001). From these, it can be noted
that: i) a number of significantly different analysis approaches were taken; ii) predictions of
strength and pre-peak response generally correlated well with the experimental results; and
iii) predictions of post-peak response were generally not as accurate and still require further
attention. Bear in mind, once again, that this was not a blind competition. Analysts had the
opportunity to calibrate parameters, optimize material models, and refine analyses. One
should also bear in mind that experimental results themselves are subject to scatter and error.
Repeating a test, particularly if conducted at different laboratories, may yield differing results.
Nevertheless, it is an inevitable conclusion that our ability to accurately model the behaviour
of reinforced concrete structures has seen significant improvement over the past three
decades. It has approached a stage of development where we may be inclined to proceed with
a certain degree of confidence.

Figure 1.19: Delft Panel competition results (Panel C): a) Variations in predicted shear strength; b) Variations
in predicted load-deformation response.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
21 !

1.4.1

R
Reasons
foor caution

Despite these sign


nificant adv
vancements in our abillity to accu
urately moddel the resp
ponse of
reinforcced concretee, the users of NLFEA
A proceduress need to bee mindful of several isssues and
potentiaal dangers:

Diversitty of theoreetical approaches


A numb
ber of ratheer diverse approaches
a
exist for NLFEA
N
mo
odelling of rreinforced concrete
structurres. Among those availlable are mo
odels built on
o nonlineaar elasticity, plasticity, fracture
mechan
nics, damagee continuum
m mechanics, endochro
onic theory, and other hhybrid formu
ulations.
Crackin
ng can be modelled
m
disscretely, or using
u
smearred crack ap
pproaches; the latter caan range
from fu
ully rotating
g crack mo
odels, to fix
xed crack models,
m
to multiple
m
noon-orthogon
nal crack
models,, to hybrid
d crack mo
odels. Som
me approach
hes place heavy empphasis on classical
mechan
nics formulaations, others draw mo
ore heavily on empiriccal data andd phenomen
nological
models.. It can geenerally be said of an
ny approach
h that it will
w be morre suited to
o certain
structurre/loading situations
s
an
nd less so to
t others. No
N one app
proach perfo
forms well over the
entire raange of stru
uctural detaiils and loadiing conditio
ons encounttered in pracctice.

Figure 1.20:- NUPEC Wall predictiion results: a)) Variations in


n predicted laateral load cappacity; b) Varriations in
predictedd lateral displlacement at ulltimate load.

22

!
1 Introduction !

Diversity of behaviour models


Reinforced concrete structures, particularly in their cracked states, are dominated in their
behaviour by a number of second-order mechanisms and influencing factors. Depending on
the particular details and conditions prevailing, a structures strength, deflection, ductility and
failure mode may be significantly affected by mechanisms such as: compression softening
due to transverse cracking, confinement effects, tension stiffening, tension softening,
aggregate interlock and crack shear slip, rebar bond slip, rebar dowel action, rebar
compression buckling, scale effects, and creep and shrinkage, to name a few. For each of
these, a number of diverse formulations can exist. In the case of tension stiffening, for
example, the stiffening effect can be ascribed to a post-cracking average tensile stress in the
concrete or, quite differently, to the load-deformation response of the reinforcement. The user
of a NLFEA software must be aware of what mechanisms are likely to be significant in the
problem at hand, be certain that it is included in the analysis model, and have some
confidence that the model being used is reasonably accurate.
Incompatibility of models and approaches
The formulation and calibration of a concrete behaviour model, as it is being developed, is
often dependent on the particular analysis methodology being used. As a consequence, some
models cannot be randomly transplanted from one analysis approach to another, or freely
combined with other models. Often, they are developed in combination with other
complementary material models, or analysis approaches, and should not be separated. As a
case in point, consider the observed and predicted behaviour of Panel PV19 tested by Vecchio
and Collins (1986); this was, in fact, Panel C from the 1981 Delft Competition, represented in
Figure 1.19. An important mechanism influencing the shear strength and deformation
response of this element was the softening of the concrete in compression due to transverse
cracking, with the panel eventually sustaining a concrete shear failure. Shown in Figure 1.21
are the predicted responses obtained using the compression softening model of Vecchio and
Collins (1986) implemented in a rotating crack formulation, and that of Okamura and
Maekawa (1991) implemented in a fixed crack formulation (i.e., each correctly matched with
the crack model for which it was first developed). Both provide equally good simulations of
response. The Vecchio-Collins formulation slightly over-estimates strength and slightly
under-estimates ductility. Conversely, the Maekawa formulation slightly under-estimates the
strength and slightly over-estimates the ductility. Either one, however, is certainly well within
the margins of accuracy we can hope to achieve with NLFEA. But consider the consequences
if one implements the Vecchio-Collins model into a fixed crack formulation, or if one uses the
Okamura-Maekawa model in a rotating crack formulation. In both cases, the results are much
less satisfactory; strength, ductility and failure mode are subject to significant miscalculation.
Experience required
Use of NLFEA for modelling and analysis of reinforced concrete structures requires a certain
amount of experience and expertise. Unlike, say, the use of plane section analysis techniques
to calculate the flexural strength of a beam cross section, the application is rarely straightforward. Decisions made with respect to modelling of the structure and selection of material
behaviour models will have significant impact on the results obtained. Again, unlike sectional
analysis techniques, two analysts may obtain widely diverging results when modelling the
same structure using the same analytical model and the same software. Decisions made
regarding mesh layout, type of element used, representation of reinforcement details, support

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
23 !

Figure 1.21: Influence of compression softening model on computed load-deformation response of Panel PV19.

conditions, method of loading, convergence criteria, and selection of material behaviour


model will produce a divergence of results. Here lies the explanation for the significant
difference in accuracy of results obtained in blind competitions as opposed to those obtained
when the desired results are known in advance. Add to this the increased likelihood of errors
in input due to the relative complexity of NLFEA.
As an illustration, consider the two simple-supported beams shown in Figure 1.22, tested by
Podgorniak-Stanik (1998); one with shear reinforcement and one without. Ten analysts, all
using the same software program and all previously experienced in its use, where asked to
independently provide predictions of the expected load capacity (Pu) and corresponding
midspan deflection (!u) for these shear-critical beams. Also requested were: the theoretical
section moment capacity (Mu) determined using hand calculations based on the common
rectangular stress block approach; the sectional shear capacity (Vu1) determined using the
Simple Method of the Canadian code specifications, which is essentially the standard 45degree truss model; and the sectional shear capacity (Vu2) determined using the General
Method of the Canadian code specifications, ostensively a more accurate calculation
involving the consideration of compatibility conditions, inclination of the stress field, and
reduction in concrete strength due to transverse cracking. The results are summarized in
Table 1.2.
The calculations of moment capacity, using simple hand methods, were consistent amongst
analysts for both beams, showing a coefficient of variation (COV) of less than 3.5%.
Calculations of the shear capacity using code specifications produced larger variations. It is
interesting to note that the calculation of Vu1 (Simple Method) wasn't any less scattered than

24

!
1 Introduction !

Figuree 1.22: Detailss of Podgornia


ak-Stanik test beams.
Table 1.22: Results of analyses of Pod
dgorniak-Stannik beams.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
25 !

that of Vu2 (General Method), due to some ambiguity present in the current code formulations
making interpretation of the code provisions subjective in certain situations. Also note that the
results were significantly more scattered for the beam without shear reinforcement (Beam
No. 2), as one might expect. Finally, let us examine the results obtained from the finite
element analyses. The estimates of load capacity (Pu) were relatively close to the codecalculated shear capacities, but showed significantly more scatter with COVs of about 17%
for the beam containing shear reinforcement, and 28% for the beam containing no web steel.
The estimates of deflection at ultimate load (!u) were widely divergent, in part because some
analysts were predicting a brittle shear failure and others a ductile flexural failure. It is worth
repeating that the same finite element program was used by the analysts, all experienced in its
use. Differences arose primarily from the selection of material behaviour models available
within the program library, and from the modelling of details such as reinforcement (smeared
or discrete) and support/loading conditions. Incidentally, the experimentally observed peak
load and corresponding deflection for Beam No. 1 were 672 kN and 20.7 mm; for Beam
No. 2, 370 kN and 5.9 mm.
Too much information
NLFEA investigations invariably produce large quantities of data, typically spawning output
files of several megabytes in size or larger for each load stage. Information may be provided
on: stresses and strains at each integration point of each element, both with respect to local
and principal axes, both for the element and for the concrete component; nodal displacements;
sectional forces per unit width at each integration point; reactions; reinforcement stresses and
strains; stiffness matrix coefficients, and more. Considering that typical problems can involve
tens of thousands of degrees of freedom, the total amount of data quickly escalates to the
point where the use of post-processors is virtually essential. Even then, the analyst must have
an awareness of what to look for and how to interpret it. Despite the use of sophisticated postprocessors and graphics capabilities, there remains the possibility for misinterpretation.
Incomplete knowledge
We must accept that we still do not understand well, let alone have accurate models for, all
aspects of reinforced concrete behaviour (see Section 1.6 Challenges Remaining).
Applications of NLFEA should be done with a healthy degree of caution and scepticism.
Wherever possible, analysis software and models should be validated or calibrated against
benchmark tests involving specimens of similar construction and loading details, dependent
on mechanisms anticipated to be significant in the analysis problem at hand (as far as this can
be done). Wherever possible, results should be supported by analyses based on different
models or approaches.
Research philosophy
Lastly, it may be said that the research community, and associated technical committees, may
have failed to meet the needs of the profession in some respects. Many working in the area
have directed their efforts to developing sophisticated models and methods of analysis, in
many cases basing their work on esoteric models or rigorous application of classical
mechanics approaches not directly suited to reinforced concrete. Consequently, advancements
have been made in developing the NLFEA concepts and methodologies, but often with
reinforced concrete being merely the application. Unfortunately, reinforced concrete is a
complex and stubborn material that sometimes refuses to act according to accepted rules of

26

!
1 Introduction !

mechanics. Researchers might do well to refocus some efforts towards better understanding
and modelling reinforced concrete behaviour, with finite element analyses being merely the
tool. Certainly, however, there is room and need to advance on both fronts.

1.5

Challenges remaining

To allay the notion that we now fully understand the behaviour of reinforced concrete, and
can accurately model it in all situations, consider two series of beams tested by Angelakos
et.al. (2001).
The first series involved five beams; all were 6000 mm in length, 1000 mm in depth, 300 mm
wide, reinforced with approximately 1.0 percent longitudinal reinforcement, containing no
transverse reinforcement, and subjected to a monotonically increasing load applied at the
midspan (Beams DB120, DB130, DB140, DB165 and DB180; see Figure 1.23 for beam
details). The only variable was the compressive strength of the concrete, ranging from 20
MPa to 80 MPa. These beams failed in a shear-critical manner upon the formation of the first
web shear crack. Current design code formulations would say that the shear strength of these
beams is directly proportional to a concrete contribution related to the tensile strength of the
concrete, which in turn is normally related back to the compressive strength (typically, the
tensile strength is taken as proportional to the square root of the compressive strength).
Hence, the 80 MPa beam would be expected to have a shear strength of close to double that of
the 20 MPa beam. Finite element analyses could also be expected to produce similar trends in
predicted strength, since the concrete tensile strength is the over-riding parameter in most
analyses of such cases. Shown in Figure 1.24 are the load-deformation responses measured
In the experiments. Note that there is little difference in the strengths and pre-peak deflection
responses observed; certainly nothing approaching the doubling of strength anticipated. In
fact, the 80 MPa beam exhibits a shear strength lower than the 20 MPa beam. At work are
mechanisms related to smoothness of the fracture plane, aggregate interlock mechanisms, and
crack slip mechanisms.
A second series of beams from the same test program involved four beams similar in
dimensions and loading to the first (beams DB120M, DB140M, DB164M and DB 180M; see
Figure 1.23). The principal difference was that these beams contained the near-minimum
amount of shear reinforcement (0.08 %). Again, the compressive strength of the concrete
ranged from 20 MPa to 80 MPa. Shown in Figure 1.25 are the load-deformation responses
recorded. Note that here, a small amount of shear reinforcement had a substantial influence on
the strength and failure mechanisms observed. Although the higher strength concrete beams
did exhibit a higher shear strength, there was still a good deal of perplexing behaviour
observed. See Angelakos et al. (2001) for a more complete description of the test program,
and a more thorough discussion of results and significance.
These two series of test results provide a stringent test of any NLFEA model. Would-be
analysts are encouraged to formulate all structural models, and select all constitutive models
and analysis parameters in advance of any preliminary computations, and to conduct the
analyses in a group and only once (i.e., forego any calibration or fine-tuning work).

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
27 !

Figgure 1.23: Deetails of Angellakos test beam


ms.

Figure 1.24: Observedd load-deform


mation responnse of Angela
akos Series I beams (beam
ms containing
g no shear
reinforceement).

28

!
1 Introduction !

Figure 1.25: Observedd load-deform


mation response of Angellakos Series II beams (beeams containing shear
reinforceement).

1.6

Objectivees

In addittion to ongo
oing work in
n the develo
opment of im
mproved co
onstitutive m
models and analysis
procedu
ures, researcch effort iss also requiired to mak
ke NLFEA procedures more ameenable to
practicaal applicatio
on and to rem
move somee of the conccerns regard
ding accuraacy and conssistency.
Specificc short-term
m objectives should incllude the following:

"

Providee design en
ngineers and
d non-experrts guidancee in the appplication of NLFEA
procedu
ures to com
mmon and prractical prob
blems.

"

Establish databank
ks and ben
nchmark prroblems, fo
or various sstructure ty
ypes and
loading
g conditionss, to facilitaate the valid
dating or callibrating off material beehaviour
modelss and analyssis procedurres.

"

Recogn
nize that div
versity in th
he analysis procedures
p
a
available
is itself valuaable, and
encouraaged it. At the same tiime, work towards
t
thee harmonizaation of con
nstitutive
modelss and anaalysis apprroaches must
m
proceeed where appropriatte. The
identifiication and
d description
n of relevaant behavio
our mechannisms, and suitable
modelss for their reepresentatio
on, should be accentuateed.

"

Providee guidance on modelliing issues relating


r
to assessmentt, rehabilitation and
forensic engineering; areas where
w
NLFE
EA procedu
ures are findding signifiicant use
today.

"

Work towards
t
dev
veloping acccurate, con
nsistent and easy-to-usee automated
d design
softwarre, with NL
LFEA being
g one of thee viable means of perfo
forming bacckground
calculaations.

Task Grroup 4.4 (Computer-Baased Modellling and Deesign), of fib


b Commission 4 (Mod
delling of
Structurral Behavio
our and Deesign) was established to work towards theese objectiv
ves. This
report constitutes
c
a initial efffort towardss achieving them.
an

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
29 !

1.7

Scope of report

This report provides an overview of concepts and techniques relating to computer-based


modelling of structural concrete. It attempts to provide a diverse and balanced portrayal of the
current technical knowledge, recognizing that there are often competing and conflicting
viewpoints. The report is written primarily for the benefit of the practicing engineer, rather
than as a state-of-the-art for researchers, concentrating more on practical application and less
on subtleties in constitutive modelling.
Chapter 2 discusses design procedures based on linear elastic modelling, the level at which
most code-based design specifications are formulated. Chapter 3 examines basic material
properties and behaviour models for concrete and reinforcement. Nonlinear modelling
concepts are explored in their application to plane frames and to two- and three-dimensional
structures in Chapters 4 and 5, respectively. Chapter 6 provides a more detailed and
comprehensive discussion of nonlinear constitutive models and behaviour mechanisms, and
their application to modelling. The use of benchmark tests and calibration procedures, a
necessary step in the use of any new modelling tool, is discussed in Chapter 7. Strut-and-tie
modelling concepts are reviewed in Chapter 8, special models for design of planar elements in
Chapter 9 and conclusions are discussed in Chapter 10.

1.8

References

ACI Committee 318. (1983), Building Code Requirements for Reinforced Concrete,
American Concrete Institute, Detroit, 111 pp.
Angelakos, D., Bentz, E. C., and Collins, M. P. (2001), The Effect of Concrete Strength and
Minimum Stirrups on the Shear Strength of Large Members, ACI Structural Journal, (in
press).
ASCE (1982), Finite Element Analysis of Reinforced Concrete, State-of-the-Art Report,
ISBN 0-87262-307-6, 545 pp.
Akiyama, H., Goto, S. and Nakazawa, T. (1996), Shear strength of Large Reinforced
Concrete Circular Slabs Under Uniformly Distributed Load, Proc. of JCI, Vol.18, No.2,
pp.1097-1102.
Bentz, E.C. (2000), Presentation at ACI Annual Convention, Toronto, October.
Braam C.R. (1990), Control of Crack Width in Deep Reinforced Beams, Heron, Vol. 4,
No. 35.
Bresler, B., and Scordelis, A.C. (1963), Shear Strength of Reinforced Concrete Beams,
American Concrete Institute J., Vol. 60, No. 1, pp. 51-74.
Cervenka, J. and Cervenka, V. (1999), Three Dimensional Combined Fracture-Plastic
Material Model for Concrete, Proceedings of the 5th U.S. National Congress on
Computational Mechanics, Boulder, Colorado, USA, August.
Cervenka, V. (2002), Computer Simulation of Failure of Concrete Structures for Practice,
First fib Congress in Osaka, October 13-19, in Japan, keynote lecture in Session 13.

30

!
1 Introduction !

Collins, M.P., Vecchio, F.J., and Mehlhorn, G. (1985), An International Competition to


Predict the Response of Reinforced Concrete Panels, Canadian Journal of Civil Engineering,
12, No.3, pp. 624-644.
Collins, M.P., Vecchio, F.J., Selby, R.G., and Gupta, P.R., (1997). The Failure of an
Offshore Platform, Concrete International, Vol. 19, No. 8, pp 28-35.
Hartl, G. (1977), Die Arbeitslinie Eingebetete Stbe bei Erst- und Kurzzeitbelastung,
Dissretation, Universitt Innsbruck.
Kawabara, Y. (2002), Shear Capacity of Rigid Base-slab Connection for LNG Storage
Tanks, JSCE annual review.
Maekawa, K., Pimanmas, A. and Okamura, H. (2003), Nonlinear Mechanics of Reinforced
Concrete:, Spon Press.
Nakano, M. (2002), In-ground LNG Storage Tanks Technological Trends and Latest
Technological Developments, Concrete Library of JSCE, No.39, June.
Nuclear Power Engineering Corporation (NUPEC) (1996), Comparison Report, Seismic
Shear Wall ISP, NUPECs Seismic Ultimate Dynamic Response Test, Report No. NUSSWISP-D014, 407 pp.
Okamura, H., and Maekawa, K. (1991), Nonlinear Analysis and Constitutive Models of
Reinforced Concrete, University of Toyko, ISBN 7655-1506-0, pp 182.
Podgorniak-Stanik, B. (1998), The Influence of Concrete Strength, Distribution of
Longitudinal Reinforcement, Amount of Transverse Reinforcement and Member Size on
Shear Strength of Reinforced Concrete Members,. M.A.Sc. Thesis, Dept. of Civil
Engineering, University of Toronto, pp 711.
van Mier J.G.M (1986), Multiaxial Strain-softening of Concrete, Part I: Fracture, Materials
and Structures, RILEM, Vol. 19, No.111.
Vecchio, F.J., and Collins, M.P. (1986), The Modified Compression Field Theory for
Reinforced Concrete Elements Subjected to Shear, Journal of the American Concrete
Institute, 83, No.2, pp. 219-231.
Vecchio, F.J., and Collins, M.P. (1990). Investigating the Collapse of a Warehouse,
Concrete International, Vol. 12, No.3, pp 72-78.
Vecchio, F.J. (2000), Disturbed Stress Field Model for Reinforced Concrete: Formulation,
ASCE Journal of Structural Engineering, 126, No. 9, pp. 1070-1077.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
31 !

[Blank Page]

!
!

Design using linear stress analysis

2.1

Introduction

Elastic based stress analysis using finite elements provides engineers with a valuable tool in
their design armory. Today, most (if not all) structural engineering companies have access to
a finite element package although of varying degrees of sophistication. In fact a number of
public domain and shareware programs are available via the web. Dimensioning of structures
based on linear analyses of frames, as one example, is commonplace. Less common is
detailing of concrete plates, shells and membrane type structures. Yet it is in some of these
structures where designers can benefit most by undertaking stress analyses, remembering
always that concrete has limited ductility. Some of the advantages of detailing based on linear
FE modelling include:
"

Linear FE modelling is well established and relatively easy to apply.

"

Multiple load cases can be accommodated quickly and with a minimal change in the
input data.

"

The greatest quantity of reinforcement is placed in the high-tension regions. That is in


regions corresponding to the initial crack locations thus helping to control crack
propagation.

The main drawbacks in using the method are:


"

No information is attained as to the collapse load of the structure (provided that


ductility demands are met, elastic analyses give a lower bound to the strength limit
design load)

"

Detailing guidelines need to be established and followed to ensure ductility and


serviceability demands are met.

"

No information is provided on inelastic phenomena such as crack widths, crack


spacings or deflections.

The above aside, however, it is not intuitively obvious how to interpret output from a finite
element analysis, in particular a linear analysis and how to detail the reinforcement. For
example, how do we detail the results of a shell element with 8 stress resultants into 4 layers
of longitudinal reinforcement plus transverse reinforcement? It is the objective of this chapter
to address this issue and to give guidance to the practicing engineer on some of the strengths
and on the limitations of using linear finite element modelling for reinforcement detailing.
Models for load-deformation analyses are treated later in this report with the focus of this
section on classical yield conditions for membrane, slab and shell elements and structures
derived from plasticity theory. This allows for a straightforward reinforcement design based
on the applied loads. Furthermore, a consistent method to account for transverse shear forces
in membranes and slabs (Marti, 1990) will be outlined.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
33 !

2.2

Membrane structures

2.2.1

Notation

As

area of reinforcing steel

Es

elastic modulus of steel

f cd

design compression strength of concrete

f c#

characteristic cylinder compression strength of concrete

fy

yield stress of reinforcing steel

f yd , f ydt

design yield stress of reinforcing steel ' f y & s

f ycd .x , f ycd . y

design yield stress of compression reinforcing steel placed parallel to the

global X and Y directions, respectively.

f ytd .x , f ytd . y

design yield stress of tension reinforcing steel placed parallel to the global

t
(

X and Y directions, respectively.


thickness of membrane
disturbance factor

)1

tension strain in the major principal direction

&c, &s

partial safety factors for concrete and reinforcing steel (taken as & c ' 1.5 ,
& s ' 1.15 )

* a , *b

reinforcement ratios in the directions of an arbitrary set of orthogonal axes

*x , * y

reinforcement ratios in the global X and Y directions, respectively.

+o

normal stress at the centre of Mohrs circle of concrete stresses

+s

stress in reinforcing steel

+ sa , + sb

stress in the reinforcing steel placed in the directions of a and b,


respectively

+ sx , + sy

stress in reinforcing steel placed parallel to the global X and Y directions,


respectively

+ xc , + yc

stress in the concrete in the global X and Y directions, respectively

+ x , + y , , xy

normal stresses in the X and Y directions and the shear stress, respectively

+1, +3

major and minor principal stress resulting from the applied loads,
respectively
major and minor principal stress in the concrete, respectively

+ 1c , + 3c

-F

34

angle between the global X axis and the major principal stress in the
concrete (measured in the anti-clockwise direction).
angle between the global X axis and the major principal stress due to the
applied loads (measured in the anti-clockwise direction).

!
2 Design using linear stress analysis !

2.2.2

General

In the output of linear elastic FE modelling of membranes the analyst is generally given
stresses and strains in a Cartesian coordinate system, which, for the analysis that follows are
chosen as collinear with the envisaged reinforcement directions. Whilst principal stresses are
of interest as they give the elastic load path, it is usually preferable to detail the reinforcement
along axes orthogonal to the axes of the structure or structural element being designed. For
example, Figure 2.1a shows normal and shear stresses on a 200 mm thick plane stress
membrane element, stress output that is typical for a linear FE analysis. From the normal
stresses it is not directly evident what quantity of steel reinforcement is needed to carry
tension stress or in what direction the steel should be placed. Plotting Mohrs circle the
principal stresses and their directions are calculated and given in Figures 2.1b and Figure 2.1c.
Assuming that the strength of the concrete is sufficient to carry the full compression stress but
carries no tension, steel reinforcement is needed such that *1 f y ' 8.5 MPa and is placed at

- ' 70.7 0 degrees to the global X axis. This requirement is satisfied with f ytd ' 435 MPa
and with 2-layers of 20-mm diameter bars at 160-mm centres.
,

5 MPa

8.5

Y
10 MPa

(-20,10)

+n

20 MPa

23 . 5

o
.-'/0/12

M Pa

MPa

8.5 MPa
-23.5 MPa

(a)

3413 o

(5,10)

(b)

Figure 2.1: a) Example output from FE modelling of a membrane;

(c)
b) and c) calculation of principal stresses.

Whilst with this solution equilibrium is satisfied, it is not usually convenient to place
reinforcement aligned with the principal directions. Rather it is normally preferred to place
the steel orthogonal to the global axes. In this case the concrete and reinforcement stress
resultants must sum to the global stresses on the membrane, as depicted in Figure 2.2. In the
compression field approaches (such as for example the diagonal compression field theory of
Mitchell and Collins, 1974) the concrete is taken to carry no tension, that is + 1c ' 0 .
Figure 2.3 shows the relationship between the applied stresses, the concrete stresses and the
reinforcement. Using this approach it is seen that there lies an infinite number of solutions to
the stress field. Some of the more interesting solutions are discussed in brief, below.
2.2.3

Reinforcement in one direction

Looking closely at the example presented in Figure 2.1, the designer may choose a stress
regime such that * x+ s ' 0 , that is to provide reinforcement solely in the global Y direction.
In this case the Mohrs circles are as shown in Figure 2.4. From the resulting geometry it can
be shown by equating the radii of the Mohrs circle of concrete stresses calculated from two
points on the circle (see Figure 2.4) that

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
35 !

+y

*55+
y sy

+3

, xy

+x

*55+
x sx

Figure 2.2: Concrete and steel reinforcement stress components.

*55+
x sx
,
$+556,555%
x xy
.-

+3c

.-F

+1c
555'4

+n

$+5567,555%
y
xy

concrete
stresses

applied
stresses

*55+
y sy
Figure 2.3: Compression field approach.

+y
(+556,555)
x xy

,xy
+x
,xy

+3c

+o

+1c

+n
(+5567,555)
y
xy

*55+
y sy

Figure 2.4: Mohrs circles for reinforcing steel placed solely in the Y-direction

36

!
2 Design using linear stress analysis !

+o '

2
+ x2 8 , xy

2+ x

(2.1)

where + o is the normal stress at the centre of the Mohrs circle of concrete stresses and + x
and , xy are the X-normal and shear stresses, respectively (shown in Figure 2.2). Given that
+ 1c ' 0 , the principal compression stress in the concrete may be obtained from + 3c ' 2+ o .
The geometry of the circles also shows that the stress in the Y-direction reinforcement
required for equilibrium is

* y+ sy ' + x 8 + y 7 2+ o

(2.2)

Further examination of Figure 2.4 shows that Eqs 2.1 and 2.2 only give a solution provided
that the inequality + x 9 0 holds true.
Similarly it can be shown that for the case where reinforcement is placed only in the global X
direction (that is * y+ s ' 0 ), then provided that + y 9 0

+o '

2
+ 2y 8 , xy
2+ y

* x+ sx ' + x 8 + y 7 2+ o

(2.3)
(2.4)

For the example presented in Figure 2.1, if it is desired to reinforce the element solely in the
Y-direction then the resulting reinforcement requirement (by Eqs. 2.1 and 2.2) is
* y+ sy ' 10 MPa . This requirement is satisfied with f ytd ' 435 MPa and with 2-layers of
20-mm diameter bars at 130-mm centres. The minor principal stress + 3c ' 2+ o ' 725 MPa .
Thus, in the example presented, the solution of providing reinforcing steel in the Y-direction
only is less efficient than when reinforcing in the major principal stress direction. This is to be
expected but likely to be more practical.
2.2.4

Isotropically reinforced panels

One of the more interesting solutions is for the case of an isotropically reinforced panel. The
Mohrs circle for this case is shown in Figure 2.5. It is seen that the circle representing the
concrete stresses is a translation of the global stresses and that * x+ sx ' * y+ sy ' + 1 . The area
of reinforcement in each of the global X and Y directions is equal to that for the case where
the steel is placed in the direction corresponding to the major principal stress. Equilibrium can
be maintained by replacing the force in the major principal direction $F1 % with any two
concurrent forces $Fa and Fb % . Letting F1 act over a unit length and taking the forces Fa and
Fb as orthogonal, as shown in Figure 2.6, then
Fa ' F1 sin :

Fb ' F1 cos :

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

(2.5)

!
37 !

*55+555'5*55+
x sx
z sy
,
$+556,555%
x xy

-'-F
.-

+3c

+1c

.-F +
1

+3

+n

$+5567,555%
y
xy
*55+555'5*55+
y sy
x sx
Figure 2.5: Mohrs circle for isotropically reinforced panels.

F1

Fb

1. c o

s:

+b
1. sin

+1

+a
Fa

Figure 2.6: Equivalent concurrent orthogonal forces and resulting stresses.

+a '

F1 sin :
' +1
t. sin :

+b '

F1 cos :
' +1
t. cos :

(2.6)

where the angle : is as shown in Figure 2.6 and t is the thickness of the element. That is
+ a ' + b ' + 1 . As the angle : is arbitrary, all orthogonal reinforcement sets are valid
solutions with the steel stress in each direction equal the major principal applied stress $+ 1 % .
Looking again at the example of Figure 2.1, assuming that the strength of the concrete is
sufficient to carry the compression but carries no tension then steel reinforcement may be
placed in any two orthogonal directions (such as, for example, in the global X and Y axis
directions) such that * a+ sa ' *b+ sb ' 8.5 MPa . This requirement is satisfied with

f ydt ' 435 MPa and with 2-layers of 20-mm diameter bars at 160-mm centres placed in each
of the X and Y directions.
The example of the isotropically reinforced plate may appear to contradict the solution
obtained when the steel is placed in the major principal stress direction without an equivalent
quantity of orthogonal reinforcement. Examination of Figure 2.6 shows that placing the steel

38

!
2 Design using linear stress analysis !

in the major principal direction corresponds to the particular solution : ' 0o giving Fa ' 0 ,
sin : ' 0 and leads to a breakdown of Eq. 2.6. This is the only condition for which this
occurs.
2.2.5

The general solution

Using the theory of plasticity, Nielsen (1964, 1971) derived a set of general solutions for
isotropically and orthotropically reinforced plates. Clarke (1976), Clyde (1977),
Mller (1978), Marti (1979, 1980), Morley (1979) and many others have also made
significant contributions to the field. The cases examined above are particular solutions of a
general failure criterion. For the most general case the strength of a membrane is limited by
the yield surface expressed by the following criteria (Kaufmann and Marti, 1998):

%$

2
Y1 ' , xy
7 * x f ytd .x 7 + x * y f ytd . y 7 + y ' 0

(2.7a)

%$

(2.7b)

2
Y3 ' , xy
7 f cd 7 * y f ytd . y 8 + y * y f ytd . y 7 + y ' 0

%$

(2.7c)

2
Y4 ' , xy
7 f cd 4 ' 0

(2.7d)

2
Y2 ' , xy
7 f cd 7 * x f ytd .x 8 + x * x f ytd .x 7 + x ' 0

%$

%$

%$

2
Y5 ' , xy
8 f cd 8 * x f ycd .x 8 + x * x f ycd .x 8 + x ' 0

(2.7e)

2
Y6 ' , xy
8 f cd 8 * y f ycd . y 8 + y * y f ycd . y 8 + y ' 0

(2.7f)

2
Y7 ' , xy
7 f cd 8 * x f ycd .x 8 + x f cd 8 * y f ycd . y 8 + y ' 0

(2.7g)

where f cd is the design value of the concrete compression strength f ytd .x and f ytd . y are the
design strengths of the steel reinforcement in tension in the global X and Y directions,
respectively, and f ycd .x and f ycd . y are the design strengths of the steel reinforcement in
compression in the global X and Y directions, respectively. The yield surface given by Eqs
2.7a to 2.7g is plotted in the three-dimensional stress space of + x , + y ,, xy in Figure 2.7 for a

panel with * x ' 0.01 , * y ' 0.005 and f yd ' 500 MPa . In Eqs. 2.7a-g, Y1 corresponds to
under reinforced elements where the steel in both the X and Y directions yields before failure
of the concrete, Y2 and Y3 correspond to web crushing failure and Y4 represents the no-yield
condition of crushing in pure shear. The criteria Y5 , Y6 and Y7 are for plates in biaxial
compression.
Whilst the application of Eq. 2.7 is relatively straightforward it is not as easy to picture the
failure surface in the 3-dimensional stress space. As the yield criterion is based on the
principal stresses we again revert to Mohrs circles of stress for an alternative graphical
solution. The case of an isotropically reinforced plate is used to demonstrate the relationship
between Mohrs circle and the general failure criterion given by Eq. 2.7. In Figure 2.8a-d

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
39 !

10

10

, xy (MPa) 5

0
-25
-20
-15
-10
-5 + x (MPa)

0
0

-5
-10

+ y (MPa)

, xy (MPa)

-15
0

-20
-25

(a)

+ y (MPa)
+ x (MPa)

-30

Regime 1

10

5
Regime 4

Regime 7

-25

(b)

Figure 2.7:

40

Yield surface for a reinforced concrete membrane with * x =0.01, * y = 0.005 and fyd = 500 MPa;
a) 3D View; b) plan view identifying yield regimes.

!
2 Design using linear stress analysis !

;,;

;,;

regime 1

+3

+n

*5f yd

+n

collapsing
Mohr's circle

;,;

*5f yd

fcd

(a)

(b)

regime 4

regimes

5&6

5&6

fcd /2

+1

+3

fcd

regimes

regimes
2&3

sliding
Mohr's
circle

fcd /2

expanding
Mohr's
circle

fcd /2

regime 4

regimes 2 & 3

regime 7

;,;

+3

fcd /2

sliding
Mohr's
circle

+1

*5f yd

+n

fcd

fcd

(c)

+1

*5f yd

+n

(d)

.5*5f yd
R = fcd /2

Region I

+n

Region II

fcd + *5fyd

*5f yd

(e)

Figure 2.8:

Isotropically reinforced plate: a-d) Mohrs circles defining the failure surface; e) admissible stress
envelope.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
41 !

Mohrs circles are plotted for an isotropically reinforced plate. Starting from the case of equibiaxial tension (that is +1 ' + 3 ' *f yd ) the circle expands with reducing + 3 until the point
where the radius reaches the limiting shear strength of f cd 2 (Figure 2.8a) with the
corresponding minor principal stress being +3 ' 7 fcd 7 * f yd . This phase represents the

regime 1 shown in Figure 2.7. In the second phase the circle shifts towards the negative,
whilst maintaining a constant radius, until + 1 ' 0 and + 3 ' 7 f cd (Figure 2.8b). This
represents regimes 2, 3 and 4 in Figure 2.7. In the third phase, corresponding to regimes 4, 5
and 6 in Figure 2.7, the circle slides further to the negative and the reinforcing steel moves
into compression (Figure 2.8c). Finally, the circle collapses with the minor principal stress
remaining constant until the equi-biaxial compression point is reached, that is the point
+1 ' +3 ' 7 fcd 8 * f yd , as shown in Figure 2.8d.

From the Mohrs circles shown in Figures 2.8a-d the admissible stress field is plotted
(Figure 2.8e). For any solution to be admissible two criteria must be met. Firstly, the Mohrs
circle of the applied stresses must lie wholly within the circular region I, where region I is
centered on the normal stress axis and has a diameter of f cd . Secondly, region I lies wholly
within region II (refer Figure 2.8e). If these two criteria are met then the membrane is safe.
Note that in practice it is sufficient to show that any two diagonally opposed points on the
Mohrs circle of stress meet these criteria to demonstrate admissibility of the solution (for
example the stress pairs ( + x ,, xy ) and ( + y ,7, xy )).
For membranes subject to tension and/or shear the zone of greatest interest is that of regime 1
in Figures 2.7 and 2.8a and the design equation given by Eq. 2.7a. Plotted in Figure 2.9 is the
Mohrs circle of stress for the applied stresses and the stresses in the concrete for a membrane
in bi-axial tension. From the upper shaded triangular region it is seen that the concrete stress
in the X direction is + xc ' + x 7 *x f ytd.x and therefore steel reinforcement is required such
that

* x f ytd . x < + x 8 , xy tan -

(2.8)

where 7 90 o 9 - 9 90 o and - = 0 .

Similarly, from the lower shaded triangular region + yc ' + y 7 * y f ytd. y and therefore to
provide sufficient capacity the Y direction reinforcement is required such that

* y f ytd . y < + y 8 , xy cot -

(2.9)

In Figure 2.9b the feasible domain of solutions to Eq. 2.7a is plotted. In this form the
relationship between Eq. 2.7a and Eqs. 2.8 and 2.9 is clear.

42

!
2 Design using linear stress analysis !

*555f
x ytd.x
,
$+5556,5555%
xc xy
+3c

$+556,555%
x xy
-

.-F

+1c

+n

concrete
stresses

$+55567,5555%
yc
xy

$+5567,555%
y
xy

applied
stresses

*555f
y ytd.y

*y fyd
+y585;,xy|5cot|-|
feasible domain

+x585;,xy| tan|-|
+y
0
0
Figure 2.9:

+x

*x fyd

a) Mohrs circle of stress for an under-reinforced concrete plate; b) solution domain for underreinforced element.

Equations 2.8 and 2.9 are identical to those derived by Nielsen (1964) using the theory of
plasticity 1. For Eqs. 2.8 and 2.9 to govern the behaviour (that is for the panel to be under
reinforced) the strength of the concrete must be such that f cd < 7+ 3c , that is the concrete
must not crush. Applying the first invariant of stresses the compression failure criterion can be
written as

+3c 8 +1c ' $+ x 7 *x f ytd.x % 8 $+ y 7 * y f ytd. y %

(2.10)

Taking + 1c ' 0 (as per the compression field theory) and with f cd < 7+ 3c crushing failure is
not critical provided that

fcd < *x f ytd.x 8 * y f ytd. y 7+ x 7+ y

(2.11)

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
1

Orthotropic membrane theory can be traced to Nielsen (1920) and Eqs. 2.8 and 2.9 to Leitz (1923) for
tan- ' 1 [according to Nielsen (1964, 1971)].

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
43 !

For strength limit design of reinforced concrete membranes the strength criteria are fully
satisfied by Eqs. 2.8, 2.9 and 2.11 and are identical to the equations presented by Kaufmann
(1999) based on the yield surface defined by Eq 2.7.
Taking once more the example presented in Figure 2.1 and with - ' 65o , Eq. 2.8 gives
*x f ytd.x ' 1.45 MPa and * y f ytd. y ' 9.67 MPa. This requirement is satisfied with

f ytd ' 435MPa and with 2-layers of 12-mm diameter bars at 330-mm centres placed in the
X-direction and 2-layers of 20-mm diameter bars at 140-mm centres placed in the
Y-direction. The required design concrete strength is given by Eq. 2.11 and is
f cd < 26.1 MPa .
2.2.6

Some comments on the angle -.

In developing solutions using linear FE modelling it is important that the designer respect
the limitations of the concrete material. In a panel subject to a constant ratio of normal and
shear stresses (with at least one tensile principal stress) before cracking the stress field in the
concrete remains relatively elastic and the stresses in the steel reinforcement are negligible.
After cracking the tension stresses in the concrete rapidly reduce while those in the
reinforcing steel increase. Assuming that the concrete does not fail in compression then the
crack directions will remain relatively stable until yield of the steel in one-direction. After
yield in one direction the forces in the structural element are continuously redistributed to
balance the applied tractions until yield in all directions has occurred. Concrete panels,
however, have a limit on the amount of redistribution that is capable of being achieved. This
is referred to as ductility demand. As a rule concrete elements such as membranes and panels
should not be pushed far beyond that which is natural. Designers must critically examine
the load path being assumed to satisfy themselves that the element has a sufficient level of
ductility to meet the demands of the structure. To this end plots of principal stress directions
from linear-elastic modelling can give the designer valuable insight. Care and engineering
common sense needs to be applied when designing any structural element but even more so
for the case of non-flexural membranes.
For a membrane element with its principal stresses in biaxial tension, steel reinforcement is
required in at least two non-concurrent directions to carry the tension stresses with the
concrete taking the shear stresses. For the case of orthogonal reinforcement - ' 45o leads to
the lowest concrete strength requirement.
Many structural elements fall into the category of tension-compression elements. Again

- ' 45o provides the lowest concrete strength requirement and in many cases provides for a
good solution. In some cases, however, - ' 45o may not be desirable. Figure 2.10a shows an
element in tension-compression where taking - ' 45o leads to tension reinforcement in the X
direction and compression reinforcement in the Y direction. This, in some circumstances, can
lead to a significant requirement for reorientation of the cracks requiring a high ductility
demand. Figure 2.10b shows that there are limits on - if compression in the steel is not
wanted. Adding to Eq. 2.8 the condition *x f yd < 0 we get the limits

44

!
2 Design using linear stress analysis !

*55+
x sx
,
.-5'5A4!

*55+
x sx
,

tension
steel

$+556,555%
x xy

$+556,555%
x xy
+3c

+1c
555'4

+n

tension
steel

+3c

.-

+1c
555'4

+n

$+5567,555%
y
xy

$+5567,555%
y
xy

applied
stresses

applied
stresses

*55+
y sy
compression steel

(b)

(a)

Figure 2.10: a) Example showing the requirement of compression reinforcement for - = 45 ; b) no


compression steel solution by limiting -.

for+ x 9 0 :

7 + x , xy ? tan - 9 >

for+ x < 0 :

0 9 tan - 9 >

(2.12a)

Similarly applying the requirement that * y f ytd. y < 0 to Eq. 2.9 gives

for+ y 9 0 :

0 9 tan - ? 7 , xy + y

for+ y < 0 :

0 9 tan - 9 >

(2.12b)

Choosing a value of - to meet the requirements of Eq. 2.12 ensures that the stresses in the
reinforcing steel do not go into compression. It is also observed for Eq. 2.12 that the lower
limit on Eq. 2.12a for + x 9 0 and the upper limit on Eq. 2.12b for + y 9 0 correspond to the
angles for unidirectional reinforcement in the x and y directions, respectively. In Figure 2.11
the solutions to Eq. 2.7a are plotted for the stresses of the example shown in Figure 2.1. The
figure shows that for - 9 63.43o the X direction steel is in compression and for - @ 63.43o
the steel is in tension. The angle - ' 63.43o corresponds to the solution of reinforcement only
in the Y direction.
Lastly, for the case of elements subject to biaxial compression, the angle - should be matched
to that of the principal stress directions due to the applied loads. In this case reinforcing steel
may be used, if desired, to carry a proportion of the compression forces.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
45 !

40

+x = -20 MPa
+y = 5 MPa
, xy= 10 MPa

*y fyd (MPa)

30
Eqn. 2.7a

20

- = 63.43o

10

0
-20

X dirn. reo. in
compression
- < 63.43o

-10

X dirn. reo.
in tenision
- > 63.43o

10

20

*x fyd (MPa)
Figure 2.11: Solution domain for element subject to plane stress.

2.2.7

The design concrete compression strength, fcd.

It has been shown by a number of researchers (Robinson and Demorieux, 1977, Vecchio and
Collins, 1982, 1986, Miyakawa et al., 1987, Belarbi and Hsu, 1991, Pang and Hsu, 1992,
amongst others) that the disturbing effect of passing tension reinforcement through concrete
in compression weakens the concrete. In addition, as concrete strength is increased the
concrete becomes more brittle. To account for the imperfect assumption that concrete is a
rigid plastic material and that the ductility demands can be met an efficiency factor is
introduced to ensure that the concrete is not overstressed. There is considerable debate on the
efficiency factor for concrete in compression in D-regions and much has been written (see for
example the quite contradictory articles of section 4.6 Nielsen, 1999, and Foster and Malik,
2001). For a history of various efficiency factors methodologies the reader is referred to the
report on recent approaches to shear design by ASCE-ACI Committee 445 (1998).
In recent approaches to quantifying of the efficiency of concrete in uniaxial compression and
subject to transverse disturbance the overall efficiency factor is separated into two
components (MacGregor, 1997, Foster and Malik, 2001). The first factor accounts for the
variation of the in-situ strength of concrete relative to that of the standard cylinder test. The
second factor accounts for the disturbance due to the transverse stress fields and the
brittleness of unconfined concrete. In planned changes to the ACI code (Cagley, 2001, p129)
the strength of struts subject to transverse tensions in strut-and-tie models is
f cd ' 0.85 ( f c# & c

(2.13)

where the 0.85 factor accounts for the variation between the in-situ and cylinder strengths and
( is a disturbance factor. Equation 2.13 is of a generally suitable form for the design of
concrete plates subject to in-plane stresses.
Whilst there are a number of variants of the efficiency factor relationship, the model by
Collins and Mitchell (1986) has generally withstood the test of time and has been
incorporated into the Canadian code of practice (CSA84, 1984). Based on the panel tests of
Vecchio and Collins (1982), Collins and Mitchell proposed that

46

!
2 Design using linear stress analysis !

('

1
? 1.0
0.8 8 170)1

(2.14)

where ) 1 is the major principal strain normal to the direction of the compression field. The
transverse strain is required to be sufficiently large for the ductility demand to be met.
Adopting a minimum transverse strain of twice the yield strain of the steel reinforcement
gives

('

1
0.8 8 340 f y Es

(2.15)

where f y and Es are the yield strength and the elastic moduli of the steel reinforcement,
respectively. For 500 MPa grade reinforcing steel ( ' 0.6 and f cd ' 0.5 f c# .
Not all members or sub-elements of members are subject to significant transverse strains and
a reduction to the degree indicated by Eq. 2.15 is unjustified. For the case where transverse
strains are small and the concrete is essentially in uniaxial compression (transverse tensions
can be carried without cracking of the concrete) and for the case of biaxial compression then
the concrete is undisturbed by crossing tension stress fields. As a guide, if the major principal
stress due to the applied loads is such that +1 9 0.33 fc# then the disturbance factor maybe
taken as

( ' 1.0

(2.16)

Examples of members and regions of members where Eq. 2.16 applies include beams and
columns outside the shear zones and at edges of members where principal stresses and strains
are aligned with the boundary of the element.
2.2.8

Example Design of a reinforced concrete squat shear wall.

Shear walls are common elements in buildings that are subject to in-plane stresses. In this
example the squat shear wall shown in Figure 2.12 is designed to carry of factored design load

H * ' 4000 kN . The load may be applied in either direction as indicated in the figure. The
wall is 200 mm thick and is bounded by 500-mm square stiff elements on both sides and a
1200-mm by 400-mm beam at the top of the wall.
A linear-elastic finite element model of the wall was set up with the mesh shown in
Figure 2.13. The mesh consists of 30 by 8-node isoparametric elements defined by 113 nodes.
The material properties were taken as E ' 25 GPa and B ' 0.18 for all elements. The model
was analysed with integration on a 3x3 Gaussian quadrature with the stress output on a 1
gauss point quadrature (that is at the centre of the element). The resulting normal stresses for
load case 1 are shown in Figure 2.13. For load case 2, the element stresses are a reflection of
the results of load case 1 about the Y-Y axis (refer Figure 2.12).
For the reasons of ductility demand, discussed above, it is important to make an assessment of
the flow of forces in the wall and adjacent boundary elements. In the design that follows, for
the wall elements the minor principal stress direction for the concrete is taken as aligned with
the diagonal of the wall (refer Figure 2.12) giving - ' 58o . For the side face boundary
elements the principal stresses are taken as aligned with the global stress directions as shown
fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
47 !

in Figure 2.12. The resulting stresses in the X and Y direction reinforcing steel are shown in
Figure 2.14.
For detailing of the reinforcement, the wall is divided into vertical and horizontal strips, as
shown in Figure 2.15. Reviewing the steel stresses in the horizontal (x) direction it is seen that
the maximum stress in each of wall strips 1X, 2X and 3X are of similar magnitude (5.45, 5.92
and 5.54 MPa, respectively). It is deemed therefore appropriate to maintain a single steel
stress in the X direction of 5.92 MPa for the full height of the wall. The area of steel
reinforcement per unit height of wall is given by
As '

* f yd t h

(2.17)

f yt & s

where t is the wall thickness, h is a unit height, f yt is the yield strength of the steel and & s is
the partial safety factor for the steel reinforcement (taken as & s ' 1.15 ). For 500 MPa grade
reinforcing steel and *x f yd ' 5.92 MPa, Eq. 2.17 gives As ' 2720 mm2 m . This is satisfied
with 2 layers of 16-mm diameter bars at 150-mm centres.
In the vertical (Y) direction the greatest stresses occur in design strips 1Y and 4Y (noting that in
strip 4Y the greatest stress occurs for load case 2). For 500 MPa grade reinforcing steel and
* y f yd ' 4.22 MPa, Eq. 2.17 (with h replaced by the unit width, b) gives an area of reinforcing
steel of As ' 1940 mm2 m . This is satisfied with 2 layers of 16-mm diameter bars at 200-mm
centres. In wall strips 2Y and 3Y the requirement is for * y f yd ' 2.22 MPa giving

As ' 1020 mm2 m and leading to 2 layers of 12-mm diameter bars at 200-mm centres.
For the boundary strips 1Y and 2Y the concrete stress angle is taken as 60 degrees giving a
maximum stress in the longitudinal (Y-direction) steel of * y f yd ' 7.5 MPa and with

f ytd ' 435MPa this gives As ' 4310 mm2 . This is satisfied with 8C28-mm bars. The stress
in the x direction of *x f yd ' 1.81MPa is carried three legged ties at 160 mm centres. The
reinforcement details for the shear wall are shown in Figure 2.16.
The minor principal stresses in the concrete $+ 3c % are obtained from Eq. 2.11 and given in
Figure 2.14. In this case the + 1c axis is taken as coinciding with the major principal axis and
thus varies through the length of the element. In the compression boundary element the
transverse tension strains are low and the disturbance factor is taken as that given by Eq. 2.16.
The maximum compression stress is 5.66 MPa and when the effect of the axial reinforcement
in the section is considered it is seen that the compression strength of the boundaries does not
control the design.
In the web the compression stresses are reasonably uniform, indicating a degree of efficiency
in the design as the whole of the wall is utilised. The minimum minor principal stress in the
concrete wall is + 3c ' 9.0 MPa and as the transverse tensions are significant the disturbance
factor is taken as that given by Eq.2.15. The minimum specified concrete strength is obtained
from Eq. 2.13 and is f c# < 26.5 MPa . A concrete strength of f c# ' 30 MPa is adopted.

48

!
2 Design using linear stress analysis !

12000 mm x 400 mm
bouundary element

Load
d Case 2

1c

400

Load Case 1

2c

500 mm squaare
boundary eleement
2c

1c

3500

58

2c

t = 200 mm
m

60

1c
varies

Y
500

5500

500

F
Figure
2.12: Finite
F
element analysis and
d design of squuat shear walll.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
49 !

+555'5
x 7F1/E5MPa
+555'5
y 741435MPa
,5555'55
xy 41.E5MPa

+555'5
x 721A35MPa
+555'5
y 7414/5MPa
,5555'55
xy 41.A5MPa

+555'5
x 7.1425MPa
+555'5
y 7414/5MPa
,5555'55
xy 41.05MPa

+555'5
x 741DA5MPa
+555'5
y 541445MPa
,5555'
xy 5541/F5MPa

+555'5
x 7/10.5MPa
+555'5
y 5/1A/5MPa
,5555'55
xy 41E05MPa

+555'5
x 7.1325MPa
+555'5
y 7414.5MPa
,5555'55
xy 213.5MPa

+555'5
x 7.1DE5MPa
+555'5
y 741.05MPa
,5555'55
xy 21A05MPa

+555'5
x 7/13D5MPa
+555'5
y 7412E5MPa
,5555'55
xy 21205MPa

+555'5
x 7413F5MPa
+555'5
y 741345MPa
,5555'
xy 55.1./5MPa

+555'55
41445MPa
x
+555'5
y 741F.5MPa
,5555'
xy 5541225MPa

+555'
x 7414/5MPa
+555'55
21E.5MPa
y
,5555'5
xy 541045MPa

+555'5
x 741235MPa
+555'55
41AF5MPa
y
,5555'5
xy 521.F5MPa

+555'5
x 741A25MPa
+555'5
y 7412A5MPa
,5555'55
xy 014D5MPa

+555'5
x 7/14/5MPa
+555'5
y 741EF5MPa
,5555'55
xy 21345MPa

+555'5
x 741D.5MPa
+555'5
y 7/1F/5MPa
,5555'5
xy 5.13A5MPa

+555'5
x 7414.5MPa
+555'5
y 7/1EE5MPa
,5555'5
xy 5410.5MPa

+555'55
41425MPa
x
+555'5
y 5D12.5MPa
,5555'5
xy 5410F5MPa

+555'55
410.5MPa
x
+555'5
y 5/1AD5MPa
,5555'5
xy 5214.5MPa

+555'5
x 741425MPa
+555'5
y 741205MPa
,5555'55
xy 213.5MPa

+555'5
x 741DE5MPa
+555'5
y 7/1/25MPa
,5555'55
xy 21FD5MPa

+555'5
x 741F45MPa
+555'5
y 7.1./5MPa
,5555'55
xy 21/05MPa

+555'5
x 7414.5MPa
+555'5
y 721D35MPa
,5555'55
xy 41DD5MPa

+555'55
41205MPa
x
+555'55
31445MPa
y
,5555'55
xy 41ED5MPa

+555'55
41AF5MPa
x
+555'55
.10F5MPa
y
,5555'55
xy .1E/5MPa

+555'55
41/D5MPa
x
+555'5
y 741/F5MPa
,5555'55
xy 21245MPa

+555'5
x 7412/5MPa
+555'5
y 7/1.E5MPa
,5555'55
xy 21235MPa

+555'5
x 741AE5MPa
+555'5
y 7.1D45MPa
,5555'55
xy 214A5MPa

+555'5
x 741235MPa
+555'5
y 7D10A5MPa
,5555'55
xy 41A05MPa

4000 kN

Figure 2.13: Finite element mesh for squat shear wall example and stress output.

4000 kN
*55f
x yd '5/12/5MPa
*55f
y yd '5.12A5MPa

*55f
x yd '521..5MPa *55f
x yd '5213.5MPa
*55f
y yd '5.12/5MPa *55f
y yd '5.1..5MPa
+555
'7E1.E5MPa +555
'7E13F5MPa
3c
3c

*55f
x yd '521DA5MPa *55f
x yd '5.1335MPa
*55f
y yd '5/1345MPa *55f
y yd '541FE5MPa
+555
'7310.5MPa +555
'701A.5MPa
3c
3c

+3c '5741335MPa

*55f
x yd '541FE5MPa
*55f
y yd '5014D5MPa

*55f
x yd '501E05MPa *55f
x yd '5D1D05MPa
*55f
y yd '521445MPa *55f
y yd '5.1/05MPa
+555
'731.05MPa +555
'7A1445MPa
3c
3c

*55f
x yd '501A.5MPa *55f
x yd '521AD5MPa
*55f
y yd '5/10F5MPa *55f
y yd '541/05MPa
+555
'7E1.25MPa +555
'7F1./5MPa
3c
3c

+555'5
3 c 7/1A35MPa

*55f
x yd '541E25MPa
*55f
y yd '5D1DA5MPa

*55f
x yd '5D1.D5MPa *55f
x yd '5D1A.5MPa
*55f
y yd '521E25MPa *55f
y yd '5/1AA5MPa
+555
'7F13/5MPa +555
'7E1.35MPa
3c
3c

*55f
x yd '5D1.F5MPa *55f
x yd '5501025MPa
*55f
y yd '5/1/D5MPa
+555
'7E1/.5MPa +555
'7F1AA5MPa
3c
3c

+555'5
3 c 721FD5MPa

*55f
x yd '5/1E/5MPa
*55f
y yd '5310A5MPa

*55f
x yd '5D10D5MPa *55f
x yd '5D10.5MPa
*55f
y yd '501..5MPa *55f
y yd '5/1A45MPa
+555
'7F1.D5MPa +555
'731225MPa
3c
3c

*55f
x yd '5D14E5MPa *55f
x yd '5521AF5MPa
*55f
y yd '541E25MPa
+555
'7310A5MPa +555
'7F1EF5MPa
3c
3c

+555'5
3 c 7D1FF5MPa

Figure 2.14: Reinforcing steel stresses for shear wall orthogonally reinforced in XY.

50

!
2 Design using linear stress analysis !

Wall Strip 4X

Wall Strip 3X

Wall Strip 2X

Boundary Strip 2Y

Wall Strip 4Y

Wall Strip 3Y

Wall Strip 2Y

Wall Strip 1Y

Boundary Strip 1Y

Wall Strip 1X

Figure 2.15: Design strips in the X and Y directions for detailing of the reinforcement.
2750

1375
400

1375

2 x C12@200
2 x C16@150

8 C.E

2 x C16@200

3 x C12@160
stirrups

500

5500

8 C.E

3500

2 x C16@200

500

Figure 2.16: Reinforcement details for squat shear wall example.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
51 !

2.3

Slabs and shells

2.3.1

General

Similar to that for membranes, the dimensioning of plate, slab and shell elements for stresses
determined from a FE-analysis is typically based on yield conditions derived from plasticity
theory. While careful consideration of the limited ductility of concrete is important in the
dimensioning of membrane elements, there is generally much less concern with this aspect in
slabs and shells as such structures are typically under-reinforced. That is, failure is governed
by yielding of the reinforcement while the concrete does not crush. Important exceptions to
this observation are concentrated transverse forces, which may result in brittle punching
failures in slabs and shells without transverse reinforcement.
Over the past decades, several investigations have been carried out on the load-deformation
analysis of shell elements, taking into account general non-linear material behaviour (see for
example Khalifa, 1986, Kirschner and Collins 1986). The resulting approaches, which
typically use a subdivision of the shell elements in finite layers, are suitable for general loaddeformation analyses of single elements, however, they do not allow for a straightforward
dimensioning. Thus, non-linear FE modelling is less suitable for practical design purposes
where many elements must be dimensioned for different load-cases and the reinforcement
quantity and arrangement is yet to be determined. The required amount of reinforcement
could be minimized by aligning the reinforcement with the principal (tensile) stress directions.
This type of reinforcement has been applied to numerous slab and shell structures in the past,
and even today, some designers prefer to provide this so-called trajectory reinforcement,
particularly in shells. However, reinforcement layouts following the principal stress
trajectories are usually impractical, requiring a high amount of complex work on site.
Moreover, if several different load-cases must be considered, principal stress directions will
vary, making the provision of reinforcement aligned with them strictly speaking
impossible. For these reasons, orthogonal reinforcement is provided in almost all slab and
shell structures, and the following considerations will thus focus on this type of
reinforcement.
2.3.2

Stress resultants

Integrating the stresses acting along the boundaries of a shell element, Figure 2.17 (a), one
obtains the elements stress resultants shown in Figure 2.17(b).
In the general case of a plane shell element, there are eight independent stress resultants, they
are, the bending and twisting moments
h2

mx '

h2

h2

G + x z dz , m y ' G + y z dz , mxy ' m yx ' G , xy z dz

7h 2

7h 2

(in kNm/m = kN) (2.18)

7h 2

the transverse shear forces


h2

vx '

G , zx z dz , v y ' G , zy z dz

7h 2

52

h2

(in kN/m)

(2.19)

7h 2

!
2 Design using linear stress analysis !

and the membrane forces


h2

n x ' G + x dz , n y '
7h 2

h2

G + y dz , nxy ' n yx '

7h 2

(a)

h2

G , xy dz

(in kN/m)

(2.20)

7h 2

(b)
h/2
y

z h/2
dz

z
1

z
x
nx

1
+x dz
+y dz

,xy dz

,yx dz

,zx dz

ny

my
mxy

,zy dz

nxy

nyx

vx

mx

myx

vy

Figure 2.17: Plane shell element: (a) Stresses; (b) Stress resultants in the general case of combined membrane
and bending action.

The stress resultants of Eqs. 2.18 to 2.20 can be subdivided into two groups: the in-plane axial
forces nx , ny , nxy (membrane action) and the out-of-plane forces bending and shear forces

Hmx , my , mxyI Hvx , vyI (bending action). For plane elements, membrane and bending action

are independent, that is, two complete sets of equilibrium conditions involving only in-plane
or out-of-plane forces can be formulated.
Curvature of the shell middle plane will result in a coupling of membrane and bending action,
such that equilibrium conditions of a curved element will generally involve all eight stress
resultants. The definition of stress resultants is also affected by curvature; in the case of thin
curved shell elements a good approximation can be obtained by multyplying the integrands in
Eqs. 2.18 to 2.20 by a factor of 1 7 z rx (stress resultants with last subscript y) or 17 z ry
(stress resultants with last subscript x). Thus, strictly speaking, for rx = ry one obtains that

mxy = myx and nxy = nyx (although , xy ' , yx still applies). However, for thin shells
z rx J 0 and z ry J 0 and thus, Eqs. 2.18 to 2.20 can usually be applied without
modification.
In the following, we will denote elements subjected to pure membrane action,
mx ' my ' mxy ' 0 and vx ' vy ' 0 , as membranes, and elements subjected to pure bending
action, nx ' ny ' nxy ' 0 , as plates or slabs. Elements subjected to combined membrane and
bending forces are called shells. While shell elements can be part of either plane or curved
structures, it is evident by these definitions that membranes and slabs must essentially be
plane structures.
Slabs are probably the most important application of structural concrete. Thus, while this

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
53 !

chapter treats the dimensioning of shells, special attention is given to slabs, including
equilibrium and boundary conditions, stress transformation and reinforcement design.
For stresses and stress resultants, the sign convention illustrated in Figure 2.17 will be used
throughout this chapter. Accordingly, positive stresses are acting in the direction of positive
coordinates on boundaries with positive outer normal; with regard to normal stresses, this
means that tensile stresses are positive. Positive membrane and transverse shear forces are
those corresponding to positive stresses, and positive bending and twisting moments
correspond to positive stresses, according to the above definition, for positive values of
coordinate z. With regards to the double subscript indices, the first index stands for the
direction of the stress, while the second denotes the normal direction of the boundary on
which the stress or stress is acting.
2.3.3

Equilibrium, stress transformation and boundary conditions for slabs

Equilibrium conditions
Formulating equilibrium of the stress resultants acting on a slab element, Figure 2.18, one gets

vx, x 8 vy, y 8 q ' 0

(2.21a)

mx, x 8 mxy, y 7 vx ' 0

(2.21b)

my, y 8 myx, x 7 vy ' 0

(2.21c)

where subscripts after a comma denote partial derivatives with respect to the corresponding
variable. Evaluating Eqs. 2.21b-c for v x , x and vy, y and substituting into Eq. 2.21a gives the
elastic plate equation

mx, xx 8 2mxy, xy 8 my, yy 8 q ' 0

(2.22)

which consists of the sum of equilibrium conditions for pure bending of unit strips in the xand y-direction plus an additional coupling term consisting of the twisting moments.

(a) !

dx

vy dx
vx dy

dy

mx5dy
my5dx

myx dy

(my+my,y dy) dx
(mx+mx,xdx) dy

mxy dx

q dxdy

(myx+myx,xdx) dy
(vx+v x,xdx) dy

(mxy+mxy,ydy) dx

(vy+v y,y dy) dx

Figure 2.18: Equilibrium conditions for slab element.

54

!
2 Design using linear stress analysis !

Transformation of bending and twisting moments


Moment equilibrium of the slab element shown in Figure 2.19(a) yields
mn ' mx cos 2 K 8 m y sin 2 K 8 mxy sin 2K

(2.23a)

mt ' mx sin 2 K 8 m y cos 2 K 7 mxy sin 2K

(2.23b)

mtn ' my 7 mx sinK cosK 8 mxy cos2K

(a)

mxysinK

my sinK

mxcosK

mn

my cosK

mtn

n
myxsinK
mxsinK

mnt

mtn
X

mt
n

myxcosK

(b)

mxycosK

2K

K1 K

(c)

(d)

vn

vxcosK

K
vxsinK

vt
1

mn

v0

n
vx

vy v

v0 vy
K0 L
2

Q (Pol)

vx
K0 K vn

vn

vy cosK

n
1

1
N

2K1

vy sinK
K

(2.23c)

3L
2

2L

Figure 2.19: Stress transformation: (a), (b) Bending and twisting moments; (c), (d) Transverse shear forces.

Equations 2.23a-c can be interpreted as equations for the transformation of bending and
twisting moments acting on any boundary perpendicular to the direction n, where the
orientation is determined by the angle K . This may be represented graphically using Mohrs
circle, Figure 2.19(b); in this case, twisting moments must be taken as positive if the
corresponding positive moment arrow is pointing towards the boundary under consideration.
The principal directions, for which twisting moments are zero, mtn ' 0 , are determined by the
condition
tan 2K1 '

2m xy
mx 7 m y

(2.24)

and the principal moments m1 and m2 in the corresponding directions are

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
55 !

m1,2 '

mx 8 m y
2

$mx 7 m y % 2 8 4mxy2
2

(2.25)

Transformation of transverse shear forces


Equilibrium of the forces acting on the slab elements shown in Figure 2.19(c) yields
vn ' v x cos K 8 v y sin K
vt ' 7v x sin K 8 v y cos K

(2.26)

These relationships can be interpreted as equations for the transformation of transverse shear
forces acting on any boundary perpendicular to the direction n, whose orientation is
determined by the angle K . The trigonometric functions can be represented graphically using
Thales circle, Figure 2.19(d). At any point of the slab, the principal transverse shear force
vo ' v x2 8 v 2y

(2.27)

is transferred in direction Ko , where

tan K o '

vy
vx

(2.28)

In the direction perpendicular to Ko , no transverse shear is transferred. Apart from some


special cases, the principal directions of bending moments and transverse shear forces are not
coincident, Ko = K .
Boundary conditions for slabs
Along the boundary of a slab, there may exist, generally, a bending moment mn , a twisting
moment mnt and a transverse shear force vn (Figure 2.20a). However, the fundamental
equation for thin elastic slabs with small deflections (Kirchhoff, 1850) cannot satisfy more
than two boundary conditions. Thus, one may introduce a further condition for simply
supported and free edges as follows: replace the twisting moments mtn by a continuous
distribution of vertical forces, Figure 2.20(b), which, at the boundaries of infinitesimal
elements of length dt , compensate each other with the exception of the increase mtn,t dt . The
increase per unit distance, mtn ,t , is now combined with the transverse shear force vn , giving
a resultant edge force of
vn 8 mtn,t ' mn, n 8 2mnt ,t

(2.29)

as shown in Figure 2.20(c). At a slab corner, the twisting moments mtn of the two joining
edges result, according to Figure 2.20(d), in a corner force of

2mtn

56

(2.30)
!
2 Design using linear stress analysis !

(a)

(b)

(c)

dt

dt

(d)
dt

dt

mtndt

mndt

(vn5+5mtn,t)5dt

mtn+mtn,t dt
5

(e)

mtn

mnt

mtn

vn5dt

dt

dn

(f)

(g)
dt
vn5dt

mndt

Vt

mtn
dt

mtn

m+ndt

+
mtn
dt
5

1
-Vt5=5mtn
25mnt

vn+5dt

Vt +Vt,t dt

Figure 2.20: Static boundary conditions and discontinuities in slabs: (a) Stress resultants at slab edge;
(b) Forces corresponding to twisting moment; (c) Resultant edge orce; (d) Corner force; (e) Edge
shear force; (f) Force flow at corner; (g) Discontinuity.

The treatment of twisting moments along slab edges outlined above, first proposed by
Thomson and Tait (1883), can be justified by Saint Venants principle. However, it is not
entirely satisfactory due to its foundation on Kirchhoffs theory and the restrictive conditions
for the application of the latter. Figure 2.20(e) illustrates an alternative explanation of the
behaviour near slab edges, based only on equilibrium considerations. In order to satisfy
equilibrium of the edge strip, there must exist a transverse shear force transferred along the
edge such that, provided the slab boundary is stress-free and stresses + t within the edge strip
do not vary along the direction t (Clyde, 1979)

Vt ' 7mtn

(2.31)

Note that the edge shear force (Eq. 2.31) is a force (for example in kN), while the resultant
edge force (Eq. 2.29) is a force per unit width (kN/m).
The condition (Eq. 2.30) for corner forces and the contribution mtn ,t of the twisting moments
to the resultant edge force (Eq. 2.29) are directly obtained from the existence of the edge
shear forces (Eq. 2.31). The corresponding boundary conditions can be summarised as
follows:
"

clamped edge: any values of mn , mtn and vn are possible

"

simply supported edge: mn ' 0 , resultant edge force vn 8 mtn,t ' mn, n 8 2mnt ,t

"

free edge: mn ' 0 , zero resultant edge force vn 8 mtn,t ' mn, n 8 2mnt ,t ' 0 .

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
57 !

These boundary conditions follow directly from equilibrium considerations and are thus valid
for any material behaviour. For thin elastic slabs more restrictive conditions can be
formulated (Timoshenko, 1959); however, these are not relevant if detailing and dimensioning
are carried out according to plasticity theory.
The edge shear forces (Eq. 2.31) must be taken into account when dimensioning free edges of
concrete slabs and shells with transverse reinforcement provided capable of resisting Vt . This
is illustrated using Figure 2.20(f), which visualises the force flow at the corner of a slab under
pure torsion. Two sets of inclined concrete compression struts, perpendicular to each other,
form at the top and bottom faces of the slab with their components normal to the slab edges.
These forces are resisted by corresponding reinforcement. The components parallel to the slab
edges are transferred to inclined concrete compressive struts along the edge strips. The
vertical components of the struts correspond to the edge shear forces (Eq. 2.31) and transverse
(vertical) reinforcement must be provided to resist these forces. This can be achieved by
providing hairpins or corresponding bends in the in-plane reinforcement (providing full
anchorage of the bent-back bar ends is achievable).
Statical discontinuities in slabs
By joining two slabs along their free edges it follows, from the equivalence of twisting
moments along slab edges and edge shear forces according to Eq. 2.31, that while bending
moments mn must be continuous (Figure 2.20g), twisting moments mnt and transverse shear
forces vn may jump across statical discontinuities within a slab. At a statical discontinuity,
8
7
7 mnt
along which a shear force Vt is transferred, the conditions mn7 ' mn8 , Vt ' mnt
and

Vt ,t ' vn7 7 vn8 must be satisfied (see Figure 2.20g).

General remarks on yield conditions


The ultimate load of concrete slabs can be investigated on the basis of plasticity theory by
considering local stresses and strains inside the slab and corresponding yield conditions and
flow rules for concrete and reinforcement. This general procedure gives correct results in all
possible cases but its application is cumbersome and can only occasionally be justified. In
most cases it is sufficient to express statical and kinematical conditions as well as yield
conditions and flow rules in terms of generalized stresses and strains. That is, to consider the
stress resultants Eq. 2.18, Eq. 2.19 and Eq. 2.20 and corresponding generalised deformations,
neglecting local distributions of stresses and strains.
Yield conditions and flow rules in terms of generalised stresses and strains can be derived
from both the kinematic or upper bound method as well as the static or lower bound method
of plasticity theory. According to the kinematic method, deformations are restricted, by means
of kinematic assumptions, to a limited class of theoretically possible deformations that can be
described by a finite set of kinematic parameters. In the next section, the normal moment
yield criterion for slabs is derived on this basis. According to the lower bound method,
statically admissible stress distributions inside an element which do not violate the yield
conditions for concrete nor reinforcement are investigated. Strictly speaking, however, the
relationships obtained in this way (such as for the sandwich model presented later in this
chapter) are no yield conditions since, generally, no compatible mechanism can be found.

58

!
2 Design using linear stress analysis !

2.3.4

Normal moment yield criterion for slabs

Yield lines
Consider the shell element given in Figure 2.21(a), with reinforcement in directions x and y,
loaded by bending and twisting moments as well as membrane and transverse shear forces.
Figure 2.21(b) illustrates an element of a yield line in arbitrary direction t where, N! n and O!n
are the relative rotation rate of the rigid slab parts and the relative extension rate at the level of
the middle plane. For this type of kinematically admissible mechanism, only the bending
moments mn and the normal forces nn contribute to the energy dissipation,
D ' mnN! n 8 nnO!n

(2.32)

For any given value of nn , the depth of the compressive zone c ' h 2 7 O!n N! n can be
determined from equilibrium of stresses in direction n, and one obtains an expression for the
ultimate moment mnu in direction n and, consequently, for the energy dissipation.
(a)

(b)

ny
my
mxy

h /2

1
nx

nxy

nyx

vx

mn

h /2

myx

nn

On

mx

Nn

vy
y

myusinK

(c)
fc
mxu

cx

' fy
asx

fc
myu

' fy
asy

nn

asx fy

asy fy

mxucosK

x
z

cy

ntn
mn

n
mtn
1

(d)

(e)

mxy

myu

k= 1

m'yu

mx
myu

mxu

my

m'xu

|mxy|
my
mxu
mx

|mxy|

Figure 2.21: Normal moment yield criterion: (a) Shell element; (b) Yield line; (c)Superposition of ultimate
moments in directions x and y; (d) Yield condition; (e) Dimensioning.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
59 !

Consideration of yield line mechanisms according to Figure 2.21(b), result in non-zero


reactions for mtn and ntn (Figure 2.21c). These do not contribute to the energy dissipation
since the corresponding kinematic parameters are zero, that is, mtn and ntn are generalized
reactions within the domain of plasticity theory. By introducing a third kinematic parameter
O!tn , representing a relative shear deformation along the yield line, a generalized yield line is
obtained in which the membrane shear forces ntn contribute to the energy dissipation
according to D ' mnN! n 8 nnO!n 8 ntnO!tn but the twisting moments mtn remain as generalized
reactions.
In the following, only yield lines with O!tn ' 0 are treated, and membrane forces are neglected,
that is nn ' 0 . Thus, only the bending moments mn contribute to the energy dissipation.
Deduction of the normal moment yield criterion
By superimposing the ultimate moments mxu and myu in the reinforcement directions while
setting mxy ' nx ' ny ' 0 , a statically admissible state of stress is obtained (Figure 2.21c).
For an arbitrary direction n it follows that nn ' ntn ' 0 , mn ' mxu cos 2 K 8 m yu sin 2 K and

mtn ' myu 7 mxu sinK cosK (see Eq. 2.23). Generally, the depths of the compression zones
in the two reinforcement directions do not coincide, that is cx = cy and there is no compatible
mechanism according to Figure 2.21(b). Thus, the value of mn determined in this way is a
lower bound for the ultimate moment mnu in direction n, mxu cos 2 K 8 m yu sin 2 K .
The differences between mn and mnu obtained for cx = cy are generally negligible and the
inequality can be suppressed, leading to the conditions
mnu ' mxu cos 2 K 8 m yu sin 2 K

(2.33a)

# ' m#xu cos 2 K 8 m#yu sin 2 K


mnu

(2.33b)

Equation 2.33b, which is valid for negative moments, follows from the same considerations as
for Eq. 2.33a.
A state of stress mx , my and mxy corresponds, according to Eq. 2.23, to bending and
twisting moments in the direction n of
mn ' mx cos 2 K 8 m y sin 2 K 8 mxy sin $2K %

(2.34)

and from the condition that

# ? mn ? mnu
7 mnu

(2.35)

one obtains, by considering all possible directions K , the yield conditions

60

!
2 Design using linear stress analysis !

$
%
2
Y # ' mxy
7 $m#xu 8 mx %$m#yu 8 m y % ' 0
2
7 $mxu 7 mx % m yu 7 m y ' 0
Y ' mxy

(2.36)

where mxu 7 mx < 0 , myu 7 my < 0 , m#xu 8 mx < 0 and m#yu 8 my < 0 . The conditions Y ' 0

and Y # ' 0 can be represented by two elliptical cones in the mx , my , mxy -space, as shown in
Figure 2.21(d).
Similarly to the yield conditions for orthogonally reinforced membrane elements, it is possible
to represent the yield conditions defined by Eq. 2.36 in parametric form:

mxu < mx 8 k mxy

m yu < m y 8 k 71 mxy

m#xu < 7mx 8 k mxy

m#yu < 7m y 8 k 71 mxy

(2.37)

Equation 2.37 is suitable for straightforward dimensioning of the four reinforcement layers,
Figure 2.21(d) where, typically, k = 1 is used.
Skew reinforcement
Yielding reinforcement layers in arbitrary, skew, directions can always be substituted by an
equivalent orthogonal reinforcement if it is assumed that all reinforcing bars are located,
approximately, in the same plane.
The effect of several reinforcement layers oriented in directions differing by angles : i from

the x-axis and with resistances nis ' as f sy per unit width corresponds to a fictitious
i
reinforcement in directions x and y with resistances

nxs ' P nis cos 2 : i ,


i

n ys ' P nis sin 2 : i ,


i

nxys ' P nis sin : i cos : i ,

(2.38)

The resistances nxs , nys and nxys can be transformed using Eqs. 2.23a-c (setting ns instead
of m) just like bending and twisting moments and there exist two perpendicular principal
directions $R ,Q % , differing by an angle ( from the x- and y-axis, for which nRQs ' 0 . The
ultimate moments mRu and mQu corresponding to the resistances nRs and nQs can be
substituted in the yield condition Eq. 2.36 together with the bending and twisting moments
mR , mQ and mRQ acting in the same direction, giving

%$

2
Y ' mRQ
7 mRu 7 mR mQu 7 mQ ' 0

(2.39)

where mRu 7 mR < 0 and mQu 7 mQ < 0 . An alternative procedure consists of transforming the
ultimate moments mRu and mQu into the directions x and y using Eq. 2.23, leading to
m xu ' mRu cos 2 ( 8 mQu sin 2 (

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

(2.40a)
!
61 !

m yu ' mRu sin 2 ( 8 mQu cos 2 (

(2.40b)

mxyu ' mQu 7 mRu sin ( cos (

(2.40c)

Comparing Eqs. 2.40a-c with the bending and twisting moments mx , my and mxy leads to

2
Y ' mxy
7 $mxu 7 mx % m yu 7 m y ' 0

(2.41)

where mxu 7 mx < 0 and myu 7 my < 0 . The conditions Y ' 0 and Y # ' 0 (which has not
been given but follows from the same considerations as outlined for positive moments) can be
graphically represented as elliptical cones in the mx , my , mxy -space, similar to the yield

conditions of Eq. 2.36 but with the vertices no longer lying in the plane mxy ' 0 .
Assuming that the effective depth is constant in all directions, it is possible to evaluate mxu ,

myu and mxyu directly from Eq. 2.38 without having to determine the principal directions

$R ,Q % and substitute them into Eq. 2.41.

Discussion of the normal moment yield criterion

The normal moment yield criterion has been established and confirmed by various researchers
(for example Johansen, 1962, Nielsen, 1984, Park and Gamble, 1980, Wolfensberger, 1964,
Wood, 1961) and corroborated with a wide range of experimental data. Due to its simplicity,
it is widely used for the dimensioning of slabs in design practice. While its application is
generally unproblematic due to the typically ductile behaviour of slabs, it does have some
weaknesses of which the designer should be aware.
Supposing that sections perpendicular to the element middle plane remain straight and
perpendicular to the deformed middle plane, the deformations of thin slabs with small
deflections are fully determined by six kinematic parameters ) x 0 , ) y 0 , & xy0 , S x , S y , S xy
. These six parameters come from elongation in x and y, distortion of middle plane, bending in
directions x and y and twist (see Figure 2.22). The consideration of yield lines is kinematically
more restrictive and the yield conditions given by Eqs. 2.36 and 2.37 are thus generally in
spite of the slight inequality suppressed in the deduction of Eq. 2.33 upper bounds for the
ultimate load. This observation holds particularly true for elements subjected to high twisting
moments with respect to the reinforcement directions (reinforcement directions deviating
significantly from the directions of the principal moments) and high reinforcement ratios
(Marti, 1980, Nielsen, 1984). Special care is required in the dimensioning of elements and
excessive reinforcement ratios should be avoided unless a more sophisticated analysis is
undertaken, for example using to the sandwich model outlined below.
Furthermore, the normal moment yield criterion does not take into account transverse shear
forces which are always present in slabs. In the case of potentially shear-critical slabs (high
shear forces, thick slabs or slabs with horizontal construction joints) it is advisable to use a
model which allows for a consistent treatment of bending and twisting moments as well as
transverse shear forces. The sandwich model approach outlined below is particularly suitable
for such situations.

62

!
2 Design using linear stress analysis !

(a)

(b)

(c)

(e)

(f)

(d)

Figure 2.22: Kinematics of a thin slab element: (a)-(c) elongations )x0, ) y0, and distortion & xy0 of element middle
plane; (d)-(f) curvatures Sx, S y and twist S xy.

2.3.5

Sandwich model for the dimensioning of shell elements

The eight stress resultants acting on a shell element, Figure 2.23(a), can be resisted by a
sandwich model. In this model the bending and twisting moments mx , my , mxy as well as

the in-plane forces nx , ny , nxy are attributed to the sandwich covers, while the sandwich

core resists the transverse shear forces vx , vy

I (illustrated in Figure 2.23b). Thus, the

sandwich model represents a lower bound solution with its basis in plasticity theory.

(a)

(b)

(c)
y

dv
z

ny
my
mxy

nxy

nx
nyx

vx

my ny
+
dv 2

mxy nxy
+
dv 2

mx nx
+
dv 2

v0

dv

dv
1

vy

my ny
+
dv 2

mxy nxy
+
dv 2

vx

vy

mx myx

K0

mx nx
+
dv 2

dvcot-

v0cot2
v0cot-

v0

v0cot2

Figure 2.23: Sandwich model: (a) Shell element; (b) Layer forces; (c) Transfer of transverse shear force in
uncracked and cracked core.

The sandwich core transfers the principal transverse shear force v0 ' v x2 8 v 2y in the

direction K 0 ' tan 71 v x v y . If the transverse shear forces are small, that is, nominal shear
stresses v0 dv are below the nominal shear cracking resistance , c, red , the core will remain
uncracked and the forces in the bottom and top sandwich covers follow directly as

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
63 !

m
n
n x inf,sup ' M x 8 x
dv
2
n y inf,sup ' M
n xy inf,sup ' M

my

dv
m xy

ny

dv

(uncracked core, v0 d v 9 , c, red )

(2.42)

n xy
2

If the principal transverse shear force v0 ' v x2 8 v 2y is high enough to produce cracking of
the sandwich core, the latter can be treated like the web of a girder of flanged cross-section
running in direction K o (Figure 2.23c). The corresponding tensile forces in the element plane
must be resisted by the sandwich covers in addition to the forces given by Eq. 2.42, resulting
in the following total forces in the sandwich covers [Marti (1990)]
mx nx
v x2
8
8
n x inf,sup ' M
2 2v0 tan dv
n y inf,sup ' M
n xy inf,sup ' M

my
dv

m xy

ny

dv

n xy
2

v 2y

(cracked core, v0 d v < , c, red ) (2.43)

2v0 tan 8

vxv y
2v0 tan -

According to Eqs. 2.42 and 2.43, a state of plane stress is obtained in the sandwich covers,
and they can be dimensioned as membrane elements. For the case of failure governed by
yielding of the reinforcement with the concrete remaining elastic in both sandwich covers
(defined as Regime 1 and given by Eq. 2.7a), the force per unit width in the reinforcement in
the x and y directions are determined with
asx f y <
asy f y <

mx nx
v x2
8
8
8
dv
2 2v0 tan my
dv

ny
2

v 2y
2v0 tan -

k
k 71

v x2

m
n
a#sx f y < 7 x 8 x 8
8
2 2v0 tan dv
a#sy f y < 7

my
dv

ny
2

v 2y
2v0 tan -

mxy
dv
mxy
dv

k# 7

8 k #71 7

8
8

m xy
dv
m xy
dv

n xy

2
n xy
2
8
8

n xy
2
n xy
2

vxv y
2v0 tan vxv y
2v0 tan 8
8

vxv y

(2.44)

2v0 tan vxv y


2v0 tan -

where as and a#s are the bottom and top reinforcement areas per unit width of the slab,
respectively.
For an uncracked core, that is v0 d v 9 , c, red , the terms in Eq. 2.44 containing vx or vy can
be omitted and no transverse shear reinforcement is required. That is * z ' 0 .
If the core is cracked, v0 d v < , c, red , transverse reinforcement of

64

!
2 Design using linear stress analysis !

*z '

v0tan dv f y

(2.45)

is required for the core to carry transverse shear force. In Eqs. 2.42 to 2.45, dv is the effective
depth of the membrane forces in the sandwich covers, k and k # are arbitrary positive factors
and - is the inclination of the diagonal compression field in the sandwich core. Different
values of k and k # can be adopted for each element of a shell or slab, avoiding abrupt changes
or providing sufficient development length in order to anchor the differential forces in the
reinforcement while in choosing the inclination angle, - . In this regard similar considerations
apply as in the design of membranes. For simplicity in design practice using k ' k # ' 1 and

- ' 45o often provides for a good solution.


Equation 2.44 is valid only if the concrete in the sandwich covers does not crush. In order to
satisfy this condition, the thickness of the sandwich covers, and consequently the effective
depth dv , should be chosen such that

$
$

f c tinf < a sx f y 8 a sy f y 7 n x inf 8 n y inf


f c t sup < a #sx f y 8 a #sy f y 7 n x sup 8 n y sup

(2.46)

where tinf , tsup = thickness of bottom and top sandwich cover and f c = effective concrete
compressive strength, respectively. Substituting Eqs. 2.43 and 2.44 in Eq. 2.46, we write

f c tinf < k 8 k 71

f c t sup < k 8 k

71

mxy
dv

%7 d

mxy
v

8
8

nxy
2
nxy
2

8
8

vxv y
2v0 tan vxv y

(2.47)

2v0 tan -

As a first and often sufficient approximation, dv can be chosen such that the middle planes of
each sandwich cover coincides with the centre between its two reinforcement layers.
The threshold value , c, red for considering cracked or uncracked behaviour of the sandwich
core should be carefully selected. Basically it could be taken as the nominal shear strength of
the slab or shell without transverse reinforcement, which, due to size effect, varies with the
element thickness; for example , c , red ' f ctd for membrane shear according to [CEB-FIP
(1990)]. Since the provision of transverse reinforcement is quite time-consuming, it is
advisable to choose the element thickness such that, apart from zones where concentrated
forces are introduced, no transverse reinforcement is required.
In cases with high axial compression nx , ny or high in-plane shear forces nxy , it is possible
to attribute part of these forces to the sandwich core and then, strictly speaking, the concrete
compression in the sandwich core must be checked taking into account the combined action
of nx , ny , nxy and v0 . The case of high axial compression is likely to occur if prestressing is

treated as forces acting on the structure. However, this is beyond the scope of this report.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
65 !

2.3.6

Dimensioning of slab and shell elements in design practice

The design equations derived from the sandwich model (Eqs. 2.44 and 2.45) are simple and
transparent, allowing for a straightforward dimensioning of in-plane as well as transverse
reinforcement in the general case of a shell element subjected to eight independent stress
resultants. In the special case of slabs and neglecting transverse shear forces, Eq. 2.44 is
equivalent to the normal moment yield criterion (Eq. 2.37) if exact values are taken for dv .
The incorporation of Eqs. 2.44 and 2.45 into a finite element program for shells does not any
present more difficulties than the implementation of the normal moment yield criterion
(Eq. 2.37) for the dimensioning of reinforcement, a post-processing option which is already
offered by a number of commercial programs today (often called Wood-Armer moments).
Usually, no explicit check of the assumed failure mode is made in slabs when applying
Eq. 2.37, assuming implicitly that the reinforcement ratios obtained are small enough to avoid
concrete crushing. While this condition is usually satisfied in slabs, a check of the assumed
failure mode should always be carried out for shells, particularly for elements subjected to
significant axial compression or cases with high reinforcement ratios. Thus, it is advisable to
implement the corresponding check of concrete resistance according to Eq. 2.46 together with
Eqs. 2.44 and 2.45.
Transverse shear forces obtained from a finite element calculation will generally be less
accurate than bending and twisting moments, since, as can be seen from Eqs. 2.21b-c, which
can be rewritten as vx ' mx, x 8 mxy, y and v y ' my, y 8 myx, x , transverse shear forces are
derivatives of the bending and twisting moments. Thus, in order to get reasonable results, a
finer element mesh is required when using Eq. 2.44 and 2.45 in their general form than in
cases where transverse shear forces can be neglected a priori.
2.3.7

Example 1

Consider the slab element shown in Figure 2.24(a); according to the normal moment yield
criterion, Eq. 2.37, using a value of k = 1, the required bending resistances in the
reinforcement directions are
m xu < m x 8 m xy ' $96 8 133% T kN ' 229 kN
J asx,inf < 1520 mm 2 m J C20 @ 200 mm
m yu < m y 8 m xy ' $7 117 8 133% T kN ' 16 kN

(2.48b)

J asy ,inf < 100 mm m J * min


m#xu < 7 m x 8 m xy ' $7 96 8 133% T kN ' 37 kN

(2.48c)

J asx, sup < 240 mm 2 m J * min


m#yu < 7 m y 8 m xy ' $117 8 133% T kN ' 250 kN
2

(2.48a)

J asy , sup < 1670 mm m J C20 @ 170 mm

(2.48d)

Note that the required reinforcement areas have been determined for uniaxial bending
according to CEB-FIP (1993).

66

!
2 Design using linear stress analysis !

(a)

(b)

mx '
x

ny
my
mxy

nx
nyx

nxy

vx

myx

mx

(c)

96 kNm/m

nx ' 7 127 kN/m

m y ' 7117 kNm/m

nx ' 119 kN/m

mxy ' 133 kNm/m

nx ' 7 138 kN/m

vy

mx '

96 kNm/m

m y ' 7117 kNm/m

f yd ' 435MPa f yk ' 500 MPa

mxy ' 133 kNm/m

f cd ' 24.0 MPa $ f ck ' 50 MPa %

v x ' 358 kN/m

, c, red ' 1.87 MPa $, c, red ' f ctd %

v y ' 624 kN/m

h ' 0.40 m, tinf,sup ' 0.08 m


dv ' 0.32 m
Figure 2.24: Examples: (a) element and materials; (b) slab element actions; (c) shell element actions. Note that
actions and strengths given are design values according to CEB-FIP (1993).

In many practical cases, one or more of the reinforcement areas resulting from the application
of Eq. 2.37 will be negative. This means that no reinforcement is required in the
corresponding direction. Also, in such cases it is basically possible to reduce the
reinforcement area in the other direction by changing the value of k such that the
reinforcement area which was negative for k = 1 equals zero or corresponds to the minimum
reinforcement to be provided. However, this is rarely done in design practice.
Consider next the shell element shown in Figure 2.24(b). With a principal transverse shear
force of vo ' v x2 8 v 2y ' 719 kN , see Eq. 2.27, cracking of the core must be expected since

vo dv ' 2.25 MPa @ , ctd . The resulting forces in the bottom and top layers follow thus from
Eq. 2.42 as nx,inf ' 7274 kN m , ny,inf ' 736 kN m , nxy,inf ' 502 kN m (bottom layer)
and

nx,sup ' 7274 kN m ,

ny,sup ' 696kN m ,

nxy,sup ' 7329 kN m

(top

layer).

Substituting these membrane forces into Eqs. 2.8 and 2.9, or directly using Eq. 2.44, gives
asx,inf < 1900 mm 2 m J C20 @ 150 mm

(2.49a)

asy ,inf < 1070 mm 2 m J C20 @ 250 mm

(2.49b)

asx, sup < 130 mm 2 m J * min

(2.49c)

asy , sup < 2360 mm 2 m J C20 @120 mm

(2.49d)
2

The transverse reinforcement resulting from Eq. 2.45 is * z ' 0.52% ' 5170 mm m , which
can be provided, for example, using 26 stirrups 16mm per square meter (16@150 x 250).
Finally, concrete compression must be checked. Rearranging Eq. 2.45 gives

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
67 !

asx,inf f y 8 asy ,inf f y 7 nx,inf 8 n y ,inf

tinf

asx, sup f y 8 asy , sup f y 7 nx, sup 8 n y , sup


t sup

' 12.6 MPa ? f cd

(2.50a)

' 8.2 MPa ? f cd

(2.50a)

and thus, the reinforcement dimensioning is valid.


If the reinforcement of the slab element of Figure 2.24(a) is dimensioned as if it were a shell
element using the sandwich model, one obtains reinforcement areas of asx,inf ' 1650,

asy,inf ' 120, asx,sup ' 270 and asy,sup ' 1800 mm2/m as opposed to that given by Eqs.
2.48a-d obtained using the normal moment yield criterion, Eq. 2.37. It is seen that
reinforcement dimensioning according to the sandwich model is slightly on the safe side. The
differences are due to the larger lever arms used in the calculation of the bending resistances
for the normal moment yield criterion as compared with the averaged value of dv used in the
sandwich model.
2.3.8

Example 2

While a comprehensive treatment of the design process and its results are beyond the scope of
this report, some basic considerations will be outlined in order to illustrate the potential of the
sandwich model. The example selected is the shell roof of the fish market at the Port of
Santander, Spain, designed by the engineer Prof. Juan Jos Arenas and built 2001-2002. The
roof is shown in Figure 2.25.
The roof consists of 23 modules measuring approximately 5.1 x 45 m each, with a typical
thickness of 0.12 m. The geometry of each module is defined by a circular arch of 30 m span
complemented by two cantilevers of 7.5 m on either side in the transverse direction, and by
5.10 m wide parabolas of variable height in the longitudinal direction. Thus, globally, each
module is acting as an arch, tied together by a post-tensioned, suspended floor or by posttensioned concrete ties depending on the module. Construction was carried out using two
complete scaffolds, each for an entire module, starting at the two extremes of the roof. The
repeated use of the scaffolds reduced the impact of their construction costs on the overall
economy of the structure to a reasonable level.
Due to the high live loads on the suspended floors, a reinforcement following the principal
stress trajectories was not possible since these directions shift considerably depending on the
load-case. Therefore, and because it was judged to be more practical, an orthogonal
reinforcement layout (in plan), aligned with the longitudinal and transverse axes of the
building, was adopted.
The reinforcement was dimensioned using the sandwich model outlined in 2.3.6, above. In a
first step, based on the different load combinations considered according to the pertinent
building codes, envelopes of all eight stress resultants were computed with a conventional
linear finite element program using eight-node shell elements (six degrees of freedom per
node). Each of the envelopes calculated consisted of the maximum and minimum values of
the stress resultant under consideration along with the concomitant values of the remaining
seven stress resultants. The results obtained in this way were then analysed according to

68

!
2 Design using linear stress analysis !

Figure 2.25: Fish market Port of Santander, Spain (designer Prof. Juan Jos Arenas).

Eqs. 2.44 to 2.47, with dv such that the middle planes of each sandwich cover coincided with
the centre between its two reinforcement layers for an assumed maximum bar diameter of
12 mm in each direction.
Using Eq. 2.45, the envelopes of the transverse shear forces were analysed to determine
whether or not the core would crack. In this case, as a result of the double curvature, loads are
transferred primarily through in-plane actions, such that transverse shear forces remain small
and the provision of transverse shear reinforcement was only required in the edge modules.
Next, the required cross-sections of the four reinforcement layers were determined using
Eq. 2.44 for a bar spacing of 150 mm. Finally, the assumed thickness of the sandwich covers
was checked according to Eq. 2.47.

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

!
69 !

Each of the steps outlined above was carried out for all of the sixteen envelopes (maximum
and minimum of each stress-resultant) at each of the roughly 1100 nodes per module. Thus, it
was decided to implement the entire design procedure in a specially written program which
was then run as a post-processor to the finite element program used in the global analysis. The
post-processor output consisted of numerical and graphical results, containing a CAD file
with all the data required for drafting of the reinforcement plans. Based on these results,
reinforcement plans such as the one shown in Figure 2.26 were produced and constructed
(Figure 2.27).

2.4

3D solid modelling

2.4.1

Introduction

As for two-dimensional modelling, linear stress analysis can be employed as an effective


design tool for the design of three-dimensional structures. Three dimensional frame analysis
programs, for example, are used in common practice with details given in numerous
textbooks. Three-dimensional solid modelling is less commonly utilised. As for twodimensional elements and shell elements it is not always intuitive on how to dimension the
reinforcing steel in 3D solids to meet the stress demands of the applied tractions. One method
is to place adequate reinforcing steel in the direction of any principal tension stress and to
ensure that the concrete has sufficient strength to meet all principal compressive stress
demands. However, placing reinforcement in principal directions is not always convenient
and placement in the local structural or global directions is preferred. The question is then
how to dimension reinforcement for any set of orthogonal axes for the 6 components of any
applied stress tensor that defines the stresses at a point. The answer to this question is
addressed below. For further reading on design using 3D stress analysis the reader is referred
to Foster and Marti (2003) where the concepts are discussed in further detail, including a
general procedure for the dimensioning of reinforcement and the determination of optimum
reinforcement.
2.4.2

Background

In 3D space the stresses at a point are completely defined by the tensor (using von Karmans
notation)

Z+ x , xy , xz W
X
U
+ ij ' X, xy + y , yz U
X, xz , yz + z U
V
Y

(2.51)

where x, y, z are any set of orthogonal axes and the stresses are defined as shown in
Figure 2.28a. For every point in a body there exist three stresses + x# , + y # and + z # on the
local x# y # z # axis system such that , x#y# ' , x#z# ' , y#z# ' 0 . These stresses are known as the
principal stresses and x# , y # , and z # the principal axes. It is well established that the principal
stresses are equal to the eigenvalues of the stress tensor. Perhaps less well recognised,
however, is that the direction cosines to the principal axes are given by the norms of the
eigenvectors of the stress tensor.

70

!
2 Design using linear stress analysis !

Figure 2.26: Reinforcement details for one module of the fish market roof Port of Santander, Spain.

!
!

fib Bulletin 45: Practitioners guide to finite element modelling of reinforced concrete structures

71

Figure 2.27: Construction of the fish market roof at the Port of Santander, Spain.

For any oblique plane (refer Figure 2.28b) having a unit normal n ' nx , ny , nz passing

through a point P, the stresses at P can be resolved into a component normal to the plane $+ n %

and a shear component parallel to the plane $Sn % . For a stress to be principal S n ' 0 which,
from Eq. 2.51, implies that

+y
,yz

,xy
,yz
,xz

+z

,xy

+n
sn

+x

,xz

(b)

(a)

Figure 2.28: a) 3D stresses at a point defined in the orthogonal xyz axis system; b) normal and shear stress for
an arbitrary plane passing through a point.

72

2 Design using linear stress analysis

+ x nx 8 , xy n y 8 , xz nz ' + n nx
, xy nx 8 + y n y 8 , yz nz ' + n n y

(2.52)

, xz nx 8 , yz n y 8 + z nz ' + n nz
Rewriting Eq. 2.52 of the form +.n = 0 it is seen that the equations are homogeneous. As all
three components of n cannot be zero the solution is non-trivial only if the determinant of the
coefficients + ' 0 , that is

+ x 7+ n
, xy
, xz
, xy
+ y 7+ n
, yz ' 0
, xz
, yz
+ z 7+ n

(2.53)

Expansion of Eq. 2.53 leads to the characteristic equation

+ n 7 I1+ n 8 I 2+ n 7 I3 ' 0

(2.54)

where I1 , I 2 and I 3 are the invariants of the stress tensor and are given by

I1 ' + x 8 + y 8 + z ' +1 8 + 2 8 +3

(2.55a)

$
%
2
2
% ' +1+ 2+ 3
I 3 ' + x+ y+ z 8 2, xy, xz, yz 7 $+ x, 2yz 8 + y, xz
8 + z, xy

2
2
I 2 ' + x+ y 8 + x+ z 8 + y+ z 7 , xy
8 , xz
8 , 2yz ' + 1+ 2 8 + 1+ 3 8 + 2+ 3

(2.55b)
(2.55c)

and where + 1 , + 2 and + 3 are the principal stresses. By common definition the principal
stresses are ordered such that + 3 ? + 2 ? +1 .
As for two-dimensions, stresses at a point in 3D can be plotted in the form of Mohrs circles
(shown in Figure 2.29) where the normal stress is plotted on the horizontal axis and the shear
stress plotted on the vertical axis. Three principal circles are possible between the principal
stress pairs + 1 7 + 2 , + 2 7 + 3 and +1 7 + 3 . In failure theorems the principal stress pair 1-3 is
regarded as the most important and the circle generated through this stress pair is referred to
as the major principal stress circle.

+55?5+55?5+
3
2
1

+3

Sn

+2

+1

+n

Figure 2.29: Mohrs circles for stresses at a point in 3D.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

73

In xyz space + x , + y and + z are, by definition, normal to the YZ, XZ and XY planes,
respectively. The shear stresses on these planes is then given by either the positive or negative
roots of
2
2
2
2
S x ' , xy
8 , xz
; S y ' , xy
8 , 2yz ; S z ' , xz
8 , 2yz

(2.56)

As the orientation of the xyz axis system is arbitrary, its interpretation represents all possible
planes. It can be then shown that all points of $+ n , Sn % must line on or between the principal
stress circles and, thus, the feasible domain of solutions lies with the hatched region of
Figure 2.29.
2.4.3

Application to reinforced concrete

In the applications that follow the xyz axes are taken to correspond with reinforcing directions.
As for two dimensions, normal stresses applied at a point in a reinforced concrete solid
element are carried by reinforcing steel and/or the concrete whilst shear stresses are carried by
the concrete alone. Given that the applied stress tensor has been determined, for example by
3D finite element solid modelling, the Mohrs circles of applied stress can be plotted, as
shown in Figure 2.30. Within the circles the stress points $+ i , Si % are also plotted where
i ' x, y, z . As the reinforcing steel cannot carry shear stress it follows that the points
$+ ci , Sci % must fall within the hatched region of the concrete stress circles where

+ ci ' + i 7 *i+ s and Sci ' Si and where *i is the volumetric ratio of reinforcement in the ith
direction and + s is the stress in the steel.

Sn

+555'5+557
ci
i 5*5+
i s

*5+
i s

$+555,
ci Sci)

+c3

+c2

+c1 +3

Concrete Stresses

$+55,
i S i)

+2

+1

+n

Applied Stresses

Figure 2.30: Compression field for 3D stress at a point.

Applying Eq. 2.51 to the stresses defined in Figure 2.30, the tensor of the concrete stresses is
written as

Z + x 7 * x f yd .x
X
+ cij ' X
, xy
X
, xz
Y

74

W
, xy
, xz
U
$+ y 7 * y f yd. y %
, yz
U
, yz
$+ z 7 * z f yd.z %UV

(2.57)

2 Design using linear stress analysis

where f yd.x , f yd. y and f yd.z are the design yield strengths of the reinforcing steels in the x
y and z directions, respectively.
The invariants of the concrete stress tensor are given by Eq. 2.55 with the appropriate
substitutions for + x , + y and + z . Comparing Eqs. 2.57 with Figure 2.30 it is seen that there
are 5 unknowns + c 2 , + c 3 , * y f yd. y , *x f yd.x and *z f yd.z (with + c1 ' 0 or another
prescribed limit such as + c1 ' fct ). As the solution to Eq. 2.53 provides for a maximum of
three real roots an infinity of solutions exist to Eq. 2.57. The designer then has the freedom to
apply two constraint equations with the three invariant equations making up the five equations
required for a solution.
The design process outlined above is demonstrated in the examples below. As the three
resulting stress points $+ ci , Sci % are not constrained to the boundary of the stress circles, as is
the case for two dimensions, graphical solutions for three-dimensional stresses tend to be
somewhat more complex than for 2D. For this reason the examples presented below are
limited to analytical solutions.
2.4.4

Reinforcement dimensioning for 3D stresses - example 1

The results of a stress analysis on a concrete structural element give the stress tensor in the xyz
dimensions as
6 7 4W
Z2
X
+ ij ' X 6 7 2 2 UU MPa
XY 7 4 2
5 UV

(2.58)

It is desired to reinforce the element in the orthogonal directions of xyz. For the stresses
defined by Eq. 2.58, the magnitudes of the shear stresses are S x ' 2 13 [ 7.21 MPa ,
S y ' 2 10 [ 6.32 MPa

S z ' 2 5 [ 4.47 MPa

and

and the principal stresses are

+ 1 ' 8.28 MPa , + 2 ' 4.32 MPa and + 3 ' 77.60 MPa . The Mohrs circle of stress for the
tensor of Eq. 2.58 is shown in Figure 2.31a. Viewing the stress plot (Figure 2.31a) it is
decided to seek the solution that gives the lowest demand on the concrete strength. This is
established by selecting the smallest diameter for the major principal stress circle. As only the
concrete carries shear stress the radius of the major stress circle $R173 % is constrained such

and thus for our example R173 ' Sx = 2 13 MPa.


Therefore, for the absolute minimum compression stress in the concrete the constraint
equations are given by

that R173 < max S x , S y , S z

+ c3 ' 72 S x ' 74 13 MPa , and

(2.59a)

* x f yd.x ' + x 8 S x ' 2 8 2 13 MPa

(2.59b)

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

75

|Sn | (MPa)
$+565;
x S x |)

$+565;
y S y |)

$+565;
z S z |)

+n (MPa)
+355= -7.60

+255= 4.32

+155= 8.28

(a) Applied Stresses

*x +sx = 9.21
$+5556;
cx S x |)

|Sn | (MPa)

*y +sy = 3.88
$+556;
x S x|)

7.21 MPa

$+556;
y S y|)

$+5556;
cy S y |)

$+556;
z S z |)

*z +sz = 9.21

$+5556;
cz S z |)

+n (MPa)

+555
c2 = -2.88

+555=
c3 -14.42

(b) Concrete Stresses


Figure 2.31: Mohrs stress circles for example 1: a) applied stresses and b) concrete stresses.

Substituting Eqs. 2.59a and b into the stress invariant equations given by Eqs. 2.55a-c we
write
I1 \

(2.60a)

+ c 2 7 4 13 ' 3 7 2 13 7 * y f yd . y 7 * z f yd .z

% $

%$

I 2 \ 74 13+ c 2 ' 72 13 3 7 * y f yd . y 7 * z f yd . z 7 2 8 * y f yd . y 5 7 * z f yd .z 7 56
I3 \

%$

0 ' 16 * y f yd . y 8 36 * z f yd .z 7 2 13 7 2 7 * y f yd . y 5 7 * z f yd .z 7 244 8 8 13

Solving Eqs. 2.60a-c gives + c 2 ' 72.88 MPa , * y f yd. y ' 3.88 MPa and
9.21 MPa. The final solution is plotted in Figure 2.31b.
2.4.5

(2.60b)
(2.60c)

*z f yd.z '

Reinforcement dimensioning for 3D stresses - example 2

In our second example we are given the stress tensor


Z 7 3 6 7 4W
+ ij ' XX 6 7 7 2 UU MPa
XY 7 4 2
0 UV

76

(2.61)

2 Design using linear stress analysis

and it is required to dimension the reinforcing steel. For the tensor of Eq. 2.61 the magnitude
of the shears are Sx ' 7.21MPa, S y ' 6.33 MPa and Sz ' 4.47 MPa and the principal
stresses are + 1 ' 3.28 MPa , + 2 ' 70.68 MPa and + 3 ' 712.60 MPa . The Mohrs circle of
stress for the applied tractions is plotted in Figure 2.32a. After reviewing the stress circles it is
decided to seek a solution such that no reinforcing steel is required in the y-direction, that is
* y f yd. y ' 0 . Substituting this constraint into Eqs. 2.55a-c gives

I1 \ + 2c 8 +3c ' 7107 *x f yd.x 7 *z f yd.z

(2.62a)

I2 \ + 2c+3c ' 7358 7*x f yd.x 810*z f yd.z 8 *x f yd.x *z f yd.z

(2.62b)

I3 \ 0 ' 288 4*x f yd.x 815*z f yd.z 7 7*x f yd.x *z f yd.z

(2.62c)

$+565;S
x |)
x

$+565;S
y |)
y

|Sn | (MPa)
$+565;S
z |)
z

+n (MPa)
+255= -0.68

+355= -12.60

+155= 3.28

(a) Applied Stresses

$+565;
x Sx |)
$+5565;
cx Sx |)

|Sn | (MPa)

*x +sx = 3.77
*z +sz = 3.77

$+5565;
cy Sy |)

$+565;
z Sz |)

$+5565;
cz Sz |)
+555
c3 = -14.57

+555
c2 = -2.98

+n (MPa)

(b) Concrete Stresses


Figure 2.32: Mohrs stress circles for example 2: a) applied stresses and b) concrete stresses.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

77

Solutions to Eqs. 2.62a-c are plotted in Figures 2.33a and 2.33b for the intermediate principal
concrete stress + c 2 versus the stress in the x and z reinforcement and for + c 2 versus + c 2 ,
respectively. From the first stress invariant it is seen that the minimum volume of
reinforcement for a unit element (for f yd.x ' f yd. y ' f yd.z ' f yd ) occurs at the point where
I1 is a minimum. In general, the optimal solutions lie at the upper end of + c 2 as shown in
Figures 2.33a and 2.33b for the example at hand. After consideration a solution is chosen such
that *x f yd.x ' *z f yd.z = 3.77 MPa, + c 2 ' 72.98 MPa and + c3 ' 714.57 MPa . The stress

circles for the chosen solution are shown in Figure 2.32b.

20

-13

*x fyd + *z fyd

-14
(MPa)

10

*z fyd

+ c3

* fyd

-20

-15

*x fyd

-16

-25

-17
-18

+c3

Ic1 (MPa)

(MPa)

15

-15

Ic1

-30

-19
0
-10

-9

-8

-7

-6
-5
+c2 (MPa)

-4

-3

-20
-10

-2

(a)

-9

-8

-7

-6
-5
+c2 (MPa)

-4

-3

-35
-2

(b)

Figure 2.33: Solutions to Eqs. 2.62a-c for + c2 versus a) reinforcement ratios and b) principal concrete stresses.

2.5

References

ASCE-ACI Committee 445 on Shear and Torsion, (1998), Recent approaches to shear design
of structural concrete, Journal of Structural Engineering, ASCE, Vol. 124, No. 12,
December, pp. 1375-1417.
Belarbi, A., and Hsu, T.T.C. (1991), Constitutive Laws of Reinforced Concrete in Biaxial
Tension Compression, Research Report UHCEE 91-2, Department of Civil Engineering,
University of Houston, Houston, Texas.
Cagley, J.R. (2001), Changing from ACI 318-99 to ACI 318-02 Whats New?, Concrete
International, Vol 23, No. 6, June, pp. 69-183.
CEB-FIP (1993), CEB-FIP Model Code 1990 for Concrete Structures. Comit EuroInternational du Bton, Bulletin dInformation No. 213/214, Lausanne, May 1993, 437 pp.
Clark, L.A. (1976), The Provision of Tension and Compression Reinforcement to Resist
In-Plane Forces, Magazine of Concrete research, Vol. 28, No. 94, March, pp. 3-12.
Clyde, D.H. (1979), Nodal Forces as Real Forces, Final Report, IABSE Colloquium on
Plasticity in Reinforced Concrete, Copenhagen, International Association for Bridge and
Structural Engineering, IABSE Vol. 29, pp. 159-166.

78

2 Design using linear stress analysis

Clyde, D.H., (1977), A General Theory for Reinforced Concrete Elements, Australasian
Conference on the Mechanics of Structures and Materials, Christchurch, New Zealand,
August.
Collins, M.P., and Mitchell, D., (1986), Rational Approach to shear Design - The 1984
Canadian code Provisions, ACI Structural Journal, Vol. 83, No. 6, Nov-Dec, 925-933.
CSA84 (1984), Design of Concrete Structures for Buildings, CAN3-A23.3-M84, Canadian
Standards Association, Rexdale, Onterio, 281 pp.
Foster, S.J., and Malik, A.R., (2002), Evaluation of Efficiency Factor Models used in Strut
and Tie Modelling of Non-Flexural Members, ASCE, Journal of Structural Engineering, Vol.
128, No. 5, May, pp. 569-577.
Foster, S.J., Marti, P. and Mojsilovi!, N. (2003), Design of Reinforced Concrete Solids
Using Stress Analysis, ACI Structural Journal, V100, N6, Nov-Dec, pp. 758-764.
Johansen, K.W. (1962), Yield Line Theory, Cement and Concrete Association, London,
181 pp.
Kaufmann, W., and Marti, P. (1998), Structural Concrete: Cracked Membrane Model,
Journal of Structural Engineering, ASCE, Vol. 124, No. 12, pp. 1467-1475.
Khalifa, J. (1986), Limit Analysis and Design of Reinforced Concrete Shell Elements,
Ph.D. Thesis, University of Toronto, Toronto, 314 pp.
Kirchhoff, G. R. (1850), ber das Gleichgewicht und die Bewegung einer elastischen
Scheibe, A. L. Crelles Journal fr die reine und angewandte Mathematik, Berlin, Vol. 40,
No. 1, pp. 51-58.
Kirschner, U., and Collins, M.P. (1986), Investigating the Behaviour of Reinforced Concrete
Shell Elements, University of Toronto, Department of Civil Engineering, Publication
No. 86-09, Toronto, 210 pp.
Leitz, H., (1923) Eisenbewehrte Platten bei allgemeinem Biegungszustande, Die Bautechnik,
Vol. 1, pp.155-157, 163-167.
MacGregor, J.G., (1997) Reinforced Concrete - Mechanics and Design, 3rd Edition, Prentice
Hall, New Jersey.
Marti, P. (1979), Plastic Analysis of Reinforced Concrete Shear Walls, Plasticity in
Reinforced Concrete, IABSE Colloquium, Kopenhagen, 21-23 May, Introductory Report,
Reports of the Working Commission, Vol. 28, pp. 51-69.
Marti, P. (1980), "Zur Plastischen Berechnung von Stahlbeton, Institut fr Baustatik und
Konstruktion, ETH, Zrich, Bericht Nr. 104, Birkhuser Verlag Basel, 176 pp.
Marti, P. (1980), Zur Plastischen Berechnung von Stahlbeton, Institut fr Baustatik und
Konstruktion, ETH Zrich, IBK Bericht Nr. 104, 176 pp.
Marti, P. (1990), Design of Concrete Slabs for Transverse Shear, ACI Structural Journal,
Vol. 87, No. 2, pp. 180-190.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

79

Mitchell, D., and Collins, M.P. (1974), Diagonal Compression Field Theory A Rational
Model for Structural Concrete in Pure Torsion, ACI Journal, Vol. 71, August, pp. 396-408.
Miyakawa, T., Kawakami, T., and Maekawa, K. (1987), Nonlinear Behavior of Cracked
Reinforced Concrete Plate Element Under Uniaxial Compression, Proceedings of the JSCE
No. 378, August, pp. 249-258.
Morley, C.T. (1979), Yield Criteria for Elements of Reinforced Concrete Slabs, Plasticity in
Reinforced Concrete, IABSE Colloquium, Kopenhagen, 21-23 May, Introductory Report,
Reports of the Working Commission, Vol. 28, pp. 35-47.
Mller, P., (1978), Plastische Berechnung von Stahlbetonscheiben und Balken, Dissertation
Nr. 83, Birkuser Verlag Basel und Stuttgart.
Nielsen, M.P. (1963), Yield Conditions for Reinforced Concrete Shells in the Membrane
State, Non-Classical Shell Problems, Procedings of the I.A.S.S. Symposium, Warsaw,
2-5 September, pp.1030-1040.
Nielsen, M.P. (1971), On the Strength of Reinforced Concrete Discs, Acta Polytechnica
Scandinavica, Civil Engineering and Building Construction Series No. 70, Copenhagen,
261 pp.
Nielsen, M.P. (1984), Limit Analysis and Concrete Plasticity, Prentice-Hall, Englewood
Cliffs, 420 pp.
Nielsen, M.P., (1999), Limit Analysis and Concrete Plasticity, 2nd Edition, CRC Press LLC,
908 pp.
Nielsen, N.J., (1920), Beregning af Spaendinger i Plader (Calculation of Stresses in Slabs),
Copenhagen.
Pang, X.B., and Hsu, T.T.C. (1992), Constitutive Laws of Reinforced Concrete in Shear,
Research report UHCEE92-1, Department of Civil Engineering, University of Houston,
Houston, Texas, 188 pp.
Park, R., and Gamble, W.L. (1980), Reinforced Concrete Slabs, John Wiley & Sons, New
York, 618 pp.
Robinson, J.R., and Demorieux J-M. (1977), Essais de Modeles dame de Poutre en Double
T, Annales de lInstitut Technique du Btiment et des Traveaux Publics, No. 354, October,
Serie: Beton No. 172, pp. 77-95.
Thomson, W., and Tait, P.G. (1883), Treatise on Natural Philosophy, Vol. 1, Part 2,
Cambridge University Press.
Timoshenko, S.P., and Woinowsky-Krieger, S. (1959), Theory of Plates and Shells, Mc
Graw-Hill, International Student Edition, 580 pp.
Vecchio F.J., and Collins M.P (1986), The Modified Compression Field Theory for
Reinforced Concrete Elements Subjected to Shear, ACI Journal Proceedings, Vol. 83,
No. 22, Mar.-Apr., 219-231.

80

2 Design using linear stress analysis

Vecchio, F., and Collins, M.P. (1982), The response of Reinforced Concrete to In-plane
Shear and Normal Stresses, The Department of Civil Engineering, University of Toronto,
Canada, March, 332 pp.
Wolfensberger, R. (1964), Traglast und optimale Bemessung von Platten, Institut fr
Baustatik und Konstruktion, ETH Zrich, IBK Bericht Nr. 2, 119 pp.
Wood, R.H. (1961), Plastic and Elastic Design of Slabs and Plates, Thames and Hudson,
London, 344 pp.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

81

[Blank Page]

Essential nonlinear modelling concepts

3.1

Introduction

In Chapter 2 it was demonstrated that, for new structures, linear elastic finite element analysis
provides a practical way of dimensioning structures, or structural elements, for determination
of strength demands and reinforcement arrangements. These designs will often require
internal stress redistribution after cracking or yielding and reinforcement detailing needs to
allow for this. In general, a design based on linear elastic analysis will provide safe solutions
to the problem of satisfying the strength limit state.
For some design and analysis problems, however, a linear analysis may not be sufficient. As a
simple example, consider the requirement of satisfying a serviceability limit states such as
calculating deflections and crack widths. For such a design-check, the local extent of cracking
is important and a linear analysis may not be sufficient to confirm that this limit state has been
satisfied. While it is practical to design structures with linear analyses, it is sometimes helpful
to check additional limit states by nonlinear analysis methods. For new structures a non-linear
analysis may be performed on the structure (or its elements) after initial proportioning using a
plasticity-based design procedure based on a linear elastic analysis (Chapter 2). In addition,
nonlinear analysis can also assist in the evaluation of complex geometry or poorly detailed
structures where the effects of localized cracking, for example, may be poorly modelled by
linear analysis. The following list provides some general cases where a nonlinear analysis
may prove useful:
1) Confirmation of safety for complex design details: Design codes offer methods to check
safety for standard members such as beams and columns but are often not sufficient for
more complex design geometries such as complicated walls systems with openings. While
the inclusion of strut-and-tie methods into modern design codes has assisted greatly in
expanding the generality of design codes, nonlinear analysis can be helpful to allow an
independent check of whether the development of cracking, for example, will be
consistent with the details of the reinforcement placement.
2) Assessment of safety of existing structures: Some structures, such as those built to older
design standards, may contain reinforcement quantities and details that are not consistent
with modern standards. In such situations there may be very significant costs associated
with strength upgrades of the structure and a nonlinear analysis can assist in providing a
better estimate of the safety factor of the as-built structure against collapse. This can be
particularly important in seismic regions where design earthquake forces have tended to
increase in severity through the years as knowledge of seismic hazards have improved.
3) Pushover analyses for structural capacity computation: Many codes of practice require a
pushover analysis of structural strength to estimate failure modes and to determine if the
collapse mode conforms to capacity design principles. That is, it is important to be able to
predict how a structure is expected to collapse so that it may be designed to do so in a
ductile mode such as flexure, rather than a brittle mode such as shear.
4) Explanation of observed distress in structures: As mentioned in Chapter 1, structural
distress seen in the field may be best investigated with the assistance of nonlinear
structural analysis. In particular, the results of nonlinear methods capable of estimating
critical events and the extent and patterns of cracking at the actual applied loads can be
directly compared to the structure and used to confirm the assessed mode of structural

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

83

behaviour. This allows a higher level of confidence to be placed in the calculated safety
factor than that obtained from a blind analysis.
5) Estimation of P-O effects: The influence of geometric nonlinearity, in the form of the P-O
effect, can be important in the moment and axial load design of frames. As concrete
material science continues to improve, it can be expected that columns of higher
slenderness will become more common and these will be more sensitive to the effect. The
combination of material nonlinearity, in the form of cracking, and geometric nonlinearity,
is a complex phenomenon that can require nonlinear finite element analysis.
6) Resistance of structures to extreme events: Design against events such as earthquakes or
sabotage must include nonlinear effects to be realistic. Ignoring the ability of structures to
redistribute stresses can result in cost prohibitive, overly conservative solutions.
7) Resistance to fire: Structures that must be designed to resist significant fire can show
major effects resulting from the thermal expansion of parts of the structure that may not be
captured by linear analysis. Considering the effects of fire and structural collapse is a
nonlinear problem that requires complex analysis.
As reinforced concrete is a nonlinear material, the salient nonlinear effects that concrete
shows are described next. Afterwards, the various frameworks with which these nonlinear
effects can be incorporated into finite element solutions are described followed by a short
discussion concerning the expectation of precision in concrete nonlinear analyses. Finally, a
short discussion of safety reliability calculations is presented. In each case, the description of
behaviour or implementation is not written to be exhaustive but simply to demonstrate to the
reader some of the more important aspects that are required for the engineer to use nonlinear
finite element tools intelligently.

3.2

Nonlinear concrete behaviour

Structural concrete is an inherently nonlinear material both at strength limit states and service
loads. Table 3.1 summarizes some of the important nonlinear effects observed in structural
concrete. These are organized in terms of effects that are primarily observed in concrete, steel
or a combination of the two. While not all nonlinear finite element analyses will need to
consider all the listed effects, it is important that engineers are aware of the different nonlinear
phenomena so that they can judge whether or not a particular set of analysis assumptions are
appropriate.
When considering nonlinear behaviour, an important aspect is whether or not strain
localization will occur as size effects are expected in such cases (see Table 3.1). Strain
localizations occur when a material stress-strain behaviour shows a decreasing stress for an
increasing strain such as concrete in tension after cracking or in compression after reaching its
peak compressive capacity. Nonlinear effects with localization can show a size effect whereby
larger elements fail at lower stresses than geometrically similar smaller ones. In addition,
material behaviour showing significant localization generally cannot be easily defined in
terms of strain but rather must be defined in terms of an absolute displacement. Thus the
direct tensile, or cohesive, stress that can be transmitted across a crack is a function of the
width of the crack and not on a strain term. This effect of localization is of particular
importance to finite element analysis as the mesh size itself introduces a fixed length scale
that can bias the results of localization sensitive problems. As an example, an analysis on

84

3 Essential nonlinear modelling concepts

Table 3.1: An overview of nonlinear behaviour in reinforced concrete structures


Types of Behavioural Effects

Tension

Compression

Macrocracking
Tension softening *
Cyclic response
Creep
Crack closing effect
Shrinkage

X
X

Crushing
Nonlinearity at high strains
Post-peak unloading
Cyclic response
Creep
Rate of loading
Bi or triaxial confinement
Poisson's ratio
Thermal effects

X
X

X
X
X
X
X
X

Examples of behaviour

X
X
X
X
X
X

X
X
X
X
X
X
X
X
X

Size dependent

Plastic strains

Bifurcation

Plain Concrete Behaviour

Stiffness

Energy balance may need to be considered to fully model listed effect


Elements must include variable stiffnesses to fully model the listed effect
Indicates a discrete change in behaviour rather than a smooth transition
Indicates that simple nonlinear elasticity may not provide a full solution
Behaviour cannot be fully modelled with strain terms alone. Modelling also requires an
absolute distance relationship such as a crack width

Energy

Energy :
Stiffness:
Bifurcation:
Plastic strains:
Size dependent:

X
X
X

X
X
X
X

(Reinhardt and Xu, 1999)

X
X
(Sakai and Kawashima, 2006)

Reinforcement Behaviour
Tension

Compression
Shear

Yielding
Strain hardening
Thermal effects
Rate of loading
Rupture
Buckling
Dowel action

X
X
X
X
X

X
X
X

X
X
X

X
X
(Ozcebe, Ersoy, Tankut, 1999)

Combined Concrete & Reinforcement Behaviour


Tension

Compression
Shear

Bond
Tension stiffening *
Tension splitting
Compression Softening
Aggregate interlock

X
X
X
X
X

X
X
X

X
X

Damage Effects

(Harajli, 2004)

Material damage
Fatigue

e.g. alkali silica reaction

* Tension softening is distinct from tension stiffening behaviour. See text sections 3.2.2, 3.2.3.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

85

plain concrete in tension will produce different load-deformation curves after cracking if
meshes with different element sizes are used and special measures are not taken (Baant and
Oh, 1983, Baant and Planas, 1998).
While plain concrete can be subjected to localization effects under increasing deformations,
these effects are often not of significant importance in structures reinforced with a minimum
quantity of ductile reinforcement. With such structures, the ductile properties of the
reinforcing steel dominate the global behaviour. Examples of behaviour where localization
effects can have a significant influence on behaviour include punching shear, shear in
members without transverse reinforcement, over-reinforced columns and beams and large
unreinforced concrete structures such as dams. In each of these cases the behaviour at failure
is more controlled by that of plain concrete than that of the reinforcement.
In using Table 3.1, and nonlinear finite element procedures in general, engineers must check
their finite element discretization and modelling assumptions used in their analyses. Note that
many of the attributes in the table are contentious in the research community and, as such, are
open to debate. Responsible engineers must use experience, judgment and intuition as to
whether, for example, localization issues will be important. While it may seem that nonlinear
finite element analyses could allow non-engineers to produce reasonable and accurate results,
this is not the case; the experience and judgment of engineers is more important in nonlinear
finite element analysis than it is with linear analysis due to the significantly higher complexity
level.
The following sections expand on a subset of the important behaviour in Table 3.1 explaining
the particularly important experimentally observed characteristic nonlinear concrete
behaviours.
3.2.1

Concrete in compression

Concrete subjected to compression acts in a nonlinear way as has been recognized for many
years (e.g. Turneaure and Maurer, 1908). Three important aspects of concrete compressive
behaviour are described in more detail here: localization in compression, confinement of
concrete, and compression softening.
Influence of localization on compression behaviour of concrete

As concrete shows a decrease in stress for increasing strains beyond the strain associated with
the compressive strength, localization of concrete in compression can be expected. That is, a
damage band forms where all the deformation associated with the post-peak region occurs
while the remainder of the specimen unloads with decreasing strain. Two primary effects will
result from this; firstly larger specimens failing primarily in a compressive mode can be
expected to show less ductility in the post-peak region than smaller specimens and, secondly,
a size effect is predicted for some member types whereby larger elements in compression will
be weaker in terms of stress than smaller elements. Both of these effects may be of particular
importance in practice as most laboratory testing is done on small-scale models and, thus,
may not be representative of the real safety of large structures in the field.
It is important to note, however, that there remains much debate on the importance of
localization effects for practical structures that contain realistic quantities of reinforcement
even within the laboratory testing field. For example, some tests on specialized structural
components confirm the presence of a size effect in unreinforced flexural compression zones
(Kim, et al. 2000). In contrast, other carefully performed tests on actual laboratory beams
86

3 Essential nonlinear modelling concepts

subjected to constant moment have not always shown the decrease in ductility expected with
size (Alca et al. 1997).
Confinement and compression softening

It is well known that concrete can carry higher compressive stresses with larger deformations
when it is laterally confined as illustrated in Figures 3.1 and 3.2. Finite element models can
require either biaxial or triaxial confinement relationships. Early experimental investigations
on biaxially loaded RC elements focused on the strength. Kupfer et al. (1969) reported on the
strength, deformational characteristics and micro-cracking of concrete under biaxial stress
conditions. Figure 3.1 illustrates the stress-strain relationships under biaxial compression. As
can be seen in both Figure 3.1a and Figure 3.1b, the maximum compressive strength of
concrete increases for the biaxial compression state. Kupfer et al.s (1969) results suggest that
a maximum strength increase of about 25% is achieved at a stress-ratio of +2/5+1 = 0.5. In
addition, concrete ductility increases under biaxial compression.

(a)
Figure 3.1:

(b)

Concrete in biaxial compression: (a) stress versus strain; and (b) strength envelope (Kupfer et al.,
1969)

It is interesting to note that during the latter stages of the loading, an increase in overall volume
occurs (dilatancy); this is typically attributed to the progressive growth of micro cracking in
concrete.
For triaxial compression, Richart et al. (1928) conducted tests at low and moderate confining
stresses with typical results shown in Figure 3.2a. Balmer (1949) conducted triaxial tests at
very high confining stress levels (Figure 3.2b). As can be observed from the figures,
depending on the confining stress, concrete may act as a quasi-brittle, plastic-softening or
plastic hardening material. This is mainly due to the fact that under higher confining stresses,
the possibility of bond failure between cement paste and aggregates is significantly reduced,
and the failure takes place through the crushing of the cement paste. As can also be seen,
pronounced increases in strength with increasing confining pressure were observed by both
Richart et al. and Balmer. In fact, experiments show that concrete has a fairly consistent
failure surface that is a function of the principal stresses under triaxial loading.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

87

(a)
Figure 3.2:

(b)

Triaxial compression stress versus axial strain: (a) low confinement (Richart et al., 1928); and (b)
high confinement (Balmer, 1949).

One of the earlier 3D failure surface models that incorporated this strength enhancement was
that of Ottosen (1977); shown in Figure 3.3. While there are many models that have been
published, the important point is not which of the models is used but whether or not the model
produces a fair representation of the failure surface. Many commercial packages will include,
for example, a Mohr-Coulomb failure model. While the Mohr-Coulomb failure criteria can be
made to represent concrete strength at very high levels of confinement, as at these levels the
failure surface for the concrete approaches circular, the model does not reproduce well the
failure surface for concrete under the levels of confining pressure typical of that in common
reinforced concrete structures. If the Mohr-Coulomb approach is to be used it must be tuned
for the problem being investigated. The better software packages for concrete analysis will
include a specific failure model for concrete but even then some care is needed as some
failure surface models require different calibration for biaxial compression than for other
triaxial compressive stress conditions.

Figure 3.3:

88

Ottosen (1977) failure surface for triaxial (+1, + 2 + 3) stresses (plotted in the Haigh-Westergaard
coordinate space).

3 Essential nonlinear modelling concepts

Many different stress-strain models exist to account for either the biaxial or triaxial behaviour
of concrete in compression. These depend on many parameters and are generally specific to
continuum modelling of triaxially loaded materials or are specific to the behaviour of
columns, for example. One model which is often used in non-linear elasticity is the Kent and
Park (1971) model later modified by Scott et al. (1982) to include the strength and ductility
enhancement due to confinement effects and the effect of strain rate (Figure 3.4). Such
models, however, ignore size and compression localization effects and can lead to erroneous
(but usually conservative) results when considering post peak behaviour. Such models are
usually sufficient, however, in adequately determining the response of a member for
increasing loading up to the maximum load. Nevertheless, as for all non-linear modelling, the
results should be scrutinized with due caution. In analyses of cracked concrete in
compression, confinement may often be ignored with little penalty as the confining stress is
almost completely released normal to the cracks.

Kfc#

) 50h

0.5Kfc#
0.2Kfc#

Unconfined
concrete
) 0=0.002 ) 50u
) 50c

Confined concrete

) 20c

)c

Figure 3.4: Modified Kent and Park (1971) model (Scott et al., 1982).

A related but not identical concept to confinement is that of weakening and softening in
compression due to the effects of transverse tensile strains beyond cracking (Collins 1978).
Figure 3.5 shows the results from many shear tests as reported by Vecchio and Collins (1993)
to quantify this effect. The specimens in this figure were subjected to combinations of shear
and axial loads and were reinforced in such a way that they were able to resist large tensile
strains transverse to the direction of the applied compression. As can be seen in the figure,
some of these specimens failed by crushing of concrete in compression at only 20-40% of the
strength of concrete control specimens. As strains perpendicular to the direction of applied
compression increase, this strength reduction effect, often called compression softening,
becomes more severe. This effect can be thought of as an extension of the confinement effect
into the tensile domain and is particularly important in understanding of the behaviour of
concrete in shear.
3.2.2

Concrete in tension

The one type of nonlinearity that is expected in all concrete structures is cracking. This
phenomenon is difficult to account for in a simple fashion, yet is vital in making realistic
estimates of structural stiffness.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

89

Figure 3.5:

Peak principal compressive stresses versus the principal tensile strain ratios (Vecchio and Collins,
1993)

When plain concrete cracks, it is able to resist residual, direct, tensile stresses across the crack
for small crack widths. These stresses, sometimes called bridging stresses or cohesive stresses
are sufficiently small that they are often ignored in traditional beam design. The decrease in
these bridging stresses as the crack width increases is called tension softening and models
based on the energy of fracture (Gf) are usually used to quantify the effect. Overall the
behaviour of cracking in plain concrete is reasonably well modelled by methods such as those
based on fracture mechanics.
In concrete that is reinforced, the process of cracking is more complex. Shown in Figure 3.6
are the multiple cracks that form in reinforced concrete around the bars (bond cracking) as
well as macro-cracking that extends to the surface of the member (Goto 1971). Each of the
cracks shown in Figure 3.6 can be expected to behave consistently with the tension softening
behaviour mentioned above for plain concrete. When the effects of all these cracks are
integrated together, including the bond cracks, and the entire concrete section is considered as
a whole, behaviour consistent with that shown in Figure 3.7 is obtained and the concrete
component is termed tension stiffening.
This tension stiffening is the difference in behaviour between the response of a bare-bar and
the observed response of the composite reinforced specimen. This tension stiffening can be
thought of as a macroscopic property applied on a large-scale compared to tension softening
which is a property on a smaller-scale. As Figure 3.8 shows, an analysis with very small
elements that considers the bond cracking and a tension-softening model does produce the
same macroscopic tension stiffening behaviour as that observed in experiments.
Macroscopic tension stiffening behaviour can be considered as being caused by an average
tensile stress in the concrete between the cracks. Figure 3.9 shows a plot of the stresses in the
reinforcement and in the concrete for a tensile element like that in Figure 3.8. It shows the
average concrete tension stiffening stresses varying along the member length are almost zero
at the location of the macro crack and are greater than zero between. The lower plot shows the
reinforcement stresses varying from a maximum at the crack, where there is no concrete
tension, to a minimum between the cracks. This is a helpful way to think of macroscopic
tensile stresses in concrete as it makes it clear that reinforcement will yield at a crack before it
does between the cracks and also that equilibrium at a crack itself must be able to be

90

3 Essential nonlinear modelling concepts

Figure 3.6: Internal cracking in reinforced concrete.

Tensile force

Member response
c

Tensile force

Bare bar

a
b
c
d

Reinforcement
(unembedded)

Average strain

Uncracked
Crack formation
Stabilized cracking
Post-yielding

R First crack
S Last crack
Y Yielding

Elongation

(a)
Figure 3.7:

(b)

Schematic of: (a) the tension stiffening effect; (b) the stress- strain relationship of embedded
reinforcing bar by the CEB-FIP Model Code 1990 (1993).

& 6-1)7-8+/01-//+49+
31)3:-8+34.31-0-

%"#
%

Macro -tension stiffness model


$"#
$
!"#
!

Plain concrete!
softening

Computed from micro analysis


#!!

$!!!

$#!!

%!!!

& '()*+, -).+/01)(.+2, (3145


bond crack

penetrating cracbond
k
deterioration zone

Figure 3.8: Modelling of tension stiffening with tension softening

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

91

maintained ignoring all concrete stresses beyond those expected by tension softening. This
equilibrium criterion is why the macroscopic concrete stresses are called tension stiffening
and not tension strengthening: they cannot increase the strength of the tension specimen in
Figure 3.9, only increase its stiffness before yield at a crack.

Figure 3.9: Average stresses in tension stiffening analysis

Considering the differences noted in the previous section between tension stiffening and
tension softening, it is perhaps not surprising that there is more than one way to include
tension in finite element analyses of concrete structures. It is helpful to divide up the level of
analysis complexity into the following three levels:
Model
Small-scale
Medium-scale
Large-scale

concrete tension
softening
stiffening / softening
stiffening

reinforcing bond
perfect bond
bond law
perfect bond

Element size
very small
medium
large

With small-scale tension modelling, as shown in Figure 3.8, it is necessary to use extremely
small elements to capture the extensive cracking that occurs due to bond and slip effects and
the analysis should be fully three-dimensional. These models may use the assumption of
perfect bond whereby the nodal displacements of the concrete adjacent to the reinforcement
are assumed to be the same as that of the reinforcement itself. Due to the number of elements,
this type of modelling is rarely practical for realistically sized structures at the present time.
With medium-scale tension modelling, the bond forces are accounted for with a macroscopic
bond model that provides a relationship between the relative displacements of the concrete
compared to the reinforcement. While this simplification means that more nodal degrees of
freedom are required, it has the advantage that much larger elements are allowable and twodimensional analysis becomes possible. Larger elements are possible as it is no longer
necessary to be able to resolve the extensive bond cracking that the small-scale tension
modelling requires. If a concrete tension softening relationship is used, however, it will still
be necessary to use small elements to capture the change in concrete stresses between the
cracks. If, instead, a concrete tension stiffening effect is included, then still larger elements
may be used, though with the cost that special routines may be needed (depending on the
formulation) to ensure that the stress in the reinforcement at a crack does not exceed the yield
stress.

92

3 Essential nonlinear modelling concepts

Large-scale tension modelling is perhaps the most practical method of including concrete
tension in nonlinear analyses currently. Perfect bond between the concrete and reinforcement
again becomes appropriate and much larger elements may be used in two or three dimensions.
Care is required to ensure that the tension stiffening relationship used is consistent with the
bond properties of the reinforcement and that local conditions at a crack are checked for
equilibrium. Due to the use of perfect bond, effects caused by insufficient development of
reinforcement must also be explicitly considered. A reasonable way to account for this effect
with perfect bond programs is to progressively increase the cross sectional area of the
reinforcement over the development length of the bar to properly model the total force that
can be transferred to the concrete at every location. Finally, again, it is necessary with all
analyses based on tension stiffening to ensure that the stresses in the reinforcement at cracks
are not allowed to exceed the yield stress with some sort of crack check. For further
discussion, see Vecchio and Collins (1986), Petrangeli and Obolt (1996), Maekawa and An
(2000) and Foster and Marti (2002, 2003).
3.2.3

Modelling of tension stiffening

Modelling of tension stiffening can be undertaken in one of two ways: the first is to modify
the stiffness of reinforcing bars; the second is to modify the concrete stiffness to carry the
tension force after generation of cracks. In the first model, the stress transfer carried by both
steel and concrete due to bond action can be expressed by changing the stiffness of
reinforcing bars, as per Figure 3.7b (Gilbert and Warner, 1978). The CEB-FIP Model Code
(1993) gives the following phases: uncracked concrete, crack formation, stabilized cracking
where only crack opening occurs and post-yielding. While such a treatment has an advantage
in terms of simple 1D calculation of response, this is not generally applicable to 2D or 3D
problems since the stiffness is unchanged even if the direction of cracks relative to the axis of
the bar changes (Okamura and Maekawa, 1991).
The second approach of modifying the concrete properties can be sub-classified as i) methods
based on a macroscopic smeared crack model, which describes spatially averaged behaviour
of RC containing multiple cracks in a defined control volume, or ii) methods based on
mesoscopic discrete crack models.
Tension stiffening models based on smeared crack concepts use a spatially averaged
constitutive relationship, which involves multiple cracks and non-uniform local stress
between the cracks in a control volume. Figure 3.10 gives three examples of models
representing average stress-strain relationships of concrete in tension. Since this type of
modelling is generally assumed independent of the spacing of cracks, direction of reinforcing
bars and reinforcement ratio, the stabilization and distribution of propagating cracks is assured
in a control volume. For this reason, it is one of the most versatile and applicable constitutive
relationships in executing structural analyses in cracked reinforced concrete.
When the stress-strain relationship of concrete is introduced in an averaged manner, the
relationship of reinforcing bars embedded in concrete should also be determined on an
average basis. Here, it has to be noted that the average stress-strain relationship of the
reinforcement is different from the corresponding relationship of bare bar after yielding
(Shima et al., 1987), since the local stress and strain of reinforcing bars in concrete is
disturbed due to the effect of bond. An example of one method of calculating an equivalent
average stress for a given average strain is presented in Figure 3.11 (Maekawa and An, 2000).
If steel remains elastic, the constitutive model can be set equal to that for the bare bars.
However, as soon as the bar yields at a crack, the elastic relationship cannot be maintained

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

93

Figure 3.10: Examples of stress-strain models of concrete in tension (from Bentz 2005).

+'

1
G +( x, y )dv
VV

+'

Local stress

1
G +( x, y )dv
VV

Local stress

Local strain

)'

Stress distribution

Stress distribution

Local strain

1
)(x,y)dv
V

G
v

Yield plateau jump!

)'

1
)(x,y)dv
V

G
v

Strain distribution

Strain distribution

(b)

(a)
+'

1
G +( x, y )dv
VV

+'

Local stress

1
G +( x, y )dv
VV

Local stress

Local strain

Strain distribution

)'

Strain distribution

Stress distribution

Stress distribution

Local strain

1
)(x, y)dv
V

Plastic zone

)'

Plastic zone

1
)(x, y)dv
V

G
v

(c)

(d)

Figure 3.11: Average stress-strain relationship of a steel bar in concrete: a) at crack initiation; b) at beginning
of yield; c)developing yield; d) full yield (after Maekawa and An, 2000).

94

3 Essential nonlinear modelling concepts

even though the parts of the bar away from the cracks are still linear-elastic (Figure 3.11a). As
the strain increases, the yielded portion of the bar is extended (Figure 3.11b and c). Then,the
average yield stress becomes lower than that of bare bar. For accurate prediction of
deformational behaviour of RC members after yielding, the constitutive relationship capable
of considering this phenomena is needed. Usually a bilinear or tri-linear relationship is
assumed for the average response of reinforcing bars.
For normal concrete combined with two-way reinforcement with a ratio of 0.1 to 2 percent in
each direction, the constitutive law based on a smeared crack approach is independent of the
control volume (Maekawa et al., 2003). In other words an average stress can be evaluated
uniquely for an average strain history regardless of density of macroscopic cracks within the
control volume and, therefore, there is no size effect in the constitutive law for reinforced
concrete in a uniform strain field. Similarly it has been shown that mean stress-strain
relationships of reinforcement in RC plates are also not influenced by the number of cracks
within an element, including in the post-yielding regime. This observation is convenient for
many structural analyses as it is not necessary to define individual macroscopic cracks within
the control volume for members that are provided with minimum, or greater, reinforcement.
This is not the case, however, where minimum reinforcement is not provided and cracks pass
through an unreinforced (or under-reinforced) concrete section. The case where the
distributed cracking assumption is not appropriate and cracks are localized is discussed in
detail in the following section.
A relatively new approach to the modelling of tension stiffening has been developed by Marti
et al. (1998) known as the tension chord model (Figure 3.12). The significance of this
approach is that an equivalent plastic bond stress-slip model (Figure 3.12c) is explicitly
included in the formulation allowing for calculations of crack spacing, crack widths and
tension stiffening directly from the model. The model was further developed into a 2D
formulation (the cracked membrane model) by Kaufmann and Marti (1998), shown in Figure
3.13, and included in a FE model by Foster and Marti (2002, 2003). With adoption of the
stepped rigid-perfectly plastic bond-slip relationship (Figure 3.12c), the stresses in the steel
and in the concrete can be determined for any point within the differential element between
cracks (Figure 3.12b). The stresses in the concrete and in the reinforcing steel at and between
cracks are then calculated directly from equilibrium.
(a)

srm
]

(b)

Ac

x
(c)

dx
dx

,b

+c+ d+c

, b0

,b

, b1
Oy

+c
+s+ d+s

,b

+s

Figure 3.12: Tension chord model: a) reinforced tension chord; b) differential element; c) bond shear stressslip relationship (Marti et al., 1998).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

95

(b) Y

^555f
y ct

srmy

-r
srmx

^555f
x ct

srmx

srmx

s rmy s rmy s rmy

sr
m

sr

cft

sr

(a)

Tension Stiffening
Stresses

Figure 3.13: 2D tension stiffening stresses in: a) material axis directions; b) equivalent orthogonal tension
chords (Kaufmann and Marti, 1998, Foster and Marti, 2003).

If a discrete modelling approach is preferred over that of a smeared crack approach, for the
modelling of reinforced concrete members, it is necessary to explicitly deal with: each
individual crack that propagates around the reinforcing bars; the distribution of local stress
and strain in steel bars; and the interaction at the interface between concrete and bar (that is,
stress transfer through bond, bond deterioration, and so on). The crack propagation around a
steel bar is caused as a result of three-dimensional strain fields and bond transfer mechanisms
and, since the fracture process in 3D stress field differs from that in 2D, a 3D analysis is
appropriate for modelling of reinforced concrete elements using discrete crack models (Morita
and Kaku, 1979, Maekawa et al. 1999). In adopting this approach, fine meshes are needed
with many degrees of freedom and, thus, the use of discrete crack models for the analysis of
reinforced concrete members has generally not gained favour over that of the more general
distributed smeared crack approach. For problems that involve tensile fracture of the concrete,
however, discrete, crack band or non-local models are needed to capture localized behaviour
and is discussed in more detail in the next section.
3.2.4

Modelling of concrete cracks

An important issue with the modelling of tension in finite element models is the way that
cracks are considered. In general they can be taken either as smeared throughout the element
or only present at finite element boundaries which is also called discrete cracking.
Modelling of reinforced concrete with dispersed cracking as a quasi-continuous material or
smeared cracking was firstly introduced by Rashid (1968) and Cervenka and Gerstle
(1971,1972) and has been widely applied for mechanics of cracked reinforced concrete. It
requires that the cracks within the element be accounted for in terms of their effect on
stiffness, strength, and energy characteristics. Generally, this is accomplished via smearing
the crack within the element (see Figure 3.14a) based upon a space averaging process in stress
and strain. Smeared crack tension stiffening models can be solely formulated based on
experimentally extracted behaviour and will, generally, lump all nonlinear processes such
as bond, aggregate interlock, tension softening and others into simple constitutive
relationships in one-dimension.

96

3 Essential nonlinear modelling concepts

The second option is to model the cracks discretely by allowing gaps to form between
different elements. This is the option taken by discrete element and lattice models discussed
below and these models are generally based on tension softening relationships.
Just as care is required when considering the stress-strain relationship between tension
stiffening and tension softening, it is also important to consider that localization problems can
occur for smeared cracking problems as well. Consider the finite element mesh shown in
Figure 3.14b. The structural cracks are shown within the elements rather than between them
indicating that this is a smeared crack analysis. However, the cracks are only within one line
of elements indicating that virtually all the displacement is localized within this region of the
member. As such, this extent of cracking in the model is consistent with a tension softening
type of formulation which is designed to account for localization. In contrast, the mesh in
Figure 3.14a shows cracking smeared throughout the tension region and thus no localization
problems are present, consistent with a tension stiffening relationship.

Element to Model

a) no localization

b) localization

Figure 3.14: Localization in Reinforced Concrete FE Analysis.

A final issue to consider with the modelling of cracking in concrete relates to the effects of a
change in principal average stress directions that can often results from redistribution. Initial
cracks are usually assumed to form at the angle of principal tensile stress in the continuous
material before cracking, but after cracking the average stress directions may no longer align
with this axis. In general there are two approaches for dealing with this: fixed angle cracks
and rotating angle cracks. Both are approximations.
Experiments on large elements subjected to pure shear have shown that the angle of principal
strains, averaged over a sufficient gauge length to include several cracks, can rotate
significantly between first cracking and ultimate limit states. In practice, new cracks form at
different angles to the previous cracks as the loading increases. This behaviour is also
observed in beams with stirrups as new shear cracks form at different angles to that of initial
shear cracks. Rotating crack models, therefore, allow the angle of cracking to change as the
analysis progresses to capture this effect. By contrast, a fixed crack approach forces the angle
of the cracks to remain constant after initial cracking with the shear stresses calculated on that
crack surface usually checked against aggregate interlock relationships.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

97

3.2.5

Modelling of reinforcement

Reinforcing bars in structural concrete are generally assumed to be one-dimensional line


elements without transverse shear stiffness nor flexural rigidity. The point-wise stress-strain
relationship can be formulated by the elasto-plastic modelling as shown graphically in Table
3.1. The plastic evolution characteristics, yield and fracture strengths and plastic elongation,
are normally dependent on the micro-properties and some high strength tendons, used in
prestressed concrete, and hard-drawn wires, used in mesh, have lower fracture to yield
strength ratios and strains at the limit of uniform elongation.
Reinforcement in a nonlinear concrete analysis can generally be treated as either discrete or
smeared. Discrete reinforcement involves the inclusion of individual axial or axial-flexural
elements into the finite element mesh that model each layer of reinforcement explicitly. This
has the advantage of providing a one-to-one correspondence between the real structure and
the model and a corresponding lower likelihood of input errors. Smeared reinforcement
involves calculating an average stress-strain relationship that applies to the entire element area
and is included directly as part of the overall concrete element stiffness matrix.
Introduction of steel into a reinforced concrete analysis, particularly one that includes smeared
concrete tension stiffening stresses, should consider the effects of those concrete stresses on
the overall behaviour. In general, this can be done by adjusting the concrete tensile behaviour
at first yield at a crack or by adjusting the steel behaviour at first yield. Figure 3.15a) shows
the option of using bare-bar stress-strain relationships for the steel summed with a smeared
concrete tension stiffening stress. At first yield at the crack, the concrete tensile forces are
explicitly reduced to zero to ensure that the steel stress at a crack does not exceed yield. In
this example, it is conceptually assumed that strain hardening at a crack will not occur. Fig
3.15b considers the opposite option of leaving the concrete stress-strain relationship
unchanged and deriving new equations for the embedded steel behaviour. Both of these
methods produce the same overall answer in this example. In any case, it can be seen that
simply adding the average concrete tension stiffening stresses to the bare bar steel stresses
would violate equilibrium at a crack after first yield at a crack. Finally, special care must be
taken in analyses that involve non-ductile reinforcement such as glass or carbon fibre or lowductility steel reinforcement. In such situations, the increase in reinforcement stresses that
occur at the cracks will result in rupture when the average strain is still well below the rupture
strain of the material. Figure 3.16 shows a comparison of the observed components
contributing to the tension force for a steel bar embedded in concrete for three tests of Shima
et al. (1987).

3.3

Nonlinear concrete modelling framework

3.3.1

Elasticity

The most familiar continuum model is the elasticity model where loading and unloading are
path-independent and follow Hookes law (Figure 3.17). Structural steel before yielding and
structural concrete before cracking can be modelled well by such a linear elastic model.
Purely elastic models can be of arbitrary complexity, and need not require isotropic
behaviour. Orthotropic modelling of concrete has been found to work well when
consideration is given to the observed tension behaviour being dramatically different from
compressive behaviour in principal directions. The most general anisotropic modelling allows
all 21 independent elastic moduli in three dimensions to be defined.

98

3 Essential nonlinear modelling concepts

fc lowered to zero
by crack check

*fy

fc not reduced

*fy
1st yield at crack

stress

stress

1st yield at crack


fc
fcr

fc

average steel stress


less than yield

fcr

fs

fs
strain

strain

(a) Bare bar

(b) Embedded steel

Figure 3.15: Two approaches for checking local yield at a crack.

Load (MN)

0.3

Specimen No.6

Specimen no.5

Specimen no.4

Bare bar

0.2

Bare bar

Bare bar

0.1

Tensile force

Tensile force

Tensile force

Stress (GPa)

0.8

Bare bar
Bare bar

0.4

Stress (MPa)

Bare bar

Average stress of
reinforcement

Average stress of
reinforcement

Average stress of
reinforcement

Average stress of
concrete

Average stress of
concrete

Average stress of
concrete

Average strain (10-2)

Average strain (10-2)

Average strain (10-2)

Figure 3.16: Contributions to the tension force in a reinforced concrete element and corresponding stresses
(Shima et al., 1987).

One direct result from the analysis of elastic bodies is that the stress distributions around the
tips of cracks appear as real values in the output. Simple elastic theory for a fine crack
predicts a stress singularity at the tip of the crack, which is clearly impossible. Concrete
responds to this by the formation of a fracture process zone just ahead of the crack and
cohesive bridging stresses after cracking. Thus, even linear elastic models contain
approximations to reality after cracking.
The accuracy of deformation by linear analysis is far from the reality for cracked concrete
structures but in many cases of ultimate limit state verification, computed section forces by
linear analysis provide reasonably close approximations to the exact nonlinear solution. This
is because the accuracy of section forces does not primarily depend on the absolute values of
section stiffness but on the relative stiffness profile. This is the reason why the linear analysis
is often adopted as a computational tool for design so to obtain the section forces even for
ultimate limit state examination.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

99

stress

stress

stress

linear elastic strain


(recoverable strain)

elastic strain
recoverable strain but
irrecoverable stiffness

plastic hardening strain


(irrecoverable)

combination of basic concepts


RC modeling
+Y

Average shear stress

, YX
Y

+X
, XY

Average shear stress

LX

, (MPa )

1
0

Average shear strain

, (MPa )

) ( 10 -3 )

Average shear strain

) ( 10 -3 )

examples of computed stress-strain

Figure 3.17: Basic concepts of elasticity, plasticity, damage and their combination.

3.3.2

Plasticity

Plasticity allows for overall nonlinear behaviour by accumulation of irrecoverable strains. It is


generally defined as an elastic stiffness component summed with an accumulating plastic
strain component. The criterion of plasticity and elasticity is generally specified in terms of
stresses and the plastic evolution generally brings about hardening of metallic materials
(Figure 3.17). This is defined as plastic strain hardening. It is often used today with a flow
theory of plasticity, which defines the direction of plastic flow or a ratio of plastic strain
increment.
Perfect plasticity with no hardening is useful when solely determining a maximum capacity
for a given structural element. This capacity is derived by assuming that all the steel yields
and the solution is therefore relatively easy to obtain. In general, the ultimate strengths
obtained from the nonlinear analysis methods in this practitioners guide will be below or at
most equal to the strengths obtained from an assumption of perfect plasticity. Note that due to
the general inclusion of an elastic component coupled with a plastic component, the concepts
of stresses at crack tips, perhaps modelled with fracture mechanics of concrete, fit equally
well with plasticity models as with purely elastic methods.
3.3.3

Damage

Methods based on damage mechanics allow the elastic stiffness of the material to be reduced
as a result of accumulation of damage from strain excursions or repeated loadings. This

100

3 Essential nonlinear modelling concepts

nonlinearity may represent the loss of mechanical volume or degraded capacity to absorb the
elastic energy, which is thermodynamically reversible in nature. Concrete shows an
irreversible reduction of elastic unloading/reloading stiffness when small lateral confinement
is provided. When higher lateral confinement is applied to concrete in compression, damage
of the elasticity properties is observed in softening of the unloading and reloading paths.
Damage is also seen in concrete tension after cracking as described above.
In many practical cases, concrete nonlinearity in compression can be simply characterized by
plasticity although plasticity modelling cannot cover all observed nonlinear behaviours of the
continuum. Elasto-plasticity is the basis of limit state design and capacity computation with
yield-lines and/or reinforced concrete strips. When it is needed to simulate the post-peak
behaviour for verification of some limit states, a damage approach is useful for collapse
simulation and assessment of the remaining performance of a damaged structure.
3.3.4

Mixed models

Models based partly on elasticity with some basic concepts from plasticity and damage
mechanics can provide simple models that can capture many important nonlinear effects in
reinforced concrete materials and reinforced concrete structural elements. One of the simplest
examples is a no-tension linear elastic concrete model coupled with an elasto-plastic steel
model. This is a well known simple approach, which has been popularly used for flexural
behavioural computations of reinforced concrete sections based on Euler-Kirchoffs in-plane
assumption. For both confined and cracked concrete in compression, combinations of
elasticity, plasticity and damage can be consistently applied in structural analysis.
Smeared-crack in-plane modelling of reinforced concrete is a further example of combining
complex combinations of compression, tension and shear across cracks in a multi-dimensional
model and can include nonlinear elastic orthotropic behaviour of concrete coupled with an
elasto-plastic hardening behaviour of steel. Such mixed models often provide relatively
simple and reasonably expressive nonlinear concrete models.
3.3.5

Discrete modelling frameworks

Rather than treating concrete as a quasi-continuous material based upon space-averaged


constitutive models, it is also possible to treat it as a series of interacting discrete bodies for
structural analysis and design (e.g., Rots and Blaauwendraad 1989). This can be interpreted in
two ways. In the first way, a framed structure can be made from elastic elements with lumped
plasticity at their ends. With this type of discrete model, nonlinear behaviour is often defined
by predefined moment rotation models, or moment curvature relationships coupled with an
assumed plastic hinge length which has much to do with the post-buckling behaviour of the
reinforcing steel.
The second way of utilizing a discrete element model is for a continuous body where the
locations of the discontinuities (cracks) are not known a priori when the mesh is created. One
option is then to explicitly re-mesh the analysis space at each load step to follow crack
propagation in a more explicit way. For fracture simulation, Ingraffea and Saoumu (1984)
first introduced the discrete approach to concrete structures. The discontinuities of the
displacement field resulting from the failure processes are introduced directly into the
numerical model. New elements are added at the crack surface to explicitly account for
tension softening and aggregate interlock. In this approach, the crack location is instantly
recognizable but it can be a lengthy procedure especially for three-dimensional analyses.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

101

An alternate approach is to implicitly assume that all elements have crack-like gaps between
them from the start that may or may not be active. Methods such as lattice models fall into
this category (e.g., van Mier et al., 1995, Bolander and Saito, 1998). The discrete crack gap
can be numerically embedded in finite element volumes and the connectivity of these
embedded crack gaps can be satisfied on the element boundaries based upon hybrid stress and
strain hypotheses. Thus, differences between smeared and discrete approaches is becoming
less distinguishable for geometrically continuous domains and, while the methods were first
applied to the analysis of material level problems, large structural problems can also be
modelled as has been demonstrated by Bolander et al. (2000) and Bolander and Hong (2002).

3.4

Solution methods

As analysis of concrete structures usually involves non-linear modelling of materials and


possibly geometry, the analysis requires at least some basic knowledge of non-linear solution
strategies. The most commonly used solvers use a Newton or modified-Newton method to
solve for the system on non-linear equations obtained. The basis of Newton solution methods
are discussed below with the reader referred to Chapter 6 for some more advanced
approaches.
Let the quantities (qo , & p) denote the displacement and current load vector at a given point
on the equilibrium path, as shown in Figure 3.18, where & is a load magnification parameter
and p is a reference vector of applied loads. The equation governing equilibrium can then be
written as

^ p 7 f (q) ' 0

(3.1)

or

K( q )q 7 f (q) ' 0

(3.2)

where f is a vector of internal forces and is a function of the current displacement state.
Eqs. 3.1 and 3.2 describe the equilibrium state of the discretized structure. While the solution
of the linear equation system
Kq f = 0

(3.3)

can be calculated directly, this is not possible for non-linear systems. Solution techniques for
non-linear systems typically require the solution of linear systems repeatedly until
convergence is obtained with the most frequent schemes used based on a Newton method.
3.4.1

Newton-Raphson method

The most frequently used iteration scheme for the solution of non-linear equations is some
form of the Newton-Raphson procedure. In the case where Eq. 3.1 cannot be solved exactly, it
can be shown that (refer Chapter 6) the solution may be progressively calculated by adding
the change in displacements O qi to the current displacement state calculated from
71

Oqi ' K Ti r ( qi )

102

(3.4)

3 Essential nonlinear modelling concepts

i
where KT is the tangent stiffness determined for the current state and r (qi ) are the out of
balance forces. The Newton-Raphson solution process is illustrated in Figure 3.18, noting that
for every step the current stiffness matrix is formed and the linearized equations solved for
O qi .

Figure 3.18: Schematic representation of Newton-Raphson method.

3.4.2

Modified Newton-Raphson method

From the above calculations it is seen that the formulation of a new tangent stiffness for each
iterative cycle, and the solution of a new system of equations, must be undertaken. In
computing terms, this can be time consuming and costly. To overcome this cost the
approximation K Ti ' K To is often made. This modifies Eq. 3.4 to
71

O qi ' K To r ( qi )

(3.5)

and resolution of the same equation set is repeatedly used. The solution at each iteration is
sped up; however, more iterations to convergence are required, as can be seen in Figure 3.19.
The overall economy of the solution procedure is dependent on the problem size and nonlinear behaviour. An updated mNR approach may be adopted where the tangent stiffness
matrix, K Ti , is updated if convergence is not obtained after a pre-determined number of
iterations, n, and continues to be updated every n iterations following.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

103

Figure 3.19: Schematic representation of modified N-R method.

3.5

Precision of nonlinear concrete FE analyses

Upon seeing the results of analysis tools that take into consideration some or all of the
complexities outlined above in modelling of reinforced concrete structures, one may be
tempted to conclude that such a computer program will be able to produce a very precise
solution to a given problem. This is, of course, an unwise conclusion.
As an example of the variability that may be expected for some analyses, consider the data
shown in Figure 3.20 of measured stiffness values of concrete as a function of compressive
strength. Note the wide scatter in the observed values. Consider that at a concrete compressive
strength of 25 MPa, limestone aggregates have can produce concretes with stiffnesses varying
from 20,000 MPa up to 35,000 MPa. Calculations sensitive to the uncracked stiffness of
concrete, such as the displacements of a beam before cracking, can be expected to be similarly
variable. No structural analysis should be expected to provide extreme precision in results
whatever the complexity of the modelling.
Engineers should perform sensitivity analyses to ensure that a given set of results is not
simply a result of mesh sensitivity or unduly controlled by a parameter in which the engineer
has low confidence. For concrete, one also should take into account the difference of strength
between control specimens and structural concrete realized in the structures concerned. The
difference is caused by size effects, concrete placing work, shrinkage, temperature hysteresis
and so on.
When seeing a figure like Figure 3.20, proponents of high precision in nonlinear finite
element analyses may counter that such scatter simply indicates that the analysis was not
properly calibrated to the experiment that was to be modelled. Of course practicing engineers
104

3 Essential nonlinear modelling concepts

know that for most design or analysis problems, there is no experiment to calibrate to. Claims
that extremely precise fits between experiment and analysis are guaranteed should be treated
with caution and not used to justify dramatically lower safety factors, for example.
Lastly, it is important that the model being adopted be verified against benchmark tests of the
problem type being considered. Benchmarking testing is discussed in detail in Chapter 7.

Figure 3.20: Modulus of elasticity as a function of compressive strength (Rashid et al. 2002).

3.6

Safety and reliability

To ensure that safety and reliability criteria are met when using numerical analyses, the model
must first be validated by experiments and benchmark tests (see Chapter 7). The verification
process should include validation using:
"

basic material tests,

"

structural tests, and

"

mesh sensitivity tests.

The validation of material models are generally done by modelling of selected standardized
benchmark or control tests (for example, tests on reinforced panels or fracture of control sized
specimens). Alternatively, validation of the material models can be performed as a part of a
standardized numerical simulation task. An assessment of the capability of the material
models to perform, as required to simulate the behaviour of a structure, is normally
undertaken when selecting and assessing the appropriateness of the software package.
A validation of structural model is undertaken for a particular model and software package in
a special-purpose study undertaken by means of benchmark calculations. In this process,
specific types of structural and material behaviour are tested. Such studies form a rational
basis for choosing appropriate material models and software for modelling of a particular
structure, structural member or element loaded under similar conditions. For example, if a

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

105

shear wall is to be simulated under a particular loading regime for a given set of boundary
constraints, a validation of the finite element software against shear wall experiments should
be performed using test data identified as being reliable. In this process of verification, it is
vital that all the potential modes of failure are identified and the model verified to show that
these failure modes are captured with accuracy.
For performance assessment, some indicators are necessary to suitably characterize the
structural responses. In general, there can be three types of indicators; force/stress related
indicators, deformation/strain related indicators and energy related indicators (force
multiplied by displacement or stress by strain). The first two indicators are practically useful
for static design and the energy-based indicator for seismic design or damage control
(Akiyama, 1985). If our problem is a perfectly linear one, the same safety allowance can be
expressed no matter what sorts of indicators are used for the safety check, because force and
deformation are linearly correlated.
The safety checking scheme in most modern codes is based on limit states and associated
partial safety factors. In this scheme, the resistance or response is based on design (extreme)
parameters, which are derived from nominal ones using a reduction by the partial safety factor
#p. By this way, uncertainty of materials, geometry and other properties is included in input
data. Similarly, the design loads or actions are considered by extreme characteristic values,
where the load factors #load are considered. In a typical static situation, material characteristic
parameters are reduced, while loads are increased, both to design values. Safety of design is
checked locally at particular points, for example in beam sections, by the limit state condition
as
R(&p ) < E(&L )

(3.6)

where R(&p) is the extreme local resistance or capacity of the structural response and E(&L) is the
extreme local action of load or response indicator. This condition assures the safety of each
local point. It does not indicate a direct measure of global safety of the structure but it
generally results in conservative and safer assessment under static loads. This method is
popular in practice, since it is easy to apply and current partial safety factors based on linear
analysis are available. However, when we work with nonlinear problems, linear correlation of
force and displacement does not hold in nature. Some codes such as Eurocode 2 (2002)
explicitly mention that the partial safety factor method is not applicable to nonlinear analysis.
The problem of applying the method of partial safety factors to non-linear modelling is
demonstrated using two extreme cases. Nonlinear forces versus deformation relationships (left
side of Figure 3.21) are common in reinforced concrete mechanics. For example, we can
imagine relationship between the transverse shear force and the strain in the stirrup
reinforcement. If the strength of concrete and amount of web reinforcement are changed, we
have different static force versus deformation relationships, as shown in Figure 3.21. Let us
consider the steel yield as a limit state safety check and an allowable safety margin in terms of
strain (i.e., a deformation related indicator). Here, the corresponding safety allowance in terms
of shear (force related indicator) differs greatly. Since a force indicator is roughly
proportional to external static loads, it is thought reasonable to specify the safety allowance
(safety factor) by the force-based index for expressing reliability of nonlinear analysis to
identify the limit state concerned. In fact, limit state design methods do not adopt stirrup strain
but apply sectional shear force as the limit state indicator for the static safety assessment.

106

3 Essential nonlinear modelling concepts

_R

Rcap

static force, stress : R

static force, stress : R

_R
small
allowance
ratio

strain, displacement based index

RC/PC nonlinear response


toward external force/stress

Rcap
strain, displacement based index

static force, stress : R

static force, stress : R

strain, displacement based index

_R

_R

strain, displacement based index

Elasto-plastic response
toward external force/stress

Figure 3.21: Safety factor for nonlinear structural problems.

For the case of elasto-plastic problems like steel structures, however, any indicator identifies
the same safety allowance (see Figure 3.21) when the initiation of plasticity is selected as the
limit state because of the proportionality of internal/external forces and deformations. Then,
in the discussion of this chapter, the safety allowance or variation is depicted by force related
indicators. There exists similar feature on fatigue limit state design for service life.
Nonlinear analysis is typically a simulation of a loading test for an entire load history up to
failure, including the serviceability and ultimate limit states. In current codes there are
different partial safety factors for different load stages and failure modes. It is difficult to
apply current safety factors according to their definitions in nonlinear analysis. A material
model based on the partial safety factors represents an imaginary, not real, material. It does
not represent an extreme behaviour with a certain probability of failure. Automatic reduction
of all parameters may cause much weaker material than required by the safety concept. In
some cases, it can be an unsafe model. Thus, considering the design values of material
parameters in nonlinear analysis does not guarantee that the target safety values, as prescribed
and calibrated by standards, are achieved.
Let us first consider a simply supported beam under uniform loading, where the bending
moment in the mid-section is completely independent of material behaviour. If we use a beam
finite element with plane section hypothesis, the nonlinear analysis is identical with the
current design methods (based on elastic assessment of internal forces) and the partial safety
factors can be applied. If we consider a statically indeterminate structure, the actions in
sections or material points depend on material behaviour and need not be proportional to the
magnitude of the external load. In general, a redistribution of internal forces due to nonlinear
behaviour can produce either positive or negative effects on local failure. An example of such
situation is concrete under a confining pressure where higher lateral pressures provide
increases in strength. In such cases, the application of partial safety factors is not justified.
Since the strong nonlinear behaviour can be expected in most of reinforced concrete
structures, it is concluded that the partial safety factor approach should be used with care in
statically indeterminate structures, especially when local post-yield states are allowed in the
design.
!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

107

Recently, static pushover analysis has been carried out in use of nonlinear FE analysis; not
only for seismic design but also for static safety design of statically indeterminate structures
composed of several members. The pushover analysis may present the whole structural
capacity under static loads and the scenario of collapse after the capacity limit state. Figure
3.22 shows a schematic of pushover analysis and the design value of loads. Up to the peak
capacity of the global structural system, some component members may fail or locally exceed
their capacity. Thus, the computed characteristic capacity of the global structural system
should be factorized by a global safety factor with consideration to the accuracy of the
nonlinear structural analysis. This factorized design value of the global system capacity has to
be greater than the overall design static load (vector). The performance assessment in terms of
the global system safety is advantageous rather than the conventional methods where all
constituent elements are required not to exceed their ultimate limit state. This means that
failure of any member is not accepted, even if the global structural stiffness remains positive.
When section failure of structural members is itemized as the limit state for structural safety,
some indicators to judge the occurrence of flexural and shear failures are needed. The
combined axial and flexural failure can be detected by monitoring the longitudinal
compressive strains of the most extreme fibre of the section, as shown in Figure 3.23. When
2D or 3D solid elements are used, point-wise local strains tend to be dependent on the sizes of
the finite elements and detailed patterns of mesh discretization (Chapters 3 and 6). Then, a
space averaged mean strain or stress indicator (sectional forces) is preferable as a design
index because the space-averaged indicator is not sensitive to the detail of the mesh
discretization.
Another way is to calculate sectional forces by integrating computed local 2D and/or 3D
stresses. This is an averaging of stress field and is compared with the sectional capacity
computed in advance by design formula or by the nonlinear analysis. Discussed is the shear
deformation based indicator to recognize the shear mode of failure accompanying diagonal
cracks. Shear deformational intensity along cracking is supposed to be a candidate but
magnitude of limit value is not uniquely decided because the shear slip/deformation response
depends on the axial compression force, dimensioning and size, etc. The axial mean
deformation of members is also expected to be an indicator for shear failure.
Regarding the restorability of structures after extreme loads, material damage based indicators
are also proposed, especially in seismic designs for practice (JSCE, 2002). Damage induced in
structural concrete has a close correlation with repair/strengthening costs. Figure 3.24 shows
an example of design modelling for concrete used in fibre modelling (Chapter 4). The local
damage indicator denoted by F is assumed to be a function of maximum experienced
compressive strain in a load- or time-step run of a nonlinear structural analysis with nonlinear
material models. The sectional averaged F value is used as a damage indicator in regards to
restorability after the application of extreme loads. The transverse displacement of the
columns corresponding to F = 0.5 is almost the same as the allowable ductility level specified
in the past design codes based on empirical macroscopic models of members. Confinement
(Chapters 3, 4 and 6) by lateral ties and steel is taken into consideration by modifying fibre
stress-strain relationships. As the lateral confinement moderates the damage intensity, the
damage evolution indicator F can be modified as shown in Figure 3.24(3). The damage
indicator was formulated to represent reduction ratio of the elastic stiffness on unloading
paths and is a measure of how much reversible strain energy is stored in the damaged
continuum.

108

3 Essential nonlinear modelling concepts

force (vector)

computed characteristic
structural capacity

design value
of loads

design value of
capacity

global safety
factored in terms of
analysis model used

partial safety margin factored


in terms of failure/collapse modes
importance of structures, etc.

force (vector)

displacement

computed characteristic
structural capacity

design value
of loads

A member reaches the limit state


in terms of stress/strain/section forces
design value of
capacity

partial safety
factored in terms of
failure/collapse modes, etc.

global safety factored


in terms of analysis
model used

displacement

Figure 3.22: Safety factor and the whole structural limit state by push-over analysis.

Axial-Flexural failure of members (safety assessment)

moment

index: bending capacity of the section (fiber/shell),


extreme fiber strain (2D-3D, fiber/shell), reinforcement strain/stress

moment

curvature

curvature

Transverse shear failure of members (safety assessment)


index: out-of-plane shear capacity (fiber/shell)
shear strain along cracking (2D-3D), strain of web reinforcement (2D-3D)
cracking pattern (2D-3D), comp. mean strain along member axis (2D-3D)
axial mean strain
of column

shear failure
occurs

drift angle of column

In-plane shear failure of members (safety assessment)


index: stress of reinforcement
compressive concrete strain along cracking, or principal axis
(shell, 2D-3D solids), in-plane shear capacity

Figure 3.23: Flexural and shear failure indicator of beam/column members.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

109

F > 0.5

Design damage indicator K

(2)

(1)

confined
cases

(3)

(4)

(5)

(6)

(7)

Figure 3.24: Structural concrete damage based indicator (Tsuchiya and Maekawa, 2006).

110

3 Essential nonlinear modelling concepts

In case of shell structures subjected to coupled in-plane and out-of-plane actions, cracking of
concrete and yielding of reinforcement are likely to develop over a wide domain of the
structures because the reinforcement is uniformly distributed and flow of stress tends to
expand over the structure. Then, the structural response level at the start of reinforcement
yield is well below the capacity. In these types of structures, compressive localization can be a
design limit state for safety and serviceability performance (Chapter 6). Figure 3.25 shows a
transient process of compressive localization after cracking of the concrete and yielding of the
reinforcement in shells. When the principal compressive strain averaged within the control
volume reaches the limit value (nearly 5000B)), some elements exhibit high strain rate while
the strain rate of neighbouring elements decreases. In case of circular shells subjected to shear
actions, a localized band accompanying shear-compression kinematics can be seen when the
principal compressive strain reaches this limit state. Then, for underground tank design, the
averaged principal compressive strain is now used as the indicator for damage control with
regard to compressive localization. This limit state almost corresponds to the structural
capacity as shown in Figure 3.25. The rationale of this criterion was checked with
reinforcement volumes from 0.4% to 2.0% (Harada et al., 2001).
The global safety condition can be written with nonlinear-based indicators as
Rm / & R < E(&L )

(3.7)

where Rm is the resistance or the capacity limit obtained by nonlinear analysis, as stated
above, based on the mean material characteristic parameters, &R is the global safety factor of
the corresponding resistance and E(&L) is the factorized external action or the response
indicator of the structural system concerned.
The global safety factor describes the safety of the system on a global level. Thus, it should
cover the uncertainties of all components of the structural system and analysis. Since safety is
related to the average resistance, it can be described by a global central safety factor. This
represents a generally accepted safety margin of usual structures produced according to
general standards. This covers wide and rather unspecified range of uncertainties with the
safety not related to specific random properties of a given structure or product. The practical
determination of global safety factor is more difficult comparing to partial safety
specification, where it is related to known random variations of dimensions and material
properties.
Therefore, the values of global safety factors are not yet proposed in the majority of codes.
For example, Eurocode 2 (2002) states that the partial safety factor concept is not applicable
to the nonlinear analysis, and does not propose a format for safety check. DIN 1045 (1998)
proposes the global safety factor for the system resistance & R = 1.3 valid only for ductile
modes of failure. Higher values should be considered in case of brittle models of failure such
as concrete shear and diagonal tension, or compression. The nonlinear analysis considered in
practical codes is mostly limited to the models based on beam-column systems. The mean
material parameters are based on nominal values that, typically, represent 5% probability.
e qud
c & gl
d

& Rd E $& G G 8 & QQ % ? Rc

b
e
` or & Sd & Rd E & g G 8 & q Q ? Rc qud
`
c & gl
a
d

b
`
`
a

(3.8)

where, & G, & Q are the global safety coefficients respectively for permanent and variable loads,

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

111

Principal compressive strain (B)

270

Element No.45

-10000

Element No.57
-7500

180

-5000
Maximum load

-2500

90
90

-1.0

;
<<
0.0

1.0

2.0

3.0

4.0

Deformation angle (%)


@!!!

Maximum Load 2871kN

Rebar Yielding 1637kN

?!!!
"

C4)8+2:D 5

%!!!

Diagonal Cracking
931kN

E
F-):

!"

E
F-):

Collapse Load
2352kN

$!!!

Flexural Cracking
735kN

>%"!

>$"!

!"!

$"!

%"!

?"!

@"!

#"!

>$!!!
Test

>%!!!

Analysis COM3
>?!!!
A -941, )0(4.+&.7*-+2B5

crack pattern and


deformation localization
in the web of RC shell tank
subjected to lateral force

Figure 3.25: Structural concrete damage based indicator (Harada et al., 2001).

& Rd is the model uncertainty coefficient on the resistance side (suggested value gRd = 1.06), & Sd
is the model uncertainties coefficient on the action side (suggested value gSd=1.15), & gl is the
global structural safety factor (suggested value gg = 1.20, but gRd ggl = 1.27 = gGl), qud is the
ultimate level of the internal actions path, reached in the incremental process of nonlinear
analysis. For the case where no model uncertainties are considered, the inequalities in Eq. 3.8
are modified to

eq
E & G G 8 & QQ ? Rcc ud
d & Gl

b
``
a

(3.9)

The JSCE (1999) LNG-Tank design code presents full 3D nonlinear analysis of soil-RC shell
interacting systems and proposes a factor of 1.3 for the global safety factor in regard to the
global deformational indicator, as shown in Figure 3.25. Here, we input the most probable

112

3 Essential nonlinear modelling concepts

values for material characteristic parameters of structural concrete and soil foundation.
According to current practice, at serviceability limit states, we can use directly for the mean
response & R = 1.0, and unfactored loads or actions. At ultimate limit states, the global safety
factor should be & R < 1.3, depending on the ductility of the response.
Concerning the ultimate limit state, Mancini (2002) and Bretagnoli et al. (2004) proposed for
nonlinear static analyses that for the evaluation of qud, the analysis should be stopped when
the ultimate strength and the corresponding deformation are reached within the most critical
region and where the whole structure is incapable of supporting any further load increments.
An example of deep beam capacity is shown in Figure 3.26. During the analysis, the first
critical element initially crushed but the structure was able to carry further load increments up
to the crushing of a second element. At this point the model was unable to reach the
equilibrium for any further load increments. This last step has been considered as the final
point of the internal actions path. Figure 3.26 shows the resisting interaction surface for the
critical second element, and the internal action path in the same element, up to the intersection
with the resisting surface. Figure 3.26 also illustrates the procedure for the application of
safety format in a vectorial combination of internal actions and, by means of definition of a
safety interaction surface, derived by the limit one by linear transformation referred to the
axes origin. If the critical element and its internal paths are not clearly identified, the external
loads magnified by the global safety factor can be applied to the structural concrete. If the
magnified loads would not cause failure of the structure, it implies the design satisfies the
safety requirement.

Critical
Element 1

450
400

P [kN]
Numerical
Experimental

350
300
250
200

Reinforcement

x
A

150
Critical
Element 2

100
50
0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
x [m]

Figure 3.26: Application of safety format of strength limit state of internal actions of a deep beam (Mancini
2002, Bretagnoli et al., 2004).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

113

In the draft Australian concrete structures standard (DR 05252, 2005), for new designs and/or
assessment of existing structures, a system safety approach is used to ensure that the structure
(or structural element) meets expected safety demands. The draft adopts the following stress
check procedure for use with non-linear frame and stress analysis:
It shall be confirmed that the design resistance of the structure or the component
member is equal to or greater than the design action -

Rd < Ed
where Rd is the design capacity of the structure based on mean strengths of
materials and Ed is the critical combination of factored actions (for example, for
dead and live load combinations the worse case of 1.25G + 1.5Q, 0.8G + 1.5Q
and 1.35G, where G = dead load and Q = live load).
The design capacity of the structure is limited by

Rd ' C sys Ru.sys


where Ru.sys is the capacity of the structure and is determined for the same
combination of actions adopted for Ed with mean values of material properties,
and C sys is a system reduction factor and is dependent on the mode of failure.

The values for C sys adopted by the draft standard are dependent on the type of failure. For
structural systems in which the deflections and local deformations at high overload are an
order of magnitude greater than those for service conditions; and yielding of the
reinforcement occurs well before the peak load is reached, C sys = 0.7. Otherwise, C sys = 0.5
(although larger values than 0.5 may be used if it can be shown that, at high overload,
adequate warning is given of impending collapse).

3.7

Statistical analyses

Statistical or risk analysis is the most conceptually rational method of safety assessment in the
present state-of-the-art. The analysis includes deterministic and probabilistic domains. The
basic structural model is generated as a deterministic one with mean (central) parameters.
Certain parameters of this model are assumed to be random variables. They can be material
parameters, dimensions, etc. Based on these input data, the structural response can be
obtained in a statistical form in which the state variables of the response, such as ultimate
load, deflection or stress state in a point, are described by a random distribution (with mean,
standard deviation and other parameters). The safety margin can be formulated by comparing
the response R and actions E. The probability of failure pf is defined as a probability of Z < 0:

Z = R - E, pf = p(Z < 0)

(3.10)

The method is illustrated in Figure 3.27.

114

3 Essential nonlinear modelling concepts

Figure 3.27: Statistical analysis of probability of failure.

This method allows an assessment of the response under given loading conditions and with
consideration of random nature of the input parameters and can be used in a rationally based
assessment of global safety. Solution of this problem can be performed numerically
combining the structural and statistical analyses. The numerical model of structure is based on
a deterministic nonlinear analysis using the finite element method. The probability of
distribution of the response can be obtained by numerical methods based on random
sampling. In these methods, the random variables of samples are generated by statistical
methods and sample response is realized by a nonlinear solver. Finally, the statistical
parameters of the response are analyzed again by statistical methods. An example of this
approach, applied to bridges, is presented in Bergmeister (2002).
The advantage of the above approach is that the reliability can be rationally evaluated by
failure probability or by a safety index. Both safety measures are well justified in reliability
engineering and prescribed by standards (eg., Eurocode 1, 2002). The resulting safety margin
(global safety factor) is based on actual input parameters, their random variation and their
mechanical relevance.

3.8

Concluding remarks

While non-linear finite element modelling can be an extremely useful and powerful approach
in determining the behavioural response of complex concrete structures, extreme care is
needed in the setting up of the models, in the verification of the model, in assessing the
models capability to correctly identify critical behaviour and in the interpretation of results.
For this, experience is needed in both computational modelling and in design and construction
of concrete structures. The output of FE models should never be considered in isolation of the
problem being investigated and a prudent engineer will always have predicted the results
using simple calculations and experience before the model is run. Finally, equilibrium checks
on the input/output are, of course, essential.

3.9

References

Akiyama, H. (1985), Earthquake-Resistant Limit-State Design for Buildings, UT-Press,


Tokyo.
Alca, N., Alexander, S.D.B., MacGregor, J.G., (1997) Effect of size on flexural behavior of
high-strength concrete beams, ACI Structural Journal, 94(1), pp. 59-67.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

115

Bergmeister, K., et. al. (2002), Structural analysis and safety assessment of existing concrete
structures, Proceedings of fib Congress - Concrete Structures in 21 Century -, Osaka, Japan,
Session 11, 47-54.
Bretagnoli, G., Carbone, V. I., Giordano, L. and Mancini, G. (2004), Safety format for nonlinear analysis, Proceeding of the fib Symposium, Avignon, France.
DR 05252 (2005), Concrete Structures, Draft for Public Comment Australian Standard,
Standards Australia, Revision of AS3600-2001.
Balmer, G.G. (1949), Shearing strength of concrete under high triaxial stress-computation of
Mohr's envelope as a curve." Tech. Rep. No. SP-23, Structure Research Laboratory, Denver.
Baant, Z.P. and Oh, B.H. (1983), Crack band theory for fracture of concrete, Materials and
Structures, 16 (1983) 155-177.
Baant, Z.P. and Planas, J. (1998), Fracture and Size Effect in Concrete and Other
Quasibrittle Materials, CRC Press, Boca Raton.
Bentz, E.C. (2005), Explaining the riddle of tension stiffening models for shear panel
experiments, ASCE J. of Struct. Engng., 131(9), pp. 1422-1425.
Bolander, J.E., and Saito, S. (1998) Fracture Analysis Using Spring Networks with Random
Geometry, Engineering Fracture Mechanics, 61, pp. 569-591.
Bolander, J.E., Hong, G.S., and Yoshitake, K., (2000). Structural Concrete Analysis Using
Rigid-Body-Spring Networks, Computer-Aided Civil and Infrastructure Engineering, 15, pp.
120-133.
Bolander, J.E. and Hong, G.S. (2002). Rigid-Body-Spring Network Modeling of Prestressed
Concrete Members, ACI Structural Journal, 99(5), Sept-Oct, pp. 595-604.
Cervenka, V. and Gerstle, K. (1971, 1972), Inelastic analysis of reinforced concrete panels:
(1) Theory, (2) Experimental verification and application, Publications IABSE, Zrich, V.3100, 1971, pp.32-45, and V.32-II, 1972, pp.26-39.
CEB-FIP Model Code 1990 (1993), Thomas Telford Services, Ltd., London, for Comit
Euro-International du Bton, Bulletin dInnformation No. 213-214, Lausanne, pp.437.
Collins, M.P., (1978), Towards a rational theory for RC members in shear, ASCE J. of
Struct. Engng., 104(4), pp. 649-666.
DIN 1045-1 (1998), Tragwerke aus Beton, Stahlbeton und Spannbeton. Teil 1: Bemessung
und Konstruktion.
Eurocode 1 (2002), Actions on structures - Part 1-1: General actions -densities, self weight,
imposed loads for buildings. CEN/TC 250/SC1.
Eurocode 2 (2002) Design of concrete structures - Part 1: General rules and rules for
buildings. CEN. 2nd draft. 2002.
Foster, S.J., and Marti, P. (2002). FE Modelling of RC Membranes Using the CMM
Formulation, Proceedings of the Fifth World Congress on Computational Mechanics

116

3 Essential nonlinear modelling concepts

(WCCM V), July 7-12, Vienna, Austria, Editors: Mang, H.A.; Rammerstorfer, F.G.;
Eberhardsteiner, J., Publisher: Vienna University of Technology, Austria, ISBN 3-9501554-06, http://wccm.tuwien.ac.at
Foster, S.J., and Marti, P. (2003). Cracked Membrane Model: FE Implementation, ASCE,
Journal of Structural Engineering, V129, N9, September, pp. 1155 1163.
Gilbert, R.I., and Warner, R.F. (1978), Tension stiffening in reinforced concrete slabs,
ASCE J. of Struct. Engng., 104(ST12), pp. 1885-1900.
Goto, Y. (1971), Cracks formed in concrete around deformed tension bars, ACI Journal
Proceedings, 68(4), pp. 244-251.
Harada, M., Onituka S., Adachi, M. and Matsuo, T. (2001), Experimental study on
deformation performance of cylindrical reinforced concrete structure, Proc. of Japan
Concrete Institute, 23(3).
Harajli, M.H. (2004), Comparison of bond strength of steel bars in normal and high strength
concrete, J. of Materials in Civil Engng., 16(4), pp. 365-374.
Ingraffea, A. R. and Saouma, V. E. (1984), Numerical modelling of discrete crack
propagation in reinforced and plain concrete, in fracture mechanics of concrete, Structural
Application and Numerical Calculation, Sih and de Tomaso (Ed), Martinus Nijho_ Publ.
JSCE (1999), Recommendation for Structural Performance Verification of LNG
Uuderground Storage Tanks, Japan Society of Civil Engineers, Concrete Library, 98.
JSCE (2002), Standard Specification of Concrete Structures Structural Performance
Verificaiton , Japan Society of Civil Engineers.
Kaufmann, W. and Marti, P. (1998), Structural concrete: cracked membrane model", ASCE
J. of Struct. Engng., ASCE, 124(12), pp.1467-1475.
Kent, D.C., and Park, R. (1971), Flexural members with confined concrete, ASCE J. of
Struct. Engng., 97(ST7), pp. 1969-1990.
Kim, J.-K., Yi, S.-T., Yang, E.I. (2000), Size effect on flexural compressive strength of
concrete specimens, ACI Structural Journal, 97(2), pp. 291-296.
Kupfer, H., Hilsdorf, H.K., and Rusch, H. (1969), Behavior of concrete under biaxial
stresses, ACI Journal Proceedings, 66(8), pp. 656-666.
Marti, P., Alvarez, M., Kaufmann, W., and Sigrist, V. (1998). Tension Chord Model for
Structural Concrete, Structural Engineering International, IABSE, 4/98, 287-298.
Maekawa, K., Pimanmas, A. and Okamura, H. (2003), Nonlinear Mechanics of Reinforced
Concrete, Spon Press, London.
Maekawa, K., An, X., and Tsuchiya, S. (1999), Application to fracture analysis of concrete
structures, recent development in mechanics of fracture for concrete structures, Concrete
Journal, Japan Concrete Institute, 37(9), pp.54-60.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

117

Maekawa, K. and An, X. (2000), Shear failure and ductility of RC columns after yielding of
main reinforcement, Engineering Fracture Mechanics, 65, pp. 335-368.
Manacini, G. (2002), Nonlinear analysis and safety format for practice, Proceeding of the
First fib Congress, Osaka, Japan, pp. 53-58.
Morita, S. and Kaku, T. (1979), Splitting bond failure of large deformed reinforcing bars,
ACI journal, Proceedings Vol.76, pp.93-110.
Okamura, H. and Maekawa, K., Nonlinear Analysis and Constitutive Models of Reinforced
Concrete, Gihodo-Shuppan Co. Tokyo, 1991.
Ottosen N. S. (1977), A failure criterion for concrete, Journal of Engineering Mechanics,
ASCE, 103, pp.527-535.
Ozcebe, G., Ersoy, U., Tankut, T. (1999), Minimum flexural reinforcement for T-beams
made of higher strength concrete, Canadian J. of Civil Engng., 26(5), pp. 525-534.
Petrangeli, M. and Obolt, J. (1996), Smeared crack approaches - Material modelling,
ASCE J. of Eng. Mech., 122(6), pp. 545-554.
Rashid, Y. R. (1968), Analysis of prestressed concrete pressure vessels, Nuclear
Engineering Design, Vol. 7(4), pp.334-344.
Rashid, M.A., Mansur, M.A., Paramasivam, P. (2002), Correlations between mechanical
properties of high-strength concrete, J. of Materials in Civil Engng., 14(3) pp. 230-238.
Richart, F.E., Brandtzaeg, A., and Brown, R.L. (1928), A Study of the Failure of Concrete
Under Combined Compressive Stresses, University of Illinois Engineering Experimental
Station, Bulletin No.185, 104 pp.
Rots, J. G. and Blaauwendraad, J. (1989), Crack models for concrete: discrete or smeared?
Fixed, multi-directional or rotating?, HERON, 34(1).
Scott, B.D., Park, R., and Priestley, M.J.N. (1982), Stress-Strain Behavior of Concrete
Confined by Overlapping Hoops at Low and High Strain Rates, ACI Journal Proceedings,
79(1), pp. 13-27.
Shima, H., Chou, L. and Okamura, H. (1987), Micro and macro model for bond behavior in
reinforced concrete, Journal of The Faculty of Engineering, The University of Tokyo (B),
39(2), pp.133-194.
Tsuchiya, S. and Maekawa, K. (2006), Cross-sectional damage index for RC beam-column
members subjected to multi-axial flexure, Journal of Advanced Concrete Technology, 4(1)
179-192.
Turneaure, F.E., Maurer, E.R. (1908), Principles of Reinforced Concrete Construction, 1st
Edition, John Wiley and Sons, Inc, New York.
van Mier, J. G. M., Schlangen, E. and Vervuurt, A. (1995), Lattice type fracture models for
concrete, Chapter 10 in Continuum Models for Materials with Microstructure (Mhlhaus,
H.B., ed.), John Wiley & Sons, Chichester, pp.341-377.

118

3 Essential nonlinear modelling concepts

Vecchio, F. and Collins, M. P. (1993), Compression response of cracked concrete, ASCE J.


Struct. Engrg., 119(12), pp. 3590-3610.
Vecchio, F. and Collins, M. P. (1986), The modified compression field theory for reinforced
concrete elements subjected to shear, ACI Journal, 3(4), pp. 219-231.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

119

[Blank Page]

Analysis and design of frame structures using non-linear


models

4.1

Introduction

In this chapter nonlinear models to be adopted in 1-D finite element procedures and/or fibre
analysis procedures for the analysis of reinforced concrete structures are treated, with the aim
of:
"

performing (conventional and seismic) assessment on existing buildings,

"

assessing rehabilitation interventions on inadequate constructions,

"

identifying (potential) causes of damage or collapse in buildings.

These models can be used in order to determine the reserve strength and deformation capacity
of existing healthy or damaged buildings. Once existing capacities are known along with
anticipated strength and ductility demands, then a structural repair or upgrade that is
consistent with the existing deficiencies of concrete structures and expected loading
conditions can be devised.
The determination of the structural behaviour of a reinforced concrete structure is based on a
correct modelling of the constituent materials, that is, the concrete, the steel and the bond
between each. Constitutive models for these materials presented in this section can be used to
clearly identify the structural capacity that can be compared with both local and global
demands. A simple example of one such case would be a reinforced concrete column
damaged during strong ground motions.
When performing accurate nonlinear analysis that leads the structure to its ultimate state, it is
fundamental to rely on detailed models that can capture the strong nonlinear behaviour of
concrete and steel reinforcing bars. As an example, reinforcing bar buckling models,
presented in this chapter, can be used in conjunction with confinement models for concrete to
quantify the available strength and ductility of a reinforced concrete element, also under
cyclic action. In these cases, both the behaviour of buckled bars and the response of confined
concrete must be modelled accurately, so that the remaining rotation capacity of a plastic
hinge and strength of earthquake-damaged columns can be correctly evaluated.
Retrofit of a corrosion-damaged concrete column would be another typical example where the
models presented in this chapter can be used. The capacity reduction in concrete columns can
be evaluated based on the amount of reinforcing bar loss due to corrosion. Once this reduction
in the capacity is determined, the models presented in this section can be used in determining
the feasibility of various repair techniques that can be used.
This Chapter is organised in two sections: Section 4.3 Nonlinear Models of Frame Elements,
and Section 4.4 Interpretation of Results. For material modelling, the reader could refer to
Chapters 3 and 6, especially to the section where issues relevant to modelling of the materials
concrete and steel are treated. Also, long-term effects of ageing and distress in both concrete
and steel should be looked at as well as buckling of steel bars and details of lap splices.
Section 4.3 Nonlinear Models of Frame Elements is devoted to the comparison among
different approaches in modelling the response at the element level. Fibre elements are
compared to strut-and-tie type of models. Modelling of shear and of bond-slip problems are
given particular attention.
!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

121

In Section 4.4 Interpretation of Results, important issues related to the localisation problem
are treated, both for displacement- and force-based frame elements.

4.2

Notation

! $ x%

array containing the derivatives of the shape functions NU(x)

D bi $ x %

bond interface force between beam and bar i

E
e ( x)
F
f ( x)
G
I
K
LIP
Lp

Youngs modulus
section deformations vector
element flexibility matrix without rigid body modes
section flexibility matrix
shear modulus
moment of inertia
element stiffness matrix
length pertaining to an integration point
plastic hinge length

M
M B $ x%

bending moment
beam section bending moment

NB $ x%

vector of axial forces in the beam

Ni $ x %

vector of axial forces in bar i

NP $x%

force interpolation functions array

NU $ x %

shape functions

P
py $ x %

element resisting forces


transverse distributed load

V
VB $ x %

shear
beam section shear force

s ( x)
U
u(x)
ub i $ x %

force distributions vector along the element


nodal displacement vector
section displacements vector
bond slip between beam and bar i

u0

longitudinal displacement at the cross-section reference axis

v0

transverse displacement at the cross-section reference axis

wh

weight factor

yi

distance of bar i from element reference axis

:0

cross-section deformation due to bending

&

cross-section deformations due to shear

)0

axial strain at section centroid

f
-

section curvature
section rotation

122

4 Analysis! and design of frame structures using non-linear models

4.3

Nonlinear models of frame elements

4.3.1

Lumped versus distributed plasticity

Because the inelastic behaviour of reinforced concrete frames often concentrates in certain
sections (typically the ends of girders and columns for seismic loads and midspan for
distributed, static loads), one option for modelling the nonlinear behaviour of reinforced
concrete frames is to used a lumped plasticity model, where nonlinear zero-length springs are
placed at the critical points, connected by linear elastic elements. Depending on the
formulation these models, they may consist of several springs that are connected either in
series or in parallel.
The earliest parallel component element was introduced by Clough and Johnston (1966) and
allowed for a bilinear moment-rotation relationship: the element consists of two parallel
elements, one elastic-perfectly plastic to represent yielding and the other perfectly elastic to
represent strain-hardening. The stiffness matrix of the member is the sum of the stiffnesses of
the components. Takizawa (1976) generalized this model to multi-linear monotonic behaviour
allowing for the effect of cracking in reinforced concrete members.
Filippou and Issa (1988) subdivide the element in different, parallel, sub-elements. Each subelement describes a single effect, such as inelastic behaviour due to bending, shear behaviour
at the interface or bond-slip behaviour at the beam-column joint. The interaction between
these effects is then achieved by the combination of sub-elements. This approach allows the
hysteretic law of the individual sub-element to be simpler, while the member still exhibits a
complex hysteretic behaviour through the interaction of the different sub-elements.
The series model was formally introduced by Giberson (1967), although it had been
reportedly used earlier. Its original form consists of a linear elastic element with one
equivalent nonlinear rotational spring attached to each end. The inelastic deformations of the
member are lumped into the end springs. This model is more versatile than the original
Clough model, since it can describe more complex hysteretic behaviour by the selection of
appropriate moment-rotation relationships for the end springs. This makes the model
attractive for the phenomenological representation of the hysteretic behaviour of reinforced
concrete members.
Several other lumped plasticity constitutive models have been proposed. Such models include
cyclic stiffness degradation in flexure and shear (Clough and Benuska, 1966, Takeda et al.,
1970, Brancaleoni et al., 1983), pinching under reversal (Banon et al., 1981, Brancaleoni et
al., 1983) and fixed end rotations at the beam-column joint interface due to bar pull-out
(Otani, 1974, Filippou and Issa, 1988). Typically, axial-flexural coupling is neglected.
Nonlinear rate constitutive representations have also been generalized from the basic
endochronic theory formulation in Ozdemir (1981) to provide continuous hysteretic
relationships for the nonlinear springs. An extensive discussion of the mathematical functions
that are appropriate for such models is given by Iwan (1978).
The dependence of flexural strength on the axial load under uniaxial and biaxial bending
conditions has been explicitly included in the modelling of beams and structural walls. In
most lumped plasticity models, the axial force-bending moment interaction is described by a
yield surface for the stress resultants and an associated flow rule according to the tenets of
classical plasticity theory (Prager and Hodge, 1951). The response is assumed to be linear for
stress states that fall within the yield surface in which case the flexural and axial stiffness of
the member are uncoupled and independent of the end loads. With the introduction of

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

123

multiple yield and loading surfaces and corresponding hardening rules, multilinear
constitutive representations that include cracking and cyclic stiffness degradation are possible
for the springs, as was originally suggested by Takayanagi and Schnobrich (1979).
A lumped model is a simplification of the actual behaviour that involves the gradual spread of
inelastic deformations into the member as a function of loading history. This modelling
deficiency was recognized in several correlation studies, particularly those related to large
resisting elements of flexural wall-frame structures (Charney and Bertero, 1982, Bertero et al.,
1984). The basic advantage of the lumped model is in its simplicity in that it reduces storage
requirements and computational cost and improves the numerical stability of the
computations. Most lumped models, however, oversimplify certain important aspects of the
hysteretic behaviour of reinforced concrete and are, therefore, limited in their applicability.
One such limitation derives from restrictive a priori assumptions for the determination of the
spring parameters. Parametric and theoretical studies of girders under monotonic loading by
Anagnostopoulos (1981) demonstrate a strong dependence between model parameters and the
imposed loading pattern and level of inelastic deformation. Neither factor is likely to remain
constant during the dynamic response. The problem is further accentuated by the fluctuation
of the axial force in the columns. Because of this history dependence, damage predictions at
both the local and the global level may be grossly inaccurate. Such information can only be
obtained with more refined models capable of describing the hysteretic behaviour of the
section as a function of axial load. Another limitation of most lumped plasticity models,
proposed to date, is their inability to describe adequately the deformation softening behaviour
of reinforced concrete members. Such deformation softening can be observed as the reduction
in lateral resistance of an axially loaded cantilever column under monotonically increasing
lateral tip displacement. Again, more advanced models are needed in this case.
The generalization of the rigid plastic theory concepts by Prager and Hodge (1951) to
reinforced concrete column stress and strain resultant variables, such as bending moment and
rotation and axial force and extension, limits the applicability of these models to well detailed
members with large inelastic deformation capacity at the critical regions. For a reinforced
concrete column section, the yield surface of the stress resultants is actually a function of a
reference strain that couples the corresponding displacement components. This contradicts
classical plasticity theory which does not account for deformation softening and assumes that
the section deformability is unlimited.
A refined resultant model has been proposed by El-Tawil and Deierlein (2001) that developed
a bounding surface plasticity model implemented in the stress-resultant space. The model is
generally applicable to steel, reinforced concrete and composite members. Two variations of
the plasticity model are considered: a finite-surface and a degenerate-surface version. The
former explicitly considers a fully elastic response region to exist within the inner surface and
is, thus, applicable to steel members that typically have such behaviour. The degeneratesurface model shrinks the elastic region to a point and the section behaviour starts out as
inelastic in any loading direction. This version is suitable for sections that have little or no
elastic response region such as reinforced concrete and composite sections. Stiffness
degradation is accounted for as a function of the plastic strain energy absorbed by the
composite member.
To overcome some of the limitations of classical plasticity theory in the description of the
interaction between axial force and bending moments, Lai et al. (1984) proposed a fibre hinge
model that consists of a linear elastic element extending over the entire length of the
reinforced concrete member and has one inelastic element at each end. Each inelastic element
is made up of one inelastic spring at each section corner, representing the longitudinal
124

4 Analysis! and design of frame structures using non-linear models

reinforcing steel, and a central concrete spring that is effective in compression only. The five
spring discretisation of the end sections is capable of simulating the axial force-biaxial
bending moment interaction in reinforced concrete members in a more rational way than is
possible by classical plasticity theory. In Lai's model, the force-deformation relationship for
the effective steel springs follows that of Takeda et al. (1970) but the parameters that define
the envelope are established from equilibrium considerations.
A more accurate description of the inelastic behaviour of reinforced concrete members is
possible with distributed nonlinearity models. These models are treated in more detail in the
following section.
4.3.2

Distributed models

In distributed models, material nonlinearity can take place at any element section and the
element behaviour is derived by a weighted integration of the section response. In practice,
since the element integrals are evaluated numerically, only the behaviour of selected sections
at the integration points is monitored. Either the element deformations or the element forces
are the primary unknowns of the model and these are obtained by suitable interpolation
functions from the global element displacements or forces, respectively.
Discrete cracks in distributed models are represented as smeared over a finite length rather
than treated explicitly. The constitutive behaviour of the cross-section is either formulated in
accordance with classical plasticity theory, in terms of stress and strain resultants, or is
explicitly derived by discretisation of the cross-section into fibres, as is the case in the spread
plasticity fibre models. Frame models are usually based on either:
i) Euler-Bernoulli beam theory in which plane sections remain plane and normal to the
longitudinal axis of the beam; that is, there are no shear deformations (see Figure
4.1a); or
ii) Timoshenko beam theory of plane sections remaining plane but not normal to the
longitudinal axis with the difference between the normal and the plane section
rotations being the shear deformation (see Figure 4.1b).
Biaxial bending is a simple extension of the case of uniaxial bending, shown in Figure 4.1.

: 0 ( x)
&

a'
Deformed

b'

Deformed

dv0
dx

dv0
dx

v0 ( x)

Undeformed

v0 ( x)

Undeformed

(a) Euler-Bernoulli beam

b'

u0 ( x )

a'

u0 ( x )

(b) Timoshenko beam

Figure 4.1: Assumptions of Euler-Bernoulli and Timoshenko beam theories.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

125

In the development of the theory that follows, the following notation is used with an element
of length dx (Figure 4.2):

u0, v0 = longitudinal and transverse displacements, respectively, at the cross-section


reference axis
dv0
= total cross-section rotation
dx

:0, & = cross-section deformations due to bending and shear, respectively.


p

M+dM

dV
dx
dM
V'
dx
p'

V+dV

dx

Figure 4.2: Beam segment of infinitesimal length.

Based on the forces acting on a segment of infinitesimal length (Figure 4.2), the following
differential equations are derived:
Euler-Bernoulli beam

d 2M
Equilibrium:
'p
dx 2

(4.1)

Section constitutive law: M ' EIf

(4.2)

Section compatibility: f '


Differential equation:

d 2v0
(curvature)
dx2

(4.3)

d 2 e d 2 v0 b
c EI 2 ` ' p
dx 2 d
dx a

Strain at any point of the cross-section: ) x '

(4.4)

d 2v
du du0
'
7 y 20
dx dx
dx

(4.5)

Timoshenko beam

g dV
hh dx ' 7 p
Equilibrium: i
hV 7 dM ' 0
hj
dx

(4.6)

Section constitutive laws: V ' GAs&

126

M ' EIf

(4.7)

4 Analysis! and design of frame structures using non-linear models

Section compatibility:

f'

d: 0
dx

(flexural deformation)

(4.8)

&'

dv0
7 :0
dx

(shear deformation)

(4.9)

g
e d 2 v0 d: 0 b
h GAs c 2 7
` ' 7p
dx
dx a
h
d
Differential equations: i
d 2: 0
e dv0
b
h
GA
7
:
7
EI
'0
0`
h s cd dx
dx 2
a
j

(4.10)

In Eqs. 4.7 and 4.10, As is the shear area. To find the expression for As in the shear stress
expression according to the Timoshenko beam theory, assume that the strain energy
U) =

1
1
, xy& xy dA ' G G , xy2 dA
G
2A
2 A

(4.11)

is the same as that found using the stresses and strains of the exact solution for an elastic
section. For a rectangular cross-section it can be shown that As ' 1.2bh ' 1.2 A . Figure 4.3
compares the sectional shear stresses from Timoshenko beam theory for a rectangular,
linearly elastic, section with that of the exact solution.
SHEAR STRESSES IN RECTANGULAR SECTION

, xy

, xy

y
x

, xy '

6V e h 2
2b
c 7y `
bh3 d 4
a

"Exact" Theory

, xy ' const '

V
GAs

Timoshenko Beam Theory

Figure 4.3: Shear stresses on a rectangular section.

Euler-Bernoulli beam elements

Two formulations are presented for frame elements, one based on the classical displacementbased approach, the other based on the force-based approach.
Displacement-based formulation
The displacement-based beam formulation uses the classical finite element approach to derive
the element stiffness matrix and the element restoring force vector. The central step is the
assumption of displacement fields along the element that are defined in terms of nodal
displacements.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

127

If U is the nodal displacement vector defined in Figure 4.4, the section displacements

u $ x % ' Hu0 $ x % v0 $ x %I are defined by:


T

u $ x % ' NU $ x % U

(4.12)

where NU $ x % are the shape functions that define a linear axial displacement u0(x) and a cubic
transverse displacement v0(x):
Z x
0
X1 7 L
X
NU $ x % '
2
3
X
exb
exb
1 7 3c ` 8 2 c `
X 0
dLa
dLa
Y

u1

0
xb
e
x c1 7 `
d La

x
L

exb
e xb
3c ` 7 2 c `
dLa
d La

v1

v2

-1

W
U
U
2
e xb
e xb U
7x c ` 8 x c ` U
d La
dLa V
0

u2

-2

U ' Hu1 v1 -1 u2

v2 - 2 I

Figure 4.4: Classical two-node 2D beam element.

Using the principle of minimum potential energy, or other equivalent variational principles,
the following expressions are obtained for the element stiffness matrix K and for the element
resisting forces P:
K = G ! T $ x % k $ x % ! $ x % dx

(4.13)

P = G ! T $ x % s $ x % dx

(4.14)

where ! $ x % is an array containing the derivatives of the shape functions NU(x) (in particular
the first row of B is the first derivative of the first row of NU, and the second row of B is the
second derivative of the second row of NU) and k(x) is the cross-section stiffness matrix. The
matrix k(x) depends on the cross-section model selected for the analyses and s(x) is the crosssection force vector, that is:

s $ x % = H N $ x % M $ x %I

(4.15)

The implementation of displacement-based frame elements in a general purpose finite element


framework is straightforward, and this is the main reason why they have been widely used.
On the other hand, the above frame element is exact only if the element cross-section is
constant and the material behaviour is linear-elastic. In the case of reinforced concrete frames,
where material nonlinearities become important, the element is approximate and this leads to
the use of refined meshes to provide a good description of a frames response.

128

4 Analysis! and design of frame structures using non-linear models

Force-based formulation
The force-based beam formulation assumes force fields rather than displacement fields along
the element. The element is developed here without rigid body modes (Figure 4.5), for
reasons that will become clear later on in the chapter.

k1, - 1

k 2, - 2

N, u

P = HM 1

M2

NI

Figure 4.5: Two-node 2D beam element without rigid body modes.

If P is the nodal force vector defined in Figure 4.5, the force distributions along the element
are written as:
s $ x% ' NP $ x% P

(4.16)

where P is the previously defined vector containing the cross-sectional forces and N P $ x % is
the force interpolation functions array given by:
0
1W
Z 0
X
NP $ x% ' e x b e x b U
Xc 7 1 ` c ` 0 U
YXd L a d L a UV

(4.17)

Again, using the principle of minimum complementary potential energy, or other equivalent
variational principles, the following expressions are obtained for the element flexibility matrix
without rigid body modes F and for the corresponding nodal deformations U:
F = G N TP $ x % f $ x % N P $ x % dx

(4.18)

U = G N TP $ x % e $ x % dx

(4.19)

where f(x) is the section flexibility matrix and e(x) is a vector containing the section
deformations. That is:

e $ x % = H) 0 $ x % f $ x %I

(4.20)

The motivation and interest for force-based elements stem from fact that the equilibrium
relationship (Eq. 4.16) is exact within the assumptions of the Euler-Bernoulli beam theory.
In other words, the axial load and the bending moment remain constant and linear,
respectively, irrespective of the beam cross-section variation and material behaviour. Element
loads are included in the force-based formulation by modifying the element force distributions
in Eq. 4.16.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

129

One problem encountered with force-based elements is with their implementation in a general
purpose finite element program. The element state determination, that is the computation of
the element stiffness matrix K and resisting forced P, become complex. As for the stiffness
matrix, this is easily computed by inverting the flexibility matrix:

K = F 71

(4.21)

and by adding the rigid body modes to K to obtain K. Computing the element resisting forces
is a much more complex problem. The complexity stems from the fact that there is no way to
directly relate section resisting forces and element resisting forces, as is the case with the
displacement-based elements. Spacone et al. (1996) propose an iterative method to compute
the exact element forces and deformations. The procedure is basically a Newton-Raphson
iteration loop under imposed nodal displacements that adjusts the element forces and section
deformations until there is compatibility between section deformations and imposed nodal
deformations, as expressed by the weak statement of Eq. 4.19. The element is computationally
more involved and expensive than the corresponding two-node displacement-based element
but its precision leads to the use of a single element per structural member, thus leading to
large savings in the global number of degrees of freedom to be solved for.
Timoshenko beam elements

Similarly to the Euler-Bernoulli beam element, two formulations are presented for the
Timoshenko beam element, one based on the classical displacement-based approach and the
other based on the force-based approach.
Displacement-based formulation
The displacement-based formulation is similar to that of the Euler-Bernoulli beam but there
are three independent fields: horizontal displacement, u0, vertical displacement, v0, and axis
rotation, :0. A typical choice for these elements is to assume parabolic displacement fields for
the three displacements. The resulting three-node element is shown Figure 4.6.

:1

v1

u1
1

:2

v2

u2
2

:3

v3

u3
3

Figure 4.6: Three-node 2D Timoshenko beam element.

In this case

U ' Hu1 v1 :1 u2

v2 : 2 u3 v3 : 3I

(4.22)

and

u $ x % ' Hu0 $ x % v0 $ x % : 0 $ x %I

(4.23)

Eq. 4.12 still holds true with the displacement interpolation vector NU(x) defined by

130

4 Analysis! and design of frame structures using non-linear models

0
0
N2 $ x %
0
0
N3 $ x %
0
0 W
Z N1 $ x %
X
U
NU $ x % ' X 0
N1 $ x %
0
0
N2 $ x %
0
0
N3 $ x %
0 U
XY 0
0
N1 $ x %
0
0
N2 $ x %
0
0
N 3 $ x % UV

(4.24)

The explicit expressions of the shape functions are given in natural coordinates in Figure 4.7.

N 2 $R % ' $1 7 R 2 %

1
N1 $R % ' R $R 7 1%
2
1

-1

Figure 4.7:

1
N 3 $R % ' R $R 8 1%
2

-1

Displacement interpolation functions for three-node 2D Timoshenko beam element in natural


coordinates ( R ' 718 2 x L ).

The remainder of the element formulation is formally identical to that of the Euler-Bernoulli
beam. The section force vector now becomes:

s $ x % = H N $ x % M $ x % V $ x %I

(4.25)

the corresponding section deformation vector is:

e $ x % = H) 0 $ x % f $ x % & $ x %I

(4.26)

and the vector B(x) changes accordingly. The resulting element is approximate even in the
linear elastic case and several elements must be used to obtain a satisfactory approximation of
the solution. Also, these elements may lock in the sense that too much energy goes into the
shear deformation mode, leading to a stiff element.
Force-based formulation
The force-base formulation for a Timoshenko beam is identical to that of the Euler-Bernoulli
beam with the exception that the expressions for the section forces s(x) and section
deformations e(x) change according to the definitions given in the displacement-based
formulations. The force interpolation functions become:
Z 0
X
ex b
N P $ x % ' Xc 7 1 `
Xd L a
X
Y 1/ L

1W
U
exb U
0
c `
dLa U
U
1/ L 0 V
0

(4.27)

The flexibility matrix F is once again exact within the assumptions of the Timoshenko
beam theory, irrespective of the variation in beam section or material behaviour.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

131

4.3.3

Section models: fibre elements vs. strut-and-tie

The definitions of the section stiffness matrix k(x) and flexibility matrix f(x) for both the
Euler-Bernoulli and Timoshenko beams depend on the section models used in the
formulations. Some of the models most commonly used in the published literature are
discussed in this section.
Rational modelling of reinforced concrete members is vital to perform structural health
assessment on existing buildings, assess rehabilitation interventions on substandard
constructions or in identification of potential causes of damage or collapse in buildings. Fibre
analysis and STM procedures presented in this section can be used to evaluate the existing
capacity of structures with or without damage.
Once the behaviour and existing capacities of structural components are established, the
deficient components can be upgraded in order to meet the capacity demands. Fibre
(sectional) analysis is an appropriate tool that can be used to estimate behaviour of the
B-regions of reinforced and prestressed concrete members. In D-regions, beam theory does
not apply as the plane sections remain plane principle of the beam theory is not accurate or,
necessarily, safe. In these regions strut-and-tie modelling (STM) can be used to estimate the
capacity. It is important to appreciate that STM can be used to model Bernoulli beams in the
way Ritter (1899) and Mrsch (1902) used concrete compression struts and tension ties
(stirrups and longitudinal reinforcement) to explain the force flow and stress fields in a typical
reinforced concrete beam.
In this section fibre analysis procedures and strut and tie modelling techniques are treated in
order to:
"

perform (conventional and seismic) assessment on existing buildings,

"

assess rehabilitation interventions on inadequate constructions,

"

identify potential causes of damage or collapse in buildings.

Structural members can be divided into B-regions (i.e. Bernoulli regions or beam regions)
and D-regions (i.e., disturbed regions or discontinuity regions.) In B-regions the beam
theory applies and flexural behaviour of reinforced and prestressed concrete members can be
established by using equilibrium, compatibility and constitutive relationships. Fibre
(sectional) analysis is an appropriate tool that can be used to establish behaviour of the
B-regions of reinforced and prestressed concrete members. In addition, through the use of the
modified compression field theory (MCFT), it is possible to extend the use of fibre models to
shear critical elements. In D-regions, strut-and-tie models, such as those proposed by
Marti (1985a, 1985b) and Schlaich et al. (1987), can be used in order to estimate the capacity.
Ritter and Mrsch planted the first seeds of what is currently known as STM by idealizing
reinforced concrete beams with a series of diagonal concrete struts and reinforcing bars (ties).
It is equally important to note that a significant amount of research has been conducted in this
area since the original research of the aforementioned pioneers. Strut-and-tie modelling is a
detailing and ultimate strength calculation procedure for discontinuity regions within
structures. Strut-and-tie modelling represents a design method for complex structural details
that has a basis in mechanics and is derived from the theory of plasticity but is simple enough
to be readily applied in design. The method involves the idealization of a complex structural
member into a simple collection of struts, ties, and nodes representing, in a general manner,
the flow of stress paths within the member.

132

4 Analysis! and design of frame structures using non-linear models

Fibre analysis and STM procedures presented in this section can be used to evaluate the
existing capacity of structures with or without damage. Once the behaviour and existing
capacities of structural components are established, the deficient components can be upgraded
in order to meet the capacity demands. The methods presented in this section are of particular
significance for this reason.
Fibre analysis

Flexural behaviour of reinforced and prestressed concrete members can be estimated by using
equilibrium, compatibility and constitutive relationships. Compatibility in fibre analysis is
achieved through the plane sections remain plane hypothesis (Hooke, 1678, Bernoulli,
1705, and Navier, 1826). This geometric assumption forms the basis of the engineering beam
theory that is used in sectional analysis of concrete members. Equilibrium is achieved by
integrating stresses at any given section and equating them to the required sectional forces.
Appropriate stress-strain relationships for concrete and reinforcing and/or prestressing steel
need to be used in order to obtain the stress distribution through the depth of the section for a
given strain profile.
Concrete strain distribution can be defined by two variables (for example, two strains, a strain
and curvature, etc.). If the strain profile is known, realistic stress-strain relationships for
materials can be used to determine the stress distribution, which can subsequently be
integrated to evaluate the resultant forces for concrete and reinforcing steel. In this way, the
moment and axial load acting on the section can be determined.
Conversely, if we know the strain profile we can evaluate the axial force and bending moment
that caused these strains. The only detail that makes the evaluation of the response of flexural
members a somewhat cumbersome task lies in the integration of the stress over the part of a
concrete section in compression. Numerical integration can be performed to simplify this task
with the section idealized as a series of rectangular (or trapezoidal) layers and with the stress
in a given layer assumed as constant over the width of that layer. Figure 4.8 illustrates a
typical example where the concrete section is subdivided into multiple layers to facilitate the
numerical integration process or fibre analysis.
The modified compression field theory (MCFT) of Vecchio and Collins (1982, 1986) has
proven to be a simple, yet powerful, computational tool that can be used in FE analysis of
reinforced concrete structures and in the sectional analysis for concrete members to predict
the deformation and load carrying capacity. The latter will be discussed in this section.

yci

fc , $ o ,

bi , hi , "ti

ysj

Asi , #$ pj , f ylj

f yt , Es

Figure 4.8: Fibre Analysis of a Reinforced Concrete Section.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

133

The MCFT was based on reinforced concrete panel tests subjected to in-plane forces.
Equilibrium, compatibility and stress-strain relationships based on test observations were used
in the formulation of the models. Constitutive laws were used to link average stress-strain
relationships both for the reinforcement and concrete. The MCFT is based on a rotating crack
model that assumes that there is a gradual reorientation of principal stresses (and strains) in
the concrete parallel to, and normal to, the direction of the cracks. Furthermore, the MCFT
assumes enforced alignment of principal stress and strain directions and links the principal
compressive stresses to principal compressive strains considering the effect of co-existing
principal tensile strains.
The use of MCFT within a layered section (fibre) analysis is discussed in Section 4.3.4. In this
way the fibre analysis, discussed above, can be extended to include rational models to obtain
shear force, axial force and bending moment interaction for reinforced and prestressed
concrete sections and members.
Strut-and-tie modelling

Strut-and-tie modelling (STM) has proven to be an efficient tool for analyzing structural
members. The efficiency arises only in areas of discontinuity where plane sections do not
remain plane, strains are not linear, and transverse strains may not be conservatively
neglected. In these areas, typical layered sectional analysis does not provide a reasonable
estimate to a sections capacity due to violation of the basic assumptions made when
analyzing the section. It should be noted, however, that any structural member may be
analyzed to find its ultimate capacity using a STM. This is possible because the method
results in a lower bound solution. For example, a typical Bernoulli beam may be analyzed
using struts and ties as shown by Ritter (1899) and Mrsch (1902) in Figure 4.9. In his classic
1902 text, Mrsch explained the truss model shown in Figure 4.10 in more detail.

Figure 4.9: Ritter (1899) truss analogy.

(a) Truss Model

(b) Stirrup Forces

Figure 4.10: Morsch (1902): (a) truss model; (b) stirrup forces.

134

4 Analysis! and design of frame structures using non-linear models

B-Regions
D-Regions

h
h-1.5h typ

h-1.5h typ

h-1.5h typ
h1-1.5h1 typ

h2-1.5h2 typ

h1-1.5h1

h
h2-1.5h2

Figure 4.11: Separation of Members into B- and D-regions.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

135

According to the theory of plasticity, any statically admissible stress field that is in equilibrium
with the applied loads and in which stress levels are on or within the material yield surface,
constitutes a lower bound solution. However, the strain capacity of the materials is a
fundamental requirement to fully satisfy that a lower bound solution occurs and, in that concrete
is not an elasto-plastic material; some care is needed in determining an appropriate model.
To begin a STM, the designer must first discretize the member into B- and D-regions. The
B-regions are areas of the member where plane sections remain plane, strains are linear and
transverse strains are negligible. D-regions are those in which sectional analysis does not
apply because the aforementioned assumptions are not reasonable. These are areas under
point loads, points of reaction, openings, re-entrant corners, frame joints, etc., as illustrated in
Figure 4.11. The D-region is assumed to be 1 to 1.5 times the member height to either side of
the disturbance, as per St. Venants principle.
Once the D-regions are isolated they can be de-coupled from the rest of the member for
analysis purposes. It is essential that the forces applied to the D-region are accurate. In some
cases this step might prove to be most challenging to designers. A clear force path must be
determined to find all external loads acting on the member. Once the external loads are
determined then structural analysis must be completed to find external reactions and internal
forces. If the member is not entirely comprised of single a D-region, then sectional analysis
must be undertaken to determine internal forces from the adjacent B-regions, as shown in
Figure 4.12.
D region

B region

Figure 4.12: Forces on Boundaries of D-regions.

Once the forces are accurately placed on the de-coupled D-region, a STM or truss model may
be applied. The truss model should accurately depict the flow of tensile and compressive
forces within a statically determinate truss. The model may be chosen by experience and
judgment of the designer, by following the elastic principal stresses produced by a finite
element model (see Chapter 8), or by guidelines given by code bodies and/or the literature.
The following considerations should be taken into account when choosing an appropriate
STM for a D-region:
"

136

Separate models may have to be conceived for different load cases. For example
one model may be used when gravity loads control, and a separate model may be
used when lateral loads control, with the region detailed to accommodate both
models (Figure 4.13).

4 Analysis! and design of frame structures using non-linear models

Tie

i. Gravity load conditions

Strut

ii. Seismic conditions

Figure 4.13: Alternate Models for Different Load Cases.

"

Serviceability of the member must be taken into account. The model chosen should
account for serviceability by limiting cracking and provide sufficient stiffness to
control deflections (Figure 4.14).

i. Service load model

ii. Ultimate load model

Figure 4.14: Model Variance on Service/Ultimate Conditions.

"

The physical geometry of the truss should be practical and free of congestion, with
no overlapping struts or nodes.

"

When analyzing in-situ structural members, consideration shall be given to the


existing condition of the concrete and physical placement of steel reinforcement.
For example, if concrete is severely cracked in a location of a modelled strut, the
strut capacity should be considerred negligible or zero. The centroid of modelled
ties should coincide with the reinforcing steel placed in the member.

In general, optimizing the model used for final design and detailing of a D-region is a
subjective and iterative process. Schlaich et al. (1987) proposed to follow the principles of
minimum strain energy after cracking. That is, the model with the least and shortest amount of
ties is the most appropriate based on the assumption that cracked concrete struts will deform
little compared to steel-reinforced ties. However, as stated in FIP (1998) and mentioned
above, this ultimate load model may not be valid when evaluating service conditions.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

137

Once the nodal geometry is determined and, hence, the tie geometry, concentration is now
given to determining the strut geometry. The struts at mid-height can be idealized using one
of the stress fields shown in Figure 4.15.

bursting forces

Prismatic stress
field

Fan stress field


(no bursting forces)

Bottle stress field

(a) Types of concrete struts


dc
C

C/2

C/2

Tb /2

Tb /2
:

l b {bursting zone}

d c /4
d c /4

d c /4
d c /4

(b) Bursting forces in bottle shaped struts


Figure 4.15: Concrete strut idealizations

The designer must exercise judgment in selecting strut geometry within the model. If a strut is
located in the pure compression field of a member, or two nodes are modelled as close
together, producing confined compression fields, a prismatic strut geometry should be
considered. If the designer wishes to model a compression field at a beam support, in
Mrschs fashion, then a fan of compression struts should be considered. If a strut is in a
location that allows the compressive stresses to spread laterally then bottle shaped geometry
needs to be considered.

138

4 Analysis! and design of frame structures using non-linear models

In the case of bottle shaped struts, further model refinement is needed to account for the
transverse tension associated the lateral spread of compression within the bottle shaped strut.
For service loads, the spread of compression may be modelled at a slope of 2 longitudinal:1
transverse (tan : = 1/2) to the axis of the strut, whereas for the ultimate condition the
dispersion angle narrows to tan : = 1/5 (Foster, 1998). When modelling the transverse
tension, the designer should be cognizant of tie reinforcing anchoring the strut, the strut angle,
and any reinforcing steel placed to confine the spread of compression; all of which influence
the dispersion of compression and hence the model (Baumann, 1998).
4.3.4

Modelling of shear

Several older structures lack the shear reinforcement to guarantee that plastic hinge will form
before shear failure of the member. Shear modelling in a frame analysis is dealt with at two
levels, the element level and the section level. At the element level, the Timoshenko beam
theory is typically used to model the shear behaviour. Displacement-based and force-based
formulations for a Timoshenko beam are presented in Section 4.3.2.
At the section level, different approaches can be followed. The simplest relies on stressresultant laws, where different nonlinear laws are given for the flexural, axial and shear
response. One such model has been proposed by Martino et al. (2000), where a fibre model is
used to model axial and bending responses, while a nonlinear law is used for the shear forceshear deformation response. While the shear response is decoupled from the other
deformations at the section level, the implementation of this section model in a force-based
element allows coupling between axial and bending responses at the element level.
A fibre section model has been proposed for a Bernoulli beam by Petrangeli et al. (1999) that
includes shear deformations. In this approach, each fibre has basically three deformations,
axial strain, transverse strain (in the direction of the stirrups) and shear deformations, plus the
corresponding stresses. Given the section deformations e(x), the axial strain and shear
deformation of each fibre can be computed through compatibility. The third condition is that
the stress in the direction of the vertical stirrups is zero; that is, the sum of the forces in the
concrete and in the stirrups is zero. Given these three conditions (axial strain, shear
deformation and vertical zero net stress), the fibre state determination consists of finding the
corresponding axial stress, shear stress and vertical deformations and the fibre stiffnesses. The
fibre stresses and fibre stiffness are added to give the section forces and the section stiffness.
A concrete constitutive law based on the microplane theory is used to describe the concrete.
The fibre section model with shear deformations proposed by Petrangeli et al. (1999) is
implemented in a force-based element and has given results that correlate well with available
experimental results.
Finally, the MCFT can also be used to describe the shear deformation of a reinforced concrete
member. The MCFT has proven to be successful in predicting the load deformation response
of reinforced concrete beams, with different amounts of longitudinal and transverse
reinforcement, using sectional analysis procedures (Vecchio and Collins, 1986). It has also
been implemented into FE programs for similar analysis (for example, Vecchio, 1989, 1990,
Foster and Gilbert, 1990, Foster, 1992, and many others). In the MCFT, constitutive
relationships are used to link average stresses and strains for both reinforcement and concrete
for analysis of reinforced concrete member behaviour. The stresses and strains used are
average values realizing that the maximum and minimum values can be different to the
average values. Conversely, local conditions may govern the failure but average conditions
are likely to be more relevant in representing the member behaviour. Figure 4.16 illustrates
the average stress and strain states employed in MCFT.
!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

139

Figuure 4.16: Stresss and Strain Conditions:


C
M
MCFT.

The axiial stresses in the reinfforcement are


a assumed
d to dependd only on thhe axial straains with
any sheear stresses acting on the reinforcement neeglected. Thhe usual unniaxial streess-strain
relation
nship is used
d for steel ass follows:

+ s ' Es) s ? f y

(4.28)

Strain hardening
h
b
behaviour
o the reinfforcement at
of
a high straain levels ccan be inco
orporated
using th
he correspon
nding stresss-strain relattionship fro
om material tests.
The aveerage stresss-strain relaationship fo
or concrete is obtainedd from paneel tests and
d can be
found elsewhere
e
(V
Vecchio an
nd Collins, 1982). Prin
ncipal stresss and principal strain axes are
assumed
d to be coin
ncident as in rotating crack modeels. The beehaviour in panel tests showed
that craacked conccrete subjeccted to hig
gh tensile strains in the directiion normall to the
compresssion is soffter and weeaker than in
i standard cylinder teests. The cooncrete stresss in the
principaal direction can be relatted to the prrincipal con
ncrete strainn as follows:
2
e )
e ) 2 b b`
c
2
c
`
f c 2 ' f c 2 max c 2 7 c ` `
c )o d )o a `
d
a

(4.29)

where fc2 is the priincipal com


mpressive strress, "2 is th
he principaal compressiive strain, and
a "o is
the straiin in concreete at peak compressive
c
e stress.

140

4 Analysis! and design of frame structures using non-linear models

A number of models have been developed to determine the effect on concrete compressive
strength, and stiffness, for elements under the influence of transverse tensile strains (fc2max in
Eq. 4.29); for example, see Vecchio and Collins (1993), Belarbi and Hsu (1995), Kaufmann
(1998) and Vecchio (2000), amongst others. One such relationship that has proved robust with
time and can be used to compute fc2 is that of Vecchio and Collins (1986):

e
b
1
``
f c 2 max ' f c# cc
0
.
8
8
0
.
34
)
)
1 oa
d

(4.30)

For concrete in tension, prior to cracking a linear elastic stress-strain relationship is used. The
principal tensile stress fc1 is given as follows:

f c1 ' Ec)1

where )1 ? ) cr

(4.31)

where Ec is the tangent modulus of concrete, ) 1 is the principal tensile strain, fcr is the
tensile strength of concrete, and ) cr is the cracking strain of concrete. After cracking, tension
stiffening of concrete can be represented by:

f c1 '

f cr
1 8 500)1

(4.32)

The stress-strain relationships summarized above can be used in a sectional analysis to


compute the load carrying and deformation capacity of reinforced concrete members. This
will be discussed in the following section.
A reinforced or prestressed concrete section can be discretized into a series of concrete strips
with superimposed longitudinal steel components (Figure 4.8). Each concrete strip is defined
by its width, b, depth, h, amount of transverse reinforcing steel, $t, and its position relative to
the top of the beam, yc. Similarly, the longitudinal steel components are defined by their
cross-sectional area, As, yield strength, fy, and position relative to the top of the beam, ys. Prestrain in reinforcement can also be considered in the analysis. Properties common to the entire
cross-section are usually the concrete strength, f c# , concrete strain at peak stress, "o, yield
strength of transverse steel, fy, and modulus of elasticity for reinforcing bars, Es. Sectional
analysis is carried out assuming plane section remains plane with equilibrium and
compatibility satisfied within each fibre. Hence, the estimates of the longitudinal strains and
the shear flow distribution through the depth of the section are required to compute the stressstrain conditions in a reinforced concrete member subjected to a given set of sectional loads.
First, an estimate of longitudinal strain distribution can be made. Strain in concrete for each
strip is assumed to be constant. The stress conditions in steel layers can be directly computed
from the assumed longitudinal strain distribution together with the uniaxial relationship of
steel from material tests. Second, a shear stress distribution is assumed for the section such
that the sum of shear stresses in each strip will be equal to the externally applied shear. Given
the longitudinal strain and shear stress for a strip, equilibrium and compatibility must be
satisfied to compute the compressive stress for the strip. The resultant of stresses for the strips
must balance the applied sectional forces, axial force, N, moment, M, and shear force, V.
For a section discretized with m concrete strips and n steel strips, equilibrium requires that

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

141

P
i '1
m

P
i '1

f li bi hi 8 P f slj Asj ' N


j '1

f li bi hi y ci 8 P f slj Asj y sj ' M

Pv b h
i '1

i i i

(4.33)

'V

where fli is the compressive stress in the longitudinal direction in the ith fibre, and vi is the
shear stress in strip i. If these conditions are not satisfied, the longitudinal strains are adjusted
until equilibrium is satisfied. To determine the correct shear stress distribution, a second
section a small distance from the one considered can be analysed. Satisfying sectional
equilibrium for each case, the assumed stress distribution can be checked using the free body
diagram of the strip.
The procedure summarized above is shown in a flow chart in Figure 4.17 with the stress and
strain state of a section for a given set of loads computed using this procedure. Also, it is
possible to analyse the overall response of a section by analysing different increments of
longitudinal strains, shear stresses, and axial forces by using the procedure outlined above.
Bayrak and Sheikh (2001) proposed a plastic hinge analysis technique that can be used in a
sectional analysis platform in order to incorporate buckling of longitudinal bars in the
analysis. This technique employs slightly different displacement compatibility requirements
along with equilibrium considerations and constitutive relationships to evaluate the plastic
hinge response of tied columns. The following is a step-by-step description of the procedure
for the analysis of plastic hinges:
1) Standard sectional analysis procedure (Figure 4.17) is followed before the initiation of
buckling of the longitudinal bars.
2) At the initiation of bar buckling, the tie forces generated as a result of core concrete
bearing against the reinforcing bars are calculated.
3) By using proper boundary conditions and an assumed shape function for the forces
acting on longitudinal bars, their outward deflection at the mid-height (e) between two
sets of ties can be calculated. Dividing this deflection by the longitudinal bar diameter
(d) yields an e/d ratio.
4) Using the e/d ratio calculated above and the ratio of the unsupported length of the
longitudinal bars (l) to the bar diameter (l/d), the relevant stress-strain curve for
reinforcing bars under compression can be selected and used as the constitutive
relationship for compressed bars in a sectional analysis. Figure 4.18 illustrates
experimentally determined stress-strain curves for reinforcing bars with initial
imperfections tested under compression (Bayrak and Sheikh, 2001).

142

4 Analysis! and design of frame structures using non-linear models

SPECIFY SEC TIO N PR O PER TIES

G IVEN ! l , v

SPECIFY C O N C RET E ST RIP PR O PERT IES

EST IM AT E ! d

SPEC IFY LO N G IT U D INAL ST EEL PRO PERT IES

EST IMAT E "

SPECIFY SECT IO N LO AD S

D ET ERMINE
! t ,! dt , # lt , # m

EST IM AT E SH EAR ST RESS DIST R IBUT IO N

EST IM AT E LO NG IT U DIN AL ST RAIN D IST RIBUT IO N

D ET ERMINE
f st , f t , f d , f dt , f l , fd , "

C O M PU T E LO NG IT UD IN AL ST R ESSES
DIST R IBUT IO N FO R EAC H REBAR LAYER
IS

" = " ?

CO MPU T E C O N C RET E ST RESSE/ST RAIN


FO R EAC H C O N C RET E LAYER

C O MPU T E RESU LT ING SECTIO N LO ADS

NO

CH ECK SECT IO N AL
EQ U ILIBRIU M
?

IS

fd = fd ?

NO

YES

C O M PU T E RESU LT ING SHEAR


ST R ESS D IST R IBUT IO N

SH EAR ST R ESS D IST .


SAME AS ASSU M ED
?
YES
O U T PUT

YES

YES

REPEAT CALC U LAT IO N S FOR


SECT IO N 2

NO

NO

Symbols:
$ l : Longitudinal strain
v : Shear stress
$ d : Concrete principal compressive strain
% : Assumed angle of inclination of principal compressive strain
$ t : Transverse tensile strain
$ dt : Principal tensile strain
& lt : Normal shear strain
& m : Maximum shear strain
f st : Transverse tensile stress
f t : Concrete transverse compressive stress
f d : Concrete principal compressive strain
f dt : Concrete principal tensile stress
f l : Concrete longitudinal stress
f d : Principal compressive stress computed using assumed shear stress
% : Angle of inclination of principal compressive strain

Figure 4.17: Sectional Analysis for Shear, Flexure and Axial Loads.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

143

Figure 4.18: Stress-strrain relationsh


hips of reinforrcing bars.

4.3.5

M
Modelling
Bond Slip in Beams

Bond plays a centtral role in the response of reinfo


orced concrrete membeers by allow
wing the
stress trransfer from
m the steel bars
b to the suurrounding concrete. Perfect
P
bondd is usually assumed
a
in the analysis off reinforcedd concrete structures and impliees full com
mpatibility between
concretee and reinfoorcement strrains. This assumption
n is valid onnly in regionns where neegligible
stress trransfer occuurs between the two com
mponents (K
Kauser and Mehlhorn, 1987) and can only
take plaace at early loading stagges and at loow strain leevels. As thee load is inccreased, craacking as
well as breaking of
o bond unaavoidably occcurs and a certain am
mount of bond-slip takees place,
all of which
w
will, inn-turn, affecct the stresss distributions in both the
t steel andd the concreete. Near
the craccks, high boond-slips deevelop caussing relativee displacem
ments betweeen concretee and the
reinforccement steeel. Due to this bondd-slip, diffeerent strainns are obseerved in the steel
reinforccing bars thhan in the surrounding
s
g concrete. At the cracks the loccalised straiin in the
reinforccing steel caan be many times greatter that the average
a
straain and this is known as
a strain
localisaation.

144

4 Analysis! and design of frame structures using non-linear models

Bond-slip affects the overall response of reinforced concrete structural members. In particular,
two phenomena are worth discussing:
1. An increase in stiffness in the regions between two adjacent cracks. This effect,
called tension stiffening, can be included by either considering the tensile
resistance of the concrete or by increasing the tensile stiffness and strength of the
steel reinforcing bars. Tension stiffening mostly affects the member response
under serviceability conditions, since the effect of bond is completely lost when
member failure occurs.
2. An increase of flexibility at the member ends, due to the pullout of the reinforcing
bars at the interface either with beam-column joints or with the footings. Similar
drops in stiffness may also be caused by insufficient lap splice lengths.
These effects become particularly important and complex under seismic loading conditions,
when bond gradually deteriorates due to large strains and damage caused by multiple load
reversals.
Rubiano-Benavides (1998) proposed the use of rotational springs at the element ends to
account for the added flexibility due to bond-slip. This approach is suitable for lumped
plasticity models but requires particular care in the selection of the rotational spring's
mechanical properties. A similar approach is followed by Filippou et al. (1999).
Another approach is followed by Monti and Spacone (2000), who modify the fibre section
model to add the effects of bond slip, modelled according to Monti et al. (1997a, 1997b). The
basic idea is fairly simple: the strain in the steel fibre is the sum of the actual strain in the
reinforcing bar and the strain equivalent to the bar slip. This model accounts not only for the
response of the reinforcing bar inside the beam but also for its anchorage outside the element,
in either a structural joint or a footing. The steel fibre strain is given by the sum of the effects
of the bar deformation and the anchorage slip. The response is still computed in terms of fibre
stress and stiffness, which are needed for the fibre section state determination. The steel fibre
strain is obtained from the section deformations using compatibility and is written as

) s8a ' ) s 8

1
ua ' ) s 8 ) a
LIP

(4.34)

where ) a is a strain-equivalent contribution of the anchorage pullout, condensed at the fibre


level through the length LIP of the integration point. That is, the total steel fibre elongation is
given by the sum of the reinforcing bar deformation and anchorage pullout. Compatibility is
maintained between the concrete strain and the total steel fibre elongation ()c = ) s+a), while
concrete and steel strains are different ()c = ) s). This procedure is illustrated in Figure 4.19 for
the case of a slice with axial deformation only and no curvature.
It is important to point out that the formulation proposed by Monti and Spacone (2000)
provides a solution to the bond-slip problem within a single beam finite element. The element
can capture the base rotation in RC columns due to bar slips and can also describe slip and
bond failure at the bar splices. The fibre section model by Monti and Spacone (2000) is
inserted in the force-based element by Spacone et al (1996).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

145

Slice length LIP

Cross Section
axial strain = )
curvature = 0

)5LIP = slice elongation


Element with perfect bond

)5'5)c55'5)s

Element with partial bond

)5'5)c5'5)s 85)a

)a LIP

$/8)s) LIP

Figure 4.19: Slice response to axial deformation only (Monti and Spacone, 2000).

A different approach to modelling bond-slip was followed by Limkatanyu and Spacone


(2002). In this model, the concrete element and the reinforcing bars are considered as
different line elements connected at the nodes to give a frame model with bond slip. The
procedure is demonstrated in Figure 4.20 with the free body diagram for a segment of length
dx of a reinforced concrete frame element with n bars and with the bond interfaces shown.
Only bond stresses tangential to the bars are considered in the formulation. Axial equilibrium
in the beam component and in the bar i lead to:

dN B $ x %
dx

8 P D bi $ x % ' 0
i '1

dN i $ x %
7 D bi $ x % ' 0,
dx

(4.35)

i ' 1, n

where N B $ x % and N i $ x % are the axial forces in the beam and in bar i, respectively, and
D b i $ x % is the bond interface force between the beam and bar i.

Vertical equilibrium of the infinitesimal segment dx yields:

dVB $ x %
7 py $ x % ' 0
dx

(4.36)

where VB $ x % is the beam section shear force and p y $ x % is the transverse distributed load.
Finally, moment equilibrium gives:

dM B $ x %
dx

146

7 VB $ x % 7

P y D $ x%
i '1

bi

'0

(4.37)

4 Analysis! and design of frame structures using non-linear models

py (x)
R/C beam
with bond-slip

beam reference axis

dx

=
py (x)

concrete beam
y1
yn

Db1(x)
Mb (x) 8 dMb (x)

Mb (x)

beam reference axis

NB (x)

NB (x) 8 dNB (x)


VB (x) 8 dVB (x)

VB (x)

Dbn(x)

bars
bar 1
y1

Db1(x)

N1(x)

beam reference axis


Dbn(x)

yn

bar n

N1(x) 8 dN1(x)

Nn (x) 8 dNn (x)

Nn (x)

Figure 4.20: Slice of RC frame element with bond slip (Limkatanyu and Spacone, 2002).

where M B $ x % is the beam section bending moment and yi is the distance of bar i from the
element reference axis. The work by Limkatanyu and Spacone (2002) follows the EulerBernoulli beam theory and, thus, the shear deformations are neglected. The shear force VB $ x %
is removed by combining the above equations to obtain:

d 2M B $ x%
dx 2

dD bi $ x %

i '1

dx

7 p y $ x % 7 P yi

'0

(4.38)

Compatibility

As for the element compatibility, the reinforced concrete element is treated as a Bernoulli
beam. On the other hand, the reinforcing bar slip are determined by the compatibility
relationship between the beam and the bar displacements:

ubi $ x % ' ui $ x % 7 uB $ x % 8 yi

dvB $ x %
dx

(4.39)

where ub i $ x % is the bond slip between the beam and bar i.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

147

Starting from the above differential equations, Limkatanyu and Spacone (2002) proposed
three formulations: a displacement-based, a force-based, and a two-field mixed formulation.
While the details of the three formulations may be involved, it is important to point out that
the displacement-based element with bond-slip, while computationally robust, is quite
inaccurate and thus several elements per structural member must be used. On the other hand,
force-based and mixed elements are much more precise and, though more complex to
implement in a general purpose FE program, they can be used in a coarser mesh leading to
computational cost savings.
4.3.6

Analysis of a section

All the tools needed for various steps described earlier are available at this point. Therefore,
the response of a section located in the plastic hinge region of a concrete column can be
predicted using the plastic hinge analysis procedure. At this point it should be appreciated that
the equations presented herein are derived for cases where buckling over one tie spacing takes
place. For cases where one or two tie sets rupture, the existence of ruptured tie sets can be
ignored and spacing can be modified accordingly so that the equations presented herein can
be used. In reality, according to the principal of minimum potential energy, the mechanism
that would require minimum energy to be stored in the system is the governing mechanism of
failure and, hence, that mechanism must be used in the analysis.
Figure 4.21 illustrates the sectional response of four specimens (AS-3, AS-17, AS-18 and AS19) tested by Sheikh and Khoury (1993). The moment curvature predictions, obtained using
the confined concrete stress-strain relationships suggested by Sheikh and Uzumeri (1982)
(SU), Kent and Park (1971) (MKP) and Mander (1988) (MAN) are also shown in these
figures. In the conventional sectional analyses performed to obtain the sectional responses, the
aforementioned confined concrete stress-strain relationships are used for the core concrete.
An unconfined concrete stress-strain relationship is used for the cover concrete and stressstrain relationships obtained from a tensile coupon test are used for longitudinal bars under
tension and compression. Predictions obtained using the plastic hinge analysis procedure, with
the Sheikh and Uzumeri model (SU+B) are also shown in Figure 4.21. The computer
program, SecRes99, (Bayrak, 1999) was used to obtain the predictions shown in Figure 4.21.
The use of the plastic hinge analysis procedure resulted in reasonably accurate predictions for
the behaviour of sections located in the plastic hinge region of the test specimens.
Conventional sectional analyses (fibre-analyses) could not provide a reasonable prediction for
the ultimate curvatures. With the use of the plastic hinge analysis technique, the reduction in
the load carrying capacity of longitudinal bars as a result of their buckling can be predicted.
The ultimate curvature and, hence, the failure of a column can therefore be determined with
reasonable accuracy.

4.4

Interpretation of results

4.4.1

Localisation problems

Reinforced concrete frame elements, similarly to concrete solid elements, can lead to
numerical inconsistencies that derive from localization problems. When the cross-section
response starts softening (as may be the case of reinforced concrete column sections), the
element response becomes non unique and depends on the number of integration points per
element and/or the number of elements per structural member. Localization issues, though
similar in nature, tend to be different in displacement-based and in force-based frame

148

4 Analysis! and design of frame structures using non-linear models

Figure 4.21: Experimental and predicted sectional responses .

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

149

elements. Localization in displacement-based solid finite elements has been studied by Baant
and Oh (1983), Baant and Planas (1998) de Borst et al. (1994), among others and the theory
extends to frame elements. Coleman and Spacone (2001), Scott and Fenves (2006) and
Valipour and Foster (2007) discuss localization in force-based elements and propose various
solution strategies.
To illustrate the numerical problems encountered in force-based frame elements when plastic
hinges form, we will consider the case of a steel cantilever beam under an imposed transverse
tip displacement. A single force-based element is used for the entire member. As the applied
tip displacement increases, a plastic hinge forms at the base where the maximum moment
occurs. Figure 4.22 illustrates the response of the force-based element for the cantilever beam
with an elastic-strain hardening section behaviour. Unloading is prescribed in the final steps
to clarify the peak displacement and curvature demands. The base shear is plotted against the
tip displacement on the right and the base curvature (i.e. the curvature of the first integration
point) is shown on the left. The response is objective at both the element and the section
levels for models with four or more integration points. Three integration points do not
accurately integrate the element integrals leading to over-prediction of the stiffness in the
strain-hardening region.
Base Shear
3 IP
3 IP

4 IP
5, 6, 7, 8 IP

4, 5, 6, 7, 8 IP

O
3, 4, 5, 6, 7, 8
Integration
Points (IP)
Elastic-strain hardening
moment-curvature
Curvature

Displacement, O

Figure 4.22: Cantilever beam with elastic-strain hardening section response (Colemanand Spacone, 2001).

Figure 4.23 shows the response of the same force-based element to an imposed tip
displacement with an elastic-perfectly plastic moment-curvature behaviour. Here, the
prediction of the element force-displacement response remains objective while the peak
curvature demand varies with the number of integration points. The loss of objective
curvature prediction is due to the localization of the inelastic curvature at the base integration
point. When this bottom section reaches the plastic moment, the column reaches its load
carrying capacity. As the tip displacement increases, the curvature of the base integration
point increases with constant (plastic) moment, while all the other integration points remain
linear elastic and do not see any change in either curvature or moment. The length of the base
integration point and, thus, the plastic hinge length, becomes a function of the number of
integration points used. As the number of integration points increases, the plastic hinge length
decreases and the curvature demand in the base integration point must increase to give the
same prescribed tip displacement. From here the non-objective section prediction of Figure
4.23 is obtained.

150

4 Analysis! and design of frame structures using non-linear models

Base Shear

3, 4, 5, 6, 7, 8 IP
8 IP

7 IP

6 IP 5 IP 4 IP 3 IP

O
3, 4, 5, 6, 7, 8
Integration
Points (IP)
Elastic-perfectly plastic
moment-curvature
Displacement, O

Curvature

Figure 4.23: Cantilever beam with elastic-perfectly plastic section response (Coleman and Spacone, 2001).

In the case of a softening moment-curvature response, the loss of objectivity is more


pronounced. Softening section responses may take place in reinforced concrete columns or in
reinforced concrete bridge piers that support a substantial dead load and are subjected to
seismic forces. Figure 4.24 illustrates the response of a reinforced concrete column modelled
with a single force-based element. Both the local base section moment-curvature response and
the global base shear-displacement response lose objectivity. As the number of integration
points increases from three to five, the length of the first integration point decreases and
increasing curvatures are required to achieve the same prescribed tip displacement. The
concrete fibre compressive strains in the hinge region quickly increase resulting in rapidly
degrading material stiffness. For larger numbers of integration points, the post-peak response
becomes brittle and, with increasing numbers of integration points, snap-back may even
occur.

P (constant)
O

3, 4, 5
Integration
Points (IP)

Base Shear

3 IP
R/C beam-column

3 IP

4 IP

4 IP

5 IP
5 IP

Curvature

Displacement, O

Figure 4.24: RC beam-column modelled with strain softening section response (Coleman and Spacone, 2001).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

151

Figure 4.25 shows the schematic bending moment and curvature distributions for three, four
and five integration points along the column height. This case refers to the elastic-perfectly
plastic section response of Figure 4.23 and helps clarify the localization process in forcebased elements. The plastic curvature Cp occurs when the plastic moment Mp is first reached.
Since equilibrium is strictly satisfied in the force-based element, the bending moment remains
linear. When the plastic moment Mp is reached at the first (base) integration point, the applied
force cannot increase and the tip displacement increases under constant applied load (and
constant base moment). Because the bending moment cannot increase beyond Mp, adjacent
integration points remain elastic and the inelastic curvature localizes at the first integration
point. The tip displacement is computed as the weighted sum of the curvatures at the
integration points. The first integration point has a finite length LIP=1 = w1L proportional to its
integration weight w1. The larger the number of integration points, the shorter the length of
the first integration point and the larger the curvature at the first integration point to obtain the
same tip displacement, as shown in Figure 4.25.

Gauss-Lobatto
integration point

Moment

Curvature

Mp C p
A)
3 Integration Points

Mp

Cp

B)
4 Integration Points

Mp

Cp

C)
5 Integration Points

Figure 4.25: Moment and curvature profiles for an elastic-perfectly plastic cantilever modelled with a single
force-based element (Coleman and Spacone, 2001).

In summary, it is the ability to capture a jump from elastic to inelastic behaviour that makes
the force-based formulation both attractive and prone to unique numerical problems. One can
observe that as the distance between the first (plastic) and second (elastic) integration point
varies, the response also varies. This implies that the number and placement of the integration
points not only influences the accuracy of the integration but also the post-peak response. For
hardening materials, plasticity usually spreads beyond a single integration point and
numerical problems are limited to a non-smooth response if too few integration points are
used (Figure 4.22). For perfectly plastic and softening cross-section responses, the curvature
tends to localize at a particular integration point and problems with objectivity arise (Figure
4.23 and Figure 4.24).
For further reading on the state of research into the effects of localisation in fibre based
elements and on methods for obtaining objective results, the reader is referred to Scott and
Fenves (2006) and Valipour and Foster (2007).
4.4.2

Physical characteristics of localised failure in concrete

In order to introduce a method for re-establishing objectivity of the force-based element


response, a brief discussion of the physical characteristics of concrete failure is necessary.
Concrete is a heterogeneous material prone to localized failure. In tension, failure is
characterized by numerous micro-cracks that bridge together to form a main crack. The
152

4 Analysis! and design of frame structures using non-linear models

discontinuity induced by the crack opening basically prevents stresses normal to the crack.
The failure is termed localized because material outside the fracture process zone remains
practically undamaged.
In compression, concrete failure is a different phenomenon to that of tension. Figure 4.26
illustrates a typical laboratory test of a concrete cylinder subjected to uniaxial compression.
The test occurs under displacement-control in order to capture the post-peak response of the
specimen. Region B in Figure 4.26 corresponds to a region of damaged concrete. Damage in
this region, initially characterized by axial splitting, progresses until a sliding shear band
forms and the cylinder stiffness rapidly degrades. The stress-strain curve in this damaged
region enters the post-peak branch, where a drop in stress is associated with increasing strains.
The concrete in the regions labelled A in Figure 4.26 is not severely damaged and has not
reached the peak strength. To maintain equilibrium of axial force, the stress-strain response in
regions A unloads elastically.
+

O
+

L
OlL
overall response

region B

region A

Figure 4.26: Uniaxial compression test under displacement control (Coleman and Spacone, 2001).

The concrete cylinder can be idealized as a series system, where one element (region B in
Figure 4.26) represents the weak link. When region B reaches the peak strength and starts
unloading, the concrete in regions A must unload in order to maintain equilibrium. From these
observations it is concluded that concrete failure in compression, as with tension, occurs in a
localized manner and requires special attention in a numerical model.
4.4.3

Regularisation techniques for force-based frame elements

Constant fracture energy criterion

The concept of constant fracture energy in tension is widely used to regularize mesh-sensitive
smeared crack displacement-based elements in continuum FE analyses (Baant and Oh, 1983,
Baant and Planas, 1998, among others). The concept is applied here to force-based beam
elements that soften in compression. While the constant fracture energy concept is not as
widely accepted for compression as it is for tension, experimental research (Lee and Willam,
1997, Jansen and Shah, 1997) and analytical investigations (Markeset and Hillerborg, 1995)
show that the theory also holds true also for localization in compression. The main idea of the
regularization process is to assume that the uniaxial stress-strain relationship for concrete is
supplemented by an additional material parameter, the fracture energy in compression Gcf ,
defined as:

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

153

G cf ' G +dui

(4.40)

where + is the concrete stress and ui the inelastic displacement. The integral represents the
area under the post-peak portion of the compressive stress-displacement curve. This
relationship mimics the tensile fracture energy with a superscript c to indicate compression.
To adapt the fracture energy concept to general use, Eq. 4.40 may be rewritten in terms of
stress and strain:

G cf ' hG +d)i ' LIP G +d)i

(4.41)

where )i indicates inelastic strain and h is a length scale. For smeared crack elements, h
represents the size of a single element in the crush band. For force-based frame elements, h
becomes the length of the softening integration point LIP.
Even though the proposed procedure is general, the regularization is applied here to the Kent
and Park (1971) law used for the concrete fibres of the fibre section. The pre-peak behaviour
is given by a parabola, followed by a linear post-peak softening branch until a stress of 0.2 f c#
is reached at a prescribed strain labelled ) 20 . The residual strength is assumed to remain
constant for strains larger than ) 20 . The fracture energy is defined here from the peak stress
fc# until the end of the softening branch (Figure 4.27). This definition is similar to that of
Jansen and Shah (1997), where the fracture energy is defined from the peak stress until
0.33 f c# . It follows that ) 20 must be calibrated to maintain a constant energy release.
Assuming that G cf is known from experimental tests (Jansen and Shah, 1997), Eq. 4.41,
together with the definition of G cf from Figure 4.27, leads to:

) 20

G cf

0.8 f c'
'
7
8 )0
0.6 f c' LIP
E

(4.42)

where E is Youngs modulus and ) 0 is the strain corresponding to peak stress as indicated in
Figure 4.27.
+

f c'

G cf
h

0.2 f c'
)o

)20

Figure 4.27: Kent and Park (1971) stress-strain law and compression fracture energy (Coleman and Spacone,
2001).

154

4 Analysis! and design of frame structures using non-linear models

Theoretically, Eq. 4.42 implies that the constitutive law must be calibrated for each separate
integration point in a structural model. In practice, plastic hinges generally form at the
element ends where the extreme integration points lie. If all elements in a model are
integrated with the same scheme and with a number of integration points, ) 20 only varies for
elements of different length, L. Linking the constitutive law to the element length is a
straightforward process.
As the number of integration points increases, the extreme integration point must furnish
larger inelastic strains to satisfy the constant fracture energy criteria. This is equivalent to
assuming a constant stress-displacement relationship (Figure 4.28b), rather than a constant
stress-strain law as shown in Figure 4.28a.

+
fc

Given: L, f c, )p, G cf
wIP varies

+
4 IP
5 IP
6 IP

4, 5, 6 IP

)o
(a)

(b)

Figure 4.28: Stress-strain curves with constant fracture energy: a) stress-strain curves for 4, 5, and 6 IP
schemes; b) stress-displacement curve (Coleman and Spacone, 2001).

Curvature post-processing

Once the global force-displacement response is regularized, in some cases there is still a need
to post-process the results to obtain an objective prediction of the curvature demand in the
plastic hinge region. This is due to the fact that the plastic zone length is the length of the first
integration point and does not necessarily correspond to the physical length of the plastic
hinge. As with the inelastic strains in a smeared crack model, the constant fracture energy
criterion is insufficient for regularization of the internal element deformations. Because
different mesh sizes in a smeared crack model must produce the same crack opening
displacements, the magnitude of inelastic strains must vary. While the inelastic strains in a
smeared crack solid finite element do not have any physical meaning and can remain nonobjective, inelastic curvature demands may be of particular interest in frame elements. A
simple procedure is presented below to obtain an objective prediction of the true curvature
demand on the member.
Figure 4.29 shows a deformed interior beam with plastic hinges forming at both ends. The
plastic rotation - i and the relevant displacements O i are indicated at the near (N) and far (F)
ends of the element. These are essentially fictitious quantities used to formulate the curvature
scaling law. The bending moment diagram and the curvature profile corresponding to elasticperfectly plastic behaviour are also shown in Figure 4.29. Inelastic curvatures, indicated by Ci,
are concentrated at the extreme integration points and spread over a length LIP.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

155

Near Node (N)

Far Node (F)

O iF

- iN

- iF
O iN

L
Mp
Moment diagram:

LIP

Mp

!i
Curvature profile:

M
EI

Ce '
LIP

Figure 4.29: Beam with plastic hinge at each end (Coleman and Spacone, 2001).

The total curvature in the plastic hinge region can be separated into elastic and inelastic
curvature components as C ' Ce 8 Ci . Considering the geometry of Figure 4.29, the inelastic
hinge rotation is - i ' Ci LIP . Neglecting the elastic curvature in the other integration points and
using a small angle approximation for - i , the inelastic curvature can be approximated as:

CiMODEL m

Oi
ZL L W
LIP X 7 IP U
2 V
Y2

(4.43)

MODEL

where Ci
indicates the inelastic curvature that results from the analysis. Substituting the
actual length of the plastic hinge Lp for LIP yields a similar approximation for the true
curvature demand:

CiPREDICT m

Oi
Z L Lp W
Lp X 7
U
Y2 2 V

(4.44)

PREDICT

where Ci
indicates the inelastic curvature demand based on the assumed plastic hinge
length Lp. Finally the model output can be scaled according to:

C ' Ce 8 (scalefactor ) CiMODEL

(4.45)

The scale factor is computed by solving for O i in Eq. 4.43 and substituting into Eq. 4.44 to
obtain:

156

4 Analysis! and design of frame structures using non-linear models

scalefactor '

wIP L2 (1 7 wIP )
Lp (L 7 Lp )

(4.46)

The double-curvature case shown in Figure 4.29 prevails in the analyses of buildings under
lateral loads. On the other hand, some structural members, such as cantilevers, experience
single curvature and the plastic hinge forms at one end only. Such is the case of bridge piers
subjected to seismic loads in the transverse direction. The double-curvature derivation is
readily recast in terms of single curvature by replacing the L 2 term in the denominator of
Eqs. 4.43 and 4.44 with the full length L. Taking this single-curvature approach results in:
scalefactor '

wIP L2 (2 7 wIP )
L p 2( L 7 L p )

(4.47)

From the previous discussion, it appears that if the length of the first integration point
corresponds to the length of the plastic hinge, that is if LIP ' Lp , no post-processing of the
curvature is needed because the curvature is objectively predicted by the element. One might
then suggest to select the number of Gauss-Lobatto integration points in such a way that
LIP [ L p . This may cause two problems:
1) the number of integration points may be too small for short elements (causing
undesirable reduced integration) or too large for long elements (increasing the
computational cost); and
2) in most cases, the length of the element would also have to be adjusted, thus
introducing an additional element in each member and greatly increasing the
computational cost of the analyses.
Different integration schemes in which the user can define the length of the integration points
would solve the issue but may compromise the accuracy of the numerical integration. The
regularization approach described above is general and does not affect either the element
formulation or the integration scheme.
4.4.4

Practical considerations

Some practical considerations are presented here to help analysts when using nonlinear
elements for static and dynamic analyses of frames. When a fibre section model is used, one
of the important question is how many fibres or layers are needed to discretize the section.
Spacone (1994) studied the sensitivity of a simple circular section cantilever beam response to
the number of section fibres (Figure 4.30). Their results, and those of similar parametric
studies, indicate that using too many fibres overkills the problem and increases the
computational cost of the analyses. While no precise indication on the number of fibres to use
exists, it appears that for a rectangular section fifteen layers are a good compromise between
accuracy and computational cost in a simple bending analysis. On the other hand, users
should always test their section discretization by comparing the results of different, more and
less refined, fibre meshes to assess the optimal fibre section discretization. Such initial tests
should be performed on simple structures such as a cantilever beam with or without axial
load.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

157

200

1000 kips COMPRESSIVE AXIAL FORCE


GAUSS-LOBATTO INTEGRATION

P,d

SHEAR P (kips)

100

55.78 ft

18 #24

4 integration points

-100

-200

-2

-1

TIP DISPLACEMENT d (ft)

Figure 4.30: Response of a cantilever beam with circular cross section for three different section mesh
refinements.

A similar problem applies to the number of integration points to use in an analysis. Two
integration schemes are typically used for line elements, Gauss and Gauss-Lobatto. The two
schemes, applied to a natural [-1,1] domain, are given by
1

I '

71
1

I '

71

f $R % dR ' P wh f $Rh %

(4.48)

h '1

m71

f $R % dR ' w1 f $R1 ' 71% 8 P wh f $Rh % 8 wm f $Rm ' 1%

(4.49)

h'2

where Eq. 4.48 is the Gauss scheme, and Eq. 4.49 is the Gauss-Lobatto scheme (Stroud and
Secrest, 1966). In Eq. 4.49, h denotes the monitored section and %& is the corresponding
weight factor. The Gauss scheme with m integration points permits the exact integration of
polynomials of degree up to (2m-1), while the Gauss-Lobatto scheme with m integration
points integrates exactly polynomials of degree up to (2m-3). The Gauss scheme is preferred
in displacement-based elements, while the Gauss-Lobatto scheme is preferred in force-based
elements where monitoring the end sections is important (Spacone et al., 1996).
Similar to the issue of the number of fibres, the selection of the number of integration points
is important, because too few points may lead to inaccurate results, while too many
integration points in computationally inefficient and can cause severe strain localization
problems. Figure 4.31 and Figure 4.32 show a study on the sensitivity of the response of a
circular cantilever beam to the number of integration points. Two to five Gauss-Lobatto and
Gauss integration points are used in a single force-based element. From the figures, it appears
that four or five Gauss-Lobatto or Gauss integration points are sufficient. Although these
results have not been generalized, experience shows that more than five integration points are

158

4 Analysis! and design of frame structures using non-linear models

200

1000 kips COMPRESSIVE AXIAL FORCE


GAUSS-LOBATTO INTEGRATION

P,d

SHEAR P (kips)

100

55.78 ft

18 #24

nGL = 2
-100

nGL = 3
nGL = 4
nGL = 5

-200

-2

-1

TIP DISPLACEMENT d (ft)

Figure 4.31: Response of a cantilever beam with circular cross-section for different numbers of Gauss-Lobatto
integration points.

200

P,d

SHEAR P (kips)

100

55.78 ft

18 #24

nG = 2
nG = 3
nG = 4

-100

nG = 5
1000 kips COMPRESSIVE AXIAL FORCE
GAUSS INTEGRATION

-200

-2

-1

TIP DISPLACEMENT d (ft)

Figure 4.32: Response of a cantilever beam with circular cross-section for different numbers of Gauss
integration points.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

159

not needed for accuracy. Analysts are, however, strongly encouraged to run preliminary
parametric studies on a simple structure (a cantilever beam, for example) to assess the number
of integration points necessary to reach a converged solution.
As discussed in the previous section on strain localization, the selection of the number of
integration points is closely related to the length of the plastic hinge that may form in the
structural member being considered. Without the inclusion non-localisation modelling (see
Valipour and Foster, 2007), plastic hinges tend to localize at a single integration point in
force-based elements, while plastic hinges tend to extend to entire elements in displacementbased elements. The details of the localization characteristics of force-based elements have
been previously discussed and should guide analysts in selecting the number of integration
points in an element that is likely to develop plastic hinges. Furthermore, it is important to
note that the plastic hinge length Lp is a quantity that is not easily computed. The following
expression suggested by Paulay and Priestley (1992) suggested that the plastic hinge length be
taken as:

Lp ' 0.08 L 8 0.022 f y db (kN, mm)

(4.50)

where L is the distance from the plastic hinge to the nearest zero moment point, fy is the steel
yield stress and db is the bar diameter. There exist other formulas in the published literature
for estimating the plastic hinge length, mostly obtained from experimental results combined
with analytical studies.

4.5

References

Anagnostopoulos, S. (1981), Inelastic Beams for Seismic Analysis of Structures, Journal of


Structural Engineering, ASCE, 107(ST7), pp. 1297-1311.
Baumann P. (1998), Die Druckfelder Bei Der Stahlbetonbemessung Mit Stabwerkmodellen,
Masters Thesis, The University of Stuttgart, Germany, June.
Bannon, H., Briggs, J., and Irvine, M. (1981), Seismic Damage in Reinforced Concrete
Frames, Journal of Structural Engineering, ASCE, 107(ST9), pp. 1713-1729.
Bayrak, O. (1999), Seismic Performance of Rectilinearly Confined High Strength Concrete
Columns, Thesis submitted in conformity with the requirements for the Degree of Doctor of
Philosophy in the University of Toronto, 339 pp.
Bayrak, O. and Sheikh, S.A. (2001), Plastic Hinge Analysis, Journal of Structural
Engineering, ASCE, Vol. 127, No. 9, September, pp. 1092-1100.
Baant, Z.P. and Oh, B.H. (1983), Crack band theory for fracture of concrete. Materials and
Structures, Vol. 16, pp. 155-177.
Baant, Z.P., and Planas, J. (1998), Fracture and size effect in concrete and other quasibrittle
materials, CRC Press, Boca Raton.
Belarbi, A., and Hsu, T.T C. (1995). Constitutive laws of softened concrete in biaxial tension
compression, ACI Struct. J., Vol. 92, No. 5, 562573.
Bernoulli, J. (1705), Histoire de lAcademie des Sciences de Paris, Paris.

160

4 Analysis! and design of frame structures using non-linear models

Bertero, V.V., Aktan, A., Charney, F, and Sause, R. (1984), Earthquake Simulator Tests and
Associated Experimental, Analytical and Correlation Studies of One-Fifth Scale Model, in
Earthquake Effects on Reinforced Concrete Structures, American Concrete Institute,
SP-84-13, Detroit, pp. 375-424.
Brancaleoni, F., Ciampi, V., and Di Antonio, R. (1983), Rate-Type Models for Non Linear
Hysteretic Structural Behavior, EUROMECH Colloquium, Palermo, Italy.
Charney, F., and Bertero, V.V. (1982), An Evaluation of the Design and Analytical Seismic
Response of a Seven Story Reinforced Concrete Frame-Wall Structure, Report EERC 82-08,
Earthquake Engineering. Research Center, Berkeley.
Coleman, J, and Spacone, E. (2001), Localization Issues in Nonlinear Force-Based Frame
Elements. ASCE Journal of Structural Engineering, Vol. 127, No. 11, pp. 1257-1265.
Clough, R. and Johnston, S. (1966). Effect of stiffness degradation on earthquake ductility
requirements, Transactions of Japan Earthquake Engineering Symposium, Tokyo.
Clough, R., and Benuska, L. (1967), Nonlinear Earthquake Behavior of Tall Buildings,
Journal of Mechanical Engineering, ASCE, 93(EM3), pp. 129-146.
de Borst, R., Feenstra, P.H., Pamin, J., and Sluys, L.J. (1994), Some current issues in
computational mechanics of concrete structures. Computer modelling of concrete structures.
Proceedings of EURO-C, Eds. H. Mang, N. Bi!ani!, R. de Borst, pp. 283-302.
El-Tawil, S. and Deierlein, G.G. (2001), Nonlinear analyses of mixed steel-concrete moment
frames. Part I - beam-column element formulation. Part II - implementation and verification,
ASCE, Journal of Structural Engineering, Vol. 127, No. 6, pp. 647-665.
Filippou, F.C., and Issa, A. (1988). Nonlinear Analysis of Reinforced Concrete Frames under
Cyclic Load Reversals, EERC Report 88-12, Earthquake Engineering Research Center,
Berkley.
Filippou, F.C., D Ambrisi, A., and Issa, A. (1999), Effects of reinforcement slip on
hysteretic behavior of reinforced concrete frame members, ACI Structural Journal, Vol. 96,
No. 3, pp. 327-335.
Filippou, F.C., and Issa, A. (1988), Nonlinear Analysis of Reinforced Concrete Frames under
Cyclic Load Reversals, EERC Report 88-12, Earthquake Engineering. Research Center,
Berkeley.
FIP (1998), Commission 3 on FIP 1996 Recommendation for Practical Design of Structural
Concrete, Fdration Internationale de la Precontrainte, May.
Foster S.J. (1992), An Application of the Arc Length Method involving Concrete Cracking,
International Journal for Numerical Methods in Engineering, Vol. 33, No. 2, January, pp. 269285.
Foster S.J. (1998), Design of Non Flexural Members for Shear, Journal of Cement and
Concrete Composites, Vol. 20, No. 6, pp. 465-475.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

161

Foster S.J., and Gilbert R.I. (1990), Non-linear Finite Element Model for Reinforced
Concrete Deep Beams and Panels, UNICIV Report No.R-275, School of Civil Engineering,
University of New South Wales, Kensington, December, 113 pp.
Giberson, M. (1967). The Response of Nonlinear Multi-Storey Structures Subjected to
Earthquake Excitations, Earthquake Engineering Research Laboratory, Pasadena.
Hooke, R. (1678), Lectures De Potentia Restitutiva, or of Spring Explaining the Power of
Springing Bodies, printed for John Martyn Printer to The Royal Society, at the Bell in St.
Pauls Church-Yard, 24 pp.
Iwan, W. (1978), Application of Nonlinear Analysis Techniques, in, Iwan W. ed., Applied
Mechanics in Earthquake Engineering, ASME, AMD, 8, New York, pp. 135-161.
Jansen, D.C. and Shah, S.P. (1997), Effect of length on compressive strain-softening of
concrete. J. Eng. Mech., ASCE, Vol. 123, No. 1., pp. 25-35.
Kaufmann, W. (1998), Strength and deformations of structural concrete subjected to in-plane
shear and normal forces, Report No. 234, Institute of Structural Engineering, ETH, Zurich,
Switzerland.
Kauser, M., and Mehlhorn, G. (1987), Finite elements models for bond problems. Journal
of Structural Engineering, ASCE, Vol. 113, No. 10, pp. 2160-2173.
Kent, D.C. and Park, R. (1971), Flexural Members with Confined Concrete, Journal of the
Structural Division, ASCE, Vol. 97, ST7, July 1971, pp. 1969-1990.
Lai, S., Will, G. and Otani, S. (1984), Model for Inelastic Biaxial Bending of Concrete
Members, Journal of Structural Engineering, ASCE, 110(ST11), pp. 2563-2584.
Lee, Y-H., and Willam, K. (1997), Mechanical properties of concrete in uniaxial
compression. ACI Materials Journal, Vol. 94, No. 6, pp. 457-471.
Limkatanyu, S. and Spacone, E. (2002), R/C Frame Element with Bond Interfaces. Part 1:
Displacement-Based, Force-Based and Mixed Formulations. Part 2: Element State
Determination and Numerical Validation. ASCE Journal of Structural Engineering, Vol. 128,
No. 3, pp. 346-364.
Mander, J.B., Priestley, M.J.N., and Park, R. (1988), Observed Stress-Strain Behaviour of
Confined Concrete, Journal of Structural Division, ASCE, Vol. 114, No. 8, August 1988, pp.
1827-1849.
Markeset, G., and Hillerborg, A. (1995), Softening of concrete in compression localization
and size effects. Cement and Concrete Research, Vol. 25, No. 44, pp. 702-708.
Marti, P. (1985a), Basic Tools of Reinforced Concrete Beam Design, ACI Journal
Proceedings, Vol. 7, No. 1, Jan.-Feb., pp. 46-56.
Marti, P. (1985b), Truss Models in Detailing, Concrete International, American Concrete
Institute, Vol. 7, No. 12, Dec. pp. 66-73.

162

4 Analysis! and design of frame structures using non-linear models

Martino, R., Spacone, E., and Kingsley, G. (2000), Nonlinear Pushover Analysis of RC
Structures. ASCE Structures Congress, Advanced Technology in Structural Engineering, M.
Elgaaly editor, Philadelphia, PA, May. 8 pp (CD-ROM).
Monti, G., Filippou, F.C., and Spacone, E. (1997a), Finite element for anchored bars under
cyclic load reversals. Journal of Structural Engineering, ASCE, Vol. 129, No. 5, May
Monti, G., Filippou, F.C., Spacone, E. (1997b), Analysis of hysteretic behavior of anchored
reinforcing bars. ACI Structural Journal, Vol. 94, No. 2, May-June
Monti, G., and Spacone, E. (2000), Reinforced Concrete Fiber Beam Element with BondSlip. ASCE Journal of Structural Engineering, Vol. 126, No. 6, pp. 654-661.
Mrsch, E. (1902), Der Eisenbetonbau, seine Theorie und Anwendung (Reinforced
Concrete, Theory and Application), Stuggart, Germany.
Navier, C-L. (1826), Resume des lecons de la resistance des corps solides, Paris.
Otani, S. (1974), Inelastic Analysis of R/C Frame Structures, Journal of the Structural
Division, ASCE, 100(ST7).
Ozdemir, H. (1981), Nonlinear Transient Dynamic Analysis of Yielding Structures, Ph. D.
Thesis, Department of Civil Engineering, University of California, Berkeley.
Paulay, T., and Priestley, M.J.N. (1992), Seismic design of reinforced concrete and masonry
buildings, Wiley, New York
Petrangeli, M., Pinto, P.E. and Ciampi, V. (1999), Fiber element for cyclic bending and shear
of RC structures. I: theory ASCE Journal of Engineering Mechanics, Vol. 125, No. 9, pp.
994-1001.
Prager, W. and Hodge, P. (1951), Theory of Perfectly Plastic Solids, John Wiley and Sons,
New York.
Ritter, W. (1899), Die Bauweise Hennebique (Construction Techniques of Hennebique),
Schweizerische Bauzeitung, Zurich, February.
Rubiano-Benavides, N.R. (1998), Predictions of the inelastic seismic response of concrete
structures including shear deformations and anchorage slip. Ph.D. dissertation, Department
of Civil Engineering, University of Texas, Austin.
Schlaich, J., Schfer, K., and Jennewein, M. (1987), Toward a Consistent Design of
Structural Concrete, Journal of the Prestressed Concrete Institute, Vol. 32, No. 3, May-June,
pp. 74-150.
Scott, M.H., and Fenves, G.L. (2006), Plastic hinge integration methods for force-based
beam-column elements. Journal of Structural Engineering, Vol. 132, No. 2, pp. 244-252.
Sheikh, S.A. and Khoury, S.S. (1993), Confined Concrete Columns with Stubs, ACI
Structural Journal, V. 90, No. 4, July-August 1993, pp. 414-431.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

163

Sheikh, S.A. and Uzumeri, S.M. (1982), Analytical Model for Concrete Confinement in Tied
Columns, Journal of the Structural Division, ASCE, Vol. 108, ST12, December 1982, pp.
2703-2722.
Spacone, E. (1994), Flexibility-Based Finite Element Models for the Nonlinear Static and
Dynamic Analysis of Concrete Frame Structures Ph.D. Dissertation, Department of Civil
Engineering, University of California, Berkeley.
Spacone, E., Filippou, F.C., and Taucer, F.F. (1996), Fiber Beam-Column Model for
Nonlinear Analysis of R/C Frames. I: Formulation. II: Applications. Earthquake Engineering
and Structural Dynamics, Vol. 25, No. 7, pp. 711-742.
Takayanagi, T. and Schnobrich, W. (1979), Non Linear Analysis of Coupled Wall Systems,
Earthquake Engineering and Structural Dynamics, Vol. 7, pp. 1-22.
Takeda, T., Sozen, M.A., and Nielsen, N. (1970), Reinforced Concrete Response to
Simulated Earthquakes, Journal of Structural Engineering, ASCE, 96(ST12), pp. 2557-2573.
Takizawa, H. (1976), Notes on Some Basic Problems in Inelastic Analysis of Planar RC
Structures, Trans. of Arch. Inst. of Japan, 240, Part I in February, pp. 51-62, Part II in March,
pp. 65-77.
Valipour H.R., and Foster, S.J. (2007), A Novel Flexibility Based Beam-Column Element
for Nonlinear Analysis of Reinforced Concrete Frames, UNICIV Report No. R-447, School
of Civil and Environmental Engineering, The University of New South Wales, September.
Vecchio, F.J. (1989), Nonlinear Finite Element Analysis of Reinforced Concrete Membrane
Elements, ACI Journal, Proceedings Vol. 86, No.1, Jan-Feb., pp. 26-35.
Vecchio, F.J. (1990), Reinforced Concrete Membrane Element Formulation, Journal of
Structural Engineering, Vol. 97, No. 1, pp.102-110.
Vecchio, F.J. (2000), Disturbed stress field model for reinforced concrete: formulation, J.
Struct. Eng., Vol. 126, No.9, 10701077.
Vecchio, F., and Collins, M.P. (1982), The Response of Reinforced Concrete to In-Plane
Shear and Normal Stresses, Publication No. 82-03, Department of Civil Engineering,
University of Toronto, March, 332 pp.
Vecchio, F.J., and Collins, M.P. (1986), The Modified Compression Field Theory for
Reinforced Concrete Elements Subjected to Shear, ACI Journal, Vol. 83, No. 2, Mar.-Apr.,
pp. 219-231.
Vecchio, F.J., and Collins, M.P. (1993), Compression response of cracked reinforced
concrete. J. Struct. Eng., Vol. 119, No. 12, 35903610.

164

4 Analysis! and design of frame structures using non-linear models

Analysis and design of surface and solid structures using


non-linear models

5.1

Introduction

As concrete can easily be moulded into any form desired by architects and/or structural
designers, concrete structures and members, normally reinforced with steel bars or
prestressing strands or wires, are usually the first choice of building materials when it comes
to constructing 2D structures such as walls, slabs and shells. Simplified methods of analysis
and design (e.g. strut and tie modelling as described in Chapter 8) for determining the ultimate
loads of such structures have proved advantageous in many cases of structural design.
However, as highlighted in Chapter 3, cases exist where more refined models are needed to
adequately assess the behaviour of structures and structural members under service and
ultimate loads and for ensuring cost effectiveness.
The transition from 1D modelling to 2D or 3D modelling gives rise to additional complexity
when dealing with a brittle material such as concrete. Examples of such complexities include:
"

The material models used for concrete not only have to take account of the bi- and
tri-axial behaviour of the materials, it is also necessary to formulate models in such a
way that they provide, in combination with the underlying numerical model (e.g. FEM
or BEM), stable and fast solution procedures.

"

More parameters are needed to determine the material behaviour. Strength limits and
their statistical scatter, measured in laboratory tests, have to be adopted for calculating
full scale structures.

"

Softening in cracked or crushed concrete can lead to solutions that depend on the kind
of discretisation chosen, as described in Chapter 4. Proper regularisation techniques
have to be applied.

"

When using displacement based finite elements in combination with smeared crack
concepts, mesh dependencies such as directional bias of cracks, spurious kinematic
modes and/or stress-locking may occur.

Depending on the task to be performed, the user has to choose the right method of
computation in order to avoid over-sophistication on the one hand, yet including the decisive
non-linear effects of a given problem on the other. This chapter deals only with the
displacement based non-linear finite element method. Some approaches that combine the
lower bound theorem of the theory of plasticity with linear optimisation algorithms and with
other numerical solution techniques are described in Chapter 9.

5.2
Ec
Eh

)c
) cu
) tu

Notation
initial elastic modulus of concrete
Complementary energy
hardening modulus of steel
concrete strain when concrete strength is reached
ultimate compressive strain of concrete
cracking strain of concrete

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

165

)t
Gf
ft
fcm
fc
R
(
H
e
C, C
e
Cs, Cs
nx, n y
mxx, myy, mxy
mp, mpx, mpy
e
f
e
u
e
K
d
W
q, p
Rd
Sd
SdL
SdNL
aS

&R
^
wd

tensile strain
fracture energy of concrete
direct tensile strength of concrete
mean value of concrete compressive strength
concrete compressive strength
vector of point loads
vector of nodal force parameters
deflection matrix
matrix containing coefficients of a linearized yield criterion
vector of yield moments
shear forces per unit of length
bending and twisting moment per unit of length
plastic moment per unit of length
vector of generalized nodal forces
vector of nodal displacements
stiffness matrix
vector of design variables
matrix of coefficients corresponding to d of a weight function
distributed loads
design value of resistance
design value of action
design value of cross-sectional action assuming linear elastic material
design value of cross-sectional action assuming material nonlinearity
area of reinforcement per unit of length
partial safety factor for structural resistance
load factor
total deformation of compressive softening zone
Poissons ratio of concrete

5.3

2D Structures with in-plane loading

5.3.1

Introduction

To demonstrate the practicalities of using non-linear FE analysis, we shall review the results
of modelling of a two span deep beam tested by Cervenka and Gerstle (1971). The
dimensions of the beam are shown in Figure 5.1a and the reinforcement arrangements for the
beam are shown in Figure 5.1b with the reinforcement consisting of #3 (USA) deformed bars
placed as shown (note that only one half of the specimen is shown with symmetry).
The properties of the concrete and the reinforcing steel as used in the analysis are shown in
Table 5.1.

166

5 Analysis and desing! of surface and solid structures using non-linear models

(a) Dimensions and loading.

(b) Reinforcement by no.3 bars.

Figure 5.1: Details of Cervenka and Gerstle (1971) panel W2.


Table 5.1: Material properties for the Cervenka and Gerstle (1971) deep beam.

Concrete

Ec
'
fcm
"c
wd
ft
Gf *

20 250 MPa
0.2
26.75 MPa
0.00264
0.5 mm
3.1 MPa
58 N/m

Reinforcement

Es
fy
Eh

190 000 MPa


353 MPa
700 MPa

Note: * Fracture energy (Gf ) calculated using the CEB-FIP Model Code 1990 (1993).

A plane stress analysis was performed using 4-node quadrilateral isoparametric finite
elements, using 4 integration points per element and with the constitutive relationship for the
concrete taken as the fracture-plastic model of Cervenka (2002). In this approach the tensile
behaviour is described by the fracture energy-based smeared crack model and the
compressive behaviour by a non-associated plastic flow rule. Softening of concrete behaviour
in both tension and compression are considered. It is worthy of note that the tension stiffening
effect is modelled indirectly through localization of strains and no specific tension stiffening
parameter appears in the model. Perfect bond between concrete and reinforcement is assumed.
The deep beam was modelled using two different FE meshes; a rectangular mesh (Figure
5.2a) and a triangular mesh (Figure 5.1b). The sensitivity of the results to the element size and
shape was compared using three rectangular meshes with 5, 10 and 20 elements across the
span and with one mesh of 10 triangular elements across the span. The results of the analyses
are shown in the form of load versus displacement plots in Figures 5.3 and 5.4 and cracking
patterns in Figures 5.5 and 5.6.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

167

(a) Quadrilateral elements.

(b) Triangular elements.

Figure 5.2: Finite element meshes (Mesh 10).


140

Force P [kN]

120
100
80
Experiment

60

Mesh 5

40

Mesh 10

20

Mesh 20

0
0

2
3
Displacement [mm]

Figure 5.3: Load versus displacement: quadrilateral mesh results compared with the experimental data.

120

Force P [kN]

100
80
60

Experiment

40

Mesh 10 Quadrilateral

20

Mesh 10 Triangular

Displacement [mm]
Figure 5.4- Effect of element type on deep beam response.

168

5 Analysis and desing! of surface and solid structures using non-linear models

Figure 5.5: Comparison of experimental and FE crack patterns at failure (Mesh 10).

(a)

(b)

Figure 5.6: a) Cracks and compressive stresses (Mesh 20); b) plastic strains in bars (Mesh 10).

The behaviour of the deep beam includes the propagation of cracks from early stages of
loading up to failure. At the maximum load, the horizontal bars at the bottom yield and
deform plastically and the concrete at the top right corner fails in compression. The analyses
show that the analytical models can reflect the real behaviour with accuracy for this failure
mode of this case type. They, thus, provide for acceptable engineering solutions for this
failure model for this case. To fully validate the model, other benchmark analyses need to be
considered for other failure modes, as outlined in Chapter 7 of this report.
In addition to strength results, the analyses can show the trends of various effects, which can
be utilized in practical cases. For example, we observe the response during the crack
propagation before maximum load is reached (that is, the ascending branch of loaddisplacement diagram). In this stage the response is dominantly affected by interaction of
partially cracked concrete and reinforcement, the so called tension stiffening effect. This
behaviour can be observed on the load displacement diagrams of Figures 5.2 and 5.3 and in
the crack patterns, Figures 5.4 and 5.5. The study shows that the larger elements give lower
stiffness in this stage and that smaller elements reflect better individual discrete cracks due to
better modelling of strain localization. If we continue to increase the number of elements, so
that the element size is several times smaller then the crack spacing, the crack pattern would
be close to the experimentally observed discrete cracks. However, in practical cases extremely
large numerical models are computationally expensive and the tension stiffening effect may
be underestimated.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

169

If we use triangular elements of a similar mesh size, the response is stiffer in the postcracking, serviceability, range (see Figure 5.4). The reason for the stiffer response of the
triangular mesh is due to the effect of the shape function of the triangular element that
produces a constant strain field, while the quadrilateral element has a semi-linear strain field.
Next we examine the effect of the mesh size on the maximum load at failure. The load
carrying capacity is exhausted by the attainment of the yield stress of reinforcement in the
bottom portion of the beam and the surpassing of the compressive strength of the concrete in
the top right corner. This is a typical bending mode of failure and larger elements give a
higher load response. The higher response is also reached by triangular (more stiff) elements.
In this particular case, the quadrilateral meshes 10 and 20 give near identical responses for the
maximum load and indicates that mesh 10 is sufficiently fine, while mesh 5 and the triangular
mesh 10 give slightly higher failure loads.
The descending shape of the load displacement diagrams after the peak indicates a limited
ductility, which is influenced by the softening model adopted for the concrete in compression.
This limited ductility is not observed in experiments, where concrete behaviour in the panel
corner may be influenced by the confining effect of the thickened central rib forming the
central support. This difference can be reduced by assuming a perfectly plastic behaviour for
the concrete, in this region, but such assumptions are not rationally founded since it applies
only to a small region of a panel and must be used with care.
This case study illustrates some of the general features valid for numerical simulation of
reinforced concrete structures. In concluding, we note that:
1. Two distinct constitutive models for concrete (smeared cracks, fracture energy-based
crack band) and for reinforcement (bi-linear) are capable of simulating reasonably
well the behaviour of our reinforced concrete panel. Other models, however, can be
adopted and it is for the user of such software to ensure that the models being used for
their problem are capable of predicting the response with accuracy. Benchmarking is
discussed in more detail in Chapter 7.
2. Finite element size and type can significantly influence the structural response. A
mesh sensitivity study is required to show the objectivity of results.
3. A coarse mesh can underestimate the structural resistance in post-cracking stage of the
analysis and overestimate the ultimate load capacity.
4. Low order elements overestimate the load response in the whole range.
These conclusions are true for reinforced concrete structures with more or less uniformly
distributed reinforcement, such as for the deep beam example illustrated. Structures with
discrete reinforcement (for example, beams without web reinforcement) are less sensitive to
the underestimation of tension stiffening but can be more dependent on the fracture model
adopted for the concrete (see Cervenka, 1998).

5.4

Plate and shell structures

5.4.1

Layered elements

One possibility for modelling of 2D Structures is the application of volume elements. The
reason why one should refrain from doing so is twofold; firstly, it is known from the nature of
the 2D problem, that stresses and strains normal to the middle surface can be neglected. Using
volume elements requires significant computational effort in calculating something that is
known a priori. Secondly, it is necessary to choose the volume elements with lateral
dimensions of the same magnitude otherwise the stiffness matrix becomes ill conditioned

170

5 Analysis and desing! of surface and solid structures using non-linear models

and resuults in an ennormous nuumber of volume elemeents as theirr characterisstic length has
h to be
of the order of the plate
p
or sheell thicknesss.
By negllecting out of
o plane noormal stressees and assu
uming a lineear variationn of the straains over
the elem
ment heightt, one arrivees at a 5 DOF
D
shell ellement. Thee strain statte of the eleement is
thus fully characterrized by 3 displacemen
d
nts and 2 in plane rotational DOF. Finite elem
ments for
plates annd shells caan be conceiived as an assemblage
a
of plane strress layers ((see Figure 5.7).

Figure 5.77: Layered sheell element

With a linear
l
strainn variation over
o
the element thickn
ness, the strrain state "xxx(k), "yy(k), #xy(k) for
layer k is
i given by:

) xx (k ) ' ) xx 8 S xxx T z (k )
) yy (k ) ' ) yy 8 S yyy T z (k )

(5.1)

& xy (k ) ' & xy 8 S xy T z (k ) T 2


where "xx, "yy, #xy, (xx, (yy annd (xy (see Figure
F
5.8) are the strrains and thhe curvaturees in the
middle plane and z(k)
z designaates the disttance of lay
yer k from the
t referencce surface. With
W the
strains given
g
assum
ming a planee stress statte, the distriibuted internnal forces oof each layeer can be
calculatted via material modeels for the reinforcem
ment and concrete. Seectional forrces and
momentts are attainned by inteegrating thee layer conttributions thhrough the element th
hickness.
The elem
ment stiffneess matrix can
c be evaluuated in a similar way by
b adding uup the layer stiffness
matrices.

Fiigure 5.8: Nottations for thee strains and curvatures


c
of the
t shell elemeent

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

171

To verify the approach described above, the orthogonally reinforced plate element M7, tested
by Marti et al. (1987), is evaluated. The specimen had dimensions of 200 x 1700 x 1700 mm
and was loaded with pure torsional moments in the reinforcement directions. Torsion was
applied by couples of equal and opposing forces, P, applied at the corners B and D of the
specimen (Figure 5.9a) with the specimen supported at the corners A and C. The
reinforcement data is shown in Figure 5.9b (the reinforcement ratio is *x = * y = 0.25%) and
the material properties are shown in Figure 5.9c.
A FE analysis of plate ML7 was performed for a single, layered, shell element. The element
was divided into concrete layers and reinforcement layers and the numerical computations
were performed both with and without including tension stiffening in the formulation. Due to
the low reinforcement ratios of the plate, tension stiffening is shown to have little influence on
the results, as shown in the moment versus curvature relationships for the plate plotted in
Figure 5.10.

Concrete:
Ec = 35500 MPa
n = 0.20
fc = 44.4 MPa
ft = 4.0 MPa
)c = 0.0025
) cu/) c = 1.5
)t/) ucr = 10
Reinforcement:
fy = 479 MPa
ES = 200 000 MPa
Eh = 0.0 MPa

(a)
(b)

(c)
Loading and deformation of test specimen ML7 (Marti, 1987):
Reinforcement; and c) material properties.

Moment - M [kNm/m]

Figure 5.9

a) applied load; b)

50
40
30
Experiment
without TSE
with TSE

20
10
0
0

10

15

20

25

30

35

Curvature 7 f [1/mm]*1E-6
Figure 5.10: Moment-curvature relationships for plate ML7.

172

5 Analysis and desing! of surface and solid structures using non-linear models

5.5

Three dimensional solid structures

5.5.1

Introduction

There are many ways that 3D solid structures can be modelled and, as for any non-linear
model, it is important that the model be calibrated to the problem being considered and
verified using appropriate procedures. In this section, three of the more common methods for
the development of models for the analysis of 3D structures are discussed: namely, models
developed based on non-linear elasticity, models developed on the theories of plasticity and
fracture and microplane models. Some examples of the implementation of such models are
presented in the sections following.
5.5.2

Models based on non-linear elasticity

Although local stress/strain fields of a reinforced concrete plate or solid element subjected to
uniform external forces are irregularly distributed, the load carrying mechanism can be
reduced to a combination of space averaged one-dimensional stress flows developing in the
cracked concrete and steel. The space-averaged constitutive models are the basis of 3D
modelling of reinforced concrete solid elements. The compression-tension field modelling of
1D has been successfully applied to 2D modelling with the same concept extended to 3D
problems.
Shear stresses on a crack plane of a reinforced concrete element cause rotation in the direction
of principal stresses from those at the initial cracked state. It has been shown that rotation of
the principal stress may lead to formation of cracks in a new direction with almost full closure
of the initial cracks (Vecchio and Collins 1982, Stevens et al. 1987). By taking the mechanical
condition of the latest developed cracks, the concept of a rotating crack approach was
implemented into FEM codes. In this approach, the local axes of cracked concrete in three
directions (1,2,3) are considered to coincide with those of principal stress axes and
information on previously developed cracks is not necessarily kept in the analysis. The
compression field theory (CFT) (Mitchell and Collins, 1974) and subsequent modified
compression field theory (MCFT) (Vecchio and Collins, 1986) provided a valuable
contribution in the implementation of two-dimensional FE analysis of reinforced concrete.
Vecchio and Selby (1991) extended the 2D rotating-crack-based constitutive models of the
MCFT to the 3D case. In their approach, prior to cracking the reinforced concrete was taken
to be a linear elastic, isotropic, material. After cracking, the concrete and reinforcement are
considered to contribute separately to the structural element stiffness and its response to
boundary tractions. Cracked concrete is modelled as a nonlinear orthotropic, pathindependent, material with the principal axes corresponding to the direction of the principal
average strains. For concrete in the direction of the largest principal compressive strain ()3),
the average stress-strain relationship is based on the MCFT (Figure 5.11a), in consideration of
the effect of the coexisting tensile strain in the 1-direction ()1) and its cause of degradation of
the concrete compressive stiffness and strength in the 3-direction. For concrete in tension and
before cracking, in the direction of principal tensile strain ()1), a linear relationship is used.
Whereas, after cracking, a decay function is adopted to represent tension stiffening effect
(Figure 5.11b). The intermediate principal stress (in the 2-direction) is evaluated according to
the two-dimensional MCFT, depending on its strain. If the strain is tensile, the tension
stiffening model is used and if compressive the compression model of MCFT is used.
Reinforcing bars are modelled using a tri-linear relationship with strain hardening.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

173

2: intermediate
principle direction

-fc3
fc

fc3max

1: direction of normal
to crack

Softening

3: direction of largest
principle compressive
strain

)0

-) 3

(a)

fc1
fcr

Tension stiffening
Local equilibrium

) c1

) cr
(b)

Figure 5.11: Concrete material models adopted in the MCFT and applied to 3D. (Vecchio and Selby, 1991).

To validate their approach, Vecchio and Selby (1991) conducted series of numerical tests
using the constitutive relationships described above. One of these tests is shown in Figure
5.12 for the analysis of hollow reinforced concrete box beam sections tested by Onsongo
(1978). The test beams, measuring 410 mm by 508 mm in section and with a span of
2290 mm, were subjected to a constant moment and torque along their length. For the
analysis, 750 mm lengths of the beams were modelled with a mesh of 1200 elements and
1760 nodes and resulting 5280 degrees of freedom. The torque was applied as nodal forces at
the end of the bars. The longitudinal reinforcement was uniformly smeared over the
respective areas. The comparison of analytical results in terms of capacity prediction and
load-deformation is shown in Figure 5.12b and Figure 5.12c. It is seen that reasonably
accurate predictions were obtained for the strength of both under-reinforced and overreinforced sections.
An alternative approach to rotating crack approach of Vecchio and Selby (1991) is the multidirectional smeared crack model. This model allows for consideration of several cracks within
a single element that need not necessarily be orthogonal to existing cracks. Here, information
of all cracks assigned for each crack direction is stored in the analysis. The number of
allowable cracks is usually limited, however, to reduce the complexity of computation (de
Borst and Nauta, 1985, Barzegar 1989, Hofstetter and Mang, 1995, Fokuura and Maekawa,
1998).
The occurrence of cracks in a 3D space can be approximated by quasi-two-dimensional stress
fields. With this assumption Maekawa et al. (1997) proposed a strain decomposition and
stress re-composition approach to obtain the 3D stress field (Figure 5.13) based on a 1D stress
flow field. In each 2D subspace (after decomposition of strains) the partial stresses rooted in
the crack projection are computed using path-dependent 2D constitutive laws (Okamura and
Maekawa, 1991, Fokuura and Maekawa, 1998).

174

5 Analysis and desing! of surface and solid structures using non-linear models

Figure 5.12: Comparison of three-dimensional analysis using a MCFT formulation with experimental results:
a) experimental layout; b) and c) comparison of results (Vecchio and Selby, 1991).

composition of
decompoded
plane-base
partial stresses

+//
+..
+/.

in-plane
constitutive
modelling

(1,2)

(1,2) decomposed plane

(1,3) decomposed
plane

+//
+22
+/2

2
1

(1,3)

(2,3) decomposed plane

+..
+22
+.2

(2,3)

Figure 5.13: Breakdown and re-composition of load carrying mechanism of 3D cracked solids of concrete

The total stress is computed on the basis of the partial stress on the three subspaces. Hauke
and Maekawa (1998) took into account the difference of concrete mechanics near and far
from the reinforcing bars through consideration of two different zones in the reinforced
concrete space (Figure 5.14a). One for the concrete near the steel bars and subjected to
tension-stiffening behaviour, the so called reinforced concrete zone (due to influence of bond
on the concrete stiffness), and a second for the concrete far from reinforcement the so called
plain concrete zone which may have a localized crack in the concrete control volume (An
et al., 2000, Hauke and Maekawa, 1998). The post cracking tension model in this approach is
expressed as
!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

175

+ t ' f t ( ) tu / ) t ) c

(5.2)

where c is a parameter representing the steepness of the descending branch and ) tu is the
cracking strain. In order to avoid mesh dependency of the results when dealing with
localization of cracks in plain concrete zone, the softening parameter in Eq. (5.2) was defined
according to fracture energy and the crack band width (Baant and Oh, 1983). On the other
hand, for the reinforced concrete zone, the softening parameter, c, is taken as constant and
equal to 0.4, irrespective of element sizes. If such a crack is located in the plain concrete zone,
the softening behaviour varies with the crack orientation. Also, for a control volume
containing smeared reinforcing bars in one or several directions, the average softeningstiffening behaviour depends on the crack inclination. Hauke and Maekawa (1998) used a 2nd
order interpolation function for considering the influence of crack inclination on the tension
softening-stiffening parameter. The normalized interpolated fracture energy (area under the
tension stiffening model as shown in Figure 5.14) in the desired direction, n, is computed as

G (n) '
*
f

n12 G *f (1) 8 n22 G *f (2) 8 n32 G *f (3)

(5.3)

n12 8 n22 8 n32

from which the tension stiffening/softening parameter is computed. In Eq. 5.3, ni (i = 1, 2, 3)


is the component of the crack normal unit vector and Gf*(i) is normalized released energy in
direction i.

(a)

(b)
2

+t / ft

+t / ft
)t

(2)

Crack
normal

n2
2
3

Mixed behaviour
Gf*(n) {Gf*(1), Gf*(2), n }
=> c(n%5or c(-)

Softening
Gf*(2) {Gf, lr,2, ft}
=> c(2)

)t

Directional
angle

RE-bar

n1

3
2
1

Stiffening
Gf*(1) {c(1) = 0.4}
+t / ft

(1)

)t

imaginary orthogonal
reinforcement system

Figure 5.14

a) Three dimensional zoning approach; b) interpolation scheme for tension stiffening model
(Hauke and Maekawa 1998).

For the reinforcing steel, the average yield stress of the bars is taken as a function of an
average uniaxial strain-stress relationship. In developing this relationship, the influence of
crack inclination with respect to direction of reinforcing bar is considered (see Hauke and
Maekawa, 1998, and Maekawa et al. 2003).
As an example, a 1/5 scale model tested by Ichikawa et al. (1998) of a pier under constant
axial load of 147 kN, applied at an eccentricity of 700 mm from the centre of cross section is
considered (Tsuchia et al., 2001). The outline of the experiment is shown in Figure 5.15 with
cyclic torsional loading applied to the specimen as shown in Figure 5.15d. In such a coupled
action, cracks developed in three directions, i.e. two-way shear cracks due to reversed cyclic

176

5 Analysis and desing! of surface and solid structures using non-linear models

torsion and shear as well as horizontal cracks due to flexural loading. The comparison of the
horizontal load versus displacement relationships recorded from the experiment and
calculated in the 3D FE analysis and, also, the torsion versus displacement relationships and
longitudinal displacement are shown in Figure 5.16. A good correlation is observed for the
numerical modelling when compared with the test results.
5.5.3

Fracture-plasticity modelling

Crack modelling based on the theory of plasticity permits a unified treatment of the
compressive and tensile behaviour of concrete. This type of concrete modelling is usually
developed from the rate constitutive equations (Hofstetter and Mang, 1995). Several
constitutive models in 3D strain space exist on the fracture-plasticity formulation of plain
concrete. For example, Pramono and Willam (1989) developed a fracture energy-based
plasticity constitutive model that accommodates strain hardening and softening. They adopted
a fracture energy release approach to describe the degradation of triaxial strength in tension
and in compression in concrete with low levels of confinement.
A unified treatment of plasticity and fracture can effectively make the return mapping
algorithm possible (Wilkins, 1964) which guarantees the solution for all magnitudes of strain
increments (Hofstetter and Mang, 1995, Cervenka and Cervenka, 1999). The main difficulty
in the development of a unified model for both the compressive and tensile regions is the
algorithm for the combination of the two models.
On the basis of plasticity theory, fracture can be modelled by using a Rankine yield criterion
(maximum tensile stress criterion) as:

Fi ' + i 7 f T ' 0

(5.4)

where +i (i =1, 2, 3) is the principal stress and fT is the tensile strength of the concrete. An
appropriate softening law should be considered and can be taken as a function of the
accumulated damage of the concrete (Feenstra, 1993) such that fT in Eq. 5.4 is given as a
function of a damage variable. One such model is the crack band model of Baant and Oh
(1983), which can be used to describe mesh size-dependent softening laws.
For hardening/softening plasticity of concrete there exists several models to describe concrete
crushing. For example, the failure surface of Drucker and Prager (1952), the five-parameter
ultimate strength surface of Willam and Warnke (1975), the four parameter model of Ottosen
(1977) and three-parameter failure surface of Menetrey and Willam (1995).
Cervenka et al. (1998) developed an algorithm for the combination of plasticity with fracture.
The model is able to handle physical changes like crack closure and is not restricted to any
particular shape of hardening or softening laws. The model can be used to simulate concrete
cracking, crushing under high confinement and crack closure due to crushing in other material
directions (Cervenka and Cervenka, 1999). The fracture is modelled by an orthotropic
smeared crack model based on the Rankine tension criterion (Eq. 5.4). Further, it is assumed
that strains, and stresses, are converted into the material directions that, in case of a rotating
crack model, correspond to the principal directions and, in the case of fixed crack model, are
given by the principal directions at the onset of cracking. The Menetrey and Willam (1995)
three-parameter failure surface was used to model concrete crushing.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

177

Figure 5.15: Experim


mental verificaation of the multi-direction
m
n fixed crackk approach: aa) applied loa
ads to the
specimenn; b) geometryy of specimen, c) cross sectiional force; d)) loading methhod (Ichikawaa et al., 1998).

(a)

(b)

(c)
Figure 5.16: Compariison of multi--directional force-displace
f
ement relationnships for a R
RC pier: a) horizontal
h
loading versus horizzontal displaccement; b) force
fo
deviatioon torsion veersus displacement; c)
displacem
ment in the traansverse and longitudinal directions
d
(Tsuchia et al., 22001).

178

5 Analysis and desing! of surface and solid structures using non-linear models

In the Cervenka et al. model, the strain is decomposed into elastic, plastic and fracture
components, as introduced by de Borst (1986). The new stress state is then computed from the
increment of each strain component based on the material used. The combined algorithm
determines the separation of strains into plastic and fracturing components in an iterative
scheme as the stress equivalence in each model must be preserved. Details of algorithm can
be found in Cervenka et al. (1998).
As an example of the verification of the model, an experiment test on pre-stress cable anchors
is simulated (Cervenka and Cervenka, 2007). The pre-stressing force is transferred from the
tendon to the concrete through special cylindrical anchors embedded into concrete. The
anchor is surrounded by spiral reinforcement and loaded by compressive forces to simulate
the action of the pre-stressing. A comparison of the load versus displacement obtained from
the FE model and the capacity of the anchorage, as obtained from the experiment, is shown in
Figure 5.17.
A second example, presented in Figure 5.18, shows plastic strains at failure for a bridge over
the Berounka River near Prague in the Czech Republic (Cervenka and Cervenka, 2007). The
model includes the curved box girder, the piers and the foundation. The box girder and piers
were modeled using higher order shell elements; the foundation by brick and tetrahedral
volume solid elements and the pre-stressing cables by string elements. Various stages at
service conditions were analysed as well as verification of the strength limit state. The
example shows how nonlinear analysis can be applied to complex large structures for
optimization of design and for verification at various limit conditions.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

179

Figure 5.17: a) FE mesh


m
for pre-sstressing cablee anchor ana
alysis; b) simuulated failure mode; and c) FE load
versus displacement
d
compared with
w
peak loa
ad recorded in the experriment (Cervvenka and
Cervenka
ka, 2007).

Figure 5.18: FE analyysis of the Berrounka River Bridge near Prague,


P
Czech Republic: vview from belo
ow ground
level shoowing plastic strains
s
at failuure (Cervenka
a and Cervenkka, 2007).

180

5 Analysis and desing! of surface and solid structures using non-linear models

5.5.4

Microplane model

In dealing with plasticity of polycrystalline metals, Taylor (1938) proposed that the stressstrain relationship be specified independently on various planes in the material. The
assumption was that the stresses on such a plane are the resolved components of the
macroscopic stress tensor, called the static constraint, or that the strain components on such a
plane are the resolved components of the macroscopic strain tensor, known as the kinematic
constraint. This idea developed for plasticity (Batdorf and Budianski, 1949), is known as slip
theory. Later, the model was extended by Baant and his co workers to concrete and
geomaterials and is referred to as the microplane model. This model is for analysis of quasibrittle materials that exhibit softening behaviour (Baant and Gambarova, 1984, Baant and
Prat, 1988a, 1988b, Baant and Obolt, 1990, Carol et al. 1992 and Baant et al., 1996a,
1996b). Baant and Prat (1988a) developed an advanced kinematic constraint version of
microplane model which was extended by Obolt and Baant (1992) to a more general cyclic
form with rate sensitivity.
The 3D version of the microplane model is constructed on the basis of uniaxial relationships
between stress and strain components on planes of various orientations within a material
(Figure 5.19). These planes may be imagined to represent the weak planes in the microstructure such as the inter particle contact planes, interfaces, planes of micro-cracks, etc.
!
!
!
!
!
!
!
!

(b)

(a)

!
!
!

(c)

Figure 5.19: Microplane model: a) integration sphere; b) microplane stress-strain components; and c) stress
transfer through a number of contact planes (microplanes).

The strain components on the microplane are taken to be projections of the macroscopic strain
tensor (kinematic constraint approach) from which the normal and shear strain component are
computed. The normal microplane strain component is then decomposed into volumetric and
deviatoric part to realistically model concrete dominated by compressive load and to control
the initial elastic value of the Poissons ratio (Baant and Prat 1988a). The simplicity of the
model is due to the fact that constitutive properties are characterized entirely by only
relationships between the stress and strain components of microplanes. Knowing the stressstrain law for each microplane component, the macroscopic stiffness and stress tensors are

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

181

calculated by the stress integration over all microplanes. Baant and Prat produced the strain
stress microplane laws for the first loading and unloading part, while Obolt and Baant
(1992) introduced more general rules at the microplane level, including rate sensitivity
(Figure 5.20).

(a)

(b)

Figure 5.20: Cyclic stress-strain relationship; a) volumetric microplane component; b) "Deviatoric microplane
component (Obolt et al. 2001).

(b)

LB15

Displacement (inch)

LB13

LB16

In 3D analysis of reinforced concrete structures, the bond between reinforcement and concrete
can be modelled by a layer of 3D elements around the reinforcing bars. Hoehler and Obolt
(2001) adopted a microplane model for these bond elements such that the shear stress-strain
relationship provides a realistic shear stress-slip response. Figure 5.21 shows the application
of the microplane based FEM program in simulation of an experiment on a beam- column
connection that was tested by Ma et al. (1976). The analysis was performed by a displacement
controlled method as set out in Hoehler and Obolt (2001). The applied displacement history
is shown in Figure 5.21b (note that in initial stage, the experiment was load-controlled). The
test and numerically simulated results are compared in Figure 5.22 and a reasonable
correlation is observed.

(a)

Figure 5.21: Details of by Ma et al. (1976) test: a) layout of experiment; b) time history of "applied displacement
in the Hoehler and Obolt (2001) analysis.

182

5 Analysis and desing! of surface and solid structures using non-linear models

Applied load (kips)

Experiment
MASA

(c)
Displacement
(inches)

(a)

(b)
(d)

Figure 5.22: 3D microplane FEM simulation and comparison with experimental data: a) section of FEM mesh;
b) comparison of load-displacements; c) distribution of principal strains from analysis; and (d)
axial strain in reinforcing bars from the analysis (Hoehler and Obolt, 2001).

A second example of the application of microplane models is that of Liu and Foster (2000)
where 3D modelling was used to examine the effects of cover spalling in the design of high
strength concrete columns under high axial loading. Figures 5.23a and 5.23b show the mesh
used for the modelling of column CS-19 tested by Razvi and Saatcioglu (1996). One eighth of
the column modelled due to symmetries about XY, YZ and XZ planes. The concrete was
modelled using 20-node isoparametric solid elements with numerical integration on a
2 o 2 o 2 Gaussian quadrature. The in-situ mean concrete strength was taken as fc = 83 MPa,
being 90% of the mean cylinder strength tested under standard conditions. The longitudinal
and transverse reinforcement were modelled using 3-node truss elements with numerical
integration on a 2 point Gauss quadrature. A perfect elasto-plastic stress-strain relationship
was used for both the longitudinal and transverse reinforcement with a modulus of elasticity
of 200 GPa. The yield strengths of the longitudinal and tie reinforcement was 450 MPa and
400 MPa, respectively.
To include the effect of cover spalling in the analysis, a phenomenon commonly observed in
laboratory tests of axially loaded columns, the cover was taken to be spalled from the crosssection when a threshold transverse tensile strain was reached at the interface between the
cover concrete and the core. Such tensile strains occur due to the effects of the restraint of the
Poissons effect provided to the core by the tie reinforcement (see Foster et al., 1998). In this
analysis, the cover was taken as spalled if this transverse tensile strain between the cover and
the core elements exceeded 750 B). The spalling was included in the analysis by setting the
cover elements to a low stiffness once the threshold strain had been reached and, thus, the
cover elements do not contribute to the axial capacity of the section once spalling of the
element was deemed to have occurred. The failure of the column was by an axial localization
of the concrete in compression around the XZ axis of symmetry, as shown in Figure 5.23c.
In the Razvi and Saatcioglu experiment, the axial strain was measured over a gauge length of
300 mm (or 150 mm from the centreline with symmetry). The load versus axial strain results
from the FE model is compared to the laboratory measurements in Figure 5.24, with and
without cover spalling included in the model. A good correlation is observed between the FE
results and the experimental data only when cover spalling is included. If the cover spalling
phenomena, as observed in the laboratory tests, was not accounted for in the numerical
analysis, the strength would be over predicted by some 20%. This example is used here to
demonstrate the care needed when undertaking non-linear FE modelling that all influences
significant to the failure condition can be captured in the analysis.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

183

P = 3840 kN
(descending)

Note:
Cover elements
not shown for
clarity

(a)

(b)

(c)

Figure 5.23: Finite element mesh for Razvi and Saatcioglu (1996) column CS-19: (a) details and dimensions;
(b) 3D view; (c) xz displacements at P = 3840 kN on the descending curve (dimensions in mm).

8000
7000

Load (kN)

6000
5000
4000
3000
Experimental
FEM - without cover spalling
FEM - cover spalling at 750 microstrain

2000
1000
0
0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

Axial strain
Figure 5.24: Finite element analysis of the Razvi and Saatcioglu (1996) column CS-19 with and without cover
spalling.

184

5 Analysis and desing! of surface and solid structures using non-linear models

5.5.5

Examples of the application of 3D FE modelling

Example 1 Optimisation of anchor system for prestressed concrete

As an example of the use of 3D non-linear modelling, the design of a pre-stressing anchoring


device (shown in Figure 5.25a) is examined. Pre-stressing anchorage systems are subjected to
three-dimensional stress fields of high intensity. The main issue is the increased compressive
strength of concrete due to confinement and appropriate reinforcement detailing which
facilitates anchoring. This example is taken from the development of Dywidag pre-stressing
system where tendons are anchored in reinforced concrete wall element. The goal was to
optimize the reinforcement of the anchoring region under a constraint of limited crack width.

(b)
(a)

Figure 5.25: a) Wall with tendon duct; and b) details of anchorage.

The model of the wall represents a typical anchoring detail in bridge structures, where the prestressing tendon is anchored in a relatively thin wall. The reinforcement arrangement,
schematically shown in Figure 5.25b, must provide sufficient resistance to a splitting action of
the anchor as well as confinement effect to concrete and ensure safe tendon anchoring. The
goal was to determine the horizontal force to be resisted by the reinforcement. The study
included variations of the amount of tie and spiral reinforcement and was used as a basis to
improve design of anchoring regions with optimal arrangements of reinforcement.
To meet the design goals, a 3D solid analysis was employed. The model was reduced to using
advantage of two symmetry axes as shown in Figure 5.26. It includes concrete, steel anchors
and all reinforcing bars that are embedded in the concrete solid elements. A fracture-plastic
constitutive model was used for the concrete and a Von-Mises plasticity model for steel of the
anchoring body. The reinforcing bars were modelled using a bi-linear elastic-plastic law with
hardening.
The loading, representing the anchoring force of the prestressing tendon, was applied by
prescribed forces at the top surface of the anchor. The force was applied in load steps up to
the value corresponding to full prestressing and then progressively increased to failure.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

185

Figure 5.26: FE mesh for prestressing anchor.

Examples of results of the analyses for the anchoring system are shown in Figure 5.27. The
stress state of the anchoring region was evaluated in detail and the concrete, compressive
stresses and plastic deformations were examined and crack widths were evaluated. The
stresses in the reinforcement were also examined.
The response of the anchor with spiral reinforcement (300mm in diameter) can be seen from
the load-displacement diagram in Figure 5.28a and the stresses in the spiral reinforcement,
Figure 5.28b, for different concrete quality (0.6 and 0.8 fc) and for different amounts of tie
reinforcement. The response can be compared with the level of service load that is applied at
the pre-stressing stage.

(a)

(b)

Figure 5.27: Results of FE analysis: a) cracks and compressive stress contours; b) stress in reinforcing bars.

186

5 Analysis and desing! of surface and solid structures using non-linear models

3.5

Anchoring force [MN]

3.0
Service load

2.5
2.0

SP300 C0.6

1.5

SP300 C0.8

1.0

SP300 C0.8 ties 1x

0.5

SP300 C0.8 ties 2x

0.0
0.0

0.5

1.0

1.5

2.0

2.5

3.0

Anchor displacement [mm]

(a)

Anchoring force [MN]

3.5
3
Service load

2.5
2
1.5

SP300 C0.8

SP300 C0.8 ties 1x

0.5

SP300 C0.8 ties 2x

0
0

100

200

300

400

Stress in spiral reinforcement [MPa]

(b)

Figure 5.28: Results from FE analyses: a) Anchoring force verses displacement diagram; b) stress in spiral
reinforcement due to anchoring force.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

187

Example 2 Axisymmetric 3D solid analysis from design to practice

In this example, the performance of a hydro-electric powerhouse to stably support a generator


is evaluated where the powerhouse is located inside the rock of a mountain (Figure 5.29). The
hydraulic pressure applied to the inside casing pipes, the thermal gradient caused by the
electricity generator and the deformation of the powerhouse restrained by the surrounding
rocks, form the design loads and the boundary conditions. The limit states corresponding to
the design requirements are specified in terms of cracking damage around the casing runner
and transmission of higher frequency vibrations caused by the production of the hydroelectric power.

vibration
protector
generator
machine
base

23.0m

casing
pile

plain concrete zone


for piping devices
hydro generator and
arrangement of
RC power house
reinforcement

GHIJ

18.0m

ax symmetric finite element mesh

Figure 5.29: 3D design of reinforcement for RC powerhouse and casing runner

For design, the high pressure of running water inside of the casing is sustained by both the
casing pipe, made of steel, and the surrounding reinforced concrete structure. If the
hydrodynamic pressure was to be carried by the steel casing alone, its thickness becomes so
large that welding quality could not be well assured. Thus, the reinforced concrete is expected
to partially bear the pressure. At the same time, concrete cracking caused by the high pressure
must be controlled so that the performance of powerhouse is maintained and able to support
the weight of the plant and to act as a foil against underground water entering through the
188

5 Analysis and desing! of surface and solid structures using non-linear models

rock joints. A conventional design, based on 3D linear analysis, shows that the quantities of
hoop reinforcement needed is 3.3% around the casing pipes, as shown in Figure 5.29. The
arrangement of the reinforcing bars in 3D is laborious and the high quantity makes high
quality concreting works problematic as compaction is difficult.
In reality, however, the thermal stress component declines significantly due to stiffness
reduction by cracking and a linear analysis based design (strength based design) tends to bring
about more reinforcement than is needed to meet serviceability and safety requirements. On
the other hand, however, the steel stresses in the composite structures tend to be
underestimated. Thus, a non-linear analysis for the performance assessment is required in
view of both safety and cost benefits.
Using a 3D non-linear FE analysis, structural engineers of an electric power firm investigated
the behaviour of the reinforced concrete powerhouse coupled with the steel casing attached to
the main body of the generator. The results of the analysis are shown in Figure 5.30 for the
propagation of cracks and for the principal strain trajectories in the radial direction. The
reinforcement ratio was 3.3%, corresponding to the strength design based on the linear FE
analysis. The reinforcement ratio was then gradually reduced using a parametric study and
finally a steel reinforcement ratio of 0.8% was adopted. As shown in Figure 5.30, the damage
situation in the reinforced concrete powerhouse of 0.8% reinforcement ratio is almost the
same as that with 3.3%. The stress level of steel casing was verified within the allowable
stress level below the yield strength. However, if the casing reinforcement is reduced further,
sharp crack localization is possible (as shown for P = 0.0) and increases the risk of yielding of

P= 3.3%

P= 0.8%

P =0.0%

Crack propagation analysis under hydro-pressure and thermal action


(micro)
900
800
700
600
500
400
300
200
100
0

Profile of mean principal tensile strain (under operational load)


Figure 5.30: Crack propagation and damage map

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

189

the steel casing. Localized damage may cause unexpected vibrations that would make the
power generator unstable. An approximately 36% reduction of costs related to the casing
runner was achieved by using non-linear 3D FE modelling.
While this example shows some of the advantages of moving to higher methods of modelling,
before adopting any non-linear analysis results great care is needed to ensure that the
modelling accurately simulates the 3D crack propagation and size effects. Thus, tension
softening of concrete around the plain concrete zone and tension stiffening of concrete close
to the casing runner were taken into account and the accuracy of the 3D nonlinear analysis
was verified by experiments (see Kato, 2000, and Kato, 2001).

5.6

References

An, X., Maekawa, K., Okamura, H. (1997), Numerical Simulation of Size Effect in Shear
Strength of RC Beams, Journal of Material, Concrete structures and Pavements, 35(564), pp.
297-316.
Batdorf, S.B., Budianski, B. (1949), A Mathematical Theory of Plasticity Based on the
Concept of Slip, Technical Note. 1871, National Advisory Committee for Aeronautics,
Washington DC.
Barzegar, F. (1989), Analysis of RC Membrane Element with Anistropic Reinforcement,
Journal of Structural Engineering, 115, pp/ 647-665.
Baant, Z.P., and Gambarova, P. (1984), Crack Shear in Concrete: Crack Band Microplane
Model, Journal of Engineering Mechanics, ASCE, 110, pp. 2015-2035.
Baant, Z.P., and Prat, P.C. (1988a), Microplane Model for Brittle-Plastic Material: I
Theory, Journal of Engineering Mechanics, ASCE, 114(10), pp. 1672-1688.
Baant, Z.P., and Prat, P.C. (1988b), Microplane Model for Brittle-Plastic Material: II
Verification, Journal of Engineering Mechanics, ASCE, 114(10), pp. 1689-1702.
Baant, Z.P., and Oh, B.H. (1983), Crack Band Theory for fracture of concrete, Material and
Structures, RILEM, Paris, 16, pp. 155-177.
Baant, Z.P., and Obolt, J. (1990), Nonlocal Microplane Model for Fracture, Damage and
Size effect in Structures, Journal of Engineering Mechanics, ASCE, 116(11), pp. 2485-2504.
Baant, Z.P., Xiang Y., and Prat, P.C. (1996a), Microplane Model for Concrete- I. StressStrain Boundaries and Finite Strain, Journal of Engineering Mechanics, ASCE, 122(3), pp.
245-262.
Baant, Z.P., Xiang, Y., Adley, M., Prat P.C., and Akers, S. (1996b), Microplane Model for
Concrete- II. Data Delocalization and Verification, Journal of Engineering Mechanics,
ASCE, 122(3), pp. 263-268.
Carol, I., Prat, P., and Baant Z.P. (1992), New Explicit Microplane Model for Concrete:
Theoretical Aspects and Numerical Implementation, International Journal of Solids and
Structures, 29(9), pp. 1173-1191.

190

5 Analysis and desing! of surface and solid structures using non-linear models

CEB-FIP Model Code 1990 (1993), Comit Euro-International du Bton, Thomas Telford,
London.
Cervenka, V. (1998), Simulation of shear failure modes of R/C structures, In:
Computational Modelling of Concrete Structures (Euro-C 98), eds. R. de Borst, N. Bicanic,
H. Mang, G. Meschke, A.A.Balkema, Rotterdam, The Netherlands, 1998, 833-838.
Cervenka, V. (2002), Computer simulation of failure of concrete structures for practice, 1st
fib Congress, Concrete Structures in 21 Century, Osaka, Japan, Keynote lecture in Session 13,
289-304
Cervenka, J., and Cervenka, V. (1999), Three Dimensional Combined Fracture-Plastic
Material Model for Concrete, Proc. 5th US National Congress on Computational Mechanics,
Boulder, Colorado.
Cervenka, J., and Cervenka, V. (2007), Personal communications.
Cervenka, J., Cervenka, V., and Eligehausen, R. (1998), Fracture-Plastic Material Model for
Concrete, Application to Analysis of Power Actuated Anchors, Proc. FRAMCOS, 3, pp.
1107-1116.
Cervenka, V., and Gerstle, K. (1971), Inelastic Analysis of Reinforced Concrete Panels: (1)
Theory, (2) Experimental Verification and application, Publications IABSE, Zrich, V.3100, 1971, pp.32-45, and V.32-II,1972, pp.26-39.
De Borst, R. (1986), Non-linear Analysis of Frictional Materials, PhD Thesis, Delft
University of Technology.
De Borst, R., and Nauta, P. (1985), Non-Orthogonal Cracks in a Smeared Finite Element
Model, Engineering Computations, 2, pp. 34-46.
Drucker D.C., and Prager W. (1952), Soil Mechanics and Plastic Analysis or Limit Design
Quarterly Journal of Applied Mathematics, 10(2), pp/ 157-165.
Feenstra, P.H. (1993), Computational Aspects of Biaxial Stress in Plain and Reinforced
Concrete, PhD Thesis, Delft. University of Technology, The Netherlands.
Foster, S.J., Liu, J. and Sheikh, S.A. (1998), Cover spalling in HSC columns loaded in
concentric compression, Journal of Struct. Engrg., ASCE, 124(12): 1431-1437.
Fukuura, N., and Maekawa, K. (1998), Multi-Directional Crack Model for In-Plane
Reinforced Concrete Under Reversed Cyclic Actions-4 Way Fixed Crack Formulation and
Verification, in de Borst et al (eds.), Computational Modeling of Concrete Structures,
Balkema, Rotterdam and Brookfield, pp. 143-152.
Hauke B., and Maekawa, K. (1998), Three-Dimensional Reinforced Concrete Model with
Multi-Directional Cracking in de Borst et al (eds.), Computational Modeling of Concrete
Structures, Balkema, Rotterdam and Brookfield, pp. 93-102.
Hoehler, M., and Obolt, J. (2001), Three-dimensional reversed-cyclic analysis of reinforced
concrete members using the microplane model, Otto Graf Journal, 12, pp. 93-113.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

191

Hofstetter, G., and Mang, H.A. (1995), Computational Mechanics of Reinforced Concrete
Structures, Friedr. Vieweg & Sohn Verlagsesellschaft mbHg, Braunschweig/Wiesbaden.
Ichikawa, H., Ogasawara, M., and Maekawa K. (1998), Experimental Study of Seismic
Performance of RC Piers Subjected to Eccentric Axial Force and Torsional Moment, Proc. of
JCI, 20(1), pp. 149-154.
Kato, S., Iizuka, K., and Nanbu, S. (2000), Nonlinear Behavior of Reinforced Concrete
Structures Subjected to High Water Pressure, Proc. of JCI, Vol.22, No.3, pp.67-72.
Kato, S., Yamaya, A., Zhao, Y., and Ozawa, H. (2001), Study on Structural Performance of
Inner concrete using Ring-gage, Proc. of JCI, Vol.23, No.1, pp.631-636.
Liu, J., and Foster, S.J. (2000), A Three Dimensional Finite Element Model for Confined
Concrete Structures, Computers and Structures, Vol. 77, No. 5, October, pp 441-451.
Ma, S.Y.M., Bertero, V.V., and Popov E.P. (1976), Experimental and Analytical Studies on
the Hysteretic Behavior of Reinforced Concrete Rectangular and T-Beams, Earthquake
Engineering Research Center (EERC), Report No. UBC/EERC 76-2, University of California,
Berkeley.
Maekawa, K., Irawan, P., and Okamura, H. (1997), Path dependent three dimensional
constitutive laws of reinforced concrete-formulation and experimental verification, Struct.
Eng. Mech., 5(6), pp. 743-754.
Maekawa, K., Pimanmas, A., and Okamura, H. (2003), Nonlinear Mechanics of Reinforced
Concrete, Spon Press, London.
Marti, P., Leesti, P., and Khalifa, W.U. (1987), Torsion Tests on Reinforced Slab Elements,
Journal of the Structural Division, ASCE.
Menetrey, P., and Willam K.J., (1995), Triaxial Failure Criterion for Concrete and its
Generalization, ACI Structural Journal, 92(3), pp. 311-318.
Mitchell, D., and Collins, M.P. (1974), Diagonal Compression Field Theory-A Rational
Model for Structural Concrete, ACI Journal Proceeding, 71(8), pp.396-408.
Okamura, H., and Maekawa, K. (1991), Nonlinear Analysis and Constitutive Models of
Reinforced Concrete, Gihodo-Shuppan Co., Tokyo.
Ottosen, N.S. (1977), A failure criterion for concrete., Journal of Engrg. Mech. Div., ASCE,
103(EM4): 527-535.
Onsongo, W. (1978), The Diagonal Compression Field Theory for Reinforced Concrete
Beams Subjected to Combined Torsion, Flexural and Axial Load, Thesis presented to the
University of Toronto, Toronto, Canada.
Obolt, J., and Baant, Z.P. (1992), Microplane Model for Cyclic Triaxial Behavior of
Concrete, Journal of Engineering Mechanics, ASCE, 118(7), pp. 1365-1386.
Obolt, J., Li, Y., and Kozar I. (2001), Microplane Model for Concrete with Relaxed
Kinematic Constraint, International Journal of Solids and Structures, 38, pp. 2683-2711.

192

5 Analysis and desing! of surface and solid structures using non-linear models

Pramnono, E., and Willam, K.J. (1989), Fracture Energy-Based Plasticity Formulation of
Plain concrete, Journal of the Engineering Mechanics, ASCE, 115, pp. 1183-1204.
Razvi, S.R, and Saatcioglu, M. (1996), Tests of high strength concrete columns under
concentric loading. Report OCEERC 96-03, Department of Civil Engineering, University of
Ottawa, Canada.
Stevens, N.J., Uzumeri, S.M., and Collins, M.P. (1987), Analytical Modeling of Reinforced
Concrete Subjected to Monotonic and Reversed Loadings, Publication No. 87-01,
Department of Civil Engineering, University of Toronto, Toronto.
Taylor, G. I. (1938), Plastic Strain in Metals, J. Inst. of Metals, London, 62, pp. 307-324.
Tsuchia, S., Tsuno, K., and Maekawa, K. (2001), Nonlinear Three-Dimensional FE Solid
Response Analysis of RC Columns Subjected to Combined Permanent Eccentric Axial Force
and Reversed Cyclic Torsion and Bending/Shear, Journal of Material, Concrete structures and
Pavements, 683(52), pp. 131-143.
Vecchio, F.J., and Collins, M.P. (1982), The Response of Reinforced Concrete to In-Plane
Shear and Normal Stresses, Publication No. 82-03, Department of Civil Engineering,
University of Toronto, Toronto.
Vecchio, F.J., and Collins, M.P. (1986), The Modified Compression-Field Theory for
Reinforced Concrete Elements Subjected to Shear, ACI Journal, V. 83, No. 2, March-April,
pp. 219-231.
Vecchio, F.J., and Selby, R.G. (1991), Toward Compression-Field Analysis of Reinforced
Concrete Solids, Journal of Structural Engineering, ASCE, 117(6), pp. 1740-1758.
Willam, K.J., and Warnke, E.P. (1975), Constitutive Model for the Triaxial Behavior of
Concrete, Proceeding of the International Association for Bridge and Structural Engineering,
19.
Wilkins, M.L. (1964), Calculation of Elastic-Plastic Flow, Methods of Computational
Physics, 3, Academic Press, New York.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

193

[Blank Page]

Advanced modelling and analysis concepts

6.1

Introduction

This chapter presents advanced details of material modelling in view of engineering


application to practical design, construction and maintenance by means of computer software.
First, the constitutive frameworks available for continuum solids will be reviewed as a basis
for dealing with detailed materials modelling, followed by a discussion of specific issues
relating to material models used for structural behavioural simulation. This chapter is also
addressed to developers of design or analysis programs for RC structures, and to engineers
who develop user options for large-scale nonlinear analysis codes. Chapter 3 served as an
introduction to this chapter for practicing engineers and as a guide to use and/or select
structural analysis software.

6.2

Constitutive frameworks

6.2.1

Non-linear elasticity

Arguably, non-linear elasticity is the most general and most popular of the approaches used to
model reinforced or prestressed concrete structures, and most commercial FE packages have
the ability for the user to use non-linear elastic models. Methods for analysis of reinforced
concrete elements, based on concepts of linear or nonlinear elasticity, encompass a wide range
of approaches, many of which have met with good success.
In uncracked or confined states, concrete is generally modelled as an isotropic or orthotropic
elastic material. In the cracked condition, concrete is considered to be an orthotropic material,
and generally represented in a smeared crack context although a discrete crack representation
is still sometimes used. Crack conditions can be represented in the manner of either fixed
cracks or rotating cracks, with each approach having the potential to provide accurate
modelling. Within the fixed crack concept, the crack direction and hence orientation of
material orthogonality are defined by the formation of the first crack. In rotating crack
models, the crack direction and material orthogonality are free to gradually realign depending
on the prevailing stress and strain conditions. The computational solution algorithm required
with both fixed and rotating crack models can be based on either an incremental-load tangent
stiffness approach, or one based on a total-load secant stiffness formulation, often iterative in
nature. Thus, the formulation of an elasticity-based constitutive model for concrete will
depend on the combination of three main modelling decisions: smeared crack or discrete
crack; fixed crack or rotating crack; tangent-stiffness or secant-stiffness.
Presented below is the basic formulation for a smeared rotating crack model using a secant
stiffness approach. An example of the formulation of a fixed crack model based on a tangent
stiffness approach is provided by Maekawa et al. (1999). A hybrid formulation, combining
elements of both fixed and rotating cracks, is described by Vecchio (1990, 2000, 2001). To
simplify the discussion, the formulation presented will be limited to the case of twodimensional plane stress; all formulations are easily expandable to the general threedimensional case.
Consider a reinforced concrete element, of unit dimensions, subjected to the plane stress
condition +. The element contains n components of reinforcement, where the i-th
reinforcement component is oriented at angle :i to the reference x-axis (0 ? i ? n). The

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

195

amount of reinforcement in the i-th direction is denoted by the reinforcement ratio *i. The
constitutive response of the element is described as
+=D)

(6.1)

where + and ) are the total stress and total strain vectors for the element, and D is the
composite secant stiffness of the material.

" ' +x

+ y , xy

qT ;

! ' )x

) y , xy

qT

(6.2)

At this point, it is being assumed that the strains are common for all components, meaning:
perfect bond between concrete and reinforcement; no initial strains in the materials (e.g., no
concrete shrinkage); and no subsequent volume changes due to thermal expansion, creep, or
other types of elastic or plastic mechanisms (these are treated later). The composite material
stiffness matrix D is comprised of contributions from the concrete and each reinforcement
component; thus:
n

D ' Dc 8 P D s i

(6.3)

i '1

For each reinforcement component, the material stiffness must first be formulated in the
direction of the reinforcement, then transformed back to the reference axes (x, y). Thus, for
reinforcement component i,
Z * i E si
X
Ds i ' X 0
X 0
Y

0 0W
U
0 0U
0 0UV

(6.4)

where Esi is the effective secant modulus of the reinforcement, evaluated as

Esi ' f si ) si .

(6.5)

The reinforcement stress f si is found from a suitable stress-strain model for reinforcing steel

(e.g. elastic-plastic with strain hardening), evaluated for the steel strain ) si . To transform
back to the reference axes, the following transformation is used:
Dsi ' TsT Dsi Tsi
i

(6.6)

where

Z
cos2 r
X
T 'X
sin 2 r
X7 2 cosr " sinr
XY

sin 2 r
cos2 r
2 cosr " sinr

cosr " sinr W


U
7 cosr " sinr U
(cos2 r 7 sin 2 r )UU
V

(6.7)

and r = :i.

196

6 Advanced modelling and analysis concepts

Prior to cracking, in low confinement states, concrete can be sufficiently well represented as
an isotropic material; thus
Z1 n
Ec X
Dc '
n 1
1 7 n 2 XX
Y0 0

W
U
0
U
(1 7 n ) / 2 UV
0

(6.8)

where Ec is the initial elastic modulus and n is the Poissons ratio.


After cracking, concrete behaves as an orthotropic material, with the directions of orthotropy
aligned with the crack direction. Thus, the concrete material stiffness matrix must be
evaluated accordingly; that is, first defined with respect to the principal stress directions
(defined by angle -, normal to the crack direction) and then transformed back to the reference
axes. Thus,

Dc ' T T Dc T

(6.9)

where T is the transformation rotation matrix as before except that r = -. The concrete
stiffness D is evaluated as
Z Ec1
X
D 'X 0
X 0
Y

0
Ec 2
0

0W
U
0U
Gc UV

(6.10)

where Ec1 and E c 2 are the effective secant moduli for the concrete in the principal tensile
stress and principal compressive stress directions, respectively, and Gc is the effective secant
shear modulus. Note that the absence of off-diagonal terms in D implies that post-cracking
Poissons ratios are ignored; these can be included in a slightly more rigorous formulation
described later. The effective secant moduli are evaluated as:

f
Ecl ' c1 ;

)1

f
Ec 2 ' c 2 ;

)2

Gc '

Ec1 T Ec2
Ec1 8 Ec2

(6.11)

where fc1 and fc2 are the post-cracking principal tensile stress and principal compressive stress
in the concrete, respectively, evaluated based on the concrete principal tensile and
compressive strains, )1 and ) 2, and using appropriate stress-strain relationships. Models for
calculating fc1 can include the effects of tension softening, tension stiffening, and other
influencing mechanisms (e.g. local stress conditions at crack locations). Models for
calculating fc2 will typically consider compression softening and/or confinement effects.
Note: The nature of the constitutive models for calculating fc1 and fc2 will depend on whether
a fixed or rotating crack approach is being used, and some caution is needed in selecting
these.
The accuracy and range of application of this basic formulation can be considerably enhanced
by introducing provisions for modelling elastic and plastic offset strains. With respect to the
concrete, this will allow the consideration of such effects as dilation (post-cracking Poissons
effects), thermal strains, shrinkage, and plastic residual strains (due to yielding and

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

197

unloading). For the reinforcement, it will permit the inclusion of prestressing strains, thermal
strains, and plastic offsets (yielding).
The general approach used is to first decompose, for each material, the total strains into elastic
or plastic offset strains and net elastic strains. Thus, for concrete,

! ' !c 8 !co 8 !cp

(6.12)

where ) are the total element strains (as before), )c are the concrete net elastic strains, ) co are
the concrete elastic offset strains (e.g. due to shrinkage, thermal, or Poissons effects), and )cp
are the concrete plastic offset strains (due to unloading, damage). Similarly, for the
reinforcement

! ' !s 8 !so 8 !sp

(6.13)

where ) si are the net elastic strains for the i-th component reinforcement, ) soi are the
reinforcement elastic offset (due to prestressing or thermal effects), and )spi are the plastic
offsets (due to yielding, unloading). It is essential to note that it is the net elastic strains, )c
and )si , that must be used within constitutive relations when determining concrete and
reinforcement stresses, and when defining the effective secant moduli.
The general constitutive relation for the element then becomes

" ' D T ! 7 "o

(6.14)

where the prestrain vector + o is defined as


" o ' Dc ( !co 8 !cp ) 8

P Ds i (! soi 8 ! spi )

(6.15)

i '1

As an example, consider the formulation required to represent dilation effects in the concrete.
Let n 21 represent the strain in the 2-direction due to a stress in the 1-direction; similarly, n 12
represents a strain in the 1-direction due to a stress in the 2-direction, where 1, 2 are the
principal stress directions. Thus, define

o
o
o
!co ' ) cx
) cy
& cxy

(6.16)

where
1
2

1
2

(6.17a)

o
) cy
' ) co1 (1 7 cos 2- ) 8 ) co2 (1 8 cos 2- )

1
2

1
2

(6.17b)

o
& cxy
' ) co1 T sin 2- 7 ) co2 T sin 2-

(6.17c)

o
) cx
' ) co1 (1 8 cos 2- ) 8 ) co2 (1 7 cos 2- )

and where, in this case, ) co1 ' n12 T ) c 2 , ) co2 ' n 21 T ) c1 and - is the inclination of the principal

198

6 Advanced modelling and analysis concepts

stress direction.
Stresses in the component materials are then evaluated as:

" c ' Dc (! 7 !co 7 !cp )


" si ' Dsi (! 7 ! so 7 ! sp )

(6.18)

Note that + si represents the contribution of the i-th reinforcement component to the total
element stresses. The actual stress in the reinforcement must be evaluated as
f si ' + si * i .

(6.19)

In this basic formulation, an iterative solution procedure is required. Some estimate of the
material secant moduli is first made, and the element composite stiffness matrix is assembled.
For a given element stress condition, the element total and net strains are then calculated.
From the net elastic strains thus determined, the material secant moduli can be re-calculated.
The process is repeated until the secant moduli have converged to stable values. This
approach is applicable to post-peak strain/strain conditions with equal numerical stability as
for pre-peak conditions.
Additional details pertaining to this basic formulation can be found in Vecchio (1990).
Equally viable formulations for a fixed crack approach, or for tangent-stiffness based models
(either fixed or rotating crack based), can be found in the literature. More esoteric
formulations based on multiple non-orthogonal cracks, microplane models, or hybrid crack
models can also be found, and these generally provide equal capacity for accurate modelling
of reinforced concrete based on elasticity principles.
6.2.2

Plasticity

The flow theory of plasticity is a natural generalization of the classical work of Tresca (1868),
Saint-Venant (1870), Lvy (1870) and von Mises (1913). Its principal ingredients are the
yield condition, the flow rule and the hardening law. The total strain is additively decomposed
into the elastic strain and the plastic strain. It is assumed that the current state of an
elementary material volume is fully described by the total strain, plastic strain, and by some
additional hardening (or softening) variables that characterize the changes of microstructure
and are collected in a vector k. The basic equations include the elastic-plastic split,

) ' )e 8 ) p

(6.20)

+ ' De) !

(6.21)

f ( + ,q ) ? 0

(6.22)

)! p ' ^! g( + , q )

(6.23)

the stress-strain law for the elastic part,

yield condition,
flow rule,

and hardening law, which consists of two parts - the definition of the hardening variables, in
general in the form of rate equations

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

199

f! ' ^!k( + ,f )

(6.24)

and the dependence of the parameters q appearing in Eqs. 6.22 and 6.23 on the hardening
variables,

q ' h( f )

(6.25)

In the above, )={)11, )22, )33, &23, &13, &12}T is the column matrix of engineering strain
components; )e and ) p represent elastic and plastic strain, respectively; +={+11, +22, +33, +23,
+13, +12}T is the column matrix of stress components; De is a square matrix of elastic moduli
(the elastic material stiffness matrix); f is the yield function; g is a function specifying the
direction of plastic flow; and ^ is the rate of the plastic multiplier. A dot over a symbol
denotes differentiation with respect to time. However, recall that the theory is rateindependent. Time plays here only the role of a formal parameter that controls the loading
process, and it does not need to have the meaning of real physical time. The rate equations,
Eqs. 6.23 and 6.24 could also be presented in an incremental form, with rates replaced by
infinitesimal increments.
The yield function defines the elastic domain in the stress space, which is bounded by the
yield surface. States for which f<0 are elastic, states corresponding to f=0 are plastic, and f>0
is not possible. Of course, as the variables q grow, the yield surface evolves - it can expand,
shrink, translate and even change its shape. Plastic flow can take place only if the current state
is plastic; this is expressed by the condition.

^! f $", q% ' 0

(6.26)

Indeed, if the material is in an elastic state (f<0), Eq. 6.26 implies ^! ' 0 and according to
Eqs. 6.23 and 6.24, the plastic strain and the hardening variables remain constant. On the
other hand, in a plastic state we have f=0 and Eq. 6.26 does not restrict the rate of the plastic
multiplier. This rate should never be negative, because g in Eq. 6.23 specifies the oriented
direction of evolution (e.g., the plastic strain rate in a uniaxial tensile test can be zero or
positive). This is described by an additional restriction, ^! < 0 . Combining it with Eqs. 6.22
and 6.23 we obtain the loading/unloading conditions in the so-called Kuhn-Tucker form as,
f ? 0, ^! < 0, ^! f ' 0

(6.27)

During plastic flow, the yield function must remain equal to zero, and so the rate of its change
is also zero. This consideration leads to the consistency condition,

^! f! ' 0

(6.28)

Suppose that the current values of all variables as well as the rate of the total strain are given.
The basic equations make it possible to compute the rates of all variables. If the current state
is elastic (f < 0), the plastic multiplier, plastic strain and hardening variables remain constant,
and the stress evolution is governed by the elastic law. If the current state is plastic (f = 0), the
plastic flow can continue or the material can unload elastically. The former case is
characterized by f! ' 0 and ^! @ 0 . In the latter case, we have ^! ' 0 and, as the subsequent
stress state must be elastic, the yield function must decrease, which corresponds to f! 9 0 . The
special case when f! ' 0 and ^! ' 0 is called neutral loading.

200

6 Advanced modelling and analysis concepts

If plastic loading takes place, the rate of plastic multiplier can be calculated from the
condition f! ' 0 . Applying the chain rule, we can write this condition as,
T

e sf b
e sf b
f! ' c ` "! 8 cc `` q!
d s" a
d sq a

(6.29)

Combining the rate form of the elastic stress-strain law with the additive split and the flow
rule we get

+! ' De ()! 7 ^! g )

(6.30)

The rate form of Eq. 6.25 combined with Eq. 6.24 yields
q! '

sh
sh
f! ' ^!
k
sf
sf

(6.31)

By substituting Eqs. 6.30 and 6.31 into Eq. 29 we have a linear equation for ^! , from which

^! '

f T+ De )!

(6.32)

f T" De g 7 f Tq Hk

where, we have denoted f+ ' sf / s+ , f q ' sf / sq , and H ' sh / sf . Substitution of this


result back into Eq. 6.30 leads to the rate form of the elastoplastic stress-strain law as
e

+! ' c De 7
c
d

b
` )!
f + De g 7 f Hk `a
De gf T+ De

(6.33)

T
q

The matrix in parentheses is the elasto-plastic stiffness matrix. It is in general non-symmetric,


except for the particular case when g ' f + . In this case, the flow rule is said to be associated,
and it can be rewritten as

)! p ' ^!

sf (+ ,q )
s+

(6.34)

As the gradient sf ls+ is normal to the yield surface, Eq. 6.34 is sometimes called the
normality rule. Yielding of metals is usually insensitive to the volumetric part of the stress
tensor, and it can be conveniently described by the von Mises yield condition

f (+ ) t J 2 (+ ) 7, 0 ? 0

(6.35)

in which , 0 is the yield stress in shear and J2 is the second deviatoric invariant of the stress
tensor. The J2-invariant can be computed from the principal stresses +1, + 2 and + 3 according
to the following formula:

1
J 2 ' (+ 1 7 + 2 ) 2 8 (+ 2 7 + 3 ) 2 8 (+ 3 7 + 1 ) 2
6

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

(6.36)

201

It is customary to present yield conditions graphically in the principal stress space with axes
+1, + 2 and + 3. The line on which all three principal stresses are equal is the hydrostatic axis,
and planes perpendicular to this line are deviatoric planes. Intersections of the yield surface
with deviatoric planes are deviatoric sections, and intersections with planes that contain the
hydrostatic axis are called meridians.
The yield surface corresponding to the Mises criterion is a cylinder rotationally symmetric
about the hydrostatic axis and is generally not acceptable for concrete structures. All
deviatoric sections are identical circles, and all meridians are identical straight lines parallel to
the hydrostatic axis. Any normal to the Mises cylinder lies in a deviatoric plane. This means
that the flow rule associated with the Mises condition gives purely deviatoric plastic flow,
which is for metals in a good agreement with reality. Volume changes are in this case purely
elastic.
In Eq. 6.35, the shear yield stress , 0 is a constant material parameter, and so the yield surface
does not evolve during plastic flow. This is the case of perfect plasticity (i.e., plasticity
without hardening or softening), which does not use any hardening parameters. More
advanced models take into account the changes in the microstructure induced by plastic flow
and the resulting evolution of the yield surface. The simplest hardening model only expands
or shrinks the yield surface radially. Such isotropic hardening can be described by a single
scalar hardening variable, k. To characterize the total amount of plastic flow, the strainhardening model uses the accumulated plastic strain, defined by the rate equation as,

f! '

2
)! p
3

(6.37)

where | . | denotes the Euclidian norm of a tensor. The role of the scaling factor 2 3 is to
make k coincident with the usual plastic strain under uniaxial loading, provided that the
plastic flow is purely deviatoric. Alternatively, the work-hardening model defines k as the
total plastic work, the rate of which is

f! ' + T )! p

(6.38)

By substituting from Eqs. 6.23 and 6.25 we can convert either of these definitions into the
general form of Eq. 6.24. The dependence of the current shear yield stress on the hardening
variable, , 0 ' h(f ) , can easily be identified from a uniaxial test. The yield function is then
defined as

f (+ ,, 0 ) t J 2 (+ ) 7, 0

(6.39)

where ,0 is the only component of the vector q.


The assumption of isotropic hardening is a rather crude simplification. If the specimen is
subjected to tension, isotropic hardening has the same effect on the yield stress in
compression as on the yield stress in tension. This does not reflect the well-known
Bauschinger effect. An improvement is achieved by kinematic hardening, which translates the
yield surface in the stress space without changing its size or shape. The current amount of
translation is described by the back stress :. For example, the Mises yield function with
kinematic hardening is written as

202

6 Advanced modelling and analysis concepts

f (+ , : ) t J 2 (+ 7 : ) 7 , 0

(6.40)

A classical evolution law for the back stress was proposed by Prager (1955) and improved by
Ziegler (1959). Realistic modelling of metals under cyclic loading requires even more
complex models (Mroz 1967, Dafalias and Popov 1975, Krieg 1975).
For pressure-sensitive materials such as concrete, rock, gravel or soil, it is important to take
into account the effect of volumetric part of the stress tensor by the first stress invariant,
I1 ' + 1 8 + 2 8 + 3

(6.41)

or by the mean (volumetric, hydrostatic) stress,

+ V ' I 1 / 3 ' (+ 1 8 + 2 8 + 3 ) / 3

(6.42)

By introducing a dependence on the volumetric stress into the Mises condition, we obtain the
Drucker-Prager criterion as

f (+ ,, 0 ) t : I1 (+ ) 8 J 2 (+ ) 7, 0 ? 0

(6.43)

If the friction coefficient : is constant, the yield surface is a rotationally symmetric cone with
straight meridians and circular deviatoric sections. Experimental measurements on concrete
specimens indicate that the meridians are curved and the deviatoric sections have the form of
a rounded triangle, whose shape changes from almost triangular for tensile and low
compressive hydrostatic pressures to almost circular for high compressive hydrostatic
pressures. As the deviatoric section is not circular, the yield function must depend on the third
deviatoric invariant as,
J 3 ' (+ 1 7 + V )(+ 2 7 + V )(+ 3 7 + V )

(6.44)

It is more convenient to transform J3 into the so-called Lode angle, which has a direct
geometrical meaning. This angle, -, is defined by the relationship
cos 3- '

3 3 J3
2 J 23 / 2

(6.45)

and its geometrical interpretation in the deviatoric plane is indicated in Figure 6.1a. A fairly
general description of the strength envelope can be presented in the form:
f ( I 1 , J 2 ,- ) ' c1 I 1 8 c 2 r (- ) J 2 8 c3 J 2 7 1 ' 0

(6.46)

where c1, c2 and c3 are material parameters and &()) is a suitable function of the Lode angle
related to the shape of the deviatoric section. A failure criterion of the form given by Eq. 6.46
was first proposed by Ottosen (1977) as
g Z1
W
hh cos X 3 arccos( K cos 3- )U
V
r (- ) ' i Y
1
L
Z
hcos 7 arccos(7 K cos 3- )W
UV
hj XY 3 3

if cos 3- < 0
(6.47)

if cos 3- ? 0

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

203

where K is a shape factor affecting the out-of-roundedness of the deviatoric section; see
Figure 6.1a. Another function &()) with the desired properties was constructed already by
Willam and Warnke (1974) but they exploited it in a somewhat different manner. Their
function,
r (- ) '

4(1 7 e 2 ) cos 2 - 8 (2e 7 1)2


2(1 7 e 2 ) cos - 8 (2e 7 1) 4(1 7 e 2 ) cos 2 - 8 5e 2 7 4e

(6.48)

is derived from an ellipse in the deviatoric plane. Parameter e is the eccentricity, again related
to the shape of the deviatoric section, as illustrated in Figure 6.1b. Mentrey and Willam
(1995) presented a failure criterion of the form of with function &()) given by Eq. 6.48.
Both definitions given by Eqs. 6.47 and 6.48 lead to a convex failure surface if the parameters
K or e are chosen within a certain admissible range. The surface is smooth everywhere except
for the intercept with the hydrostatic axis. If the parameters are determined from the same set
of experimental data, there is usually only slight difference between the surfaces generated by
the Ottosen criterion and the Mentrey-Willam criterion. As an example, Figure 6.2 shows the
deviatoric sections of the Ottosens surface constructed with parameters c1 = 0.16 MPa-1,
c2 = 0.5 MPa-1, c3 = 0.005 MPa-2 and K = 1. These values correspond to the recommendation
of the CEB Model Code 1990 (1993) for concrete of grade C20.
Finally, let us point out that associated flow rules usually mispredict plastic changes of
volume in frictional materials such as concrete, and they have to be replaced by nonassociated ones, often written in the form

)! p ' ^!

sg
s+

(6.49)

where g(),q) is a new function called the plastic potential. We might say that the flow rule is
associated if the yield function is used at the same time as the plastic potential. A general
plastic potential defines a set of equipotential surfaces such that the plastic strain always
grows in the direction normal to the surface on which the current stress state is located. It is
often possible to assume that the flow is associated in the projection onto the deviatoric plane
and only its volumetric part is non-associated. A suitable plastic potential can derive from the
yield function by modifying only the term that reflects the pressure sensitivity of the material.
From the Drucker-Prager criterion (Eq. 6.43), we could derive a plastic potential as
g ( I1 , J 2 ) ' :r I1 8 J 2

(6.50)

where : r is the dilatancy coefficient, in general different from the friction coefficient : that
appears in the yield function. The most difficult component of a plasticity model for concrete
is a realistic description of hardening and softening. This issue will be addressed later in the
sections dedicated to particular models.

204

6 Advanced modelling and analysis concepts

e=0.5
e=0.6
e=0.7

K=1.0
K=0.9
K=0.8

(b)

(a)
Figure 6.1:

Influence of the shape parameter on the deviatoric section correspondingto zero mean pressure
according to a) Ottosen, b) Mentrey and Willam.

80

60

250

0 MPa
-10 MPa
-20 MPa
-30 MPa

200

-30 MPa
-50 MPa
-100 MPa
-200 MPa

150

40
100

20

50
0

50

20

100

40
150

60

200
250

80

(b)

(a)

Figure 6.2: Deviatoric sections of Ottosens surface at a) low and b) high levels of mean pressure.

6.2.3

Continuum damage mechanics

The structure of a typical model based on damage mechanics will be explained using the
simplest example - the isotropic damage model with a single scalar parameter. The stressstrain equations are written in the total form of

+ ' (1 7 N) De)

(6.51)

where N is the damage parameter. Initially, N is equal to zero, and the response of the material
is linear elastic. As the material deforms, the initiation and propagation of microdefects such
as voids or microcracks decreases the stiffness, which is reflected by the growth of the
damage parameter. The model described by Eq. 6.51 is based on the simplified assumption

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

205

that the stiffness degradation is isotropic, i.e., stiffness moduli corresponding to different
directions decrease proportionally, independently of the direction of loading.
Similar to plasticity, we introduce a loading function f specifying the elastic domain and the
states at which damage grows. In damage theory it is natural to work in the strain space. The
loading function is therefore chosen to depend on the strain, ), and on an additional parameter
k that describes the evolution of the elastic domain. States for which f(), k) < 0 are supposed
to be below the current damage threshold. Damage can grow only if the state reaches the
boundary of the elastic domain. This is described by the loading/unloading conditions that
again have the Kuhn-Tucker form of,

f () , f ) ? 0,

f! < 0,

f! f () , f ) ' 0

(6.52)

It remains to link the damage threshold k to the damage parameter N. It would be possible to
use a damage evolution law in a rate form. However, as both variables grow monotonically, it
is more convenient to postulate an explicit relation between their total values as

N ' g (f )

(6.53)

The function g affects the shape of the stress-strain diagram and can be directly identified
from a uniaxial test.
An important advantage of the damage model is an easy evaluation of the stress
corresponding to a given evolution of strain. The loading function usually has the form,

f () , f ) ' )# () ) 7 f

(6.54)

where, )~ is the equivalent strain. The formula defining the equivalent strain plays a similar
role to the yield function in plasticity, because it directly affects the shape of the elastic
domain. The simplest choice is to define the equivalent strain as,

)# ' )

or

)# ' ) T De) / E

(6.55)
(6.56)

where, the normalization by E is introduced in order to obtain a strain-like quantity. Every


definition of equivalent strain corresponds to a certain shape of the elastic domain in the strain
space. For illustration, Figure 6.3a shows the elastic domains in the principal strain plane and
in the principal stress plane for the case of plane stress and Poissons ratio n = 0.2. The
domains are elliptical and symmetric with respect to the origin. Consequently, there would be
no difference in the response to tensile and compressive loadings. For concrete and other
materials with very different behaviours in tension and in compression, it is necessary to
adjust the definition of equivalent strain. Microcracks grow mainly if the material is stretched,
and it is natural to consider only normal strains that are positive and neglect those that are
negative. This leads to the modified definitions

)# ' )

206

(6.57)

6 Advanced modelling and analysis concepts

Figgure 6.3: Loaading surfacess for various definitions


d
of equivalent
e
straain.

or,

)# '

De ) / E

(6.58)

where McAuley
M
brackets < > denote thee positive part of. For
F scalars, < x > = maax(0 , x),
i.e., < x > = x for x positive annd < x > = 0 for x neg
gative. For symmetric
s
ttensors, succh as the
strain teensor ), the positive paart is a tensoor having th
he same prinncipal axes as the origiinal one,
with priincipal valuues replacedd by their poositive partss. Consequeently Eq. 6.57 can be rewritten
r
as

)~ '

P)
i '1

2
i

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

(6.59)

207

where )i, i = 1, 2, 3 are the principal strains. The elastic domains corresponding to Eqs. 6.57
and 6.58 are shown in Figure 6.3b. If a model corresponding to the Rankine criterion of
maximum principal stress is desired, one may use the definitions

)# '

1
max De)
E
i '1,2,3

(6.60)

or,

)# '

1
1
De) '
E
E

P
i '1

De)

2
i

(6.61)

where < De) >i, i = 1, 2, 3, are the positive parts of principal values of the effective stress
tensor De). The former definition exactly corresponds to the Rankine criterion while the latter
rounds off the corners in the octants with more than one positive principal stress; see Figure
6.3c.
An important advantage of damage models is that the stress evaluation algorithm is usually
explicit, without the need for an iterative solution. For a loading function in the form of
Eq. 6.54, the variable f has the meaning of the largest value of equivalent strain that has ever
occurred in the previous deformation history up to its current state. For a prescribed strain
increment, the corresponding stress is computed simply by evaluating the current value of
equivalent strain, updating the maximum previously reached equivalent strain and the damage
parameter, and reducing the effective stress according to Eq 6.51. Depending on the definition
of equivalent strain, we may have to extract the principal strains or principal stresses. This can
be done relatively easily, since closed-form formulas for the eigenvalues of symmetric
matrices of size 2x2 or 3x3 are available.
Finally, let us emphasize that the purpose of the present brief introduction was to illustrate
some of the basic aspects of the approach based on damage, using the simplest possible
model. Damage mechanics literature contains a large number of models with various levels of
complexity and sophistication, e.g., two-parameter scalar damage models in the compliance
form (Ladevze 1983, Mazars 1985, Borderie 1991) or in the stiffness form (Mazars 1985,
Mazars 1986), models with two second-order damage tensors and one scalar parameter
(Ramtani 1990, Papa and Taliercio 1996), a model characterizing damage by a fourth-order
damage tensor (Chaboche 1979), or models characterizing damage directly by the compliance
tensor or the stiffness tensor (Ortiz 1985, Simo and Wu 1987, Yazdani and Schreyer 1988) to
name only a few.
6.2.4

Smeared crack models

Smeared crack models present the total strain as a sum of two parts - one corresponds to the
deformation of the uncracked material, and the other is the contribution of cracking. The
response of the uncracked material can be governed by a general nonlinear material law but
usually is assumed to be linear elastic. The strain decomposition is written as

) ' )e 8 )c

(6.62)

and the elastic strain is related to stress by the linear law

208

6 Advanced modelling and analysis concepts

+ ' De) e

(6.63)

The crack strain, )c, represents in a smeared manner the additional deformation due to the
opening of cracks. A crack is initiated when the stress state reaches a certain failure surface.
Models aiming at the description of fracture under shear and compression exploit more
general criteria (Weihe 1995, Peekel Instruments 1994). The initiation criterion specifies also
the orientation of the crack. According to the Rankine criterion the crack plane is normal to
the direction of maximum principal stress. In a local coordinate system with axis n normal to
c
the crack and axes m and l lying in the crack plane, the only nonzero crack strains are ) nn,
c
c
& nm and & nl. In other words, opening of the crack contributes to the normal strain in the
direction normal to the crack plane, and sliding of the crack contributes to the shear strains in
planes normal to the crack plane. The corresponding strain components in global coordinates
are obtained by the transformation as
g )11c u Z n12
h c h X 2
h) 22 h X n2
c
2
hh) 33 hh X n3
'
X
i cv
h& 23 h X 2n2 n3
h& 13c h X 2n1n3
h ch X
jh& 12 wh XY 2n1n2

n1m1
n2 m2
n3 m3
n2 m3 8 m2 n3
n1m3 8 m1n3
n1m2 8 m1n2

W
U
n2l2 U c
g ) nn u
n3l3 U h c h
U i& nm v
n2l3 8 l2 n3 U h c h
& nl
n1l3 8 l1n3 U j w
U
n1l2 8 l1n2 UV
n1l1

(6.64)

where ni, mi and li, i=1,2,3 are the components of unit vectors in the direction of axes n, m and
l. In a compact notation, transformation rule (Eq. 6.64) can be written as

) c ' Tec

(6.65)

The components of ec are assumed to be directly related to the traction transmitted by the
crack, s = {snn, snm, snl}T which is a projection of the stress on the crack plane. As s and ec
form a work-conjugate pair, the same as + and )c, the stress transformation formula

s ' T T+

(6.66)

contains the transpose of the strain transformation matrix T from Eq. 6.65.
The relationship linking s to ec is a smeared counterpart of a traction-separation law exploited
by discrete (cohesive) crack models. Early smeared crack models assumed that the traction
transmitted by the crack immediately drops down to zero. As explained in Chapter 3, such an
approach leads to results that are not objective with respect to the mesh size. To ensure proper
energy dissipation, and also to avoid unrealistic stress jumps, it is necessary to describe the
loss of cohesion as a gradual process. Under pure Mode-I conditions, the normal traction can
be considered as a decreasing function of the normal crack strain,
c
s nn ' f () nn
)

(6.67)

where the function f is easily identified from a uniaxial experiment.


The cracking law (Eq. 6.67) for normal components is sufficient for the rotating crack model,
which assumes that the crack normal rotates and remains aligned with the current direction of

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

209

maximum principal strain. In contrast, the fixed crack model freezes the crack direction at the
moment of crack initiation. The crack in general transmits shear that produces relative sliding
of the crack faces, represented by shear components of crack strain. In simplistic versions of
the fixed crack model, the shear traction is taken as proportional to the shear crack strain, with
a proportionality factor (G, where G is the shear modulus of elasticity and ( < 1 is the socalled shear retention factor (Suidan and Schnobrich 1973). This is not very realistic because
such a model crack can transmit large shear tractions even when it is widely open. If the shear
retention factor is constant, a very small value has to be set, e.g. ( = 0.01, to limit the spurious
stress transfer that may lead to the stress locking 2. A better remedy is to make ( variable and
decrease it to zero as the crack opening grows (Cedolin and Poli 1977). It is also possible to
formulate the relation between s and ec in the spirit of damage theory, with a scalar damage
parameter depending on an equivalent crack strain that is computed from ec. Whatever choice
is made, the relations can always be transformed into the incremental form with tangential
stiffness matrix Dc as+
$
s! ' Dc e! c

(6.68)

The rate forms of Eqs. 6.62, 6.63 and 6.65, can be combined to yield

+! ' De)!e ' De ()! 7 )!c ) ' De ()! 7 Te!c )

(6.69)

By substituting this into the rate form of Eq. 6.66 and comparing with Eq. 6.68 we obtain a set
of equations for e!c from
e!c ' (T T DeT + D c ) -1T T De )!

(6.70)

It is possible to show that D e ' T T De T is a sub-matrix of the elastic stiffness matrix


expressed in the local coordinates. If the elastic properties are isotropic,
D e '

Z1 7 n
E
X 0
(1 8 n )(1 7 2n ) X
XY 0

W
U
(1 7 2n ) / 2
0
U
0
(1 7 2n ) / 2UV
0

(6.71)

is the same in any coordinate system, and no matrix multiplication is needed. Moreover, if the
cracking laws for the normal components and for the shear components are decoupled, then
the crack stiffness matrix D c is diagonal, and the inversion of the diagonal matrix D e 8 D c is
straightforward. Substituting Eq. 6.70 back into Eq. 6.69 we finally obtain the relation
between the increments of stress and total strain in the form

+! ' D)!

(6.72)

D ' De 7 De T ( D e 8 D c ) 71 T T De

(6.73)

is the tangent material stiffness. An alternative expression is


!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
2

In the present context, stress locking means spurious stresses build up around the band of cracking elements.
This pollutes the numerical results and leads to an overestimated energy dissipation and residual strength of a
cracked structure.

210

6 Advanced modelling and analysis concepts

D ' (C c 8 TC c T T ) 71

(6.74)

with C e ' De71 = elastic compliance matrix, and C c ' D c71 = tangent crack compliance matrix.
Eq. 6.74 is formally simpler than Eq. 6.73 but it requires the inversion of a 6x6 matrix while
in Eq. 6.73 we have to invert only a 3x3 matrix (sometimes even diagonal).
So far we have tacitly assumed that the crack opening is monotonically increasing. However,
it can also happen that the crack starts closing (unloading), and the tangent stiffness has to be
replaced by the stiffness valid for unloading. Usually it is assumed that unloading takes place
to the origin (Figure 6.4a), so that the crack strain completely disappears when the applied
stress is removed. A model with some permanent strain (Figure 6.4b) is more realistic but in
most practical simulations there is little difference between the two unloading models. When
the applied normal stress changes sign to compression, the crack closes completely and
becomes locked. The crack strain should not assume negative values because the crack
faces cannot overlap. This is reflected by the vertical part of the diagram in Figure 6.4a. Upon
complete crack closure, the material restores its original stiffness, and the overall response is
purely elastic as long as the normal traction remains negative. If this traction changes sign
again, the crack starts re-opening at constant secant stiffness, and when the crack opening
reaches its previous maximum, the basic curve described by Eq 6.67 is followed.

"c

"c

(a)

(b)

Figure 6.4: Cracking law a) with unloading to the origin, b) with some permanent strain.

The approach outlined above can be generalized to the case of m cracks with different
orientations. The basic equations remain formally the same but the column matrices s and ec
now consist of blocks that correspond to individual cracks, the transformation matrix T is
augmented accordingly, and the tangent crack stiffness matrix becomes block-diagonal:

s ' $s1 s2 .... sm %T ,

T ' pT1 T2 % Tm q ,

ec ' $ec1 ec 2 .... ecm %T


Z D c1
X
Xx

Dc ' X
X &
Xx
Y

x
D c2

(6.75)

x W
U

x U

& UU
% D cm UV

(6.76)

However, for m cracks the number of possible combinations of loading and unloading is 2m,
and it may be difficult to find the correct one. Another difficulty is a proper generalization of

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

211

the crack initiation criterion. If the standard criterion of maximum principal stress was kept,
the normals of individual cracks could be arbitrarily close to each other. The number of
generated cracks could be theoretically infinite. Then, it was proposed to impose additional
restrictions such as the minimum threshold angle between a new crack and the already
existing ones (Borst and Nauta 1985).+
As an alternative to the multiple fixed crack model, the rotating crack model takes into
account the possible existence of cracks with various orientations by a continuous adjustment
of the orientation of a single crack. The crack normal is assumed to always coincide with the
direction of maximum principal strain. Here, the directions of principal stress and strain
remain aligned and no shear tractions or shear crack strains appear in the formulation. This
simplifies the cracking law because a relation between the normal components of type
Eq. 6.67 is needed. On the other side, it is necessary to take into account that the
transformation matrix T does not remain constant because the crack normal can rotate. For
example, when differentiating Eq. 6.65 we have to write
! 8 Te!
)!c ' Te
c
c

(6.77)

This complicates the derivation of the tangent stiffness matrix but the resulting formula is
remarkably simple (Baant 1983, Willam et al. 1987). In principal coordinates, we have:
Z(C# + C# c ) -1
D'X e
x
Y

xW

(6.78)

U
D# s V

This formula covers the general case, in which cracking can take place in all three principal
planes. Matrix
7n
1

Z 1
1X
~
C e ' X7 n
E
XY7 n

7n

7n W
7 n UU
1 UV

(6.79)

is the block of the elastic compliance matrix that corresponds to normal components, and Cc
is the diagonal crack compliance matrix with current compliances of three mutually
orthogonal cracks on the diagonal. If some of the principal directions is not cracking, the
corresponding crack compliance is set to zero. The block of the stiffness matrix (Eq. 6.78)
that corresponds to shear is

Z s2 7 s3
X
X 2$e2 7 e3 %
~
0
Ds ' X
X
X
X
0
XY

0
s3 7 s1
2$e3 7 e1 %
0

W
U
U
U
0
U
s1 7 s2 UU
2$e1 7 e2 % UV
0

(6.80)

where the symbols si and ei, i = 1, 2, 3, stand for principal stresses and principal strains. The
tangent stiffness matrix D can be rotated into the global coordinates using the standard
transformation rule.

212

6 Advanced modelling and analysis concepts

The shear coefficients in Eq. 6.80 are properly defined only if the principal strains ei,
i = 1, 2, 3, are mutually different. The case of two or three equal principal strains is a weak
point of this particular form of the rotating crack model (Jirsek and Zimmermann 1997). An
alternative formulation that avoids this deficiency is embodied within the Disturbed Stress
Field Model, described by Vecchio (2000, 2001).+
6.2.5

Microplane models

Unlike conventional tensorial models that relate the components of the stress tensor directly to
the components of the strain tensor, microplane models work with stress and strain vectors on
a set of planes of various orientations (so-called microplanes). The basic constitutive laws are
defined on the level of the microplane and must be transformed to the level of the material
point using certain relations between tensorial and vectorial components. The most natural
choice is to construct the stress and strain vector on each microplane by projecting the
corresponding tensors, i.e., by contracting the tensors with the vector normal to the plane.
However, it is impossible to use this procedure for both the stress and the strain and still
satisfy a general law relating the vectorial components on every microplane. The original slip
theory for metals worked with stress vectors as projections of the stress tensor; this is now
called the static constraint. Most versions of the microplane model for concrete and soils have
been based on the kinematic constraint, which defines the strain vector e = (e1, e2, e3) on an
arbitrary microplane with unit normal n = (n1, n2, n3,) as
ei ' ) ij n j

(6.81)

where, ) ij, i, j=1, 2, 3 are the components of the strain tensor. When dealing with tensorial
components, we use the Einstein summation convention implying summation over twicerepeated subscripts in product-like expressions. For example, subscript j, on the right-hand
side of Eq. 6.81 appears both in ) ij and in nj, and so a sum over j running from 1 to 3 is
implied.
The microplane stress vector, s = (s1, s2, s3,), is defined as the work-conjugate variable of the
microplane strain vector, e. The relationship between e and s is postulated as a microplane
constitutive equation. A formula linking the microplane stress vector to the macroscopic stress
tensor follows from the principle of virtual work, written here as

+ ij O) ij '

3
2L

s i Oei d y

(6.82)

where O) ij are components of an arbitrary (symmetric) virtual strain tensor, and

Oei ' O) ij n j

(6.83)

are components of the corresponding virtual microplane strain vector. Note that the
summation convention implies summation over i and j on the left-hand side of Eq. 6.82,
summation over i on the right-hand side of Eq. 6.82, and summation over j on the right-hand
side of Eq. 6.83. Integration in Eq. 6.82 is performed over all microplanes, characterized by
the components of their unit normal vectors, ni, i = 1, 2, 3. Because of symmetry, the
integration domain y is taken as one half of the unit sphere, and the integral is normalized by
the area of the unit hemisphere, 2L/3.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

213

Substituting Eq. 6.83 into Eq. 6.82 and taking into account the independence of variations O),
we obtain (after certain manipulations restoring symmetry) the following formula for the
evaluation of macroscopic stress components:

+ ij '

3
4L

( si n j 8 s j ni ) dy

(6.84)

In summary, a kinematically constrained microplane model is described by the kinematic


constraint Eq. 6.81, the stress evaluation formula Eq. 6.84, and a suitable microplane
constitutive law that relates the microplane strain vector, e, to the microplane stress vector, s.
If this law has an explicit form (Carol et al. 1992)+

s'~
s (e, n)

(6.85)

then the resulting macroscopic stress-strain law can be written as

+ ij '

3
4L

G p~s (e, n)n


y

8 ni ~
s j (e, n) dy

(6.86)

Realistic models for concrete that take into account the complex interplay between the
volumetric and deviatoric components of stress and strain (Baant and Prat 1988, Obolt
1995, Baant et al. 1996) usually lead to more general microplane constitutive laws of the
type

s ' s# (e,n;+ )

(6.87)

that are affected by some components of the macroscopic stress, +, for example by its
volumetric part. Instead of a direct evaluation of the explicit formula Eq. 6.86, the
macroscopic stress is then computed as the solution of an implicit equation, and the stressevaluation algorithm involves some iteration.

6.3

Solution strategies

6.3.1

Introduction

As mentioned in Chapter 3, the most common solution algorithm used in FE analysis of


concrete structures is one based on a Newton or modified-Newton technique. A few of the
possible methodologies are discussed in this section.
In Section 3.4, we started with a point on the equilibrium path (qo , & p) , as shown in
Figure 3.18, where qo is the current displacement vector, & is a load magnification parameter
and p is a reference vector of applied loads. Equilibrium of the discretized structure was then
written as

^ p 7 f (q) ' 0

(6.88)

K( q )q 7 f (q) ' 0

(6.89)

or

214

6 Advanced modelling and analysis concepts

where f is a vector of internal forces and is a function of the current displacement state. The
solution of the equation system is then obtained from
Kq f = 0

(6.90)

and can be calculated directly for a linear system. For a non-linear system, however, an
iterative approach is needed with successive solutions of the linear system until convergence
is obtained.
A number of algorithms have been introduced into finite element programs to trace
equilibrium paths. The majority of these use either the Newton-Raphson (NR) or modified
Newton (mNR) techniques as their basis. The NR and mNR algorithms work well when linear
or bi-linear material relationships are being used, however, become more and more inefficient
as higher degrees of non-linearity are introduced. A further disadvantage of the NR and mMR
method is that without the addition of special techniques a falling load path cannot be
handled. An early method of obtaining limit points was to use displacement control as given
by Batoz and Dhatt (1979). This method has been used successfully on many occasions;
however, it has a number of drawbacks, many of which are highlighted in Crisfield (1981). A
further problem when trying to apply this method to finite elements which have a high degree
of material nonlinearity (such as those modelling concrete) is that the initial solution, which
forms the basis of further iterations, may be well away from the final equilibrium state and
divergence and/or algorithm failure may occur.
A major improvement to the standard NR mNR techniques was introduced by Riks (1972)
and Wempner (1971) and later modified by Crisfield (1981) and Ramm (1981) and involves
control of the load/displacement path. The constant arc length method was later modified to
include line searches and accelerations (Crisfield, 1983).
6.3.2

Newton-Raphson method

The most frequently used iteration scheme for the solution of non-linear equations is some
form of the Newton-Raphson procedure. In the case where Eq. 6.88 cannot be solved exactly
the residual forces at the ith iteration can be written as
!

'()* + , -./ 0 1()* +

(6.91)

Eq. 6.91 may be differentiated with respect to q to obtain


df (qi )
d
(r (qi )) ' 7
' 7 K (qi )
dq
dq

(6.92)

If an approximate solution q ' qi is obtained, then the truncated Taylor expression may be
written as
r (qi 81 ) ' r (qi ) 8

dr
(qi ) O qi ' 0
dq

(6.93)

where O qi ' _ qi 81 7 _ qi and, therefore,

r (qi ) ' K (qi ) O qi

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

(6.94)

215

Substituting Eq. 6.94 into Eq. 6.93 and rearranging yields

Oqi K (qi )71 (_^p 7 f (qi ))

(6.95)

Denoting K $qi % , the tangent stiffness matrix for the current displacement state, as K Ti , then
71

Oqi ' K Ti ( _^ p 7 f ( qi ))

(6.96)

The Newton-Raphson solution process is illustrated in Figure 3.18, noting that for every step
the current stiffness matrix is formed and the linearized equations solved for O qi .
6.3.3

Modified Newton-Raphson method

From the above calculations it is seen that the formulation of a new tangent stiffness for each
iterative cycle and the solution of a new system of equations must be undertaken. In
computing terms, this can be time consuming and costly. To overcome this difficulty the
approximation K Ti ' K To is often made. This modifies Eq. 6.96 to

O qi ' K To

71

( _^ p 7 f ( qi ))

(6.97)

and resolution of the same equation set is repeatedly used. The solution at each iteration is
sped up; however, more iterations to convergence are required, as can be seen in Figure 3.19.
The overall economy of the solution procedure is dependent on the problem size and nonlinear behaviour. An updated mNR approach may be adopted where the tangent stiffness
matrix, K Ti , is updated if convergence is not obtained after a pre-determined number of
iterations, n, and continues to be updated every n iterations following.
6.3.4

Incremental displacement method

The incremental displacement method was developed to overcome problems with load control
solution schemes and their inability to deal with structural stability problems or problems
where the loading decreases with increasing displacements.
A discrete representation of a linear problem is given by Eq. 6.90 where the internal force
vector is defined by Eq. 6.88. In a usual linear analysis the load vector, ^ p , is given and
Eqs. 6. 88 and 6.90 can then be solved directly for the displacements, q. If, however, the jth
component of q is chosen as unknown, the standard form of Eq. 6.90 needs to be modified to
solve for & and (n-1) unknown displacements of q.
Batoz and Dhatt (1979) developed an algorithm such that the structures stiffness matrix, K, is
not modified, instead a solution of two vectors q a and q b is sought such that
K q a ' r uh
v
K q b ' p hw

(6.98)

where r is the residual vector defined in Eq. 6.91. By using this method the structural stiffness
matrix remains banded and symmetric.

216

6 Advanced modelling and analysis concepts

If the stiffness matrix, K, of Eq. 6.90 is non-singular, then Eq. 6.98 maybe solved. The
solution for & and q is given by

q ' q a 8 ^ qb

(6.99)

q j ' q aj 8 &q bj

(6.100a)

where for the jth component

^'

q j 7 q aj

(6.100b)

q bj

For a purely linear system, r = 0 and thus Eq. 6.99 leads to

q ' ^qb
^'

(6.101a)

qj

(6.101b)

q bj

The algorithm of Batoz and Dhatt can be applied to the Newton-Raphson solution procedure.
Letting the quantities (qo , ^o p) denote the displacement and load state at a point on the
equilibrium path, as shown in Figure 6.5, then instead of increasing the load parameter, the jth
component of q is incremented by _q j . The initial displacement vector, q, is modified so that

qi ' qo where qij ' qij 8 _q j .


The residual force vector due to the modified initial displacements is computed and the
displacement vectors due to the residual force, r, and unit load, p, are computed
simultaneously from Eq. 6.98, that is,

p_q & q q' K (q )


a
i

b
i

71

pr (qi )& pq

(6.102)

where K $qi % is the non-linear tangent stiffness matrix corresponding to the current
displacement state, q i .
Eq. 6.100 maybe rewritten in the form

O qi ' _ qiq 8 O ^i qib

(6.103)

j
It is desired that the jth element of O qi is equal to zero (that is, O qi ' 0) . This leads to

O ^i ' 7 c _ q aj q 8 _ q j `
j
d

ai

(6.104)

The structure displacement vector can now be computed together with the loading parameter
using
!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

217

Figure 6.5: Schematic representation of incremental displacement method using mNR iteration.

_ qi 81 ' _ qi 8 O qi
_^i 81 ' _ ^i 8 O ^ i

(6. 105)

qi 81 ' qi 8 O qi
If the algorithm is used with the modified Newton-Raphson method, then the structure
stiffness matrix, K $qi % is updated only at selected iterations. A summary of the solution
method is depicted in the schematic outlined in Figure 6.5.

218

6 Advanced modelling and analysis concepts

6.3.5

The constant arc length method

The constant arc length method was developed to overcome some of the problems associated
with relying on merely load or displacement control. The principle behind the arc length
method is to increment the load-displacement path in such a manner as to derive the benefits
of both load and displacement control. This was first introduced by Riks (1979) using a
normal to the tangent stiffness and later modified by Crisfield (1981) to use a circular path;
the latter method being slightly less likely to fail.
Let the quantities $qo , ^o p % denote the displacement and current load vector at a given point
on the equilibrium path, as shown in Figure 6.6. The equation governing equilibrium is then
given by Eq. 6.88 and the residual force vector, r, given by Eq. 6.91.
Since the arc length method treats both & and q as variables, an extra equation is necessary to
enable solution of Eqs. 6.88 and 6.91. This extra equation comes from the constraint
relationship

_ qiT81 _ qi 81 ' _ qiT _ qi ' _ ' 2

(6.106)

where _' is the prescribed incremental length and _ qi is the incremental displacements after
the (i-1)th iteration (see Figure 6.6).

Figure 6.6: Schematic representation of constant arc length method.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

219

The solution procedure starts by incrementing the load parameter by _^ , and forming the
current tangent stiffness pKa q . The initial change in displacement vector, O qo can be
calculated by

Letting

HK

71
a .

Oqo ' _ ^1 K a71 . p

(6.107)

O q1 ' O qo ' _^1 # T

(6.108)

p ' # T gives

By now applying the constraint equation 6.106 the arc length can be set as
#TT . #T

_' ' _^1

(6.109)

and remains constant for the remainder of the iterations within this load step.
From Eq. 6.91 the residual force vector, r i , may now be calculated. The solution continues
with the calculation of O q i from

O qi ' K a71 . p ri 8 O ^i . p q

(6.110)

Note that the load parameter & must be modified by O ^i in order not to violate the constraint
equation (that is, _^i 81 ' _^i 8 O^i ). Letting
the form of

HK

71
a . ri

I ' # , Eq. 6.110 may be rewritten in


i

O qi ' # i 8 O ^i #T

(6.111)

Writing the updated displacement vector in the form of

_ qi 81 ' _ qi 8 Qi O qi

(6.112)

where Qi is an acceleration parameter obtained by line searches (if line searches are not used
then Qi ' 1.0) and substituting Eqs. 6.112 and 6. 111 into Eq. 6.106 gives

p _q 8Qi p #i 8 O ^i #T qq T p _q 8Qi p#i 8 O ^i #T qq7 _'2 ' 0

(6.113)

Resolving Eq. 6.113 leads to the quadratic expression

a1 O ^i2 8 a2 O ^i 8 a3 ' 0

(6.114)

where

220

6 Advanced modelling and analysis concepts

a1 ' Qi #TT #T

a2 ' 2 #TT _qi 8 Q1#TT #i


T

a3 ' 2 #i _ qi 8 Qi # i # i

u
h
h
h
v
h
h
h
w

(6.115)

The two roots of Eq. 6.114 correspond to the two points of intersection between the
equilibrium path and the surface described by Eq. 6.113. The appropriate root is the one that
gives forward progression and may be obtained by ensuring an acute angle ) exists between
_ qi and _ qi 81 where
cos - ' 1 8

Qi Z T
# _qi 8 O^i #TT _qi W
2 XY i
UV
_'

(6.116)

In most cases Eq. 6.116 will yield one positive value of cos ) (corresponding to an acute value
for )) and one negative value and hence there is no problem finding the root that gives
forward progression. In the event that both values for cos ) are positive, the appropriate root
is the one that is closest to the linear solution a3 a2 . Once the value for O^ i has been
obtained, substitution into Eq. 6.111 yields Oqi and subsequently into Eq. 6.112 for O^ i 81 .
As yet application of the arc length method to the modified Newton-Raphson technique has
only been discussed. The constant arc length method can be applied equally well using the
Newton-Raphson technique or any updated mNR method by recalculating the #T vector
whenever the tangent stiffness matrix [Ka] is reformed. The arc length, _' , however, is not
recalculated until the end of the current load/displacement increment.
6.3.6

Line searches

Irons and Elsawaf (1970) and Elsawaf (1979) first applied line search concepts to non-linear
finite element problems. Since then many researchers have adopted the line search technique
as a method of accelerating towards a solution. Crisfield (1983), as a method to avoid
numerical difficulties caused by concrete cracking, introduced line search techniques into the
arc length method.
The line search concept seeks a scalar Qi such that the energy z at qi 81 is stationary in the
direction of Qi , that is;
sz
sz
sq
'
' 7 riT81 #i ' S j ' 0
sQi i 81 sqi i 81 sQi

(6.117)

In practice Eq. 6.117 cannot be met. Instead it is satisfactory to satisfy


S j 9 r So

(6.118)

where

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

221

u
h
v
T
S j ' 7 # i . r i 81 hw
S o ' 7 #Ti . r i

(6.119)

and * is the required solution tolerance. The work of Crisfield (1982, 1983), Wolfe (1975),
Dixon (1972), Shanno (1978) and Foster (1992) indicate that a slack tolerance of * = 0.8 is
optimum. For a complete mathematical formulation of the line search methods see Crisfield
(1982, 1983).
6.3.7

Convergence criteria

In a finite element analysis of the non-linear behaviour of a structure, it is almost impossible


to solve Eq. 6.88 exactly $ ie; r $qi % ' 0 % using the iterative solution techniques discussed. For
this reason a measure of whether the solution has reached a predetermined accuracy is
required. A simple method of monitoring convergence is to check either the out-of-balance
force vector or the change in displacements after each iteration. A load step cycle may be
considered complete whenever

r i 9 ) max r

( j ' 1,2,3... i )

(6.120)

T
where r i ' O q i . O q i or r i ' r (q i )T . r (q i ) , as appropriate.

When modelling concrete structures, due to the changing of constitutive relationships from
the uncracked to cracked states and the change in the material strain relationship as a function
of the principal stress ratio, the displacement criteria to set the solution tolerance is often used.
6.3.8 Load-displacement incrementation

Ideally when using one of the Newton techniques to solve non-linear problems, the increase in
load or displacement, as the case may be, from a converged state to a new state, should reflect
the current degree of non-linearity. If the load step is too large then the problem will be slow
to converge. If the load step is too small, then a larger number of steps is required to define
the load-displacement state than needed. Crisfield (1983) remarked that, for modelling of
reinforced concrete it may not be wise to relate the load increment size to the number of
iterations required to achieve results at the previous load step. This is due to unpredictable
nature of concrete and the large redistribution of forces when cracking occurs. Substantial
change in the model occurs frequently and the number of iterations required to give
convergence may increase dramatically. This, however, does not necessarily affect subsequent
load steps and no apparent advantage is obtained by reducing the initial load increment (+&o)
for the following load steps. In fact, substantial computer time may be taken to map a small
aberration in a few elements that ultimately do not affect the results of the model. A more
reliable method of automatic load incrementation is to monitor the performance of a key
displacement component within the model and modify the load increments (for load control
methods or the constant arc length method) within a set maximum and minimum in order to
obtain optimum displacement increments.

222

6 Advanced modelling and analysis concepts

6.4

Other issues

6.4.1

P peak response
Post
r
off compressiion elemen
nts

There are
a many othher compliccating issuees related to specific orr special struuctures not covered
in this report andd it is for the designner to pre-aassess theirr importancce before adopting
a
computeed results. In
I the case of
o columns,, for examp
ple, if inform
mation on a post-peak response
r
is desirred issues such
s
as spaalling of coover concreete and bucckling of thhe reinforceement in
compresssion membbers needs consideratio
c
on. Figure 6.7
6 illustrates typical ffailure in th
he plastic
hinge reegion of a cyclically
c
looaded colum
mn specimen
n tested by Bayrak (1999). Since buckling
b
of the longitudinal
l
l bars is geenerally noot considereed in conveentional anaalytical pro
ocedures,
behavioour at large inelastic cuurvatures is generally over-predict
o
ed with resppect to stren
ngth and
ductilityy.

Figuure 6.7: Longiitudinal bar buuckling in speecimen RS-12H


HT (Bayrak, 11999).

6.4.2

E
Effects
of ageing
a
and distress in concrete

Concrette is an ageeing materiaal, and its mechanical


m
properties change witth time. At an early
age, concrete structure may be subjected to interrnal actionss due to innternal therm
mal and
moisturre gradientss, and it may
m be affected by ex
xternal condditions of lloads, restraint and
surrounnding enviroonment. Thhe coupling of these actions
a
leadds to non-unniform defo
ormation
inside structural
s
m
members,
whhich resultss in the gen
neration of internal strress and sometimes
causes early
e
age cracking. Suuch initial defects
d
are likely to reeduce durabbility perforrmances,
load carrrying capaccity, and duuctility of a concrete
c
strructure.
The vollume changee of the matterial at an early age iss mainly cauused by tem
mperature risse due to
hydratioon reaction and autogeenous and drying
d
shrin
nkage rootedd in the mooisture behaaviour in
micro pores.
p
Here,, the internaal stress gennerated by the volumee change is strongly deependent
on creeep behaviouurs. In addition, for a PC design
n, considerration of crreep is esseential to
evaluatee the loss of
o prestress. Here, a coouple of evaaluation meethods for sshrinkage an
nd creep
are briefly reviewed.
Several practical models
m
have been develloped to preedict creep and
a shrinkaage deformaations. In
such moodels, it hass been a customary proccedure to deefine the tottal strain off a uniaxially
y loaded
specimeen at time t after the casting of conncrete as folllows,

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

223

)$t % ' ) e $t % 8 ) c $t % 8 ) s $t % 8 ) T $t %

(6.121)

where, )(t) is the total strain, )e(t) the instantaneous strain, ) c(t)the creep strain (which is
divided into basic and drying creep), )s(t) is the shrinkage strain and )T(t) is the thermal
dilatation. For stresses up to approximately 40% of the strength of concrete, creep can be
assumed to be proportional to the stress (known as Davis-Granville relationship). Based on
this assumption under constant stress +, the above equation can be transformed as,
)$t % ' + T J $t , t #% 8 ) s $t % 8 ) T $t %

(6.122)

where, J(t, t) is defined as a compliance function or creep function, which represents the
strain at time t produced by a unit constant stress acting since time t. It can be written as,

J $t , t #% '

1
1 8 C$t , t #%
8 C $t , t #% '
E $t #%
E $t #%

(6.123)

where, E(t)is the elastic modulus of concrete at time t # , C $t , t # % is the specific creep that
represents the creep strain at time t produced by a unit constant stress acting since time t # and
C(t, t # ) is a creep coefficient which represents the ratio of the creep strain to the elastic strain
at time t. These creep functions have been formulated based on concrete properties, structural
dimensions, and atmospheric conditions by many past researches and design codes.
CEB-FIP Model Code 1990
In the CEB-FIP model code, the concrete creep function is
) c $t , t #% '

+ c $t #%
+ $t #%
T C$t , t #% ' c pC 0 T ( c $t 7 t #%q
Ec
Ec

(6.124)

where, +c: constant stress applied at time t, Ec: modulus of elasticity at the age of 28 days, C0
(t, t # ): notional creep coefficient and it depends on the ambient relative humidify, size of the
member, and the mean compressive strength at the age of 28 days, and (c (t, t # ): coefficient
describing the development of creep with time after loading and it depends on the ambient
relative humidity and size of member.
In the code, shrinkage is calculated by the following equation as,
) s $t , t s % ' ) so T ( s $t 7 t s %

(6.125)

where, ) so is the notional shrinkage coefficient and it depends on the ambient relative
humidity, the mean compressive strength at the age of 28 days, and the type of cement, (s is
the coefficient to describe the development of shrinkage with time, which depends on the size
of member, t is the age of concrete and ts is the age of concrete at the beginning of shrinkage
or swelling.

224

6 Advanced modelling and analysis concepts

National codes of practice


All national codes of practice include simplified creep and shrinkage models in their contents
that can be used in modelling. For example, the Japan Society of Civil Engineers (JSCE)
Code 2002 has the following model for creep:
)#c ' C T +#cp E ct

(6.126)

# the applied stress and Ect the modulus


where, ) c# is the creep strain, C a creep coefficient, + cp

of elasticity at the loading age. In the code, the following equation to obtain specific creep is
introduced as,

) #c $t , t #, t 0 % +#cp ' 1 7 exp 7 0.09$t 7 t #%

0.6

IqT )#

cr

) #cr ' ) #bc 8 ) #dc

(6.127)

# is the ultimate value of specific basic creep strain, which is a function of unit
where, ) bc
# is the ultimate value of specific
weight of water and cement and water-to-cement ratio and ) dc
drying creep strain, which is a function of ambient relative humidity, unit weight of water and
cement, water-to-cement ratio, exposure surface area, and structural dimension.

The above models are generally used to predict a mean value of shrinkage and/or creep over
the cross section of a certain member. For practical purposes, such simple treatments are easy
to use, but their applicability should be clearly kept in mind. That is to say, they dont give
local properties within the cross section of a concrete member, such as the variation of
internal stresses, moisture states, and generation of micro cracking. In addition, these models
adopt the conventional separation between the shrinkage, drying creep, and basic creep in
their formulations. Strictly speaking, however, these behaviours should not be treated
separately but they are rather several aspects of one physical phenomenon.

Other models
Recently, several models based on the microphysical phenomena in concrete, such as cement
hydration, moisture transport/equilibrium, and microstructure of cement paste, have been
proposed (Baant and Prasannan, 1989, Baant et al., 1997, Lokhorst and Breugel, 1997,
Maekawa and Ishida, 2001). Since they try to simulate the actual rheological phenomena in
the material and structure from the mesoscopic viewpoint, it is expected that they are a
breakthrough giving a solution to classic research topics in the concrete engineering, i.e.,
creep and shrinkage. Figure 6.8 shows an example of such an approach for a creep test by
Ross (1958) undergoing a complex loading and unloading history and modelled by Chong
(2004) using the solidification formulation of Baant and Prasannan (1989).
6.4.3

Effects of ageing and distress in reinforcing steel

As an effect of ageing/distress of reinforcement, corrosion may be main influential factor on


structural performance. Corrosion of reinforcement is associated with formation of rusts. The
volume of the oxide increases two to ten times at the steel/concrete interface (Broomfield
1997). This leads to cracking and spalling of cover concrete, and eventually it affects overall
structural performances, such as reduction of load carrying capacity and ductility.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

225

Figure 6.8: FE modelling of Ross (1958) creep test (Chong, 2004).

Figure 6.9 shows a schematic representation of the effect of corrosion on structural


performance. When a reinforcing bar corrodes, it causes damage affecting structural
behaviour, including: 1) damage of reinforcement itself (loss of bar section, reduction of
strength, etc.); 2) damage of bond between reinforcement and surrounded concrete; and 3)
damage of concrete around corroded reinforcement due to the expansion of the oxide
(cracking and spalling of cover concrete). In order to evaluate the structural performance of a
structure having corroded reinforcement, the above phenomena should be appropriately taken
into account.
Steel corrosion

Volumetric expansion
accompanied with rust

Loss of bar section


Reduction of strength
Acceleration

Crack formation

Bond

Spalling
Loss of concrete cross section

Load carrying capacity

Ductility

Structural performance
Figure 6.9: Effect of corrosion on structural performance

226

6 Advanced modelling and analysis concepts

6.4.4

Second ord
der effects

Where second ordder effects are


a importaant in assesssing the sttrength or sservice limiit states,
these neeed to be inncluded in thhe modellinng. Many, iff not most, commerciaal packages used for
modelling of concrrete structurres will havve some cap
pability for modelling geometricaally nonlinear problems
p
toogether with the non-linear mateerial modellling descriibed in the various
chapterss of this repport. An exxample of a common sttructural eleement wherre designerss will be
familiarr with the issue
i
of seccond order effects in their
t
treatm
ment of slennder column
ns where
theeffecct of geometric non-linnearity is to magnify beending mom
ments. A seccond, less common,
c
examplee is in creeep buckling of a beam
m-column, su
uch as thatt shown in Figure 6.10
0, where
geometrric non-lineearity combbines with time effeccts and othher material non-lineaarities to
engendeer failure.
Other seecond orderr effects incclude the reesponse of concrete
c
eleements subjected to hig
gh strain
rates suuch as, for example, blaast and impaact. The treeatment of thhese speciaal cases, how
wever, is
not withhin the scoppe of this repport.

Steel ch
hannel section
eT
Eccentric loadiing
Strong wall

1250

Ten
nsioning cable

Dial gaugee

Stirrups 10 at 15
50
Clearr cover 15 mm
1250

2N12
5000

Dial gaugee

Dial gaugee

2N12

1250

150
Section A-A

1250

Test co
olumn

eB

Hydraaulic jack
I-sectio
on loading arm
Load cell

Figuree 6.10: Modellling of concreete columns suubjected to creeep buckling (Chong,


(
2004,, Chong et al.,, 2008).

6.5

R
Reference
es

Batoz J.,
J and Dhattt G. (1979)), Incremenntal displaceement algorrithms for nnonlinear prroblems,
Int. J. Num.
N
Meth. Engng., Vool 14, pp. 12262-1266.
Bayrak,, O. (1999),, Seismic Performanc
P
ce of Rectilinearly Connfined Highh Strength Concrete
C
Columnns, Thesis submitted in conformitty with the requiremennts for the D
Degree of Doctor
D
of
Philosopphy in the University
U
o Toronto, pp.339.
of
Baant Z. P. (19883), Comm
ment on orthotropic models forr concrete and geomaaterials,
Journal of Engineeering Mechaanics, ASCE
E, 109, pp.8
849-865.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

227

Baant Z. P. and Prat P. (1988), Microplane model for brittle plastic materials. I: Theory, II:
Verification, Journal of Engineering Mechanics, ASCE, 114, pp.1672-1702.
Baant Z.P. and Prasannan S. (1989), Solidification theory for concrete creep. I.
Formulation, II. Verification and application, Journal of Engineering Mechanics, 115(8),
pp.1691-1725.
Baant Z. P., Xiang Y., and Prat P. C. (1996), Microplane model for concrete. I: Stress-strain
boundaries and finite strain, Journal of Engineering Mechanics, ASCE, 122, pp.245-254.
Baant, Z. P., Hauggaard, A. B., Baweja, S., and Ulm, F.-J. (1997). Microprestresssolidification theory for concrete creep. (I: Aging and drying effects, II: Algorithm and
verification). Journal of Engineering Mechanics, ASCE, Vol. 123, No. 11, pp. 1188-1201.
Borderie C. La (1991), Phnomnes unilatraux dans un matriau endommageable:
Modlisation et application lanalyse de structures en bton, PhD thesis, Universit Paris
VI., France.
Borst R. and Nauta P. (1985), Non-orthogonal cracks in a smeared finite element model,
Engineering Computations, 2, pp.35-46.
Broomfield, J.P. (1997), Corrosion of steel in concrete, Understanding, investigation and
repair, E&FN SPON.
Carol I., Prat P., and Baant Z. P. (1992), New explicit microplane model for concrete:
Theoretical aspects and numerical implementation, International Journal of Solid and
Structures, 29, pp.1173-1191.
CEB-FIP Model Code 1990 (1993), Thomas Telford Services, Ltd., London, for Comit
Euro-International du Bton, Bulletin dInnformation No. 213-214, Lausanne, pp.437.
Cedolin L. and Poli S. Dei (1977), Finite element studies of shear-critical R/C beams,
Journal of Engineering Mechanics, ASCE, 103(3), pp.395-410.
Chaboche J. L. (1979), Le concept de contrainte e.ective, appliqu llasticit et la
viscoplasticit en presence dun endommagement anisotrope, Grenoble, Editions du CNRS,
Number 295 in Col. Euromech 115, pp.737-760.
Chong, K.T. (2004). Numerical Modelling and Time Dependant Cracking and Deformation of
Reinforced Concrete Structures, PhD Thesis, The University of New South Wales, School of
Civil and Environmental Engineering, Australia,
Chong, K.T., Foster, S.J., and Gilbert, R.I. (2008). Time-dependent modelling of RC
structures using the cracked membrane model and solidification theory, Computers &
Structures, 86, pp.1305-1317.
Crisfield, M.A. (1981), A fast incremental/iterative solution procedure that handles snapthrough, Comput. Struct., Vol 13, pp.55-63.
Crisfield, M.A. (1982), Accelerated solution techniques and concrete cracking, Comp. Meth.
Appli. Mech. Engng. Vol 33, pp.585-607.

228

6 Advanced modelling and analysis concepts

Crisfield, M.A. (1983), An arc length method including line searches and accelerations, Int. J.
Num Meth. Engng. Vol 19, pp.1269-1289.
Dafalias Y. F. and Popov E. P. (1975), A model of nonlinearly hardening materials for
complex loading, Acta Mechanica, 21, pp.173-192.
Dixon L.C.W. (1972), Nonlinear optimization, The English University Press Ltd, London.
Elsawaf, A. (1979), The conjugate-Newton method for non-linear finite element problems.
PhD. Thesis, University of Calgary, Canada.
Foster S.J. (1992), An application of the arc length method involving concrete cracking, Int.
J. Num. Methods Eng., Vol. 33, No. 2, pp. 269-285.
Irons, B.M., and Elsawaf, A. (1970), The conjugate-Newton algorithm for solving finite
element equations, in formulations and algorithms in finite element analysis, (Eds. K.J.
Bathe, J.T. Oden and W. Wunderlich) MIT Cambridge, MA, pp. 656-672.
Jirsek M. and Zimmermann T. (1997), Rotating crack model with transition to scalar
damage: I. Local formulation, II. Nonlocal formulation and adaptivity, LSC Internal Report
97/01, Swiss Federal Institute of Technology, Lausanne, Switzerland.
Krieg R. D. (1975), A practical two-surface plasticity theory, Journal of Applied
Mechanics, ASME, pp.641-646.
Ladevze P. (1983), Sur une thorie de lendommagement anisotrope, Rapport interne
L.M.T. 34, E.N.S. de Cachan, Universit Paris VI.
Lvy M. (1870), Mmoire sur des quations gnrales des mouvements intrieures des corps
solids ductiles au del des limites ou llasticit pourrait les ramener leur premier tat, C.
R. Acad. Arts Sci., 70, pp.1323.
Lokhorst SJ and Breugel K. (1997), Simulation of the effect of geometrical changes of the
microstructure on the deformational behaviour of hardening concrete, Cement and Concrete
Research, 27(10), pp.1465-1479.
Maekawa K and Ishida T. (2001), Service-life evaluation of reinforced concrete under
coupled forces and environmental actions, Materials Science of Concrete, special volume,
Ion and mass transport in cement-based materials, pp.219-238.
Maekawa K, Chaube R. and Kishi T. (1999), Modelling of Concrete Performance, London:
E & FN SPON.
Mazars J. (1985), A model of a unilateral elastic damageable material and its application to+
concrete , In Fracture toughness and fracture energy of concrete, Elsevier, New York.
Mazars J. (1986), A description of micro and macroscale damage of concrete structures,+
International Journal of Fracture, 25, pp.729-737.
Mentrey Ph. and Willam K. J. (1995), A triaxial failure criterion for concrete and its
Generalization, ACI Structural Journal, 92, pp.311-318.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

229

Mroz Z. (1967), On the description of anisotropic work-hardening, Journal of the


Mechanics and Physics of Solids, 15, pp.163.
Ortiz M. (1985), A constitutive theory for the inelastic behavior of concrete, Mechanics of
Materials, 4(1), pp.67-93.
Ottosen N. S. (1977), A failure criterion for concrete, Journal of Engineering Mechanics,
ASCE, 103, pp.527-535.
Obolt J. (1995), Mastabseffect und Duktilitt von Beton- und Stahlbetonkonstruktionen,
IWB Mitteilungen 1995/2, Universitt Stuttgart, Germany.
Papa E. and Taliercio A. (1996), Anisotropic damage model for the multiaxial static and
fatigue behaviour of plain concrete, Engineering Fracture Mechanics, 55(2), pp.163-179.
Peekel Instruments, Rotterdam, The Netherlands. (1994). SBETA Program Documentation.
Prager W. (1955), Probleme der Plastizittstheorie, Birkhuser, Basel, Stuttgart.
Ramm E. (1981), Strategies for tracing non-linear response near limit points, in
formulations and algorithms in finite element analysis, (Eds. K.J. Bathe, J.T. Oden and W.
Wunderlich) MIT Cambridge, MA, pp. 656-672.
Ramtani S. (1990), Contribution la modlisation du comportement multiaxial du bton
endommag avec description du caractre unilatral, PhD thesis, Universit Paris VI.
Riks E. (1972), The application of Newtons method to the problem of elastic stability, J.
Appl. Mech., Vol. 39, pp.1060-1066.
Ross, A.D. (1958), Creep of Concrete under Variable Stress, ACI Proceedings, 54(9),
pp. 739-758.
Saint-Venant B. (1870), Mmoire sur ltablissement des quations diffrentielles
mouvements intrieures oprs dans les corps solids ductiles au del`a des limites ou
llasticit pourrait les ramener leur premier tat, C. R. Acad. Arts Sci., 70, pp.473.
Shanno D.F. (1978) , Conjugate gradient methods with inexact searches, Math O.R., Vol. 13,
pp. 71-113.
Simo J. C. and Wu J. W. (1987), Strain and stress based continuum damage models: Part IFormulation, International Journal of Solids and Structures, 23(7), pp.821-840.
Suidan M. and Schnobrich W. C. (1973), Finite element analysis of reinforced concrete,
Journal of the Structural Division, ASCE, 99(10), pp.2109-2122.
Tresca. (1868), Mmoire sur lcoulement des corps solides, Mmoires prsents par divers
savants, 18, pp.733-799.
Vecchio, F.J. (1990). Reinforced Concrete Membrane Element Formulations, ASCE
Journal of Structural Engineering, Vol. 116, No. 3, pp. 730-750.
Vecchio, F.J. (2000), Disturbed Stress Field Model for Reinforced Concrete: Formulation,
ASCE J. Structural Engineering., Vol. 126, No. 9, pp. 1070-1077.

230

6 Advanced modelling and analysis concepts

Vecchio, F.J. (2001), Disturbed Stress Field Model for Reinforced Concrete:
Implementation, ASCE J. Structural Engineering., Vol. 127, No. 1, pp. 12-20.
Von Mises R. (1913), Mechanik der festen Korper im plastisch-deformablem Zustand,
Gttinger Nachrichten (math.-phys. Klasse), 1, pp.582-592.
Wolfe M.A. (1975), Numerical methiods for unconstrained optimization an introduction,
Van Nostrand-Reinhold, London.
Yazdani S. and Schreyer H. L. (1988), An anisotropic damage model with dilation for
concrete, Mechanics of Materials, 7(3), pp.231-244.
Weihe S. (1995), Modelle der fktiven Ribildung zur Berechnung der Initiierung und
Ausbreitung von Rissen, PhD thesis, Universitt Stuttgart.
Wempner, G.A. (1971), Discreet approximations related to nonlinear theories of solids, Int.
J. Solids Struct., Vol. 7, pp. 1581-1599
Willam K. J. and Warnke E. P. (1974), Constitutive model for the triaxial behavior of
concrete, Concrete Structures Subjected to Triaxial Stresses, volume 19 of IABSE Report,
pp.1-30, International Association of Bridge and Structural Engineers, Zurich.
Willam K., Pramono E., and Sture S. (1987), Fundamental issues of smeared crack models,
In S. P. Shah and S. E. Schwartz, editors, Fracture of Concrete and Rock, pp.192-207, Society
for Experimental Mechanics, Bethel, Connecticut.
Ziegler H. (1959), A modification of Pragers hardening rule, Quarterly of Applied
Mathematics, 17, pp.55-65.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

231

[Blank Page]

Benchmark tests and validation procedures

7.1

Introduction

Non-linear finite element analysis (NLFEA) programs for the design and analysis of structural
concrete have not yet matured to the point where a user can simply develop a geometrically
accurate model of the structure, specify basic material parameter values, and then expect that
the program will accurately predict the response (capacity, deformation, distribution of
internal stresses, etc.) of any structure to any prescribed loading. The goal of this chapter is to
provide guidance for the efficient and safe use of NLFEA programs for the design and
analysis of concrete structures. The development of this guidance is complicated by the
variation in the types and capabilities of different programs and in the reasons for which
numerical investigations are conducted.
NLFEA programs range from concrete-specific to large commercial programs. These
programs (or tools) employ a large variety of multiple material models, element types, and
solution schemes. The level of required validation is program and task specific. If a concretespecific tool is used to examine the type of behaviour of a structure for which this tool was
explicitly developed, then it is closer to being pre-calibrated. Some two-dimension frame and
continuum analysis programs are close to being this mature. By contrast, large commercial
programs are often not specifically designed to deliver reliable predictions of the behaviour of
concrete structures but do provide the user with a selection of material models to choose from
that need to be calibrated by user-defined values. In the use of these programs, the user must
accept a greater responsibility for calibrating and validating the program in order to produce
reliable estimates of the response of a modelled structure.
NLFEA tools can be used for the design of a structure as well as for conducting a detailed
investigation of a particular aspect of its behaviour such as the maximum stress in
reinforcement, crack orientation and width, or energy absorbing capacity. For the purpose of
general design and analysis, it is necessary that the practitioner evaluate the programs ability
to predict the strength and behaviour of a well-selected group of experiments (benchmark
tests) in order to determine a global safety (or strength reduction) factor to use in his or her
design. The predicted capacity of a structure is then multiplied by this global safety factor to
obtain a safe design capacity. Thus, a higher global safety factor is appropriate when the user
has reliable material data for setting material characteristic values and when the program
provides accurate predictions for the behaviour of physical experiments that capture the range
in behaviours that are expected in the structure being modelled. The more comprehensive the
validation activity, the greater confidence the designer can have in any calculated global
safety factor. In the absence of less reliable material data, or when the NLFEA program does
not provide a good estimate of the capacity of representative physical test specimens, or when
an extensive validation procedure is not conducted, the results of the analysis should be
penalized by the use of a lower global safety factor.
Chapter 7 provides guidance for a designer who is trying to assess the appropriate global
safety factor to apply to the results from a particular finite element prediction as well as for
the user who is trying to investigate a particular aspect of the behaviour of a concrete
structure. The most important segment of this chapter is in Section 7.2 which describes the
steps in the calibration and validation of NLFEA programs. Section 7.3 discusses the factors
to consider in the determination of a global safety factor. Section 7.4 addresses several
important issues in the selection, use, and validation of NLFEA programs. The remainder of
the document is devoted to the presenting case study examples. Section 7.5 presents an
example of the design of a squat concrete shear wall with openings in which the global safety
factor is first determined through a comparison of experimental test data with the predictions
of a specific NLFEA program. In the second case study (Section 7.6), a comparison is made
!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

233

between two different NLFEA programs for investigating the detailed behaviour of a deep
beam. One of these programs is a large commercial program in which several behavioural
models are available for the user to select between. The use of such a program requires
additional care by the user as is illustrated in this example. This chapter concludes in
Section 7.7 with a discussion of future trends in model validation.

7.2

Calibration and validation of NLFEA models

7.2.1

Overview of model calibration and validation process

The steps and flow of information in the calibration and validation of a NLFEA program is
shown in Figure 7.1. The case of the design of a shear wall using a 2-Dimensional continuum
finite element analysis program is used in this example in order to provide more specific
details for the types of information that need be exchanged at each step.

Calibration

Constitutive Models for Uniaxial State


of Stress for Concrete and Steel Reinforcement
(User may need to choose from a number
of models for compressive, tensile, and
shear behavior)

Level 1

Mechanical Characteristics
of Concrete and Reinforcement
[Material Level Benchmark Tests]

Modelling
Smeared Crack Model for Two-Dimensional
State of Stress for Membrane Element

Calculation
(Prediction)

Experimental Data from Panel


Elements Under Biaxial State
Validation
& Calibration* of Stress (Collins 1982)
[Element Level Benchmark Tests]

Used to Predict Capacity of Selected


Membrane Elements
Results from Comparison with Element Test
Data may be used to Calibrate Additional
Element Level Parameters Such as Crack
Spacing
Minor Adjustment
to Element-Level
Parameters*

Level 2

Application in
2-D FEM

Complete 2-Dimensional Continuum


Model for Design and Analysis of
Planar Walls

Calculation
(Prediction)

Results from Comparisons Used to


Assess Accuracy of 2-D Continuum
Model for Predicting Strength and Stiffness
Performance of Planar Walls; Global Safety
Factor Selected Based on Comparisons

Validation
Calibration*

Results from Lateral Load Tests


on Reinforced Concrete Planar
Walls
[Structural-Level Benchmark
Tests]

Level 3

*Any Calibration based on Element or Structural Level Experiments Must be Completed with Caution to Avoid
Tuning a Model to Fit One Specific Class of Experiments; Importance of Validating Using a
Comprehensive Series of Test Data that Captures the Range of Critical Behavior Characteristics cannot be
Overemphasized

Figure 7.1: Overview of procedure for model calibration and validation (after Okamura and Maekawa, 1991).

234

7 Benchmark tests and validation procedures

7.2.2

Level 1: model calibration with material properties

In the use of all NLFEA programs, it is necessary to specify the mechanical characteristics of
materials. This is likely to include initial elastic stiffness as well as peak stress and strain
values, concrete shrinkage strain, thermal expansion coefficients and other standard material
properties. The more accurately the user can specify the required information, the more
reliable will be the results of any analysis. In selecting material values, the user should use
mean (average) material properties and recognize that the material properties in structural
concrete may differ from those in material test specimens. For example, due to additional
shrinkage in thin members and the restraint due to the presence of reinforcement it is common
for structural concrete to crack under stresses that are considerably below what is obtained
from material test data. It is also important to recognize that there can be a wide variation in
some material properties and the reliability of analysis results is reduced if assumed values are
used. To illustrate this latter point, Figure 7.2 plots the tensile cracking stress (fct) as a
function of cylinder compressive strength (fcm). This wide variation illustrates the importance
of using test data and not simply relying on typically assumed relationships, such as the one
shown in Eq. 7.1, if the results of the analysis are expected to be sensitive to this value.

fct ' 0.33 fcm

(7.1)

where the values are in MPa.

Figure 7.2: Variation in tensile splitting strength as a function of cylinder compressive strength.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

235

7.2.3 Level 2: validation and calibration with systematically arranged elementlevel


benchmark tests

Since a NLFEA program does not provide the exact capacity of a structure, it is necessary to
begin the use of any investigation with an evaluation of the program through a comparison of
predicted and measured response of test structures that have been load-tested to failure. By
examining the ratio of measured to predicted capacity of a well selected set of experiments,
the user can determine a global safety factor to apply to the results of an analysis. This global
safety factor, which will be labelled ', will be a fraction less than or equal to 1.0. The safe
capacity predicted by the program for the design or analysis being conducted will then be the
capacity predicted by the NLFEA program multiplied by this global safety factor, '. The
global safety factor may be considered to be the end product of the model validation process.
While Level 1 was purely a calibration activity in which material parameter values were set,
Level 2 it not always a pure validation exercise. This is because the experimental test data that
is most suitable for model evaluation is often the most suitable test data for setting structurallevel parameters that account for such effects as compression softening, tension stiffening,
shear retention, loss in bond, crack spacing, etc. The tests that can serve this dual purpose are
simple tests where there is little chance for misinterpretation of the test data. These tests may
be referred to as fundamental or element-level tests. When evaluating a tool, or for that matter
calibrating structural-level parameters, the user needs to predict the behaviour of a
systematically arranged set of element-level benchmark tests that encompass the range in
possible behaviours that may be experienced by the structure being designed or studied.
Consider the following example to illustrate the evaluation of NLFEA programs at the
element level. The construction of the Panel Element Tester at the University of Toronto
(UofT) in 1979 allowed uniform tests on reinforced concrete shear panels to be performed for
the first time. Based on the results of an IABSE colloquium in June of 1981, it was decided to
hold an international competition (Collins, 1985) to test the quality of various numerical
predictions for a collection of four panel elements (A-D). Panel D is shown in Figure 7.3.

Figure 7.3: Panel D after failure in the panel element tester.

236

7 Benchmark tests and validation procedures

The tests were performed in August of 1981 and while the measured material properties were
made available, the actual experimental results were kept confidential to ensure that the
results of the contest would be true predictions. A total of 27 entries were received from 13
countries. The tests had been selected such that they would be difficult to predict; none of the
elements failing in biaxial shear. It was found that the predictions varied substantially from
the observed behaviour. Figure 7.4 presents the details of the panels. The best entry was able
to predict all 4 panels to within 17% of the measured capacity. However, other predictions
were in error by as much as a factor of 3 in the measured strength, as illustrated in Figure 7.5.
What made the UofT competition so valuable is that each of the four selected panels evaluates
an important aspect of the biaxial response of structural concrete. Panel A was selected so that
the accuracy of the shear transfer relationship and the compression softening model would
have little effect on the shear response of the test specimen. In Panel B, the softening of the
compression model dominated the response. In Panel C, it was most important to accurately
model both tension stiffening and shear transfer across cracks. In Panel D, it was important to
accurately model all aspects of behaviour in order to make a good prediction of behaviour.
When reviewing the accuracy of the 27 entries it was clear that using only 2 or 3 of the panel
tests would have led to an inaccurate assessment of the capabilities of the numerical tools;
thereby illustrating the importance of the selection of a systematically arranged set of
validation experiments (benchmark tests).
Since 1982 there has been significant advancements in the development of NLFEA programs
for structural concrete. Figure 7.6 provides a comparison of the predicted versus measured
shear behaviour of the four panels using a more recent NLFEA program and labelled as
NLFEAP-1.
Y

PANEL A
t

PANEL B
t
sx

0 9 ( 9( u

+ x ' 70.7,
+ y ' 70.7,
0 9 ( 9( u

sy

f c' ' 20.5 MPa

$ 0 ' 0.00190

f c' ' 20.5 MPa

$ 0 ' 0.00190

f xy ' 442 MPa

* x ' 0.01785

f xy ' 442 MPa

* x ' 0.01785

f yy ' 442 MPa

* y ' 0.01785

f yy ' 442 MPa

* y ' 0.01785

PANEL C
t

PANEL D
t
sx

0 9 ( 9( u

+ x ' 7(, 7 3.9) 9 0


+ y ' 7(, 7 3.9) 9 0
0 9 ( 9( u

sy

f c' ' 19.0 MPa

$ 0 ' 0.00215

f c' ' 21.7 MPa

$ 0 ' 0.00180

f xy ' 458 MPa

* x ' 0.01785

f xy ' 441 MPa

* x ' 0.01785

f yy ' 299 MPa

* y ' 0.00713

f yy ' 324 MPa

* y ' 0.00885

Figure 7.4: Description of panels A-D for the UofT competition.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

237

12

Contest Results for Panel B

Number of Predictions

Number of Predictions

Contest Results for Panel A

10
8
6

*x5'5*y

4
2
0

0.7

0.8

0.9

1.1

1.2

1.3

8
7
6
5
4
3
2
1
0

,
,

+y
*x5'5*y
0.3

1.4

0.5

0.7

Contest Results for Panel C

1.1

1.3

1.5

1.7

Contest Results for Panel D


6

Number of Predicitons

Number of Predictions

0.9

Predicted/Observed

Predicted/Observed

9
8
7
6
5
4
3
2
1
0

+x

*x5'52.55*y

0.5

0.6

0.7

0.8

0.9

1.1

1.2

1.3

1.4

+x

+y
*x5'5 25*y

2
1
0
0.6 0.8

1.2 1.4 1.6 1.8

Predicted/Observed

2.2 2.4 2.6 2.8

Predicted/Observed

Figure 7.5: Distribution of measured to predicted capacities of four test specimens.

10
Observed-A
Analysis-A
Observed-B

Shear Stress(MPa)

Analysis-B
Observed-C
Analysis-C

Observed-D
Analysis-D

Panel A

Panel B

Panel C

Panel D

Measured v test (MPa)

6.29

9.02

3.94

5.61

Predicted v nlfea (MPa)

6.597

8.244

4.136

7.2

Tatio v test v nlfea

0.95

1.09

0.95

0.78

0
0

10

15

20

25

Shear Strain (x0.001)

Figure 7.6: Comparison of measured and predicted behaviour of four test panels.

7.2.4

Level 3: validation and calibration at structural level

In order to make a thorough evaluation of a particular NLFEA program and thereby determine
the global safety factor, it is necessary to make detailed comparisons with carefully selected
test data of both element level tests (discussed in Section 7.2.4) and tests of complex
structures. These complex or structural-level tests can account for the effects of complex
geometries, loadings, and edge effects. In evaluating a model with structural-level tests, many
types of comparisons can be made. The most common and often important comparisons are
initial stiffness, capacity, deformation at peak load, and deformation at failure. The overall
load-displacement response captures all of these measures. It is also possible to compare a
wide range in other performance measures including the development, propagation, width and
orientation of cracks, the strains in reinforcement, and the distribution of compressive strain in
238

7 Benchmark tests and validation procedures

concrete. These other comparisons are determined by the specific needs of a particular
investigation.

7.3

Selection of global safety factor

In codes of practice, it is common to use a strength reduction factor, ', to determine the
reliable design strength as a fraction of the code calculated nominal (or predicted) strength.
The range of typical values for ' is between 0.6 and 1.0. This factor aims to account for
variability in actual material strengths, imperfections in construction, the influence of load
duration on structural behaviour, and inaccuracies in the relationships for calculating nominal
strength when applied to structures that may be outside of the range of those experiments used
to develop these relationships. For example, it is more common to use a lower safety factor in
calculating the reliable shear strength as code provisions typically use coarse empirical
relationships for calculating shear capacity. By contrast, a much less severe strength reduction
factor (up to 0.95) is used for flexural capacity for which the model for behaviour is well
established and verified.
An example is now used to illustrate how global factors of safety are selected in a code-ofpractice; the equivalent approach can be used for determining ' for any particular NLFEA
program. Figure 7.7 plots the ratio of the experimentally measured versus ACI318-02 code
calculated shear strength of reinforced concrete beams with shear reinforcement, where the
code calculated nominal shear capacity (Vn) is determined by

Vn ' Vc 8 Vs '

fc#
6

bwd 8

Av f y d

(7.2)

where bw is the width of the web, d is the member depth, Av is the area of shear reinforcement,
and s is the spacing of the transverse reinforcement (units are in N, mm, and MPa).

3.0
2.5

Vtest / VACI

2.0
1.5
1.0
0.5
0.0
0

20

40

60

80

100

120

140

f'c (MPa)

Figure 7.7: Shear strength ratio of RC members with shear reinforcement.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

239

As shown in Figure 7.7, Eq. 7.2 does not provide a very accurate estimate of the measured test
strength nor does it always produce a conservative result. However, with the use of a suitable
value of C, then this somewhat inaccurate equation can be safely used for design with the safe
design strength being taken as CVn.
In the development of codes of practice, the global safety factor is selected so that there is a
small probability that the actual capacity is less than the design strength (C o nominal
capacity). Based on an assumed normal distribution, there is a direct relationship between the
mean (,), the standard deviation of the strength ratio (s), the strength reduction factor C and
the percentage of the design strengths (or the Fractile level, F) that are expected to be less
than C o nominal capacity. This relationship between these values is given by

eC 7u b
F ' zc
`
d s a

(7.3)

C ' z 71 ( F ) T s 8 u

(7.4)

where z is a Gaussian cumulative distribution function.


To illustrate this relationship, consider the data described in Figure 7.7 for which the mean is
, = 1.35 and the standard deviation is s = 0.37. The statistical values in Table 7.1 can be
calculated using Eqs. 7.3 and 7.4. If no strength reduction factor is used (that is C = 1.0), then
in 17.2% of the cases it would be expected that the actual strength is less than the design
strength. When a strength reduction of C = 0.75 is applied, which is the current value for shear
used in ACI318, then for only 5.24% of the cases is the actual strength expected to be less
than the design strength (see Figure 7.8).
Table 7.1: Example of relationships between global safety factor and fractile level.

Item

C
F

1.00

0.95

0.1721 0.1398

Statistical Values
0.85
0.80
0.75

0.90
0.112

0.70

0.65

0.60

0.0883 0.0686 0.0524 0.0395 0.0293 0.0213

B = 1.35
s = 0.37
COV = 0.28

Fractile level
F

C = 0.75 C = 1.00

B = 1.35

F= 0.0524 F= 0.1721

Figure 7.8: Relationship between level of safety and strength reduction factor.

240

7 Benchmark tests and validation procedures

7.4

Other issues in the use and validation of NLFEA programs

7.4.1

Problem definition and model selection

In many cases, the practitioner has limited choice in the NLFEA program that is available to
apply to a particular problem due to financial, time, or technical constraints. In other cases,
the practitioner has some time to evaluate and select the most suitable program for their short
or long-term needs. There are many factors that influence which model is most appropriate to
use in an analysis, ranging from technical to practical issues, as discussed below.

Capability: Are the analytical models (constitutive models and failure theories) incorporated
in the NLFEA program capable of capturing the types of behaviour most relevant to the
design challenges (or problems) to be studied? Some of the behaviours that are difficult and
infrequently modelled well in structural concrete include the benefit of passive confinement,
damage accumulation due to cyclic loading, dowel action across crack interfaces, fracture in
plain concrete regions, interface shear transfer and compression softening.
Transparency of Underlying Behavioural Models: Are the models, algorithms, assumptions,
and solution routines employed in the NLFEA program clearly defined? Some programs,
employ classical mechanics approaches, use a few well defined models for behaviour and
employ reasonably strict convergence criteria. While this provides a transparent and wellfounded methodology, the inability to account for complexity of behaviour of cracked
structural concrete coupled with frequent convergence difficulties limit the range of
applicability of such programs. In addition, the user can be daunted by the impact that
selected parameter values and convergence criteria have on the predictions of these tools. By
contrast, less flexible NLFEA programs that are specifically designed to capture the
complexity of structural concrete behaviour can provide reliable predictions of structural
response without the need for extensive calibration or validation.
Guidance for Selecting Parameter Values: What level of documentation and guidance does
the NLFEA program provide for the determination of model input parameter values? It is
frequently the case that there is insufficient material and structural test data information for
defining all input parameters and, thus, guidance for selection of parameters values is often
necessary. It is also useful if documentation is provided of the sensitivity of typical solutions
to input parameter values.
Benchmark Tests: What is the availability of benchmark tests for the type of problem for
which the NLFEA program is to be utilized? Has the developer of the program provided a
large and systematic set of experiments and analysis examples? If there is a good selection of
benchmarks tests to use for the validation (and limited calibration) of the NLFEA program,
then the user will be better able to more easily evaluate the global safety factor.
Other Capabilities and Features of NLFEA Program: What is the ability of the NLFEA
program to model other materials, physics, and complex geometry? What are the pre and
post-processing capabilities of the program? Can the program automatically generate an
appropriately dense finite element mesh for the structure under investigation?
Many NLFEA programs are freely available while others cost up to tens of thousands of
dollars per year.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

241

7.4.2

Working within the domain of the programs capability

A common problem in the application of NLFEA is that the program is used to investigate the
type of behavior or type of structure that is outside the range of problems for which it was
designed. For example, a program that uses a smeared-cracking model may not be able to
predict the propagation of cracking and capacity of an unreinforced region. The range of
application of a particular tool can be determined by a careful review of the documentation
that should be delivered with the program.
A less obvious source of error is that the practitioner has not properly assessed the accuracy of
their program using a systematically selected set of elements and structural-level benchmark
experiments. For example, research has shown that the common assumption that shear
strength is proportional to the square root of the concrete strength does not hold when the
concrete strength exceeds around 60 MPa. In this case, it would be unwise to scale up the
results from tests on lower strength concrete by the square root of the compressive strength
for the design of a structure for which high-strength concrete is to be used.
Another example is the importance of scale. It is unsafe to evaluate the accuracy of a NLFEA
program using small scale experimental test results when the structure to be designed is
massive. This is illustrated in the following example. For many years the research community
assumed that the shear stress at failure was not influenced by the depth of a member until the
classic set of experiments were conducted on members up to 3 meters deep conducted by
Shioya (1989) as shown in Figure 7.9. These tests demonstrated that the shear stress at failure
of very large members could be less than half the shear stress at failure of smaller members
(Figure 7.10).
When selecting a set of experiments to use in any validation activity, it is important to use
highly regarded and carefully conducted experiments the results of which are reliable
measures of the actual performance of the test structure. Not all tests are equal and in many
instances the researchers have failed to provide the information necessary to ensure that the
results of the experiments are reliable. As an example, joint ASCE/ACI Committee 445
Shear and Torsion oversaw the establishment of a database of shear test results for
reinforced concrete members that did not contain shear reinforcement. While more than 1000
experiments were identified, after criteria were applied to assess the reliability of the test data,
only about 300 test results remained for use in the evaluation of code provisions. The main
reason that beams were disqualified were that they were either too small, calculated to fail in
flexure, calculated to have an anchorage failure, only specified material strengths were
available, or neither of the supports was a roller.

242

7 Benchmark tests and validation procedures

Figure 7.9: Tests of Large Scale Bridge Girders by Shioya (1989).

d (mm)
0
3.2

1000
PCA Tests of
Model Air Force
Warehouse
d
Beams

2.8
2.4
V
bwd fc#

2.0

(psi)

1.6

12

Air Force
Warehouse
Beams

" = As = 0.4%
bwd
fc# = 3500 psi
(24.1 MPa)
fy = 52 ksi
(386 MPa)

a = 2.5 mm
(0.1 in.)

0.8
8
d = 4 in.

20

0.20
0.15

V
bwd fc #
(MPa)

0.10
b = 39

20
39

24

0.25

12d

max. aggregate size


a = 25 mm (1 in.)
a=10 mm
(0.39 in.)

3000
d

V calculated here

1.2

0.4

2000

d = 79

b = 59
d=
118

25.4 mm = 1

40

60
80
100
d (inches)

120

0.05
0

Figure 7.10: Size effect in shear.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

243

7.5

Case 1: Design of a shear wall with openings

7.5.1

Objective

The variation in strains in the squat wall with opening described in Figure 7.11 is very
complex. Since neither engineering beam theory nor codes-of-practice provide detailed
guidance for the design of this type of structure, the strut-and-tie method (STM, See
Chapter 8) can be used for the design of this wall (Paulay, 1992). The designer may,
therefore, wish to evaluate the capacity of the designed structure and to investigate its
condition at service loading using a NLFEA program. In this example the design of this shear
wall is completed using the 2-D continuum program (NLFEAP-1) that was used to predict the
behaviour of the four Toronto Panels discussed in Section 7.2.3. As the performance of this
program for predicting the response of element-level benchmark tests has already been
completed for the grade of concrete and amount of reinforcement that is expected to be used
in this wall, the next step is to validate the program using a full structural level benchmark
test, or tests. Following this, all of the information obtained is used to establish the global
safety factor. The reinforcement determined from the strut-and-tie design is shown in
Figure 7.12.
7.5.2

Level 1 calibration

For this program, only Level 1 calibration is required, which is the setting of the mechanical
characteristics of the concrete and reinforcement. This involved setting the following material
parameter values: f-c = 31 MPa, "-c = 0.002, Es = 210000 MPa, fy = 414 MPa, "sh = 0.0025,
"rupt = 0.1 and fu = 621 MPa
7.5.3

Level 2 and 3 validation

Level 2 element-level validation of the program was conducted in Section 7.2.3. Prior to using
NLFEAP-1 for the design of the shear wall with multiple openings, a single structural level
validation is conducted. The shear wall with an opening, as described in Figure 7.13 and
which had been load tested to failure, was modelled in NLFEAP-1 as shown in Figure 7.14.
A comparison of the measured and predicted load versus deformation response of this shear
wall is shown in Figure 7.15 from which it can be determined that the shear strength ratio
(measured strength / predicted strength) is 164.3/163.8 = 1.00.
7.5.4

Evaluation of global safetyf

Table 7.2 presents the ratio of the measured to predicted capacity of the four panel elements
and the wall tested by Taylor (1998). The mean of the strength ratios B = 0.96 and the
standard deviation of these ratios s = 0.114.
Table 7.2: Measured and Predicted Capacities

244

Panel A

Panel B

Panel C

Panel D

Wall

Measured Capacity

6.29

9.02

3.94

5.61

164.3

Predicted Capacity

6.597

8.244

4.136

7.2

163.8

Ratio Measured/Predicted

0.95

1.09

0.95

0.78

1.00

7 Benchmark tests and validation procedures

15

30

15

3Q

15

3Q

15

30

15

30

60

30

2Q

30

15

15

30

15

60

30

2Q

30

6.00
2.52

8.83

7.10

4.25

Figure 7.11: Shear wall to be designed using NLFEAP-1.

D10-300

6-D16
0
0
7
R6 ties-300
0
0
1
2

2-D12
D12-300

0
0
1
sie
t
6
R

4-D20
4-D12
0
0
1
-s
ite
6
R

4-D10

4-D16

0
0
7

0
0
1
2

2-D12

8-D20
0
0
7
0
0
1
-s
e
ti
6
R

4-D16

8-D10

4-D10

0
0
1
2

2-D12
18-D12 14 -D10
4200

2100

R6 ties at 100

D 12 D10 D10

D12

700

700

D12 D16 D12

D 12

D16

Figure 7.12: Reinforcement in squat shear wall based on STM design (Paulay, 1992).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

245

48 (1 21 9 mm)

We b Bars

FOU RTH FLOOR

H oop s

2 .5 (64 m m) 6 .5 (16 5 mm)


8@3 .75 =3 0
( @95 =7 62 m m)

THIRD FL OOR
11 @7 .5 =8 2.5
(@ 19 1=20 96 mm )
A

A
9@ 7.5=67 .5
(@19 1=17 15 mm)

SEC OND FL OOR

14 4
(36 58 mm)

FIR ST FLOOR

4@3 .75 =1 5
( @95 =3 81 mm)
C

27 .5 @2 =5 5
(@ 51 =1 39 7 mm)

5@ 5=25
(@1 27 =6 35 m m)

GR OUN D LE VEL
27
(68 6 mm)
76 (1 93 0 mm)

RW3-O: Elevation view showing reinforcing

(a)
48 (1 21 9 mm)
3@ 2=6
(@51 =1 52 m m)
0.7 5
6
(1 9 mm )
(15 2 mm)

3@ 2=6
(@51 =1 52 m m)
0.7 5
(1 9 mm )

6
(15 2 mm)

8-# 3

3@ 7.5=22 .5
(@1 91=5 72 mm )

#2 @7 .5
(@1 95 mm)

lap spl ice = 6 (15 2 mm)

SECTION A-A
3@ 2=6
(@51 =1 52 m m)
0.7 5
(1 9 mm )

3.3 8
(8 6 mm )

8-# 3

2.6 3
6 7 mm)
3@ 5=15
(@1 27 =3 81 mm)

#2 @5
(@12 7 mm)

2-# 3

3@ 2=6
(@51 =1 52 m m)
0.7 5
3@ 4.5=13 .5
(1 9 mm )
(@ 114 =3 43 mm )

3/16 (5 mm ) h oo ps &
cr oss tie s@2( 51 mm )

SECTION B-B

3.3 8
(8 6 mm )

8-# 3

2.6 3
6 7 mm)
3@ 5=15
(@1 27 =3 81 mm)

#2 @5
(@12 7 mm)

0.7 5
(1 9 mm )

2-# 3

3@ 2=6
(@51 =1 52 m m)
0.7 5
(1 9 mm )
0.7 5
(1 9 mm )

12
(30 5 mm)

3/16 (5 mm ) h oo ps &
cr oss tie s@2( 51 mm )

SEC TION C -C
RW3-O: C ross sections

2.5
(6 4 mm )

0.7 5
(1 9 mm )

0.7 5
(1 9 mm )

3@ 2=6
(@51 =1 52 m m)

(b)
Figure 7.13: Shear wall experiment (Taylor, 1998): (a) front element view and (b) reinforcement details.

246

7 Benchmark tests and validation procedures

(b) Vertical reinforcement ratio

(a) Horizontal reinforcement ratio

Figure 7.14: Finite element model of shear wall experiment.

180

Lateral load (KN)

150
120
90
60

NLFEAP-1

30

Experiment

0
0

10

20

30

40

50

60

70

Top displacement (mm)


Figure 7.15: Predicted and measured load-deformation response of shear wall.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

247

Using the method in Section 7.4, the global safety factor was evaluated based on it being
acceptable in 5% of situations that the design strength (C o capacity) determined by the use of
NLFEAP-1 were less than the actual strength. Adopting an assumed normal distribution of
strength ratios, the global safety factor is evaluated by Eqs. 7.3 and 7.4 to be C = 0.76.
Since only a few benchmark tests were used to evaluate the performance of NLFEAP-1
program, it may be desired to use a lower global safety factor than that calculated above. As
additional comparisons are made, the updated value of the global safety factor will be more
reliable.
It is not surprising that NLFEAP-1 had such good success at predicting the capacity of the
benchmark tests as this program was specifically written for the design of structural concrete
with distributed reinforcement in two directions and thus the practitioner can anticipate that
the developers of this program took considerable effort to model the complexities of the
behaviour of structural concrete and to validate and calibrate this program based on large
numbers of systematic element-level and structural level benchmark experiments.
It is not surprising that NLFEAP-1 had such good success at predicting the capacity of the
benchmark tests as this program was specifically written for the design of structural concrete
and thus the practitioner can anticipate that the developers of this program took considerable
effort to model the complexities of the behaviour of structural concrete and to validate and
calibrate this program based on large numbers of systematic element-level and structural-level
benchmark experiments.
Design using a NLFEA program is of course an iterative procedure where the user proposes a
selected pattern of reinforcement and then uses the program to calculate the capacity. In this
case, the trial reinforcement pattern shown in Figure 7.12 was designed using the strut-and-tie
method The finite element model used in this analysis is shown in Figure 7.16 and the
predicted load deformation response of the squat wall with openings in shown in Figure 7.17.
Based on the results of the analysis using NLFEAP-1, the yield strength of the wall is 108 kN
while the ultimate capacity is approximately 136 kN. Using the global safety factor previously
evaluated (C = 0.76), the design capacity of the wall is determined to be 0.76 o 108 = 82 kN
which is just larger than the design load of 80 kN.
With the availability of the completed analysis, the designer can also investigate the condition
of the structure under serviceability loading or continue to refine the design to most efficiently
use the reinforcement or to provide a desired level of overall ductility.

248

7 Benchmark tests and validation procedures

(a) Horizontal reinforcement ratio

(b) Vertical reinforcement ratio

Figure 7.16: Finite element model used to model squat wall with openings.

160

Q (KN)

120

80

40

NLFEAP-1
Factored design load(Qu)

0
0

10

12

14

Top displacement (mm)


Figure 7.17: Load deformation response of squat wall by NLFEAP-1

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

249

7.6

Case study II: design of simply supported deep beam

7.6.1

Objective

The objective of this example is to illustrate the differences between the steps and capabilities
of two different NLFEA programs in predicting the behaviour of a deep beam that supports a
point load at mid-span. The first of these programs is NLFEAP-1 which was the concrete
specific program used in Case Study I. The second of these is a large commercial finite
element program and is denoted as NLFEAP-2. The commercial program provides several
material models for structural concrete and uses a classical mechanics approach.
7.6.2

Calibration and validation of NLFEAP-1

For this program, only Level 1 calibration is required, which is the specification of the
mechanical characteristics of the concrete and reinforcement. This involved setting the
following material parameter values: f c# = 33.5 MPa, "-c = 0.002, Es = 210 000 MPa
fy = 498 MPa, "sh = 0.01 ,"rupt = 0.1 and fu =549 MPa.
7.6.3

Calibration and validation of NLFEAP-2

The second program to be used for predicting the behaviour of a deep beam is a large
commercial NLFEA Program. The process of designing and analyzing a structure using this
type of program is substantially different from the use of the concrete specific tool as often
these comprehensive programs were not designed to produce accurate predictions of the
behaviour of concrete structures. Rather, they provide a choice of material models, solutions
schemes, and convergence limits to be selected by the user and thus the user is primarily
responsible for calibrating and validating these programs to ensure that they will produce
reliable predictions. This requires that the user becomes familiar with the underpinnings of the
program and evaluates the sensitivity of the program predictions to non-basic material
parameter values. The analysis methods employed in NLFEAP-2 and the results of a
sensitivity analysis are provided below. The calibration and validation of NLFEAP-2 follows
these discussions.

Overview of concrete model used in NLFEAP-2


There are two different analysis methodologies that are used in NLFEAP-2. The implicit time
integration scheme uses a traditional stiffness-based approach in which the tangent stiffness
equation is solved using the Newton-Raphson method (refer Chapter 3). In the iterative
approach, the number of time steps and convergence tolerances are used to control the
solution process. In the explicit time integration scheme, both static and dynamic FEA are
performed with the use of a stiffness matrix. At each time step, the process involves the
solution of multiple vector equations in which no iterations are required or convergence
criteria need be specified. A large number of very small time steps (more than 10,000) are
often required in the use of an explicit analysis methodology.
NLFEAP-2 supports two different concrete models, the smeared-crack model and the damage
plasticity model. The smeared crack model is most applicable when concrete is subjected to
monotonic loading and either tensile cracking or compressive crushing dominates the material
response. The effect of cracking is taken into account by adjusting how the stresses and
material stiffness values at integration points are calculated. Cracking is defined to occur
when the tensile stress reaches a failure surface that NLFEAP-2 refers to as a crack detection

250

7 Benchmark tests and validation procedures

surface, shown in Figure 7.18. After cracking, anisotropic behaviour is assumed. Plastic
straining in compression is controlled by a classical compression yield surface, as shown in
7.19.
5q

5+ u
5.

Crack detection surface

51

Compression surface

51

5.

5p

52

5+u

Figure 7.18: Tension failure surface.

Crack detection surface


Uniaxial tension

5+2
5+1

Uniaxial compression
Compression
surface

Biaxial
tension

Biaxial compression

Figure 7.19: Yield and failure surface for in-plane stress.

In order to achieve the best results from a NLFEA, it is often necessary to carefully select
values for many parameters as well as to choose from a variety of material models and
solution methodologies. The parameter values are used to control the behaviour and influence
of many effects including:
- crack spacing

- tension stiffening effect

- compression softening effect

- benefits of passive confinement

- interface shear transfer

- bond strength

- shear model

- shape of stress-strain response

- fracture energy and critical length

- dowel action

It can be very challenging to select values for the parameters that control the effects listed
above unless specialized test data is available. Since the results of an analysis can be greatly
influenced by the values selected for these parameters, it is necessary to conduct a sensitivity
analysis to understand the effect of parameter values on the predicted response. In
!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

251

NLFEAP-2, the concrete and reinforcement are modelled separately. Reinforcement is


modelled as truss elements in which a one-dimensional strain model is used. The effects of
tension stiffening and dowel effects are not accounted for in modelling the behaviour of the
reinforcement but rather are accounted for in the concrete model through tension stiffening
and shear retention parameters.

Sensitivity Analysis
A sensitivity analysis was conducted on the parameter values that control tension stiffening
and shear retention on the predicted response by NLFEAP-2 of a reinforced concrete panel
subjected to in-plane shear loading. As presented in previous chapters, the tension stiffening
effect accounts for the contribution of cracked concrete in tension for carrying tension in the
concrete between cracks. The tension stiffening relationship used in NLFEAP-2 is given in
Figure 7.20 in which the average tensile stress is shown to decrease linearly from the tensile
strength of concrete at "cr to zero at "ts. It is possible to have this tension degrade in a several
ways, but a linear relationship was selected for this sensitivity analysis. The shear retention
parameter, which is expressed as a ratio from S.R. = 0 to S.R. = 1.0, is used to define shear
stiffness after cracking as a fraction of the uncracked shear stiffness. It decreases from the
selected S.R. value to zero at a tensile strain at $max. See Figure 7.21. Note that the user must
select values for .cr, "cr, S.R., and "max .

Stress, +
Failure point

5+cr

Tension stiffening
curve

5)

5+cr

) ts

Strain, )

5E

Figure 7.20: Tension Stiffening relationship used in NLFEAP-2.

S.R.
1.0

0.5

)max )
Figure 7.21: Shear Retention relationship used in NLFEAP-2.

252

7 Benchmark tests and validation procedures

As shown in Figure 7.22, the results of an NLFEAP-2 prediction can be quite sensitive to the
selected tension stiffening strain limit. In this evaluation, the shear retention factor was set to
0.5 and the corresponding maximum shear retention strain was set to "max = 0.006.
The sensitivity of the shear response predicted by NLFEAP-2 as a function of the values used
in the shear retention parameters is shown in Figure 7.23. In this comparison, the value of the
tension stiffening parameter was held constant at 0.0012. Four different shear retention values
are chosen. In three of these four cases, the shear retention parameter is set so that the shear
modulus just after cracking is set to 50% of the uncracked modulus (S.R. = 0.5) and the
tensile strain at which the modulus reduces to zero ($max) is varied from 0.003 to 0.012. The
results in Figure 7.23 illustrate that as the maximum strain is increased, the stiffness of the
post cracking response and the progression in the convergence also increases.
6

Shear stress(MPa)

T-S=0.0012, S-R=0.5, 0.006


T-S=0.0018, S-R=0.5, 0.006

T-S=0.0036, S-R=0.5, 0.006


T-S=0.0048, S-R=0.5, 0.006

0
0

Shear strain(x10^3)

Figure 7.22: Sensitivity of the NLFEAP-2 prediction to the tension stiffening model.

Shear stress(MPa)

T-S=0.0012, No S-R

T-S=0.0012, S-R=0.5, 0.012


T-S=0.0012, S-R=0.5, 0.006

T-S=0.0012, S-R=0.5, 0.003

0
0

Shear strain(x10^3)

Figure 7.23: Sensitivity of NLFEAP-2 prediction to the shear retention parameter.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

253

Now that the models used in NLFEAP-2 have been reviewed and a sensitivity analysis has
been performed, the next step is the calibration of NLFEAP-2 for the deep beam example.
The basic material parameters for the deep are first selected, as illustrated for the use of
NLFEAP-1 in Section 7.6.2. Since NLFEAP-2 was not specifically designed for modelling
the behaviour of structural concrete, it is necessary to calibrate the program in Level 2 and as
the test data is used for model calibration, it cannot then be used for model validation.
One element-level calibration example is given below for NLFEAP-2 where the implicit
solution scheme and smeared cracking model is calibrated so to fit the measured shear stress
versus shear strain response of panel PV20 of Vecchio and Collins (1982), shown in
Figure 7.24, by adjusting the tension stiffening and shear retention values. The values for
these parameters that create a good fit with the test data are a strain at which the tension
stiffening effect drops to zero of "ts = 0.0036, a shear retention factor S.R. = 0.1, and a strain
at which the shear modulus goes to zero of ) max = 0.011. The measured and computed (by
NLFEAP-2) shear stress versus shear strain response for PV20 are compared in Figure 7.25.
v

Loading: Monotonic pure shear


Specimen Dimension = 890 mm x 890 mm x 70 mm
fc =19.6 MPa,

)c = 0.00180

v Maximum Aggregate Size = 6 mm

Yonung's Modulus of the steel, Es = 200,000 MPa for all reinforcement


*x = 1.79 %, (f6.35 mm @ 50 mm on each face), Clear Cover = 6 mm, fxy = 460 MPa
*y = 0.89 %, (f4.47 mm @ 50 mm on each face), Clear Cover = 12.35 mm, fyy = 297 MPa
Observations:

Plan View

vcr

= 2.21 MPa

vu

= 4.26 MPa

Shear failure of concrete prior to yielding of longitudinal steel


Cross Section

Figure 7.24: Description of Panel Element PV20.

Shear stress(MPa)

3
Experiment

T-S=0.0036, S-R=0.1, 0.011

0
0

Shear strain(x10^3)
Figure 7.25: Calibration of NLFEAP-2 using results from PV20.

254

7 Benchmark tests and validation procedures

Since the required number and diversity of experiments for model calibration increases as the
number of parametric values to set increase, it is desired to fix as many of the parameter
values as possible using available material test data and from other experience. In some cases,
it may be desired to calibrate one computational model, such as a comprehensive commercial
program, with the predictions of more specialized program.
One of the noticeable challenges to the use of commercial packages is that the softening
behaviour and stiffness degradation in concrete materials can lead to convergence difficulties
when implicit solution routines are utilized. One way of overcoming this difficulty is by using
damaged plasticity models in which a viscoplastic regulation of the constitutive relationships
is employed, but this topic is not further explored in this report.
7.6.4

Analysis of deep beam

The goal of this analysis is to compare the behaviour of the deep beam described in
Figure 7.26 by the two NLFEA programs. This side-by-side comparison is a useful exercise
for the practitioner in choosing between analysis programs. In this evaluation, a comparison is
made of the ability of the programs to predict the capacity of the beam (P), the deformation at
failure (+), the diagonal compressive strain just above and inside of the support ("d), the strain
in the longitudinal reinforcement just inside of the support ("ls) and the strain in the stirrups
mid-way between the point of loading and the support ("esv).
P
2-10M bars
10M Stirrups

300

4 sp.
@ 40

200

1000

300

PL 300x200x25

12-#6 bars

Cross Section

120

P/2

PL 200x200x25

P/2

2800 mm

120

Elevation View
Loading: A single mid-span load
Specimen Dimension = 3040 mm x 1000 mm x 200 mm
fc =33.5 MPa, ) c = 0.0024

Maximum Aggregate Size = 10 mm

*l = 2.01 %, (12 - # 6 Bars + 2 - 10M bars), Clear Cover = 30 mm


*v = 0.50 %, (2 - 10M Bars @ 200 mm), Clear Cover = 15 mm
Observations: vcr = 6.79 MPa (Vcr = 1100 kN)

Bar As
Size (mm )
10M 100
#6 284
2

fy

Es

(MPa)

(MPa)

529
498

197,300
216,800

vu = 11.94 MPa (V
u = 1935 kN)

Crushing failure of the entire end block over the west support followed by a complete loss
in re-bar anchorage and spalling of large pieces of concrete

Figure 7.26 :Description of deep beam (Lee, 1982).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

255

Modelling of the behaviour is made difficult by the complex state of stress and distribution of
cracking that is expected to occur in the structure. Examining the behaviour from a strut-andtie perspective, the spreading of the diagonal compressive strut between the point of loading
and support is expected to produce cracks along this axis and well as to result in a
concentration of compressive stresses near the points of loading and reaction. In addition,
there will be a heavy demand placed on main tension reinforcement at the inside of the
supports. The behaviour of this region may also be viewed from a sectional perspective in
which this is a deep beam subjected to a combination of shear and flexure in which a
combination of flexure, web-shear, and flexure shear cracking is to be expected.
Based on the considerations described above, and in order to contrast the use and capabilities
of two very different types of NLFEA programs, NLFEAP-1 and NLFEAP-2 are both used to
predict the structural response of the deep beam described in Figure 7.26. The specific
comparisons that should and can be made depend on the objective for the use of the NLFEA,
the characteristics of the modelling activity, and the availability of data from the experiment.
In this example, five items will be compared:
(i)

capacity (point load P at midspan)

(ii)

the deformation at peak load (98% of ultimate)

(iii)

the diagonal compressive strain just above and inside the face of the support. This
is interesting as it is expected to be the large compressive strain in the test structure
and an indicator of local crushing that could lead to the failure of the deep beam.

(iv)

the strain in the longitudinal reinforcement just inside of the face of the support.
This is an important measure as the level of straining in this location would be
dramatically different if one is looking at the behaviour from a strut-and-tie
perspective or from a sectional behaviour perspective.

(v)

the strain in the stirrups near mid-height and between the support and the point of
load application.

These five comparisons are presented below.

(i) & (ii) Capacity (P) and deformation at midspan


The relationship between the mid-span loading (P) and the mid-span displacement (/) is
presented in Figure 7.27. There are two different measured responses shown. From the test
data, it appears that the member was cracked prior to the recording of the data. That is, the
uncracked response was not recorded. Based on the cracking load reported in the test report, a
second measured response is shown in which the data points are shifted to the left by an
amount corresponding to the additional displacement with is estimated to have occurred in the
pre-cracked member.
The predicted versus measured/calculated capacity and deformation at peak load is
summarized in Table 7.3.

256

7 Benchmark tests and validation procedures

2500

Point load (kN)

2000

1500

1000
Experiment(Original)
Experiment(accounting for precrack)

500

NLFEAP-1
NLFEAP-2

0
0

10

12

14

Midspan displacement (mm)


Figure 7.27: Load-deformation response of deep beam.
Table 7.3: Comparison of Results for Deep Beam
Procedure
TEST Result (Original)
TEST Result (Adjusted)
NLFEAP-1
NLFEAP-2
ACI318-99
EuroCode 2

Capacity P (kN)
1935
1935
1950
2120
695
1062

Deformation at Midspan at Failure (mm)


12
10.5
8.4
8.2
N.A.
N.A.

It is useful to note that the prediction of capacity by the three NLFEA methods ranged from
only 1950 to 2120 kN (8%). Program NLFEAP-1provided the best estimate of the capacity of
the deep beam to within 1% of the measured strength. NLFEAP-2 overestimated the strength
by 10%. By contrast, the strength of the member calculated by code relationship were only
35% and 55% of the measured strength. Therefore, it can be concluded that the NLFEA
programs provided a more accurate, albeit less conservative, estimate of the capacity.
The stiffness of the load-deformation response predicted by the analysis tools were quite
similar - differing by only a few percent. However, they both overestimated the stiffness.

(iii) Diagonal compressive strain


Figure 7.28 compares the predicted versus measured diagonal compressive strain just above
and inside the face of the support.
All analysis programs provided a similar and accurate prediction of the compressive straining
up until the cracking of the test specimen in the region of the measured diagonal strain.
NLFEAP-2 did not effectively predict the influence of cracking on the concentration of
straining at this location and in this direction. NLFEAP-1was better able to predict the
concentration of straining along this direction with the total strain predicted by NLFEAP-1
providing a reasonable estimate of the measured behaviour.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

257

2500

Point Load(KN)

2000

1500

Experiment
NLFEAP-1(Concrete net strain)
NLFEAP-1(Total strain)

1000

NLFEAP-2

500

P/2

0
0

P/2

0.5

1.5

Compressive strain (millistrain)


Figure 7.28: Compressive strain above left side support.

(iv) Strain in reinforcement at inside face of support


It is interesting to examine the strain in the reinforcement at the inside face of the support as
this would be poorly predicted by engineering beam theory. A comparison between the
predicted and measured straining at this location is presented in Figure 7.29.
Program NLFEAP-1 provides two values for the strain in the reinforcement, one which
represents an average value and assumes perfect bond between the reinforcement and the
concrete in which the concrete is supporting an average tensile stress determined by the
tension stiffening relationship using the MCFT. The other is the strain at a crack location,
which is calculated by assuming that there is no tensile stress in the concrete at the location
that the strain is being calculated. Since it is not known if the strain gage on the reinforcement
reported in this figure is at a crack location, it is to be expected that the measured strain is
somewhere between that predicted at a crack and on average. Both analyses provided a
reasonable estimate of the straining in the reinforcement at this location.

(v) Strain at location of shear reinforcement in left side of deep beam


The predicted and measured straining at the location of a stirrup in the deep beam is compared
in Figure 7.30.
The strain in the shear reinforcement was not directly measured in the test, but the surface
strain over a 200 mm gauge on the surface of the concrete, immediately above a stirrup, at the
location shown in Figure 7.30 was measured at each loading stage. NLFEAP-1was found to
provide a reasonable estimate of the strain at this location while NLFEAP-2 underestimated
the strain.

258

7 Benchmark tests and validation procedures

2500

Point load (KN)

2000
Experiment(HS2)

1500

NLFEAP-1(average)
NLFEAP-1(at crack)
NLFEAP-2

1000

Longitudinal Strain

500
P/2

P/2

0
0.0

0.5

1.0

1.5

2.0

Tensile Strain (millistrain)


Figure 7.29: Strain in reinforcement at inside face of left support

2500

Point load (KN)

2000

1500
P

Experiment(S4)

1000

NLFEAP-1(average)
NLFEAP-1(at crack)
NLFEAP-2

500
P/2

P/2

0
0

10

15

20

Tensile Strain (millistrain)


Figure 7.30: Strain at stirrup location on left side of deep beam.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

259

The observations from this validation activity are as follows:


(i)

NLFEAP-1 was able to accurately predict the strength of the deep beam and
provide a reasonably accurate estimate of the overall deformation at failure even
when the default values were used for parameters. The program was also able to
predict well the compressive straining and reinforcement stresses at the locations
examined in the example.

(ii)

NLFEAP-2 provided a reasonable estimate of overall load-deformation response


up until peak loading. Prior to using this program to predict the response of
structures with similar types of behaviour, it would be desirable to compare the
prediction of the model used in this comparison (with the same parameter values)
with other tests. This is particularly important in programs where predictions are
sensitive to defined parameter values and selected convergence criteria, as was
found to be the case for NLFEAP-2.

Since a large number of mechanisms of resistance control the behaviour/strength of structural


concrete, it is important to validate NLFEA programs for the type of behaviour anticipated in
the structure being designed or analyzed. This is best achieved by the development of a
database of personally conducted and well-documented model calibration and validation
efforts on a suite of NFLEA programs. Over time, the user will develop a measure of the
strengths and limitations of each NLFEA program and develop appropriate global safety
factors for each program.

7.7

Summary and future trends in model validation

7.7.1

Summary

The first step in the use of a computational program for the design and analysis of concrete
structures is a thorough evaluation of the capabilities and limitations of the selected program.
This requires a review of the underlying principles and behavioural models upon which the
program is based. To this end, the analyst should examine the user options, including what
constitutive relationships and failure theories are available, and make an assessment of the
intended applications of the program and identify what parameter values are to be set by the
user and the guidance provided in manuals for their selection. If the program is deemed to be
potentially suitable for the analysis of the structure under examination, then the next step is to
evaluate the accuracy and reliability of the program by using it to predict the response of
structures that are expected to exhibit similar types of behaviour to that being designed and
for which well documented experimental test data is available. This will provide the user with
a quantitative assessment of the programs ability to predict overall strength, stiffness,
ductility and more specific information such cracking characteristics and reinforcement
strains.
While there is a spectrum of computational programs that provide a wide variety of features
and capabilities, it is useful to characterize them as being of two types. One type of program is
specifically developed for the analysis of certain forms of structural concrete in which the
authors of the program have utilized their expertise to set default parameter values. To use
these tools, the user should follow the guidance provided in the program manuals and need
only input basic material property and geometric data. At the other end of the spectrum is a
general-purpose analysis tool in which the user can greatly influence the predicted response
by selected parameter values. These programs require a more thorough evaluation, which
should involve a sensitivity study to evaluate the effect of selected values on predicted
260

7 Benchmark tests and validation procedures

response. In this chapter, the steps used in the setting parameter values and in evaluating the
accuracy of a computational program were referred to as calibration and validation as briefly
summarized below.
At the material level, the mechanical characteristics of materials are specified based on the
results of standard compressive, tensile, and fracture tests. At the element level, calibration
and validation can be done for specific behaviour such as tension stiffening, compression
softening, crack spacing, as well as stiffness and strength degradation with cyclic loading. For
this, a systematically arranged set of benchmark tests should be utilized that covers the range
in behaviours that are expected to be exhibited by the structure under investigation. From
these comparisons and for the purpose of making capacity predictions, the accuracy of the
computational tool can be assessed as the ratio of the experimental measured strength to the
predicted strength. This data can then be used to evaluate the mean test/predicted strength
ratio and the coefficient of variation from which a global safety factor can be calculated to
obtain the level of safely desired from the design. This is equivalent to what is done in the
development of strength relationships in codes of practice.
This calibration and validation procedure was illustrated in two separate examples; for the
design of a shear wall and the analysis of a deep beam in which a different type of
computational program was used in each example. NLFEAP-1 was a concrete-specific tool in
which the user may rely upon the element level calibration predefined by the developers
whereas NLFEAP-2 utilized a general type of material model that required an evaluation of
the sensitivity of the results to non-material parameter values and calibration at the element
level. The shear wall example was used to illustrate the use of computation tools for strength
evaluation while the deep beam example was also used to illustrate how computational
program can be used to also predict deformations and local strains.

7.8

Future trends in model validation

Model validation is the process of evaluating the suitability and accuracy of a particular
computational program for a specific design and analysis task. This is done through a
comparison of predicted and experimentally measured behaviour. Major challenges in model
validation are a lack of suitable and well-documented test data, uncertainty as to what
quantities to compare other than overall strength and deformation, and how to weigh the
significance of a programs effectiveness for each of these comparisons. Many efforts are
underway that will lead to improved and more reliable validation procedures, some of which
are described below.
In 2001, the American Society of Mechanical Engineering (ASME) established a committee
on Verification and Validation in Computational Solid Mechanics (VNVCSM). The charter of
this group is to develop standards for assessing the correctness and credibility of modelling
and simulation in computational solid mechanics. The membership includes individuals from
US National Laboratories, professional practice, and from academia, including civil
engineers. This group is focused on establishing a formal architecture for the process of model
validation and for the development and use of benchmark tests. This group uses the
interrelationship between three entities to describe the issues and methods for VNVCSM, as
described in Figure 7.31. The entities consist of the real world problem, the conceptual model,
and the computer model. In the field of structural engineering, this is the structure, a particular
theory such as, for example, the modified compression field theory, and a specific non-linear
finite element program such as NLFEAP-1 which was used in an example presented in this
chapter. The relationships between the entities define questions: is the conceptual model

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

261

appropriate to use for the structure under analysis, is the conceptual model properly
implemented in the computer model (verification) and can the computer model be expected to
provide accurate predictions of the behaviour of the real work problem (validation)? At the
core of the process is the data required for making these assessments.

Conceptual Model
Validation

Reality of Interest

Abstraction and Modeling


Validation or
Prediction

Conceptual Model

Simulation
Software Implementation

Verification

Computational Model

Figure 7.31: Verification and validation in CSM Figure 1 (after ASME, 2006).

The data for validation consists of experimental test results, such as experiments on beams,
columns, walls, and joints. There are several shortcomings to test data including its
unavailability, lack of basic material data and geometric details, breadth of tests, and density
of measurements. In addition, only a brief summary of experimental test results are made
available in technical journals. While more detailed information is usually presented in a
longer report, typically a student thesis, this document is often more difficult to obtain, and
often not all of the collected information provided. Another shortcoming in that basic material
test data is not always collected and all aspects of experiments and experimental test set-ups
may not be well described. Examples of missing basic information include compressive
stress-strain relationships for concrete, fracture strengths, cover between longitudinal
reinforcement and side walls, support conditions, and location of measurement devices.
Another concern with laboratory test data is that what researchers test in laboratories does not
represent the size and complexity of structures that are built in the field; very similar tests are
often repeated in many institutions at the expense of conducting the types of experiments that
are more greatly needed. This is often due to a lack of awareness of what has previously been
conducted and the tendency to address similar issues to what has been studied by others. A
final shortcoming is that few measurements are made of the detailed internal deformation and
straining within structures such that for the majority of tests little more than overall loaddeformation response is available and this is insufficient for the types of validations that are
necessary for assessing the accuracy of comprehensive models.
Fortunately, progress is being made in many ways to improve experimental research
practices. This includes the development of national and international databases of test results
that are managed by design code and other committees. This provides the opportunity for
262

7 Benchmark tests and validation procedures

serious scrutiny of previous test data, the evaluation of design code requirements, and the
identification of areas of greatest research need. These databases are also available for the
validation of computer-based numerical models, but are typically only useful for comparison
predicted and measured strengths. There are however some database efforts that are more
sophisticated in which all collected data from an experiment is archived in a searchable digital
repository. One of the these databases is being developed as part of the US Network for
Earthquake Engineering Simulation in which there is the requirement that the information
collected from all funded projects be ingested into this digital archive with appropriate
metadata in accordance to an predefined model such that all content is searchable. The
development of similar archives and data-models are under development by other nations and
there are international efforts to coordinate these activities.
One of the most critically important advances has been in measurement systems. Coordinate
measurement machines, digital photoelastic methods, photogrammetric methods, laser
scanners, other optical methods, and image analysis methods are enabling researchers to
collect an unprecedented density and accuracy of test data. One of the challenges has been to
develop effective data processing and visualization methods that can deal with the difficulties
of data from multiple sources that provide different densities, accuracies and, sometimes,
conflicting information. Efforts to develop the necessary post-processing tools are also
underway. As these databases become populated, the information will be available for the use
of more comprehensive validation procedures.
A further and very relevant development is that design code committees are now discussing
how to facilitate the use of computational programs in structural design and analysis. As
discussed in Section 7.3, the basis for design code provisions is experimental test data in
which design code relationships are selected so that there is a specific low probability that the
design strength will be less than the factored load. Through the establishment of consensus
databases of benchmark tests, code committees can directly facilitate the validation and use of
computer-based numerical models.
As has been described in this chapter, validation of computational tools is essential for their
safe and effective use in practice. While examples for assessing global safety factors for
different types of computational tools were provided, efforts have only recently been started
to develop more formal validation procedures, test databases, and means of supporting the use
of computations tools in practice.

7.9

References

ACI Committee 318-02. (2002), Building Code Requirements for Structural Concrete,
American Concrete Institute, Detroit, 443 pp.
Collins, M.P., Vecchio, F.J., and Mehlhorn, G. (1985), An International Competition to
Predict the Response of Reinforced Concrete Panels, Canadian Journal of Civil Engineering,
12, No.3, pp. 624-644.
Lee, D.K., (1982) "An Experimental Investigation in the Effects of Detailing on the Shear
Behavior of Deep Beams ", Master Thesis, Toronto, Ontario
Okamura, H., and Maekawa, K. (1991), Nonlinear Analysis and Constitutive Models of
Reinforced Concrete, University of Toyko, ISBN 7655-1506-0, pp 182.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

263

Paulay, T., and Priestley, M.J.N. (1992), Seismic design of reinforced concrete and masonry
buildings, New York: Wiley.
Shioya, T. (1989), Shear Properties of large reinforced concrete member, Special Report of
Institute of Technology, Shimizu Corporation, No. 25, 198 pp.
Taylor, C.P., Cote, P.A., and Wallace, J.W., (1998), Design of Slender Reinforced Concrete
Walls with Openings, ACI Structural Journal, Vol. 95 No.4, pp. 420-433.
Vecchio, F.J. (2000), Disturbed Stress Field Model for Reinforced Concrete: Formulation,
ASCE Journal of Structural Engineering, 126, No. 9, pp. 1070-1077.
Vecchio, F.J., and Collins, M.P. (1986), The Modified Compression Field Theory for
Reinforced Concrete Elements Subjected to Shear, Journal of the American Concrete
Institute, 83, No.2, pp. 219-231.
Vecchio, F.J., & Collins, M.P. (1982), The Response of Reinforced Concrete to in-plane
Shear and Normal Stresses, Publication No. 82-03, Department of Civil Engineering,
University of Toronto, Canada, March, 332 pp.
ASME (2006), American Society of Mechanical Engineers PTC 60 Committee, Guide for
Verification and Validation in Computational Solid Mechanics, American Society of
Mechanical Engineers, New York.

264

7 Benchmark tests and validation procedures

Strut-and-tie modelling

8.1

Introduction

Since being promoted by Marti (1985) and Schlaich et al. (1987), strut-and-tie modelling has
been increasingly accepted as a powerful tool for the analysis and design of reinforced and
prestressed concrete structures. It has been used simply as a tool to identify and describe the
flow of forces in a cracked concrete continuum as well as a quantitative tool for performing a
complete design, that is, to evaluate force distributions and to proportion and detail structural
concrete members. It has served as a basis in full-member design procedures, which is the
focus of this chapter, and is referred herein as the strut-and-tie method (STM).
Although the STM can be applied to all parts of a structure, it is usually applied to structural
concrete regions near statical or geometrical discontinuities, commonly referred to as D(disturbed or discontinuity) regions. This is because the STM provides a more rational and
consistent design procedure than empirical approaches historically applied to those regions. It
was originally developed to replace these empirical rules, and at the same time was introduced
as a method that is simpler than other procedures for D-regions, such as those based on finite
element analysis. The other reason for using the STM in D-regions is that the other portions of
a structure, commonly referred to as B- (Bernoulli or beam) regions, have been traditionally
designed using sectional design procedures which are based on well-established principles,
for example, the Bernoulli beam theory for flexural design and the truss analogy method for
shear design.
Provisions for designing using the STM have been incorporated into major international codes
for structural concrete. These include, among others, the Canadian Building Code (1984),
CEB-FIP Model Code 1990 (1993), and AASHTO LRFD Bridge Design Specifications
(1994). Most recently, STM design provisions have been included as an alternative design
procedure in the 2002 edition of the ACI318 Building Code Requirements for Structural
Concrete (2002). Guidelines for using the STM are also available in the FIP report entitled
Practical Design of Structural Concrete (1999).
The design concept in the STM is simple but the implementation can be time consuming. As
described later in Section 8.4, the design process is iterative and involves graphical
representations of strut-and-tie models. A few special purpose computer applications
implementing the STM have been developed to address the time consuming aspect of the
calculations and therefore enable engineers to focus on proportioning and detailing the
concrete members. However, most of the programs were developed in universities and have
been primarily used for research purposes; only a few of them are currently available for
practitioners. It is anticipated that additional computer-based STM tools will be available for
the practicing community as the method is becoming well established.
Another critical drawback of the STM is that the method focuses on design for only strength;
serviceability requirements, such as deflections and crack widths, are not explicitly accounted
for in the procedure. Extension of the STM that considers serviceability and predict the
behaviour for the entire loading range would be an added value to the STM, and therefore the
computer-based STM, as the same model used in proportioning and detailing may also be
used in response evaluation.
This chapter begins with an overview of the STM (Section 8.3), which includes brief
discussions on strut-and-tie models, steps in STM design and complications in using the STM.
Additional considerations in the use of the STM are then discussed in Section 8.5. Section 8.6
presents an overview of available computer-based STM tools and Section 8.7 describes some
aspects in strut-and-tie modelling using the tools. The latter section is also intended to provide
!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

265

some insight on various procedures adopted in the available computer-based STM to obtain
valid solutions. The discussion is limited to planar structures with uniform thickness in the
out-of-plane dimension. A design example using a computer-based STM is provided to
complete the discussion (Section 8.8).

8.2

Notation

Ac

effective cross-sectional area of at one end of a strut

As

area of non-prestressed reinforcement

A ps

area of prestressed reinforcement

b1 , b2 , ...

effective flange width of a structural member

C , C1 , C 2 , ...

compression force in a strut-and-tie model

Eco

initial tangent modulus of elasticity of concrete

Es

modulus of elasticity of steel

f c , f c1, fc 2 , ... normal stress in the concrete


fc#

specified compressive strength of concrete

fcu

uniaxial compressive stress limit of a strut

ft#

specified tensile strength of concrete

fy

specified yield strength of non-prestressed reinforcement

f py

specified yield strength of prestressed reinforcement

F , F1 , F2 , ...

forces in strut-and-tie model components (general notation)

Fcu

capacity of a strut

Ftu

capacity of a tie

h, h1 , h2 , ...

depth of a structural member

T , T1, T2 , ...

length of the struts or ties


thickness (out-of-plane dimension) of D-region
tension force in a strut-and-tie model

vc

shear stress in concrete

w, w1 , w2 , ...

effective width

_f p

remaining stress in non-prestressed reinforcement to resist tension

effectiveness factor
angle between a strut and a tie

-s
+1 , + 2

principal (maximum and minimum) stresses (general notation); + 1 is always


algebraically greater than + 2

+ x, c , + y ,c

normal stresses carried by concrete and defined in the x-y coordinate system

, xy, c , , yx, c

shear stresses carried by concrete and defined in the x-y coordinate system

266

8 Strut-and-tie modelling

8.3

Overview of the STM

8.3.1

Strut-and-tie models

The STM is based on the lower-bound theory of limit analysis (for example, Nielsen 1971, Chen
and Han 1988, Muttoni et al. 1997). In the STM, an internal truss is envisioned to carry the
applied loading through the D-region being considered to its supports or boundaries. This truss is
termed a strut-and-tie model and is a statically admissible stress field in lower bound solutions.
Examples of strut-and-tie models for a few typical D-regions are shown in Figure 8.1.

(b)
(a)

(c)

(d)

(f)
(e)
Strut

Figure 8.1:

Tie

Node

Examples of strut-and-tie models for common D-regions: a) beam-column joint; b) coupling


beam; c) non-slender wall; d) dapped-beam end; e) pile cap under combined axial load and
moment; and f) prestressed concrete beam end.

A strut-and-tie model consists of struts, ties, and nodes, where struts are the compression
members, ties are the tension members and nodes are the meeting points of struts and ties. In
this chapter, struts are represented using broken lines and ties by solid lines.
8.3.2

Components of strut-and-tie models

A strut represents a concrete compressive stress field where the principal stresses are
predominantly along the centreline of the strut. The shape of strut stress fields in planar Dregions may be prismatic, bottle-shaped, or fan-shaped, as shown in Figure 8.2 (Schlaich et al.
1987).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

267

!
fc1<fc
fc

fc

Figure 8.2:

fc

fc

fc

(a)
(b)
(c)
Idealized stress fields in struts: a) prismatic; b) bottle-shaped; and c) Fan-shaped (after Schlaich
et al., 1987).

In general, the stress limit of a strut is not the same as the uniaxial concrete compressive
strength obtained from cylinder tests, f c# . It is defined as f cu ' n f c# , where: fcu = stress limit
of a strut, commonly referred to effective strength, and n = effectiveness factor less than or
equal to 1.0 (refer to Section 8.5.3). The critical extent of a strut stress field perpendicular to the
strut centreline in the plane of the structure under consideration, usually at the ends of the strut,
is termed the effective width, w. The extent of the stress field in the out-of-plane dimension is
the same as the thickness of the structure, t. If uniformly distributed stress field across the
effective area is assumed (see Section 8.3.3), the capacity of struts is simply Fcu = Acfcu, where
Ac is the effective cross-sectional area of the strut and is given as Ac = wt.
Ties typically represent one or multiple layers of reinforcing steel or prestressing steel or, on
occasions, concrete tensile stress fields. The stress in the ties is limited by the yield strength of
ordinary reinforcing steel, fy, or prestressing steel, fpy. The effective prestressing force and its
effects may be introduced as a set of statically equivalent external forces acting on the D-region
under consideration (Schlaich et al. 1987). When this approach is applied to bonded prestressing,
the capacity of a tie consisting of non-prestressed and prestressed reinforcement is given as
Ftu = fy As + _fp Aps, where As = area of reinforcing steel, Aps = area of bonded prestressing steel,
and _fp = remaining prestressing steel stress available to resist tension.
Nodes (nodal zones or nodal regions) represent regions in which forces are transferred
between struts and ties. Depending on the types of forces being connected, there are four basic
types of nodes; namely CCC, CCT, CTT, and TTT, as illustrated in Figure 8.3, where C and T
denote compression and tension forces, respectively. In 2D problems, the stresses in the nodal
zones are biaxial and are limited to the yield criteria for plane stress problems. The stress
distribution in the nodal regions depends on the idealized shapes that, in turn, depend on the
effective width and direction of the strut or tie stress fields framing into the nodes. The
simplest idealized shape of a nodal zone is that formed by the intersection of actual
dimensions of struts and ties whose centrelines coincide at that node. An example of this
nodal zone construction is shown in Figure 8.4. Variants of constructing the nodal zone
shapes have been developed to simplify the state of biaxial stresses in the nodal regions.
These include hydrostatic approach (Marti, 1985, Nielsen, 1999) or modified hydrostatic
approach (Schlaich and Anagnostou, 1990). Further discussion about these last two
approaches is presented in Section 8.7.6.

268

8 Strut-and-tie modelling

C
C

T
C

(a)

(b)

(c)

(d)

Figure 8.3: Basic types of nodes: a) CCC; b) CCT; c) CTT; and d) TTT.

fc2

!
fc3
C3

t
ru
St

C1

fc3

fc4

fc1

fc1

fc1

fc4 fc1
PStrut 1
1

fc2 fc3 fc5

3
PStrut 3

fc2

fc4

C
ut
Str

C2

PStrut 2

fc5

PNode

fc5

Strut 1

fc2

,
fc3

(a)

(b)

Figure 8.4:

8.3.3

(c)

Example of a node with three struts meeting: a) force acting on the node; b) simple nodal region
shape; and c) Mohrs circle describing biaxial stresses in the nodal zone.

Admissible strut-and-tie models

As a statically admissible stress field in the lower bound method, a strut-and-tie model has to
be in equilibrium externally with the applied loading and reactions (the boundary forces) and
internally at each node. Reinforcing or prestressing steel is selected to serve as the ties, the
width of each strut is selected, and the shape of each node is constructed, such that the yield
criteria for all components are not violated.
Lower-bound solutions of limit analysis permit various stress distributions across the effective
width of stress fields. The simplest and most commonly selected distribution is one that is
uniformly distributed. An example of this distribution is illustrated in Figure 8.5a. With this
selection, a stress discontinuity exists along the boundary of a stress field. This boundary is
commonly referred to as a line of stress discontinuity. As can be observed from Figure 8.5b,
the stress in Strut 3 of Figure 8.5a changes abruptly from a stress of fc3 to zero and vice versa.
Nevertheless, equilibrium at any point along the lines of discontinuity is still always satisfied
(Figure 8.5c).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

269

fc1

w1

Stress-Free
Region

fc2

fc1

(1)

Boundary of
the Structure

(4) Strut 2

(2)
t3
ru
St

Line of Stress
Discontinuity

w2

fc3

fc2

(5)

Stress-Free
Region

w3

Lines of Stress
Discontinuity

fc3

fc3

Stress-Free
Region

Stress-Free
Region

(3)

Stress Region
of Strut 3

Line of Stress
Discontinuity

Section A-A

w3

(a)

fc3

fc3

(b)
fc4

fc5

vc
vc

vc
vc

(1)

fc2
fc4

fc5

fc1

fc3

fc2

fc2

fc2

fc3
fc1

(2)

(3)

(4)

(5)

(c)

Figure 8.5:

8.4

Examples of stress discontinuities in a typical D-region: a) free-body diagram showing the stress
fields underneath the applied loading; b) stress distribution across the effective width of Strut 3
(along Section A-A); and c) equilibrium state of stress in the selected infinitesimal regions near the
stress discontinuity lines.

STM design steps

Design using the STM for a D-region may be summarized into the five steps illustrated in
Figure 8.6. These are:

270

"

Define the D-region to be considered and then evaluate the boundary and body
forces. Boundary forces include the concentrated and distributed forces acting on the
D-region boundaries. Boundary forces can also come from sectional forces (moment,
shear, and axial load) at the interface of D- and B-regions. Body forces include those
that result from D-region self-weight or the reaction forces of any members framing
into the D-region.

"

Sketch a strut-and-tie model and solve for the truss member forces.

"

Select the reinforcing steel that is necessary to provide the required tie capacity and
ensure that this reinforcement is properly anchored.

"

Evaluate the dimensions of the struts and nodes, i.e., select the effective width of the
struts and construct the shapes of the nodal regions, such that the capacity of these
components is sufficient to carry the design forces.

"

Ensure sufficient ductility capacity in the D-region by providing distributed


reinforcement.

8 Strut-and-tie modelling

STEP 2
383 kN

kN

454 kN

356 kN

kN

356 kN

596 kN
356 kN

309 kN-m

44
2

407 kN

71 kN

74
7k

kN

48
9

2
32

kN

121 kN

50
3

!STEP 1

71 kN

810 kN

h
STEP 5

m
m

50

m
m

75
m

8 #16

4 #19
12
5

mm

2 #13 bars
two-legged
closed stirrups

8 #13

75

37.5 mm

STEP 3 & 4

4 #16 two-legged,
4 #16 two-legged
closed stirrups
stirrups @ 100 mm
@ 50 mm

Framing bars 2 #19


2 #13 U bars
4 #19 bars
welded to angle
2 #19 U bars
3 #25 bars

3 #25

Figure 8.6: Steps in STM design process (after Tjhin and Kuchma 2002).

8.4.1

Complications in STM design

Several complications can arise in executing this five step design process. The selection and
sketching of the initial strut-and-tie model is sometimes a challenge, particularly when dealing
with complex D-regions. The selected strut-and-tie model often needs to be adjusted and
refined graphically in order to satisfy stress limit criteria in the strut-and-tie model
components, to fit the truss dimensions within the boundaries, to capture secondary loadresisting elements, or to optimize the design. Furthermore, truss dimensions, i.e., the widths of
struts, the shapes of nodal regions and the area and configuration of steel ties, often need to be
iteratively adjusted for similar reasons. Further, complications in the design using the STM
may arise from the need to solve for truss member forces in statically indeterminate strut-andtie models. In many designs, multiple load cases and load combinations have to be
considered. The latter may increase the time required to complete a design using the STM by
a considerable factor because different strut-and-tie models have to be prepared to handle
each different loading situation and load cases usually cannot be superimposed directly to
form load combinations as a result of strain compatibility requirements. All these problems
can make hand-based solutions prohibitively time consuming. Computer-based strut-and-tie
design programs, as described in Section 8.6, can overcome many of these challenges.

8.5

Some considerations in using the STM

8.5.1

Rules in defining D-regions

Saint-Venants principle is usually used to determine the extent of the D-region under
consideration. Based on this principle, the extent of a D-region is taken within a distance
equal to the largest of the depth or width of the member from the discontinuity or the size of
the disturbance. If two D-regions determined in this manner touch or overlap, they are
considered one D-region. Once the D-region is determined, the internal forces at the interface
between the B- and D-regions become boundary forces for that D-region.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

271

Examples of division of B-regions and D-regions are shown in Figure 8.7. In the figure, the
unshaded area with notation B indicates a B-Region and the shaded area with notation D
indicates a D-Region. The notations h1 , h2 , ... and b1, b2 ,... are used to denote the depths and
effective flange widths of structural members, respectively.

h1

D B

h2

h2 h2

h2

h3
h4

h2

B
D

h1

h1

h3
h3

h1

D
h4 h4

B
h4

D
D
h5

h3

B
h3

h1

h5

h5

(a)
b1

b1 > h1

b2

h1
b1

D
D

B
h1

h2

Section 2-2

b1

h4

Section 1-1

b1

h2

D
D

B
b2

h3

h3
h2

b2 > h4

D
h4

b2

B
b2

D
b2

B
h3

D
D

(b)

Figure 8.7:

8.5.2

Examples of division of B- and D-regions: a) building structure b) bridge structure (after Tjhin
and Kuchma, 2002, and Kuchma and Tjhin, 2002).

Two- and three-dimensional D-regions

D-region problems may also be classified as either 2D or 3D problems. The associated strutand-tie models are also 2D or 3D. Examples of 2D D-regions with their strut-and-tie models
have been given in Figure 8.1. Examples of 3-D D-regions are pile caps with more than one
row of piles in each direction and slabs nearing the vicinity of columns.
272

8 Strut-and-tie modelling

Some 3D D-regions may be conveniently separated into several 2-D regions. In the case of
the simply supported T-section under a vertical load as in Figure 8.8, for example, the
structure can be decomposed into two 2D regions, one associated with the web and the other
associated with the flange. The strut-and-tie model for each region is also shown in the figure.
In this case, equilibrium at the interface between the web and flange has to be maintained, that
is, the top struts of the web become boundary forces for the strut-and-tie model in the flange
region and vice versa.

(a)

(a)

(b)

(b)

(c)

Figure 8.8:

8.5.3

Example of separating a 3-D D-region into multiple 2-D regions: a) 3D view of the beam; b)
strut-and-tie model in the web region; and c) strut-and-tie model in the flange region (after
MacGregor, 1997).

Capacity of struts

Limit analysis was developed based on a rigid perfectly plastic material. This is a reasonable
assumption for reinforcing steel in tension but it is a significant simplification of the typical
uniaxial stress-strain relationships of concrete. Under increasingly uniaxial compressive
stress, concrete softens and then weakens after reaching its post peak stress. In addition, its
deformation capacity is usually limited. To account for this behaviour the STM, a parameter
called the effectiveness factor, n , is introduced in the determination of the capacity of struts.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

273

The effectiveness factor, n , is also used to account for strength reductions caused by the
shape of strut stress fields and disturbances due to cracks and tensile strains along the path of
a strut. It is also used to consider strength enhancements as a result of the use of distributed
reinforcement and confinement surrounding struts (see Figure 8.9). Strength degradation in
struts may also arise in the case of reversed cyclic loading. This degradation may occur, for
example, when two strut-and-tie models associated with different loading arrangements share
a common region (Figure 8.10) or when forces change direction from compression to tension
or visa-versa.

(f)

!
(d)

(e)

Figure 8.9:

(c)

(a)

(b)

Types of struts in a typical D-region: a) prismatic strut in an uncracked field; b) prismatic strut in
a cracked field where struts are parallel to cracks; c) prismatic strut in a cracked field where
struts are not parallel to cracks;d) bottle-shaped strut with bursting control reinforcement; e)
bottle-shaped without bursting control reinforcement; and f) confined strut (after Tjhin and
Kuchma, 2002).

Figure 8.10: Strength degradation of struts in the shaded region due to reversed cyclic loading.

Many of the factors influencing the compressive strength of concrete struts have been
identified. However, the variety of factors considered in quantifying the strength of struts has
resulted in different strength values specified in codes and proposed by researchers (e.g., Yun
and Ramirez, 1996, ASCE-ACI Committee 445 on Shear and Torsion, 1998, and Foster and
Malik, 2002).

274

8 Strut-and-tie modelling

8.5.4

Uniqueness of strut-and-tie models

As a strut-and-tie model is a statically admissible stress field, it follows that it is required only
to satisfy equilibrium and the yield criteria. The third requirement in solid mechanics
framework, that is, strain compatibility, does not have to be satisfied and, thus, the method is
a lower bound plastic method.
As a result of these relaxed requirements, there is no unique strut-and-tie model for a given
problem. More than one admissible strut-and-tie model may be developed for each load case
as long as the selected truss is in equilibrium with the boundary forces and the stresses in the
struts, ties, and nodes of the truss are within acceptable limits. In theory, the lower-bound
method should dictate that the capacity obtained from all statically admissible stress fields is
lower than or equal to the collapse load. This conservative prediction of STM designs has
been verified experimentally (e.g., Maxwell and Breen, 2000, Aguilar et al., 2002, Chen et al.,
2002, Mitchell et al., 2002). In practice, however, the capacity may be over-calculated
because of limited ductility in structural concrete. This limits the number of viable solutions
for each design and suggests that feasible strut-and-tie models should be those that best match
strain conditions and require small load redistribution before failure. Because it approximates
the principal stress flow in a structure and is in equilibrium with the boundary forces, a strutand-tie model may be thought as an idealization of the actual flow of forces in the structure.
Based on this consideration, feasible strut-and-tie models can be identified by examining the
linear-elastic load path in the structure.
In Figure 8.11, the importance of selecting appropriate strut-and-tie models in STM is
illustrated. The figure shows nonlinear finite element prediction of load-deformation
responses associated with three STM designs of a short cantilever subjected to a point load at
its tip (Ali and White, 2001). The strut-and-tie model selected for each design is different but
the design load capacity was the same for all. Figure 8.11 demonstrates, however, that a high
variability in the predicted response and load capacity may exist between the designs. Design
1 gives the worst load and deformation capacity because the selected strut-and-tie model is
not compatible with strain conditions; excessive cracking occurs in the top region near the
support after the diagonal tie yields and the structure fails without much load redistribution.
Design 2 shows an improved behaviour in which load redistribution from tie to the diagonal
strut can occur before the structure fails due to diagonal splitting. Design 3 gives the largest
capacity of the three designs as the additional loading is taken by both the diagonal tie and
diagonal strut after yielding of the horizontal tie.
Figure 8.12 illustrates a further example in which one solution is preferable to another. Due to
the point load at the tip of the cantilever portion, the upper part of the beam is likely to
develop horizontal tensile stresses along the beam. Therefore, the model with the upper
horizontal tie (Figure 8.12a) is preferable to that shown in Figure 8.12b. The latter only
effectively resists the tension in the upper region near the middle support.
Further discussion on stress redistribution and ductility demand is presented in Section 8.5.9.
Guidance to select appropriate strut-and-tie models is presented in Section 8.7.1.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

275

!
Load

Design 3

Severe Cracking
Design 2
Severe Cracking
(extending to
the whole section
at failure)

Design 1

Tip Deflection

Figure 8.11: Predicted load-deformation responses of a short cantilever with different strut-and-tie models
having the same computed capacity (after Ali and White, 2001).

(a)

(b)

Figure 8.12: Two statically admissible strut-and-tie models for a cantilevered deep beam under vertical
loading: a) workable truss; b) less favourable truss due to excessive ductility demands.

8.5.5

Strain incompatibility of struts and ties

In constructing a strut-and-tie model, the angle between struts and ties acting on a node, -s,
should be reasonably large to minimize the level of strain incompatibilities that are caused by
strut shortening and tie lengthening in almost the same direction (Figure 8.13a). Similar
situation occurs when a tie crosses a strut at an angle (Figure 8.13b). As the angle decreases,
the transverse tensile strain in the vicinity of the struts increases and this results in a reduction
of the effective strength of the struts, as observed by Vecchio and Collins (1986).

276

8 Strut-and-tie modelling

!
-s

-s

(a)

(b)

Figure 8.13: Examples of strain incompatibility between a strut and a tie: a) near a nodal zone region; and b)
case where the tie crossing the strut.

The importance of the selecting the angle between struts and ties acting on a node has been
recognized and discussed by many researchers (e.g., Rogowsky and MacGregor, 1986,
Muttoni et al., 1997). In general, there are two approaches to handle this problem. Rogowsky
and MacGregor (1983), for example, recommend that the angle be between 25 and 65 degrees
for slender beam design. Vecchio and Collins (1986) include the angle parameter in the
formulation of effective strength of struts.
8.5.6

Tension stiffening in ties

It has been long recognized that concrete surrounding ties in the region between cracks
contributes to carrying tension forces. This is known as the tension stiffening effect. This
effect reduces as the level of tensile strain in the ties increases.
Tension stiffening is normally small when a tie reaches its yield strength and tension stiffening
does not contribute to equilibrium at the crack. Because strut-and-tie models represent the strength
limit state, tension stiffening is neglected in calculating the tie capacity. However, it should be
considered if used for predicting load-deformation behaviour.
8.5.7

Influence of tie anchorages

In the STM, the steel ties have to be properly and effectively anchored beyond the nodal
regions so that tensile forces can be fully developed and forces in the nodes can be fully
transferred. There are still uncertainties in the current STM about the anchorage requirements
and the need to distribute the tie reinforcement throughout the nodal region. Figure 8.14
illustrates how the distribution and anchorage of tie reinforcement clearly influence the ability
to transfer the horizontal component of a diagonal strut to the tie at the end of a simply
supported member.
In addition to end anchorage of bars, other known factors that influence tie anchorage
requirements include bar size and roughness, lateral and vertical spacing between bars, the
angle of the incoming strut, width of bearing plate in nodal zone, amount of confinement,
length of bar and use of fibres.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

277

(a)

(b)

(c)

(d)

Figure 8.14: Examples of end anchorage of ties in a simply-supported deep beam: a) straight bar; b) standard
90-degree hook; c) T-headed bar; and d) bearing plate (after Tjhin and Kuchma, 2002).

8.5.8

Size, geometry, and strength of nodal zones

As discussed above, nodal zones are regions where the forces in a strut-and-tie model are
redirected. Consequently, these regions are subject to multidirectional states of stress. The
anchorage detail discussed in the preceding section is another example that illustrates the
complexity of load transfer in nodal regions.
Many factors may influence the strength of nodal zones. These include the shape of the nodal
zone, the type of truss members (struts or ties), number of intersecting truss members,
distribution of tie reinforcement, confinement and use of fibres, level of transverse straining,
volume and condition of surrounding concrete and anchorage conditions of ties. Some work
in quantifying the effects of these factors has been initiated (e.g., Lee, 1982, Jirsa et al., 1991,
Polla, 1992), but it is still limited. As a result, different strength values are specified in codes
and proposed by researchers, as summarized, for example, by Yun and Ramirez (1996) and
ASCE-ACI Committee 445 on Shear and Torsion (1998).
8.5.9

Load redistribution and ductility requirements

Under lower load levels (before cracking), the distribution of stress throughout a D-region is
well predicted by linear elastic methods. As the loading increases, the concrete cracks,
softens, and may weaken, and the reinforcing steel may yield and strain harden, resulting in
continuous adjustments of stress distribution to maintain equilibrium within the continuum.
This phenomenon is termed load redistribution. The ability of a structure to perform load
redistribution after yielding of reinforcing steel is known as ductility.
As previously discussed, inappropriate selection of strut-and-tie model may require excessive
ductility demand and thus reduce the load capacity. Failure of tie anchorages and
inappropriate estimation of effective strength may also lead to premature failure as excessive
load redistribution occurs to create new equilibrium systems. Therefore, it is necessary that all
structures be detailed to possess sufficient ductility capacity. In this regard, the capacity of the
strut-and-tie models should be governed by the capacity of ties. Also, the struts and nodes
must have sufficient strength to include the over-strength of the tie steel due to strain
hardening.

278

8 Strut-and-tie modelling

8.6

Computer-based STM

At the time of writing this report, most of the work on computer-based STM in Europe has
been done at the Swiss Federal Institute of Technology (ETH) and the University of Stuttgart.
In North America, a series of computer-based STM tools have been developed at Cornell
University, Purdue University and the University of Illinois.
A computer-based STM tool typically consists of one or more main modules listed below.
The modules are linked together and share the same database.
Graphical User Interface (GUI) module
This module serves as the interface between users and computer. It is an essential module in a
computer-based STM tool because the design process in the STM is graphically extensive. In
this module, users are allowed to interactively define the boundary of D-regions, loading and
restraint conditions and strut-and-tie models. It also allows user to interactively modify the
strut-and-tie model and other design parameters.
The GUI module also serves as postprocessor to enable users to visually check the model and
display the analysis and design results. It also acts as an interface with other output units or
other computer applications, e.g., CAD software (Tjhin and Kuchma, 2002, 2007a).
Module for identifying strut-and-tie models
This module is developed to provide guidance in selecting the appropriate strut-and-tie
models. Thus, it can be considered as the pre-processor of a computer-based STM tool.
All guidance available is based on the use of elastic solutions for identifying strut-and-tie
models, a technique suggested by Schlaich et al. (1987). The elastic solutions are given in the
form of principal stress plots, optimal topographies, or automatic generation of simple strutand-tie models. Principal stress trajectories for this purpose have been obtained from linear
elastic finite element analysis (e.g., Rckert, 1991, Mish et al., 1995, Alshegeir and Ramirez,
1992) and nonlinear finite element analysis of plain concrete (e.g., Yun and Ramirez, 1996).
The optimal topographies are obtained from topology optimization of continuum structures
(e.g., Liang et al., 2002). The automatic generation of simple strut-and-tie models is obtained
from statistical results of principal stress trajectories (Harisis and Fardis, 1991) or topology
optimization of trusses (e.g., Kumar, 1978, Biondini et al., 1999, Ali and White, 2001).
Module for obtaining truss member forces and for dimensioning of model components
This module consists of a truss analysis solver coupled with the routine for dimensioning
struts, selecting tie reinforcement and constructing nodal regions.
Truss analysis solvers that have been developed include elastic truss analysis (e.g., Alshegeir
and Ramirez, 1992), nonlinear truss analysis (e.g., Yun, 2000), and combination of lowerbound approaches and plastic truss analysis (Anderheggen and Schlaich, 1990).
Module for predicting load capacity and load deformation responses
This module performs nonlinear analyses of the STM designed structure to obtain load
deformation responses and confirm that the design loads are achievable.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

279

8.7

Modelling aspects using computer-based STM

8.7.1

Identifying strut-and-tie models

As discussed in Section 8.5.4, there is no unique strut-and-tie model for a design but it has to
be selected appropriately and used consistently throughout the design process. In general, a
strut-and-tie model for a loading system can be identified by the following techniques:
Following common and similar models
Strut-and-tie models of various D-regions have been developed by researchers as part of their
research studies. These strut-and-tie models have usually been verified and are available in
the literature. Designers can benefit from the available strut-and-tie models by selecting
models that match the loading conditions of their design and use them with appropriate
adjustments.
In some cases, an appropriate strut-and-tie model may also be taken from one that has been
derived for a different structure but having similar conditions. Figure 8.15 shows an example
in which the strut-and-tie model for the cantilevered deep beam (Figure 8.15a) is used for the
design of a single corbel (Figure 8.15b).

(a)

(b)

Figure 8.15: Example of selecting a strut-and-tie model used in different structure: a) Deep beam shown in
Figure 8.12; and b) Single corbel utilizing the same strut-and-tie model.

Developing load paths


This approach requires some intuition and experience on how the forces flow within the Dregion. The concept that forces tend to take the shortest path to reach the support is utilized. A
curvature in the load path introduces unbalanced equilibrium system and this must be
balanced by introducing forces in the vicinity of the curvature. The load path and the
balancing forces are then replaced by struts, if the forces are compression, and or ties, if the
forces are tension.
Identifying a strut-and-tie model in this way is shown in the example shown in Figure 8.16 of
a deep beam under two point loads C1 and C2. The loads are positioned such that the left and

280

8 Strut-and-tie modelling

right reaction forces are C1 and C2, respectively. As shown in Figure 8.16a, two load paths,
each has two curvatures, are developed: one is associated with the compression force C1 and
the other with compression force C2. The curvature near the support system is balanced with a
tension force T and the curvature near mid-height of the beam is balanced with force C3.
Figure 8.16b shows the strut-and-tie model after replacement of the load paths and balancing
forces by struts and ties.

C1

C 1> C 2

C3

C2

C2

C1

C3

Load Path

Load Path

C1

C2
(a)

C1

C2
(b)

Figure 8.16: Example of identifying a strut-and-tie model through load path: a) selected load paths; and b)
strut-and-tie model (after Schlaich et al. 1987).

Fitting crack patterns


This technique is based on the observation that principal compressive stresses are parallel to
cracks because the principal tensile stresses that cause the cracks are always orthogonal to the
principal compressive stresses (Figure 8.17).

Figure 8.17: Principal stresses of an infinitesimal stress field region passing through a crack.

In using this technique, pictures of crack pattern of experimentally tested or similar D-region
under consideration must be available and the locations of ties must be known. The crack
pattern pictures are used as the backdrop for drawing strut-and-tie models. The struts are
arranged to be parallel between cracked regions. Figure 8.18 shows an example of a strut-andtie model constructed in this way. In the absence of this information, figures of crack pattern

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

281

in the D-region
D
unnder a loading system may
m still bee obtained from
f
visuallizing the deformed
d
configuuration and then
t
identiffying the pootential cracck locationss. The latterr, however, requires
some faamiliarity with
w the behaaviour of loaaded structu
ures.

Figure 8.18: Examplee of identifyingg a strut-and-ttie model by fitting


f
crack paattern Schlaicch and Schfeer, 1991).

Following elasttic solutionss


Originallly proposedd by Schlaiich et al. (19987), this ap
pproach is based
b
on thhe premise that
t
both
serviceaability and ultimate
u
lim
mit states of a D-region will
w be satissfied if its sttrut-and-tie model is
arrangedd accordingg to elastic stress
s
distribbution. An advantage
a
inn this approoach is that only
o
one
model is required to
t satisfy booth limit staates. The dev
viation of thhis arrangem
ment should
d be kept
within 15
1 degrees of
o the elasticc solution, as
a suggested
d by Schlaichh et al. (19887).
To use this technique, the elaastic stress distribution
n needs to be presenteed in the fo
ormat of
stress trrajectories or plots off principal stresses. Liike the crack pattern pictures, th
he stress
trajectorries or plotts of princippal stresses are used as
a a backdroop for draw
wing an app
propriate
strut-annd-tie modell. The strutts are arrannged paralleel with the principal
p
coompressive stresses
and the potential tiee locations are identifieed by the prrincipal tenssile stressess.
In compputer-basedd STM toolls, isotropicc linear finite elementt analysis oof plain con
ncrete is
usually used to obbtain the pllots of prinncipal stressses. Exampples of strutt-and-tie models
m
of
various D-regions are
a providedd in Figuress 8.19 to 8.2
22.
It is woorth nothingg that plots of
o principall stresses fro
om nonlinear finite eleement analy
ysis, also
based on
o plain conncrete, are also
a used foor laying ou
ut strut-and--tie models (Yun and Ramirez,
R
1996). This is useeful when the
t actual capacity
c
off a D-regionn using struut-and-tie model
m
is
requiredd to be betteer assessed. However, the
t accuracy of such ann approach is arguable because
the nonlinear stresss distributioon is affecteed by the reesponse of the
t reinforcing steel ass soon as
the conccrete crackss.

282

8 Strut-and-tie modelling

(a)

(b)

Figure 8.19:

Strut-and-tie models of a deep beam supported by a column and a squat wall: a) load on top;
and b) load on bottom.

Model 1

C
L

Model 2

Figure 8.20: Strut-and-tie model of a multicolumn bridge pier.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

283

Model 1

C
L

Model 2

Figure 8.21: Strut-and-tie model of a prestressed hammerhead bridge pier.

Figure 8.22 - Strut-and-tie model of a three storey coupled shear wall subjected to lateral loading.

284

8 Strut-and-tie modelling

Recently, topology optimization of continuum structure has been introduced to the STM to
identify the appropriate strut-and-tie models. As an example, Liang et al. (2002) proposed this
type of optimization based on assumption that the concrete continuum is linear elastic. The
optimal solution is obtained by minimizing the weight of the structure while maintaining the
elastic strain energy within a prescribed value. A performance index as a function of the strain
energy and structural weight was devised and is used to determine when the optimization
process is terminated. A finite element procedure for plain concrete (plane stress problem) is
used to calculate the elastic strain energy in the structure at each step in the optimization
process. To achieve the optimization objective, a small number of finite elements that have
the lowest strain energy density are removed from the continuum at each step in the
optimization process. An example of this work is presented in Figure 8.23 with the associated
performance index given in Figure 8.24.

2500 mm

2500 mm

2500 mm

500 mm

3750 mm

2250 mm

500 mm

3000 mm
(a)

(b)

(c)

(d)

Figure 8.23: Example of identifying a strut-and-tie model of a bridge pier under vertical loading using topology
optimization scheme: a) geometry and loading; b) topology at the 20th iteration; c) topology at the
40th iteration; and d) optimal topology at the 49th iteration (after Liang et al., 2002).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

285

! Performance
Index

1.5

0.5

0
0

20

40

60

80
Iteration

Figure 8.24: Performance index of the bridge pier example of Figure 8.23!(after Liang et al., 2002).

Automatic generation of strut-and-tie models


The use of principal stress plots for guiding the construction of strut-and-tie models has also
been extended for automatic generation of strut-and-tie models.
Harisis and Fardis (1991) used statistical analysis of principal stress data obtained from linear
finite element analysis to identify locations of struts and ties. The strut-and-tie models
generated using this approach consists of triangles.
Another example of automatic strut-and-tie model generation is the work of Rckert (1991).
Based on the observation that regions bounded by the mesh of principal stress trajectories
represent finite elements subjected to only normal stresses on each side, a strut-and-tie model
can be formed using these finite elements. One example of a strut-and-tie model generated
using stress trajectories is shown in Figure 8.25.
Kumar (1978), Biondini et al. (1999) and Ali and White (2001) used topology optimization of
trusses for automatic generation of strut-and-tie models. In this optimization technique, the
so-called ground truss (or basic truss) is developed for the D-region under consideration. This
ground truss consists of user-defined locations of nodes and truss members which
interconnect the nodes. The truss members in the ground truss represent all possible struts or
ties. The optimal solution in this work is said to give the most appropriate strut-and-tie model.
In the work by Kumar (1978), the optimal solution is obtained from minimizing the elastic
strain energy of the ground truss. Biondini et al. (1999) use similar approach, except that
optimum criterion is improved by introducing weighted functions to account for different
deformability characteristics in the struts and ties and to account for deviation in the
orientation of struts and ties from the corresponding principal stresses. In the work by Ali and

286

8 Strut-and-tie modelling

!
Strip
Stress
Trajectories

(a)

(b)

Figure 8.25: Automatic generation of the strut-and-tie model of a deep beam under uniformly distributed load:
a) stress trajectories; b) strut-and-tie model (after Rckert 1991).

Figure 8.26:

(a)

(b)

(c)

(d)

Example of automatic generation of a strut-and-tie model of a two-span deep beam: a)


ground truss; b) topology at the 3rd iteration; c) topology at the 6th iteration; and d)
topology at the 30th iteration (after Ali and White, 2001).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

287

White (2001), optimal truss solution can be generated following the elastic, minimum
reinforcement volume, or composite criterion. Using an energy approach, they introduce the
elastic strain compatibility error (SCER) concept to measure how a selected strut-and-tie
model deviates from the elastic solution; the lower the SCER value, the closer it is to the
elastic stress distribution. An example of obtaining the strut-and-tie model of a two-span deep
beam is illustrated in Figure 8.26.
8.7.2

Refining strut-and-tie models

An appropriate strut-and-tie model should at least capture the key load-resisting elements in
the problem under consideration. In many cases, strut-and-tie models need to be refined to
capture more elements contributing to carrying the loading system.
A representative example of refining a strut-and-tie model to higher degrees is shown in Figure
8.27. Another example of a refined a strut-and-tie model is shown in Figure 8.28 for the double
corbel example shown in Figure 8.18. In the latter example, the vertical ties are utilized in the
model to provide a better load capacity estimate. Figure 8.29 shows similar examples in which
strut-and-tie models used in design are adapted to give a higher load capacity.
8.7.3

Other considerations

Besides the various techniques of choosing strut-and-tie models described above, Schlaich et
al. (1987) suggest selecting a strut-and-tie model in which the total length of ties is a minimum.
This recommendation is based on the principle of minimizing strain energy and the observation
that ties are more deformable than struts. For practical reasons, ties should be laid out to form an
orthogonal reinforcing net with the bars parallel to the boundaries of D-regions.
Stress peaks much higher than the average values are normally observed in the corner regions
of a structure. These peaks can be several times higher than the tensile strength of concrete,
causing cracks to be easily initiated. Since corner cracks may have definite influence on the
load capacity and serviceability behaviour, Almsi (1992) recommends that the corner regions
be reinforced to enhance the behaviour at both ultimate and serviceability states. The
appropriate positions of this reinforcement can be identified by including the corresponding
ties in the selected strut-and-tie model.

(a)

(b)

(c)

Figure 8.27: Various degrees or refinement of strut-and-tie models for a prestressed anchorage zone: a) simple
model; b) refined model; and c) further refined model (after FIP Commission 3, 1999).

288

8 Strut-and-tie modelling

Figure 8.28: Refined strut-and-tie model of Figure 8.18 (after Schlaich and Schfer, 1991).

Section A-A

(a)

Section A-A

(b)

Figure 8.29: Different selection of strut-and-tie models for a deep beam example under top loading and
another deep beam under bottom loading: a) models used for design; and b) models for estimating
capacity (after Leonhardt and Walther, 1966).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

289

8.7.4

Static indeterminacy of strut-and-tie models

Like steel trusses, depending on how the reaction and member forces are obtained, strut-andtie models can be classified as either statically determinate or statically indeterminate.
Statically indeterminate cases are further subdivided into internally indeterminate and
externally indeterminate.
A strut-and-tie model is said to be statically determinate when all of the reaction forces and
member forces can be obtained solely from statics. If the total number of strut-and-ties
exceeds the number required by statics, the strut-and-tie model becomes internally statically
indeterminate. The extra number of struts and ties is commonly referred to the degree of
internal redundancy. As an example, the strut-and-tie model in Figure 8.30 is internal
redundant to one degree. Similarly, when a strut-and-tie model has more than three nonconcurrent supports in the restraint system, it becomes externally statically indeterminate; the
extra number of support components is commonly referred to as the degree of external
redundancy. In all cases, as a result of equilibrium requirement in the STM, sufficient
restraints are to be provided in all directions and the struts and ties are arranged properly.
!

Figure 8.30: Example of a statically indeterminate strut-and-tie model redundant to one degree.

Besides statics, additional requirements are usually needed to solve for the reaction and
member forces in a statically indeterminate truss. These requirements are related to the
relative elastic stiffness between the truss members. Member forces are then analyzed
considering the strain compatibility between truss members and boundary conditions.
However, difficulties arise in determining appropriate relative stiffness values for the struts
and ties. These uncertainties have lead to the development of several approaches for obtaining
the distribution of forces, some of which are described below.
8.7.5

Procedures to solve statically indeterminate strut-and-tie models

Linear elastic analysis


To obtain the relative elastic stiffness for each strut-and-tie component, Yun and
Ramirez (1996) and Yun (2000) proposed to use a truss-bar elastic stiffness of Eco Ac/L for a
strut and EsAs/L for a tie, where Ac = the effective cross-sectional area of the strut, As = the
required area of the tie, Eco = initial tangent modulus of elasticity of concrete, Es = modulus
of elasticity of steel and L = length of the struts or ties. The strut effective cross-sectional area
and required tie area are determined iteratively to satisfy the corresponding stress limits.

290

8 Strut-and-tie modelling

Plastic truss method


The plastic truss method is a classic way to handle a statically indeterminate case in lower
bound methods of limit analysis. In this method, several ties in the truss, usually those that are
most heavily loaded, are assumed to have yielded and the redundancies removed so that the
tie forces become known and the truss becomes statically determinate.

A s fy

As fy

To apply this method to the strut-and-tie model of Figure 8.30, the vertical tie is chosen as the
yielded tie. If the reinforcement area is taken as As, the yield strength of the tie is As fy. The
statically determinate truss system of the beam is shown in Figure 8.31. In analyzing the
structure, the member is removed and replaced with a statically equivalent set of forces.

Figure 8.31: Plastic truss model of the strut-and-tie model shown in Figure 8.30.

Another way of applying this method is to decompose the statically indeterminate truss under
consideration into several statically determinate trusses. Figure 8.32 illustrates the truss
decomposition process for the strut-and-tie model of Figure 8.30.

Truss (1) + (2)

Truss (1)

Truss (2)

Figure 8.32 - Decomposition of the strut-and-tie model of Figure 8.30 into two statically determinate trusses.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

291

Procedures based on lower bound solutions


These procedures include rigid-plastic optimal design (RPOD) and shakedown optimal design
(Anderheggen and Schlaich, 1990). In the RPOD, member forces of a statically indeterminate
strut-and-tie model are determined in such a way that minimum tie resistances, corresponding
to the minimum weight of steel ties, are obtained. The process is done using linear
programming. Determining the statically redundant forces by minimizing steel weight has
been traditionally done in limit analysis. For beam problems, it can be shown that this leads to
minimum strain energy (Nielsen, 1999).
The second procedure is based on the static shakedown theorem. This procedure is similar to
RPOD except that elastic truss solutions have to be considered. It is particularly useful for
multiple load cases as the computational time is less than for RPOD. The solution obtained
lies between the elastic solution and the rigid-plastic solution.
8.7.6

Dimensioning nodal regions

There are several rules in defining the idealized geometry of a nodal zone. The dimensions
and shape of a nodal zone affect its in-plane stress distribution and are dependent on the
number, direction, and the effective widths of struts and ties framing into the node. The
effective width of a strut is usually selected in order that the effective strength is not
exceeded. The rules in determining the effective width of a tie are less clear and usually
follow those for constructing nodal zones. Some methods for constructing the nodes are:

Simple method
In this method, the shape of a node is formed by the intersection of actual dimensions of struts
and ties whose centrelines are concurrent and meet at the node. An example of this node is
shown in Figure 8.4. In this example, the node is triangular as the node is formed by the
intersection of three members. The biaxial stress distribution in the node is uniform and can
be presented in the form of Szmodits Mohrs circle, as illustrated in Figure 8.4c. For nodes
formed by more than three intersecting members (Figure 8.33), the biaxial stress distribution
is no longer uniform. In these cases, the adequacy of the node strength may be simply
checked by comparing the in-plane stresses acting on all sides of the nodes with the
corresponding stress limits (Schlaich and Schfer, 1991). This approach, however, is only
applicable for nodes with typical configuration and must be used discriminately for complex
node configurations.
The finite element method becomes a choice to check the adequacy of complex node region
shapes constructed using the simple method. The generic model for the finite element analysis
is shown in Figure 8.34. The forces of the struts and ties framing into node become the
external forces acting on the node. Tie forces are treated as compressive forces acting from
behind the node. Alshegeir and Ramirez (1992) proposed to verify the bearing capacity of the
nodal zones by comparing the principal stresses obtained from the finite element analysis with
the Mohr-Coulomb failure criterion. In the later version of their program, nonlinear finite
element analysis was used with failure criteria determined from experimental test data of a 2D
plain concrete (Yun and Ramirez, 1996).

292

8 Strut-and-tie modelling

fc2

C1

C2

C3

fc1

fc3

C4

fc4
(a)

(b)

Figure 8.33: Simple nodal region bounded by four struts (after Tjhin and Kuchma 2002).

ru
St

St
ru

Tie

t
Figure 8.34: Checking of nodal regions using finite element analysis (after from Alshegeir and Ramirez 1992)

The state of stress of node zones having triangular shapes (i.e., those formed when three struts
or ties intersecting) is constant, as shown in Figure 8.4. For nodal zones consisting of more
than three sides, similar stress conditions may be obtained by breaking them down into
several stress triangles. The triangles are separated by lines of stress discontinuity (refer to
Section 8.3.3) and are arranged such that the state of stress in all triangles is constant and
equilibrium along the lines of stress discontinuity is satisfied. The lines of stress discontinuity
must be introduced at the vertices of the nodal zone polygon to achieve these conditions. This
technique was implemented by Tjhin and Kuchma (2007b). Figure 8.35 shows how this
approach is applied to the nodal zone of Figure 8.33a.
Any boundary forces acting on nodal zones divided into constant stress triangles are treated as
if the force components acted on the sides of the nodal zones. An example of how this is done
is illustrated in Figure 8.36. With this assumption, a body force that acts on a node will not
have direct influence on the stress distribution in the nodal zone itself but it will affect the
stress distribution in all other nodal zones and in all of the struts and ties. The assumption that
body forces have no direct influence on the stress distribution of nodal zones carrying them
can be justified by the fact that body forces mostly come from conservative lumping of selfweight or are reaction forces of other members framing into and forming integral parts within
the D-region being considered.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

293

fc6

fc7
fc2

fc5

fc8

fc1

fc1

fc9

fc3
fc4

fc10

Lines of
Stress Discontinuity
fc4

Figure 8.35: Nodal zone of Figure 8.33 divided into constant stress triangles.

ft2

ft2
F2

ft1

fc3

F3

F4

ft1
F

F1 F2
F4

F3

fc3

F1
fc4
(a)

fc4
(b)

(c)

Figure 8.36: Example of how a body force acting on a node consisting of constant stress triangles is treated: a)
nodal zone; b) decomposition of the body force according to the head-to-tail rule; and c) statically
equivalent forces acting on the sides of the node.

An equilibrium approach of limit analysis is used for estimating the contribution of concrete
and reinforcing steel in carrying the stresses in the constant stress triangles of CCT, CTT, and
TTT nodal zones. The reinforcing steel is assumed to extend inside and beyond the nodal
zones. This steel is also assumed to be evenly distributed in the nodal zones, i.e., smeared
steel is assumed. See the CCT nodal zone example provided in Figure 8.37.
The Mohr-Coulomb-Rankine yield condition, as shown in Figure 8.38, is used to check the
strength adequacy of the constant stress triangles of CCC nodes (Schlaich and Anagnostou,
1990). A linearized version of the Mohr-Coulomb yield condition (Hajdin, 1990) is used for
checking the adequacy of constant stress triangles of CCT, CTT, and TTT nodes. (Figure
8.39). For reinforcing steel, an elasto-plastic yield condition is used. The stress distribution
between the concrete and reinforcing steel follows the approach by Mller (1975).

294

8 Strut-and-tie modelling

Figure 8.37: Example of how the steeel distributionn assumed in nodal zone co
onsisting of coonstant stress triangles.

+1

Sliding Failure

fcu

Se
eparation Failure

(-1, 0)
-11/15

ft
,
fc

+2
fcu

ft
,
fc

-11/15
(0
0, -1)

Figure 8.38: Concretee yield criteriaa for checkingg the adequaccy of CCC nod
dal zones dividded into constant stress
triangless.

,xy,c

!
(-3/4, -1/4 , 1/2)
(-1, 0, 1/6)

fcu

+y,c
fcu

(-1/4
4, -3/4, 1/2)
0
(0,, -1, 1/6)

+x,c
fcu
c

(-1, -1, 0)

Figure 8.39: Concrette yield criteriia for checkingg the adequaccy of CCT, CT
TT, and TTT nodal zones divvided into
constantt stress trianglles.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

295

Hydrostatic node construction


This is the classic method for dimensioning a node. In this method, the shape of a node is
arranged such that the stresses on all sides of the node, from the truss member forces as well
as from the boundary forces meeting at the node, are equal. The biaxial state of stress inside
the node is hydrostatic in two dimensions, that is, the in-plane stresses are isotropic,
homogeneous, and equal to those on the sides.
Arranging the node in this shape can be done by sizing the boundaries of the node so that they
are proportional and perpendicular to the forces acting on them. In defining the effective
width of a tie, the tie force is treated as a compressive force acting from behind the node
(Figure 8.40). This type of node is termed a hydrostatic node. Figure 8.41a shows how the
CCC node of Figure 8.4a looks if it is arranged using the hydrostatic approach. As shown in
Figure 8.41b, the Mohrs circle for the nodal zone collapses to a point because the stresses are
hydrostatic. However, the forces framing into the node may not be concurrent (Figure 8.42).

fc

fc

fc

fc

fc

fc

fc
fc

fc

fc
fc
fc

fc

fc

fc
fc

fc

fc

fc

T
(d)

(c)

(b)

(a)

Figure 8.40: Hydrostatic construction of nodes shown in Figure 8.3: a) CCC; b) CCT;c) CTT; and d) TTT
(after Tjhin and Kuchma 2002).

!
w3

ru
St

A
Strut 1

w1

fc

t3

fc
fc

fc

ut
Str

fc

fc

w2
(a)

(b)

Figure 8.41: a) Construction of nodal zone shape of Figure 8.4 using the hydrostatic procedure; b) Mohrs
circle representing the stresses in the nodal zone.

296

8 Strut-and-tie modelling

Figure 8.42: Examplee of the node in


i Figure 8.333a constructed
d using the hyydrostatic procedure (after Tjhin and
Kuchma,, 2002).

Modifieed hydrostatic node con


nstruction
This appproach waas introduceed by Schlaaich and Anagnostou
A
(1990) and basically
y utilizes
hydrostaatic node ruule whilst maintaining
m
g a set of co
oncurrent forces.
fo
A noode with more
m
than
three members
m
inttersecting iss handled by
b breaking
g down the node into several hyd
drostatic
nodes of
o triangularr shapes coonnected byy short prissmatic struts. Figure 8.43 shows how the
CCC noode of Figurre 8.33a is arranged
a
usiing this app
proach.
Trapezooidal transittion stress zones
z
betweeen a node and the inttersecting trruss membeers were
also forrmulated using this appproach, alloowing differrent stress intensities
i
oof truss mem
mbers to
act on thhe node. Thhe length off these transsition zones is determinned using thhe separation failure
criteria of the Mohr-Coulomb--Rankine yiield conditio
on (Figure 8.38).
8

Figure 8.43:
8
Examplee of modified hydrostatic
h
no
ode (after Tjhiin and Kuchm
ma, 2002).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

297

8.8

Design example using computer-based tools

8.8.1

Problem statement

The propped cantilever deep beam with an opening shown in Figure 8.44 is to be designed
using the STM. The beam supports a factored column reaction force, P, of 5000 kN and a
factored uniformly distributed load of 120 kN/m. The distributed load includes the self weight
assumed, for simplicity, to act on the top of the beam. The compressive strength of concrete,
f c# , and the yield strength of the steel reinforcement, f y , are taken as 35 MPa and 420 MPa,
respectively.

4000 mm

P = 5000 kN

w = 120 kN/m

1000
mm

t = 600 mm

1500
mm

4000 mm

1000 mm

Shear
Wall
1500 mm

2750 mm

500 mm
11000 mm

Figure 8.44: Propped cantilever deep beam with an opening.

8.8.2

Solution

The deep beam is statically indeterminate to the first degree, as schematically shown in Figure
8.45. For simplicity and to fit into the strut-and-tie model to be selected, the distributed load is
lumped at locations indicated in the figure.
A 2D linear finite element analysis was conducted to estimate the reaction forces. The
principle stress vectors for the FE solution are plotted in Figure 8.46. A comparison with a
beam model, with and without considering the shear deformation along the beam length, is
given in Table 8.1 and shows that the vertical reaction at end A, VA, is not very sensitive to the
analysis conducted. In the subsequent discussion, the value of VA equal to 3000 kN will be
used and treated as an additional external load acting on the structure, thus removing the
external redundancy. Force redistribution was assumed to be adequate to handle any deviation
in the distribution of reaction forces.

298

8 Strut-and-tie modelling

480 kN

5000 kN 330 kN

330 kN

156 kN

VA

MA

2000 mm

2000 mm

1375
mm

1375
mm

4250 mm

VB

Figure 8.45: Magnitude and location of the lumped uniformly distribution loading.

Figure 8.46: Plot of principal stresses from linear finite element analysis.

Table 8.1: Reaction forces from four different analyses

Analysis

Reaction
VA (kN)

VB (kN)

MA (kN-m)

Linear finite (quadrilateral) element

3000

3310

9430

Beam (flexure deformation only, uniform stiffness)

2890

3420

10530

Beam (flexure deformation only, considering opening)

2930

3380

10000

Beam (flexure and shear deformations, uniform stiffness)

2970

3340

9620

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

299

The selected strut-and-tie model is shown in Figure 8.47a. The model is statically determinate
and was laid out based on the plot of elastic principal stresses shown in Figure 8.46. The force
transfer from the column loading to the left support is provided by two diagonal struts AF and
BG, connected by a vertical tie AG and horizontally equilibrated by strut FG at the top and
ties FG and GH at the bottom. The main force transfer in the right side of the column force is
also provided by diagonal struts, except that a smaller truss system extending from node C to
node I was used in the vicinity of the opening, replacing the single diagonal strut CI. The truss
system around the opening is shown in a greater scale in Figure 8.47b. For ease of
construction, the ties of this truss system were laid out such that the resulting reinforcement is
parallel to the boundaries of the opening.
A stress limit equal to 0.6 f c# was used for dimensioning struts and nodes. A strength
reduction factor of 0.75 was applied to all struts, ties, and nodes to account for variations in
construction quality and workmanship. This corresponds to an effectiveness factor of
0.75 o 0.6 f c# ' 0.45 f c# . Figure 8.48 shows the distribution of the forces in the strut-and-tie
model. Table 8.2 lists the required effective width and the corresponding capacity of the
struts. The reinforcement for the ties is summarized in Table 8.3.
To increase ductility capacity in the beam, additional distributed reinforcement 0.14% of the
cross-sectional area is provided in the horizontal direction. A standard 90-degree hook is used
to anchor tie FG while straight lengths are used for anchoring all other ties. The final
reinforcing details are shown in Figure 8.49.
Table 8.2: Strut forces, selected effective widths and strut capacities.

Strut ID
IJ
BC
AB
AF
BG
BH
DK
UI
TU
QT
NQ
CN
SI
PS
MP
LM
CL
QR
RU
OP
LO
FF'

300

Force
(kN)
-1327
-2180
-1714
-3455
-2902
-2665
-4192
-2947
-1396
-1974
-2712
-2209
-2378
-2928
-2111
-1492
-2986
-288
-2084
-330
-2000
-3000

Effective Width
(mm)
400
270
212
427
358
329
517
364
172
244
335
273
294
361
261
184
369
200
257
200
247
370

Capacity
(kN)
3240
2184
1714
3455
2902
2665
4192
2947
1396
1974
2712
2209
2378
2928
2111
1492
2986
1620
2084
1620
2000
3000

8 Strut-and-tie modelling

250 mm

5000 kN

480 kN

330 kN
78 kN

P
K

S
U

kN

3917 kN

92
41

3500 mm

F
250 mm

330 kN

1327 kN

78 kN
2000 mm

1375
mm

2000 mm

1375
mm

4250 mm

3000 kN
(a)

1275 mm

850
mm

850
mm
78 kN

850
mm

L
P

1300 mm

1300 mm

1150 mm

955
mm

R
U

78 kN
950
mm

850
mm

1150 mm

850
mm

850
mm

1275 mm

Figure 8.47: a) Selected strut-and-tie model and boundary forces; b) detail ofstrut-and-tie model around the
opening.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

301

Figure 8.48: Force distribution in the struts and ties.

Table 8.3: Tie forces, required and provided reinforcing steel, and tie capacities.

302

Tie ID

Force
(kN)

Required
Steel Area
(mm)

FG
GH
HI
CD
DE
GA
HC
TR
OM
ID
NO
RS
QL
UP

1714
3154
2180
1327
3917
2520
2480
1474
1414
2966
1084
1185
730
798

5442
10014
6921
4212
12434
8000
7873
4678
4490
9416
3442
3762
2319
2535

Provided Steel
Configuration

Area
(mm)

3 Layers of 4 #29
4 Layers of 4 #29
3 Layers of 4 #29
2 Layers of 4 #32
4 Layers of 4 #32
2-Legged Stirrups #13 @100 mm
2-Legged Stirrups #16 @100 mm
2-Legged Stirrups #16 @100 mm
2-Legged Stirrups #16 @100 mm
2-Legged Stirrups #16 @100 mm
1 Layer of 5 #32
1 Layer of 5 #32
1 Layer of 3 #32
1 Layer of 3 #32

7740
10320
7740
6552
13104
8772
7960
5572
5572
9552
4095
4095
2457
2457

Capacity
(kN)
2438
3251
2438
2064
4128
2763
2507
1755
1755
3009
1290
1290
774
774

8 Strut-and-tie modelling

!
Two-Legged Stirrups
#16 @100 mm

3 Layers of
4 #29
500 mm

1750 mm

2 Layers of 4 #32

2000 mm

5 #32

3 Layers of 4 #29

1375 mm

4250 mm

Two-Legged Stirrups
#16 @ 100 mm

3 #32

5 #32

4 Layers of 4 #29

4 Layers of 4 #32

3 #32

Two-Legged Stirrups
#16 @ 100 mm

4000 mm

Two-Legged Stirrups
#13 @ 100 mm

2 #32 (Framing Bars)

4 #29
(Framing Bars)
1125 mm

10750 mm
Note: Column reinforcement is not shown
Horizontal Web Reinforcement
#13 @ 300 mm Each Face (Typical)

Figure 8.49: Reinforcing details for deep beam example.

8.9

References

ACI Committee 318 (2002), Building Code Requirements for Structural Concrete (ACI 31802) and Commentary (ACI 318R-02), American Concrete Institute, Farmington Hills,
Michigan, 443 pp.
Aguilar, G., Matamoros, A.B., Parra-Montesinos, G.J., Ramirez, J.A., and Wight, J.K. (2002),
Experimental Evaluation of Design Procedures for Shear Strength of Deep Reinforced
Concrete Beams, ACI Structural Journal, Vol. 99, No. 4, July-August, pp. 539-548.
Ali, M.A., and White, R. N. (2001), Automatic Generation of Truss Model for Optimal
Design of Reinforced Concrete Structures, ACI Structural Journal, Vol. 98, No. 4, JulyAugust, pp. 431-442.
Almsi, J. (1992), Cracks as Important Constituents of Strut and tie Models, Periodica
Polytechnica Civil Engineering, Budapest University of Technology and Economics, Vol. 36,
No. 3, pp. 251-270.
Alshegeir, A., and Ramirez, J. A. (1992), Computer Graphics in Detailing Strut-Tie
Models, Journal of Computing in Civil Engineering, Vol. 6, No. 2, April, pp. 220-232.
American Association of State Highway and Transportation Officials (1994), AASHTO LRFD
Bridge Specification, 1st ed., Washington, DC, 1091 pp.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

303

Anderheggen, E., and Schlaich, M. (1990), Computer Aided Design of Reinforced Concrete
Structures Using the Truss Model Approach, Proceedings of the Second International
Conference on Computer Aided Analysis and Design of Concrete Structures, N. Bicanic and
H. Mang, eds., Zell am See, Austria, April 4-6, pp. 539-550.
ASCE-ACI Committee 445 on Shear and Torsion (1998), Recent Approaches to Shear
Design of Structural Concrete, Journal of Structural Engineering, ASCE, Vol. 124, No. 12,
December, pp. 1375-1417.
Biondini, F., Bontempi, F., and Malerba, P.G. (1999), Optimal Strut-and-Tie Models in
Reinforced Concrete Structures, Computer Assisted Mechanics and Engineering Sciences,
Institute of Fundamental Technological Research, Polish Academy of Sciences, Vol. 6, No. 34, pp. 279-293.
Comit Euro-International du Bton (1993), CEB-FIP Model Code 1990, Thomas Telford
Services, Ltd., London, 437 pp.
Chen, B. S., Hagenberger, M.J., and Breen, J.E. (2002), Evaluation of Strut-and-Tie
Modeling Applied to Dapped Beam with Opening, ACI Structural Journal, Vol. 99, No. 4,
July-August, pp. 445-450.
Chen, W.F., and Han, D.J. (1988), Plasticity for Structural Engineers, Springer-Verlag, Inc.,
New York, 606 pp.
CSA Technical Committee on Reinforced Concrete Design (1994), Design of Concrete
Structures, A23.3-94, Canadian Standards Association, Rexdale, Ontario, December, 199 pp.
FIP Commission 3 (1999), Practical Design of Structural Concrete, fdration internationale
du bton (fib), Lausanne, September, 114 pp.
Foster, S.J., and Malik, A.R. (2002), Evaluation of Efficiency Factor Models used in Strutand-Tie Modeling of Nonflexural Members, Journal of Structural Engineering, ASCE, Vol.
128, No. 5, May, pp. 569-577.
Lee, D.D.K. (1982), An Experimental Investigation in the Effects of Detailing on the Shear
Behaviour of Deep Beams, M.A.Sc. Thesis, Department of Civil Engineering, University of
Toronto, 138 pp.
Hajdin, R. (1990), Computeruntersttzte Berechnung von Stahlbetonscheiben mit
Spannungsfeldern, Report No. 175, Institut fr Baustatik und Konstruktion, Eidgenssische
Technische Hochschule, Zrich, August, 115 pp.
Harisis, A., and Fardis, M.N. (1991), Computer-Aided Automatic Construction of Strut-andTie Models, Structural Concrete, IABSE Colloquium, Stuttgart 1991, International
Association for Bridge and Structural Engineering, Zrich, March, pp. 533-538.
Kuchma, D.A., and Tjhin, T.N. (2002), Design of Discontinuity Regions in Structural
Concrete Using a Computer-Based Strut-and-Tie Methodology, Paper No. 02-3000,
Transportation Research Record 1814, Transportation Research Board of the National
Academies, pp. 72-82.
Kumar, P. (1978), Optimal Force Transmission in Reinforced Concrete Deep Beams,
Computers and Structures, Vol. 8, No. 2, pp. 223-229.

304

8 Strut-and-tie modelling

Jirsa, J.O., Bergmeister, K., Anderson, R., Breen, J.E., Barton, D., and Bouadi, H. (1991),
Experimental Studies of Nodes in Strut-and-Tie Models, Structural Concrete, IABSE
Colloquium, Stuttgart 1991, International Association for Bridge and Structural Engineering,
Zrich, March, pp. 525-532.
Leonhardt, T., and Walther, R. (1966), Wandartiger Trger, Deutscher Ausschuss fr
Stahlbeton, Bulletin No. 179, Wilhem Ernst & Sohn, Berlin, 159 pp.
Liang, Q.Q., Uy, B., and Steven, G.P. (2002), Performance-Based Optimization for Strut-Tie
Modeling of Structural Concrete, Journal of Structural Engineering, ASCE, Vol. 128, No. 6,
pp. 815-823.
MacGregor, J. G. (1997), Reinforced Concrete: Mechanics and Design, Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey, 3rd ed., 939 pp.
Marti, P. (1985), Basic Tools of Reinforced Concrete Beam Design, ACI Journal,
Proceedings, Vol. 82, No. 1, January-February, pp. 45-56.
Maxwell, B.S., and Breen, J.E., Experimental Evaluation of Strut-and-Tie Model Applied to
Deep Beam with Opening, ACI Structural Journal, Vol. 97, No. 1, January-February, 2000,
pp. 142-148.
Mish, K., Nobari, F., and Liu, D. (1995), An Interactive Graphical strut-and-Tie
Application, Proceedings of the Second Congress on Computing in Civil Engineering,
Mohsen, J. P., ed., American Society of Civil Engineers, New York, pp. 788-795.
Mitchell, D., Cook, W.D., Uribe, C.M., and Alcocer, S.M. (2002), Experimental Verifcation
of Strut-and-Tie Models, Examples for the Design of Structural Concrete with Strut-and-Tie
Models, SP-208, K.-H. Reineck, ed., American Concrete Institute, Farmington Hills,
Michigan, pp. 41-62.
Mller, P. (1978), Plastische Berechnung von Stahlbetonscheiben und balken, Report No.
83, Institut fr Baustatik und Konstruktion, Eidgenssische Technische Hochschule, Zrich,
July, 160 pp.
Muttoni, A., Schwartz, J., and Thrlimann, B. (1997), Design of Concrete Structures with
Stress Fields, Birkhauser, Boston, Massachusetts, 143 pp.
Nielsen, M. P. (1971), On the Strength of Reinforced Concrete Discs, Acta Polytechinca
Scandinavica, Civil Engineering and Building Construction Series No. 70, Copenhagen, 261
pp.
Nielsen, M. P. (1999), Limit Analysis and Concrete Plasticity, CRC Press LLC, 2nd ed.,
908 pp.
Polla, M. (1992), A Study of Nodal Regions in Strut-and-Tie Models, M.A.Sc. Thesis,
Department of Civil Engineering, University of Toronto, 130 pp.
Rogowsky, D.M., and MacGregor, J. G. (1983), Shear Strength of Deep Reinforced
Concrete Continuous Beams, Structural Engineering Report No. 110, Department of Civil
Engineering, University of Alberta, Edmonton, November, 178 pp.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

305

Rogowsky, D.M., and MacGregor, J. G. (1986), Design of Reinforced Concrete Deep


Beams, Concrete International, August, pp. 49-58.
Rckert, K.J. (1991), Design and Analysis with Strut-and-Tie Models Computer-Aided
Methods, Structural Concrete, IABSE Colloquium, Stuttgart 1991, International Association
for Bridge and Structural Engineering, Zrich, March, pp. 379-384.
Schlaich, J., Schfer, K., and Jennewein, M. (1987), Toward a Consistent Design of
Structural Concrete, Journal of the Prestressed Concrete Institute, Vol. 32, No. 3, May-June,
pp. 74-150.
Schlaich, J., and Schfer, K. (1991), Design and Detailing of Structural Concrete Using
Strut-and-Tie Models, The Structural Engineer, Vol. 69, No. 6, March, pp. 113-125.
Schlaich, M., and Anagnostou, G. (1990), Stress Fields for Nodes of Strut-and-Tie Model,
Journal of Structural Engineering, ASCE, Vol. 116, No. 1, January, pp. 13-23.
Tjhin, T.N., and Kuchma, D.A. (2002), Computer-Based Tools for Design by the Strut-andTie Method: Advances and Challenges, ACI Structural Journal, Vol. 99, No. 5, SeptemberOctober, pp. 586-594.
Tjhin, T. N., and Kuchma, D. A. (2007a), Integrated Analysis and Design Tool for the Strutand-Tie Method, submitted to Engineering Structures, Journal of Engineering Structures
[accepted for publication].
Tjhin, T. N., and Kuchma, D. A. (2007b), "Limit State Assessment of Nodal Zone Capacity in
Strut-and-Tie Models, Vol. 4, No. 4, August, pp. accepted for publication to the journal of
Computers and Concrete
Vecchio, F.J., and Collins, M.P. (1986), The Modified Compression Field Theory for
Reinforced Concrete Elements Subjected to Shear, ACI Journal, Vol. 83, No. 2, MarchApril, pp. 219-231.
Yun, Y.M., and Ramirez, J.A. (1996), Strength of Struts and Nodes in Strut-Tie Model,
Journal of Structural Engineering, ASCE, Vol. 122, No. 1, January, pp. 20-29.
Yun, Y.M. (2000), Computer Graphics for Nonlinear Strut-Tie Model Approach, Journal of
Computing in Civil Engineering, ASCE, Vol. 14, No. 2, April, pp. 127-133.

306

8 Strut-and-tie modelling

Special purpose design methods for surface structures

9.1

Introduction

Up to and including Chapter 7 this Guideline has discussed nonlinear analysis procedures on
the basis of realistic constitutive material models for reinforced concrete and finite element
techniques. It appears that the approach is of interest for both the design of structures and for
purposes of final performance checking. Chapter 8 about strut-and-tie modelling, primarily
focussing on design, differs fundamentally from the previous chapters. Apart of these
methods some other design-oriented approaches are advocated for special types of surface
structures like plates loaded-in-plane (shear-walls, etc.) and plates loaded out-of-plane (slabs).
Three such approaches are discussed in this chapter. They all apply a computational elementbased technique to construct an equilibrium stress-state, of which two (developed in Denmark
and Switzerland) offer an automated design on basis of optimizing procedures in combination
with ideal plasticity and one (the Netherlands) an interactive design tool which links up with
the modified compression-field theory. The Danish method is discussed for both plates loaded
in their plane and slabs, the Swiss one for slabs only, whereas the Dutch one is restricted to
plates loaded in their plane.

9.2

Notation

In this chapter a notation is applied which is as much as possible the same for the three
methods. As a consequence the notation of the original documents will differ from the one
chosen here.

d
f
f

fel
fi
B

set of design variables


load vector; global vector of all f
element nodal forces
elastic global vector f

c se

global element force vector corresponding with ( i


kinematic matrix
follows from vector of yield limits

matrix that relates ( to C s ; global matrix of all C e

Ce

matrix that relates f e to c se

Cd

matrix that relates d to C s ;


matrix depending on the linearization of the yield criteria
array of yield moments for all nodes; global vector of all c se

Cs
Ec
fc
fi

complementary energy
compressive strength
panel edge forces (i = 1, 2, 3, 4)

ft x

tensile strength in x-direction

tensile strength in y-direction

ft

F
H
K

matrix of all f i ; flexibility matrix; fictitious plastic strain distributions

matrix that relates ( to Rc 8 ^ R


global stiffness matrix

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

307

Ke
l
mxx
m yy

element stiffness matrix


span of slab
bending moment in x-direction
bending moment in y-direction

m xy

twisting moment

mp

plastic yield moment

m8px

positive yield moment in x-direction

m7px

negative yield moment in x-direction

m8py

positive yield moment in y-direction

m7py

negative yield moment in y-direction

M
Mp

moment in beam
positive plastic moment in beam

M 7p

negative plastic moment in beam

pL
N
Nc
Nt
R
Rc
ui

limit distributed load on slab


normal force in bar
limit compressive normal force in bar

ue

element nodal displacements


elastic global vector u

uel
vx
vy

V
W

(i
^
r
+xx

limit tensile normal force in bar


set of variable concentrated loads
set of fixed concentrated loads
panel edge displacements (i = 1, 2, 3, 4)

shear force in x-direction


shear force in y-direction
set of dual variables related to vector Re
matrix to compute weight of structure from design variables d ;
matrix depending on the linearization of the yield criteria.
vector of unknown stress and/or moment parameters;
vector of all free amplitudes (i
free amplitude of strain distribution;
generalized panel stress (i = 1, 2, 3, 4, 5)
scale factor
set of variables corresponding to the yield moments C s

+ yy

stress in x-direction for in-plane loads


stress in y-direction for in-plane loads

+ xy

shear stress for in-plane loads.

308

9! Special purpose design methods for surface structures

9.3

Design of slabs and shear walls: perfect plastic approach

Plates loaded in their plane (shear-walls) and loaded perpendicular to their plane (slabs) can
be considered as perfectly plastic solids and designed using limit analysis. Two main
approaches exist, based on either an assumed stress state or an assumed displacement mode,
resulting in a lower bound solution or an upper bound solution, respectively. Solutions have
been published for many plate types but in general cases hand calculations are time
consuming. In this section, a Danish approach is described in which use is made of a
computer analysis. The aim of the method is to provide a safe lower bound solution and
information on the deformations at failure. The approach can be used to determine not only
the ultimate load for a given geometry and the reinforcement but also to optimize the
reinforcement for a given loading. First, the highlights of the method will be explained for
plates subjected to bending and then for plates acting as membranes. For more details of the
methodology the reader is referred to Damkilde and Krenk (1997), Krabbenhoft et al. (2002),
Poulsen et al. (2002).
9.3.1

Slabs subjected to bending loads

For slabs subjected to bending loads, similar to finite elements, the plate is divided into
elements triangular in plan. Within each element, a moment field is assumed which is linear
for both the bending moments mxx and myy and for the twisting moment mxy. Consequently, the
shear forces vx and vy are constant over the area of an element. Nine moment parameters are
required to define this stress state, three in each corner. An equilibrating bending moment
field is constructed for which the bending moment normal to the edge is continuous at the
inter-element boundaries. The same is done for the twisting moments, which, strictly
speaking, is not necessary. Figure 9.1 shows one element and the moments on the boundary,
which are made continuous. As a result, a vector $ with unknown moment parameters is
achieved for the whole plate.

Figure 9.1: Continuous moments on the boundary of triangular elements.

It should be noted that a triangle, with a linear moment field and constant shear forces, can be
loaded only by constant distributed loads along the edges and these loads are replaced by
equivalent corner point loads. One could say a homogeneously distributed load on the slab is
replaced by a statically equivalent set of point loads at the joints. This is fully acceptable if the

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

309

mesh is sufficiently fine. The set of point loads is named Rc + &R, where Rc is a fixed load and
&R a variable load, scaled by the factor &. For equilibrium, the bending moments are related to
the nodal loads by

H( ' Rc 8 ^R

(9.1)

where the matrix H is assembled from element matrices in a similar way as a global stiffness
matrix is assembled in the standard finite element method from element stiffness matrices.
The matrix H is rectangular, because the number of moment parameters in $ is three times the
number of nodes, which defines the length of the vectors Rc and R.
The moments must obey the yield criterion for reinforced concrete slabs. For this purpose the
yield criterion as shown in Figure 9.2 is applied, which was published by Nielsen (1964) and
Wolfensberger (1964). The yield criteria is then represented by
2
7 (m8px 7 mxx )(m8py 7 myy ) 8 mxy
?0
2
7 (m7px 7 mxx )(m7py 7 myy ) 8 mxy
?0

(9.2)

where m 8px and m 7px are positive and negative yield moments respectively in the x-direction,
and m 8py and m 7py are positive and negative yield moments in the y-direction. The yield
criterion can be linearized using section planes. The minimum linearization contains eight
planes, and a more accurate one is a replacement by sixteen planes. After such a linearization
the yield condition can be written in the form:

C( ? Cs

(9.3)

where the values in C depend on the linearization planes and the global array Cs contains the
yield moments for all nodes, based on the material strengths.

Figure 9.2: Yield criterion for reinforced concrete slab.

310

9! Special purpose design methods for surface structures

9.3.2

Ultimate load determination

A slab of given dimensions and reinforcement is considered for which we want to determine
the maximum value of the load parameter &. In language of mathematical programming, ) is
the objective function and Eqs. 5.13 and 5.14 are conditions to be met.. The objective function
is to maximize:

g( u
ZYOT 1WV i v
j^ w

(9.4)

7 R W g ( u ' g Rc u
i v i v
O UV j ^ w ? j C s w

(9.5)

with the conditions:

ZH
XC
Y

From this mathematical programming problem, we receive a solution for $ and &. The
procedure to solve this linear programming problem is not discussed here. Rather, the reader
is referred to the reference source literature Limit Analysis and Optimal Design of Plates
with Equilibrium Elements by Krenk et al. (1994). Here it suffices to mention that the
number of unknown 0-values can be reduced to the number of static indeterminacies of the
structure. The computing time on modern PCs is in the order of tens of seconds. It is reported
that a linearization of the yield condition with eight planes yields almost the same results as
with sixteen planes; the differences are in the order of 5%.

Example 1
The first example is the simply supported square slab under uniform load shown in Figure 9.3
and the slab is to be isotropically reinforced equally for both the upper and lower layers. The
plastic yield moment is mp. For this case the exact solution for the upper-bound limit load on
basis of the yield line method is:
p L ' 24.0

mp
l2

l
Figure 9.3: Simply supported square slab.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

311

When an element mesh is chosen that contains the crossing diagonals, the same solution is
obtained independently of the number of elements. Now it is a lower-bound solution. If no
negative moment capacity in the corner region is available, the exact upper-bound solution is:
p L ' 22.0

mp
l2

A linearization of the yield function with eight planes yields a coefficient 21.9 for a negative
moment capacity that is one-eight of the positive capacity, which is about five percent
smaller. For a zero negative moment capacity the coefficient will even be smaller (expectation
21.3). It is reasonable that the value is less than 22.0, because the element-based method is
based on the lower-bound theorem and the yield-line method on the upper-bound theorem.

Example 2
In this example, a square plate is considered that is clamped at all edges (see Figure 9.4). The
exact upper bound solution for a homogeneous reinforcement arrangement in two directions,
top and lower layers, is given by Fox (1974) as
p L ' 42.85

mp
l2

The results of the optimal design yields a coefficient of 40.8 for eight yield planes and 41.7
for 16 yield planes.

l
Figure 9.4: Square slab with clamped edge.

Example 3
The element-based method is of course not intended for elementary cases as were discussed in
the first two examples. These examples were done to get confidence in the approach. In this
example a more complex slab is analysed to illustrate the use of the element-based method. It
is a structure, which originally was designed by a traditional yield-line calculation. The
geometry, supported edges and yield-line pattern are shown in Figure 9.5. Where the edges
are supported, they are simply supported. The load is uniform and the reinforcement
arrangement is isotropic. The limit load from a hand calculation is called 100% here.
312

9! Special purpose design methods for surface structures

Damkilde and Krenk (1997) later referred to a fully automatic yield-line calculation, in which
a limit load of 89.6% of the hand calculation resulted. The element-based method gives a
result of 83.5% if eight planes are used to linearize the yield function and 88.3% for a
linearization with 16 planes. Finally, a calculation with the non-linear yield criterion without
any approximation gives 88.8% of what was originally calculated by hand. This is very close
to the 89.6% of the fully automated upper-bound calculation.

(b)

(a)

Figure 9.5: Slab of irregular shape: a) yield line pattern; and b) FE mesh.

9.3.3

Failure mode determination

In the preceding section the optimization problem was stated as follows:


maximize:

with the conditions:

g( u
ZYOT 1WV i v
j^ w
ZH
XC
Y

7 R W g ( u ' g Rc u
i v i v
O UV j ^ w ? j C s w

(9.6)

(9.7)

This is called the primal linear programming problem. In the linear programming theory there
corresponds a dual problem to each primal problem. The variables of the dual problem
correspond to each of the equations of the primal problem much like the principal of virtual
work in mechanics. Thus, there is a set of variables V corresponding to the load vector Rc and
another set of variables r corresponding to the yield moments Cs. The dual problem is
obtained from the primal by transposing the matrix and interchanging the roles of the righthand side and the vector in the objective function. In the present case this leads to the dual
problem:
minimize

ZY RcT

g 7V u
CsT WV i v
jr w

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

(9.8)

313

Z HT
conditions: X T
Y7 R

C T W g 7V u ' gO u
U i v i v;
OT V j r w ' j 1 w

r <O

(9.9)

Only the dual variables % associated with the inequalities of the primal problem are restricted
to positive values. In the present problem, V can be interpreted as the displacements of the
nodes, while % can be interpreted as rotation discontinuities associated with the corresponding
linear part of the yield surface.
It is established within the theory of linear programming that the solutions of the primal and
dual problems produce the same value of the objective function and that the solution of the
one may be constructed directly from the other. This fact is important in the present context,
because it establishes equivalence between the static model and a corresponding kinematic
model. Thus, the notion of upper and lower bound solutions in classical plasticity theory is
replaced by a common approximate solution that may give preference to the representation of
the static or kinematic fields. For the structural engineer it means that he does not just receive
the limit load but also the deformation mode that is associated with the failure load.
The displacement field is shown in Figure 9.6 for the structure of Example 3. It is seen that
the assumed yield-line pattern of the original hand calculation is not confirmed. The
deformation mode associated with the limit load is quite different.

(a)

(b)

Figure 9.6: Irregularly shaped slab: a) original yield-line pattern; and b) displacement lines.

9.3.4

Material optimization

In case of material optimization the load Rc is kept constant and no & occurs. We add a
number of design variables d and make the yield criteria linear in the design variables:

C $ + Cd d * Cs

(9.10)

where Cd depends on the linearization of the yield criteria. The quantities C and Cs have the
same meaning as in the previous section. Now the equilibrium equations and yield conditions
combine to the linear programming problem:

g( u
T
minimize: ZYO W WV i v
jd w
314

(9.11)

9! Special purpose design methods for surface structures

ZH
conditiions: X
YC

O W g ( u g Rc u
i v' i v;
Cd UV j d w j C s w

r <0

(9.12)

where W dependss on the chhoice made regarding the quantiity to be m


minimized (volume,
(
blem gives the design parameterss and the
weight, etc.). The solution off this optimiization prob
T method
d can easilyy be extendded to seveeral load
correspoonding bennding momeent field. The
cases buut this will not
n be discuussed here.

Examplle 4
In this example,
e
ann optimal reeinforcemennt arrangem
ment is sought for the sllab in Figurre 9.7. It
is rectanngular and a/b = 2/3. One
O side a is
i clamped and one sidde b is simpply supporteed, while
the rem
maining two sides are free.
f
At thee corner off the two free sides, thhe slab supp
port is a
column. A uniform
m load is appplied. The cost of eacch yield mooment is assumed prop
portional
with itss magnitudee and the saame proporrtionality faactor is usedd for all yieeld momen
nts. If all
four plaastic momeents (layers in the x- and
a y-directions; uppeer layer andd bottom laayer) are
taken eqqual and hoomogeneous over the full
f plate, only
o
one deesign param
meter occurss. In this
case thee total costt is called 100%. If the
t four plastic momeents can bee different but still
homogeenous over the
t full plate, four design parameters occur. Now
N the cosst decreases to 89%.
A thirdd optimization run is done
d
in whhich three regions
r
are distinguishhed in the plate,
p
as
shown in
i Figure 9.7 and each with its own material group.
g
In eaach region ffour differen
nt plastic
momentts can deveelop, so 12 design paraameters now
w occur. Now the costt reduces fu
urther to
64%. The total cosst of four layers reinfoorcement peer unit areaa is for the three regio
ons 4.07,
1.73 andd 3.02, resppectively. Inn region I all
a four plasttic momentts are substaantial (highest layer
cost 1.226), in regioon II mainlyy the bottom
m layers are important (highest
(
layyer cost 0.77
7) and in
region III
I only a hiigh amountt of reinforccement is reequired in the
t upper laayer in the direction
d
of side b (layer cosst 2.98).

Figure 9.7:
9 Element mesh
m
and diffe
ferent materiall groups.

Remarkk
In somee codes of practice, ruules are givven about th
he distance that is perrmitted betw
ween the
elastic and
a plastic solution in order to guuarantee sufficient defformation caapacity. This fits in
directlyy in the LP-sscheme just as another restriction.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

315

9.3.5

Plates subjected to in-plane loads

Shear-walls and deep beams are plates with in-plane loads. For this type of structure a similar
procedure can be followed. Here we restrict ourselves to a short indication of the possibilities.
Now triangular elements are combined with bar and beam elements. The triangles represent
the distributed reinforcement and the bar elements the concentrated reinforcement. The
stresses in the triangles are .xx, .yy and .xy. The stress resultant in the bar is a normal force N
and in the beam a bending moment M. Again, equilibrium is fully satisfied by choosing a
linear stress state in the triangles, a parabolic distribution for the normal force in the bars and
a cubic moment distribution in the beams. The linear stress distribution in the triangle
(parameters .) results in four generalized forces on each side, as shown in Figure 9.8.
Through these generalized forces, transfer takes place to adjacent plate elements, bars (see
Figure 9.9) and beams (see Figure 9.10). The unknown stress parameters . in the triangles, N
in the bars and M in the beams are assembled again in the vector $.

Figure 9.8: Triangular element with generalized forces on the edges.

Figure 9.9: Stress parameters and generalized forces on the edges.

Figure 9.10: Stress parameters and generalized forces for a beam element.

316

9! Special purpose design methods for surface structures

In the case of in-plane loads the yield criterion is chosen proposed by Nielsen (1984) and
given by the equations

7( f t x 7 + xx )( f t y 7 + yy ) 8 + xy2 ? 0
7( f c 8 + xx )( f c 8 + yy ) 8 + xy2 ? 0

(9.13)

where fc is the compressive strength and f t x and f t y are tensile strengths (limit tensile forces in
reinforcement per unit length divided by the thickness of the plate). The yield criterion is
drawn in Figure 9.11.

Figure 9.11: Yield criterion for a plate loaded in plane stress.

Here we skip the discussion how to linearize the yield criterion. It suffices to say that it is
done differently for load optimization and for material optimization. The linear yield criteria
for a bar and a beam , respectively, are given by

7 Nc ? N ? Nt
7 M p7 ? M ? M p8

(9.14)

where Nc is the limit compression force, Nt the limit tension force, M p8 the positive plastic
moment and M p7 the negative plastic moment. In a beam with both normal and bending forces,
a combination of N and M is necessary. In the case of load optimization where the strengths
are known, a linearization as shown in Figure 9.12a is used. Whereas for the case of material
optimization, the linearization shown in Figure 9.12b is used in order for the restriction to be
linear in the material parameters.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

317

(a)

(b)

Figure 9.12: Linearization of the yield surface for:(a) load optimization, and (b) material optimization

9.4

Design of slabs using the reinforcement field approach

In this Swiss modelling methodology, the aim is to provide a safe lower bound solution and
providing a practical reinforcement arrangement with a minimal reinforcement weight. Only
the two basic concepts and their application to slabs subject to bending loads are discussed
here and the reader is referred to Anderheggen et al. (1995) and Steffen (1996) for more
details of the method. Furthermore the discussion is restricted to the subject of material
optimisation. For ultimate load determination, an incremental approach is used on basis of
non-linear constitutive materials law, which is outside the scope of this report. A related
method for plates loaded in plane is not discussed here. For this application is referred to
Despot (1995).
9.4.1

Linear yield conditions for element nodal forces

The approach is based on the standard linear finite element method. Each finite element is
regarded as an independent dimensioning unit that has to withstand the forces transmitted to it
from its neighbouring elements (Figure 9.13).

Figure 9.13: Nodal forces for a four node plate bending element

318

9! Special purpose design methods for surface structures

This means that unlike the usual method of formulating dimensioning constraints on the stress
components, the dimensioning conditions are imposed directly upon the (generalized) element
e
e
e e
nodal forces f . With f = K u these forces, which satisfy equilibrium with the external
e
(nodal) loads exactly, are easily obtained from the nodal displacements u and the element
e
stiffness matrix K . Starting from a linearized yield criterion, the yield condition for a finite
e e
e
e
element can now be written as C f ? cs , with cs corresponding to a vector of yield limits
e
e
e
based on the material strength. Assembling C , cs and f of all finite elements into the global
matrix C and the global vectors Cs and f,the yield condition for the slab now reads:
C f * Cs

(9.15)

In Eq. 9.15, f corresponds to the stress state in the structure. In the case of a linear-elastic
analysis f is equal to fel, which is easily calculated from the elastic nodal displacements uel.
9.4.2

Material optimisation through stress redistribution

According to the lower bound theory of plasticity, the superposition of any self-equilibrating
stress state on a stress state, which satisfies equilibrium with the external loads does not alter
the load-carrying capacity of the structure. This opens the way to optimisation through stress
redistribution; as long as equilibrium and yield conditions are satisfied, the stress state and
thus the global nodal force vector fel obtained by linear-elastic analysis can be modified at
will to reduce the total amount of steel. This is done with the introduction of fictitious plastic
strain distributions within the finite elements. For each considered strain distribution of free
amplitude $i, a global element nodal force vector fi is determined by linear-elastic analysis.
Assembling fi and $i of all strain distributions into the matrix F and the vector $ the condition
of Eq. 9.15, including stress redistribution, now reads:
C fel + CF 0 * Cs

(9.16)

Adding a number of design variables d, corresponding to the steel content of considered


reinforcement domains, the total reinforcement weight of the structure is written as Wd. With
Cs = Cd d we now make the yield criteria linear in the design variables and get the following
optimisation problem:

g( u
T
minimize: ZYO W WV i v
jd w
conditions: pCF

g( u
7Cd q i v ? C fel
jd w

(9.17a)

(9.17b)

This method is easily applied to multiple load combinations if instead of Cfel the
corresponding envelope values of all considered load combinations are used. Minimum values
for the design parameters can also be specified, if required for strength limit or serviceability
conditions. It should be noted here that the concepts presented above are very general and can
be applied to all structure types like slabs, walls or shells. For the procedure to derive linear
yield conditions for different types of finite elements and the strategy of generating useful
fictitious plastic strain distributions is referred to Anderheggen et al. (1995).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

319

9.4.3

Slab subjected to bending loads

The reinforcement layout of a slab consists of a series of domains of reinforcement, called


reinforcement fields, with known locations and directions of reinforcing bars but unknown
steel content d. The possibly overlapping reinforcement fields can be independently defined
for the top and bottom steel. The yield condition Eq. 9.15 is derived from the yield criteria of
Figure 9.2 through linearization. Assuming a constant inner lever arm per finite element and
reinforcement direction, Cd d in Eq. 9.17b then corresponds to the plastic yield moments in
each of the elements.
9.4.4

Dimensioning procedure

Following the goal of the method, which is providing a method for practicing engineers, the
dimensioning procedure is implemented as a postprocessor to a standard linear-elastic finite
element analysis. The three steps of the procedure are the following:

Step 1: Standard linear-elastic analysis


The structure is first analysed for all independent load cases by means of linear-elastic
analysis. Then, based on the condition Eq. 9.15, the required elastic reinforcement content in
every finite element is calculated in both directions on the top and bottom layers.

Step 2: Elastic reinforcement layout


Next, the engineer defines the conceptual reinforcement layout by graphically entering the
location and direction of a series of reinforcement fields for the top and bottom steel, in
certain zones possibly specifying the minimum reinforcement content according to the code.
The choice of reinforcement fields should mainly be based on his experience and knowledge
of a practical reinforcement layout and not exclusively on the results of Step 1as, at this point,
no fictitious plastic strain distributions are to be considered, Eq. 9.17 can be written as
minimise:

Wd

conditions: - Cd d # - C fel

(9.18a)
(9.18b)

The solution of this optimisation problem is the elastic reinforcement layout that, due to
undesirable elastic stress peaks, in general, results in an excessively large reinforcement
content.

Step 3: Optimised reinforcement layout


In order to optimise the elastic solution the program now automatically generates fictitious
plastic strain distributions F and solves the optimisation problem of Eq. 9.17, minimizing the
total steel weight. The optimisation algorithm works in steps, meaning that the full F is not
introduced at once. In fact, starting from the elastic solution new plastic strain distributions
are added in increments. After each carefully selected increment, an optimal solution is
obtained. New optimisation increments are added until the reinforcement appears to be
satisfactory to the engineer or until relevant savings are no longer possible.
The methodologies described above have been incorporated into commercial software and is
just one example of how optimisation techniques can be implemented into the design of slab

320

9! Special purpose design methods for surface structures

and wall structures. As for any non-linear approach to design, however, the methods should
only be applied to appropriately verified problem types.

9.5

Design of shear-walls: the stringer-panel approach

The stringer panel method, developed in the Netherlands, is intended for the design of plane
structures like shear-walls, beam-column joints, beams with openings and dented beam. The
method can be used for the serviceability limit state (crack-widths) and the ultimate limit state
(ultimate strength). Moreover it can handle multiple load conditions.
Basically, just two elements are used, a stringer element (straight bar) and a panel element
(rectangle or quadrilateral) as shown in Figure 9.14. This corresponds to the observation that
main reinforcement often occurs in bundles at the edges of structures and around holes
(represented by stringers) and distributed net reinforcement is applied between the bundles
over large parts of the structure (represented by panels). The idea to apply stringers and panels
has been used earlier in limit design for plane stress states, see Nielsen (1999).

reinforcement net

Panel

panel normal force


panel shear force

reinforcement bar
Stringer

stringer force

Figure 9.14: Stringer and panel element

9.5.1

Linear-elastic version

In a simple orthogonal version of the method, the material behaviour of stringers and panels
is kept as linear-elastic; constant shear forces occur in the rectangular panels and normal
forces occur in the stringers. Full equilibrium is assured in the panels, the stringers and in
the interface between panels and stringers. The latter implies that constant shear panels must
go together with linearly varying normal forces in the stringers. The method is generalised
to quadrilateral panels, which gives more freedom for irregular structure shapes. Figure 9.15
shows this general panel configuration.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

321

f3
u3
y
f4

u2

u4
x

: Q

f2

u1
f1

Figure 9.15: Quadrilateral constant shear panel.

The stress state in the total structure is fully equilibrating and the principal of minimum
complementary energy (Ec) is applied. This principal is modified in order to work in the
framework of the stiffness method by adding the inter-element equilibrium equations,
multiplied by a Lagrange multiplier, to the energy functional. This extended functional must
be minimised with respect to both the stress parameters and the Lagrange multipliers. The
multipliers can be interpreted as average displacements u in the interface between the
panels and stringers and displacements in the joints where adjacent stringers meet. For the
total system, the modified energy functional is written as:
Ec '

1 T
( F ( 7 ( T Bu 7 f T u
2

(9.19)

where $ is a vector of stress parameters, F is a flexibility matrix, B a kinematics matrix which


is dependent on the topology of the element mesh and f is the load vector. Load can be
applied in the direction of each degree of freedom, so f and u are associated. The functional
must be stationary with respect to the stress parameters $ and displacements u:
sE c
' F( 7 Bu ' 0
s(

(9.20)

sE c
' BT ( 7 f ' 0
su

(9.21)

From these equations, we solve the relationship between the stress parameters and the
displacements and the relationship between the stress parameters and the load:

( ' DB u

(9.22)

BT ( ' f

(9.23)

where the rigidity matrix D is the inverse of the flexibility matrix F. Substitution of the first
equation into the second yields the set of equations and the stiffness matrix K of the global
system:

322

Ku ' f (global set of equations)

(9.24)

K ' BT DB (global stiffness matrix)

(9.25)

9! Special purpose design methods for surface structures

The global stiffness matrix can be assembled form the individual element stiffness matrices.
After solving the equations Ku = f, the stress parameters can be calculated from the equation
$ = DB u. For details reference is made to Blaauwendraad et al. (1996) and Hoogenboom
(1998).
9.5.2

Non-linear version

In the non-linear version of the model, the panel concrete material can both crack and crush
and the panel reinforcement can yield. In this version, the normal forces must be included in
the panels. Again, equilibrium is maintained within the panels, the stringers and on the
interface of the elements. This is achieved by choosing an appropriate stress state in the
panels as shown in Figure 9.16 for a square element. The stress field varies linearly in the xand y-directions giving
2x
Z
1 0 0
Z+ xx W X
a
X
U X
0
X+ yy U ' X0 1 0
X+ xy U X
Y
V X0 0 1 7 2 y
XY
a

WZ ( W
0 UX 1 U
(2
2 y UX U
U X (3 U
b UX U
2x (4
7 UU XX UU
b V Y (5 V

(9.26)

In Eq. 5.38, +xx and +yy are normal stresses and +xy is the shear stress in the panel material.
Variables (1 to ( 5 are generalised stresses and a, b are panel dimensions. Five independent
stress modes occur that are in equilibrium. A generalisation to quadrilaterals has been made.

Figure 9.16: Stress modes of a square panel with non-linear behaviour.

The non-linear stringer behaviour in tension can account for tension stiffening after cracking
and yielding of the reinforcement at ultimate, see Eurocode (1991). The non-linear panel
behaviour is according the modified compression field theory of Vecchio and Collins (1986).
In Figure 9.17a the characteristic force-strain curve for a stringer in tension is drawn. In
Figure 9.17b an impression is given of the relationship between the shear stress and the shear
strain in the panels at given values of the normal strains.
9.5.3

A three-step design procedure

The dimensioning procedure for shear-walls and/or beam details consists of three steps, one
of which is elastic and two are non-linear, Hoogenboom (1998), Blaauwendraad et al. (2002).
Hereafter the subsequent three steps are explained.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

323

force

step 1

step 2
step 3

shear force

step 1

step 2 and step 3

yielding
cracked
reinforcement only
uncracked

shear strain

strain

(a)

(b)

Figure 9.17: Non-linear characteristics for stringers and panels: a) characteristic force-strain curve for a
stringer in tension; b) relationship between the shear stress and the shear strain at each steps.

Step 1: Elastic analysis


In this step the simple linear model is used. Thus, normal forces occur in the stringers and
shear forces occur in the panels, which is fine for the elastic phase. For this first step it does
not matter so much what sizes of cross-sections are assigned to the stringers. Application of
rules of thumbs is sufficient. The panels have the wall thickness and a linear-elastic analysis
for all load combinations must be performed. The force distribution achieved in this way is a
first indication for the determination of the reinforcement in the tensile stringers and the mesh
reinforcement in the panels.

Step 2: Non-linear
Now one can prepare for a non-linear analysis in which account is taken of cracking in the
tensile stringers and cracking and yielding in the panels. In this second step, the non-linear
models are used and, thus, the stress state in the panels is extended to shear and normal
stresses. In the non-linear step it does matter that one enters the correct cross-sections of the
stringers. In compressed stringers the cross-section determines the compressive force, which
has to be compared with the ultimate design strength. In tensile stringers, the assigned
concrete area determines the contribution of the concrete to the tension stiffening of the
stringer. For this analysis, the results from the elastic step are used as a first estimate.
All input quantities being determined now and entered into the program, the non-linear
calculation is performed. The loading is increased incrementally until the ultimate design load
is reached (load factor is 1) using a Newton-Raphson procedure to solving the non-linear
equations. Because cracking in the stringers and yielding in the panels occur, a non-linear
load-displacement diagram will result. In this step no yielding of the reinforcement in tensile
stringers is considered in order to receive at a robust design tool. In case a tensile stringer
would reach its tensile yield strength, the cracked branch in the force-strain diagram of the
stringer is artificially extended. From the analysis results it will become clear whether or not
the ultimate tensile strength of a stringer has been surpassed for any of the load combinations
and, if so, the reinforcement has to be increased. The crack-widths at service loads can also be
inspected and the reinforcement adapted, if needed. Due to redistribution of stresses and the
enriched capacity of the panels in this second step, it also may occur that the reinforcement in
a tensile stringer can be reduced.

324

9! Special purpose design methods for surface structures

Step 3: Final simulation


All decisions made on basis of the non-linear analysis are entered in the model and then the
final step is started. The load is incremented until failure occurs. In this final simulation, no
restrictions on the non-linear responses are made; the reinforcement in panels and stringers
can yield and concrete can crush. If the ultimate load factor is greater than 1.0 and the crack
widths are within serviceability bounds, no further iterations are needed.
If done in this way, a robust and fast program can be made. The elastic analysis (step 1) is
done more or less instantaneously and the non-linear analysis (step 2) and final simulation
(step 3) only require a few minutes on moderns PCs. In fact, the time involved with the initial
modelling of the structure and the professional decisions to be made by the engineer are
determining for the duration of the design process.
Finally, particular attention must be paid to detailing and anchoring of the steel reinforcement
in both the stringer and panel elements.
9.5.4

Example

The stringer-panel method is demonstrated with the design of a structure, which has been
previously used in a Swiss study, see Despot (1995). The structure is shown in Figure 9.18.
The stringer-panel model is shown in Figure 9.19. In principle only stringers are chosen along
edges of the structure, along edges of holes, where line loads or point loads are applied and
where supports occur.

0.85

3.00

1.00

8.00

2.15

150 kN

1000 kN
150 kN/m
2.15

3.00
100 kN/m

5.00
100 kN/m

2.35

2.00

0.50
100 kN/m
0.50

1.00 1.50

6.30

0.90

4.80

15.00 m

Figure 9.18: Deep beam example (dimensions in m).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

325

113 kN

240 kN

1356 kN

332 kN

281 kN

324 kN

356 kN

190 kN

190 kN
110 kN

310 kN
185 kN

238 kN

221 kN

398 kN

236 kN

Figure 9.19: Stringer-Panel model for the deep beam of Figure 9.18.

The first elastic step in the analysis results in the stress distribution shown in Figure 9.20,
which shows linearly varying normal forces in the stringers and constant shear stresses in the
panels. On basis of this elastic stress distribution reinforcement is chosen. The reinforcement
in the stringers is concentrated and in the panels a two-way mat reinforcement is adopted.
This reinforcement scheme is given in Figure 9.21. In general, codes of practice will prescribe
minimum reinforcement percentages for wall-type structures that is sufficient for large parts
of the structure. Only local additional panel reinforcement may be needed.
Next step 2 is started, in which cracking of the stringers is permitted and yielding and
crushing in the panels is allowed for. It is important to have selected proper cross-sectional
areas for the stringers at the conclusion of step 1, at least for the tensioned stringers, because
the tension stiffness depends on the cross sectional area.
Figure 9.22 shows the resulting load-displacement diagram for the structure. The load factor
is stepwise increased until a load factor of 1 is reached. The resulting stress distribution is
shown in Figure 9.23. Substantial redistribution does occur if compared with the stress
distribution for the elastic step in Figure 9.20. This means that the reinforcement has to be
adapted. More reinforcement is required at the top of the hole, while in the lower edge of the
structure the reinforcement can be reduced. Figure 9.24 shows crack-widths for the
serviceability state. Clearly too large crack-widths do occur if one wants to limit them to
0.3 mm. This is another reason to adapt the reinforcement.
Figure 9.25 shows a revised (adapted) reinforcing scheme. On basis of this reinforcement, the
final simulation of step 3 is made in which the load factor is increased until failure. Figure
9.26 shows the load-displacement curve for the resulting analysis and. Figure 9.27 shows the
crack-widths at the serviceability load. Finally, in Figure 9.28 the deformation mechanism at
failure load is plotted. As the crack-widths are within the specified permissible range and the
ultimate load factor is larger than 1, the design is completed.
In this example, more than half of the reinforcement weight is due to the distributed
reinforcement in the panels. Other examples also suggest that the distributed minimum
reinforcement is, largely, the determining factor of the amount of reinforcement. Therefore,
the total amounts of reinforcement in the elastic layout and after redistribution do not differ
substantially. However, the position where the concentrated reinforcement is required changes
substantially.

326

9! Special purpose design methods for surface structures

552 kN
88

659

353

354

36

142

779 kN
313 kN
582

235

749 kN
342

26

647

189

359

649 kN

Figure 9.20: Elastic stress distribution.

216

12-150

320+216

420+216

216
A

12-180

220

216

12-300 at both faces

420+16

Figure 9.21: Reinforcement scheme on basis of elastic stresses.

1.1
1.0

load factor

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

displacement [mm]
2

10

12

14

16

18

20

Figure 9.22: Load-displacement diagram in non-linear design step 2.

566 kN
3.9

522 kN
2.5 MPa
1.6

347 kN
3.1

1233 kN

2.9

2.0
540 kN

Figure 9.23: Non-linear stress distribution in design step 2.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

327

0.04 mm

1.10
0.34
0.90
0.44

0.25 mm

Figure 9.24: Crack-widths in SLS in design step 2.

12-150

620+416

220+316

416

12-180

216

220

12-300 at both faces

616

A: 7-150 in both directions on both faces

Figure 9.25: Reinforcement arrangement on the basis of the non-linear stresses of step 2.

1.1
1.0

load factor

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

displacement [mm]
2

10

12

14

16

18

20

Figure 9.26: Final load-displacement curve in design step 3.

328

9! Special purpose design methods for surface structures

0.13

0.04 mm
0.19

0.28
0.21

0.28 mm

Figure 9.27: Crack-widths at serviceability state in design step 3.

Figure 9.28:Deformation mechanism at failure load in design step 3.

9.6

References

Anderheggen, E., Despot, Z., Steffen, P., Tabatabai, S., and Tabatabai, M.R. (1995), Finite
Elements and Plasticity Theory: Integration in Optimum Reinforcement Design, 6th Intl
Conference on Computing in Civil Engrg and Building Engrg, Berlin, pp. 653-660.
Blaauwendraad J, and Hoogenboom P.C.J. (1996), "Stringer Panel Model for Structural
Concrete Design", ACI Structural Journal, Vol. 93 No. 3, pp. 295-305
Blaauwendraad, J. and Hoogenboom, P.C.J. (2002), Design instrument SPANCAD for Shear
Walls and D-regions, Proceedings of the first fib Congress 2002, Concrete Structures in the
21st Century, Osaka, Japan, October, 2003, Vol 2, p. 411 - 416, in CD ROM.
Damkilde, L., and Krenk, S. (1997), LimitS A System for Limit State Analysis and
Optimal Material Layout, Computers and Structures, Vol. 64, Issues 1-4.
Despot, Z. (1995), Methode der finiten Elemente und Plastizittstheorie zur Bemessung von
Stahlbetonscheiben (Finite Element Method and Plasticity Theory for Dimensioning of
Reinforced Concrete Disks), Institut fr Baustatik und Konstruktion, ETH Zrich, IBK
Bericht Nr. 215, Birkhuser Verlag, Basel, (In German).
Eurocode (1991), Common Unified Rules for Concrete Structures, ENV1992-1-1: pp. 171174.
Hoogenboom, P.C.J. (1998), Discrete Elements and Non-linearity in Design of Structural
Concrete Walls, doctoral thesis, Delft University of Technology.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

329

Krabbenhoft, K., and Damkilde, L. (2002), Lower Bound Limit Analysis of Slabs with
Nonlinear Yield Criteria, Computers and Structures, 80 (220), pp. 2043-2057.
Krenk, S., Damkilde, L., and Hoyer, O. (1994), Limit Analysis and Optimal Design of Plates
with Equilibrium Elements, J. Engrg. Mech., Vol. 1200, Issue 6, pp 1237-1254.
Nielsen, M.P. (1964) Limit Analysis of Reinforced Concrete Slabs, Copenhagen, Acta
Polytech. Scand., Civil Eng. Build. Constr., Serial No. 26, 167 pp.
Nielsen, M.P. (1999), Limit Analysis and Concrete Plasticity, second edition, ISBN 0-84939126-1, Prentice-Hall, London, (first edition 1984).
Poulsen, P.N., and Damkilde, L. (2000), Limit State Analysis of Reinforces Concrete Plates
Subjected to In-plane Forces, International Journal of Solids and Structures, 37, pp. 60116029.
Steffen, P. (1996), Elastoplastic Dimensioning of Reinforced Concrete Slabs by Finite
Dimensioning Elements and Linear Programming, PhD-thesis ETH Zurich, Switzerland (in
German; original title: Elastoplastische Dimensionierung von Stahlbetonplatten mittels Finiter
Bemessungselementen und Linearer Optimierung).
Vecchio, F.J., and Collins, M.P. (1986), The Modified Compression-Field Theory for
Reinforced Concrete Elements Subjected to Shear, ACI Journal, V.83, No. 2, pp. 219-231.
Wolfensberger, R. (1964), Traglast und optimale Bemessung von Platten, Zurich, PhD
dissertation.

330

9! Special purpose design methods for surface structures

10 Concluding remarks
10.1 Introduction
This Practitioners Guide to Computer Based Modelling of Structural Concrete is intended to
summarize the basic knowledge required for use of nonlinear analysis methods as applied to
practical design, construction and maintenance of concrete structures. The contents of this
guideline will be reviewed, once again, using several examples of performance based design
schemes to illustrate how each chapter is applied. Note that in this guide, short-term static
loads under normal ambient conditions are primarily considered and the mechanical
properties of structural concrete and steel are assumed unchanged during the service life of
the structure. Note also that nonlinear analysis and mechanical models of structural concrete
can also be applied to the performance assessment of existing structures in service, although
the input analysis data and boundary conditions would differ from those for newly constructed
or future planned structures. This chapter summarizes the manner in which nonlinear
modelling can be practically applied; design limit states and safety factors are also briefly
discussed in relation to nonlinear structural analysis.

10.2 Structural performance based design in practice


Figure 10.1 shows a flowchart of a limit state design or performance assessment scheme in
which nonlinear modelling and analysis may function. In general, there exists a set of design
requirements that must be satisfied for the various design loads acting on the structure
(including natural ambient conditions such as temperature, humidity, radiation, rainfall and
water supply). Structural shapes, sizes and dimensioning must be selected such that they
satisfy requirements relating to safety, serviceability, durability and fatigue. Elegance (i.e.,
aesthetics) and environmental issues must also taken into account. Basically, this decision
making process proceeds on a trial-and-error basis since there may exist an infinite
solutions meeting the specified requirements and design conditions. If we begin with overlysafe details in the initial stage of the design cycle or, on the contrary, if very unsafe ones are
chosen, then it may take many iterations to reach the solution that enables us to achieve a
technically, economically, and environmentally efficient design. If a particular requirement of
safety is known to be the governing one, we can apply strut-and-tie and stress-flow based
nonlinear modelling to decide the arrangement of the reinforcement and associated details in a
reasonable manner (Chapters 2, 8 and 9).
Mechanics-aided design procedures present a means of realizing the load-carrying mechanism
which was assumed in advance. The inversely decided design details regarding reinforcement
and thickness of members can be shown to provide sufficient capacity that is not below the
assumed static design loads, provided that some ductility is accomplished. Then, if other limit
states are inconsequential, subsequent performance assessment by nonlinear analyses with
respect to safety limit states is not necessary. There is a possibility of over-design in capacity
if an appropriate stress flow is not assumed, and, in some cases, dense reinforcement may
cause heavy construction of concreting works and elevate costs. At this point, the capacity can
be verified by running nonlinear analyses and re-design is possible to provide a refined
assessment of strength and improved detailing and dimensioning. As the successful strut-tie
modelling under static loads is carried out under a single specified load, we have to check
other performances required if those are thought to be possibly substantial. For example,

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

331

Nonlinear Modelling $Performance Assessment


load+required
performance 1

load+required
performance 2

load+required
performance N

Chapter 2,5,8

design process (creative jobs)


stress-flow imagination, experiences, free-hand drawing, etc.

shapes,
shapes,dimensions,
dimensions,detailing,
detailing,B.C.,
B.C.,
materials,
arrangements,
etc..
decided
materials, arrangements, etc.. decided
performance
check needed?

No

END

Yes
load+required
performance 1

load+required
performance 2

load+required
performance N

selection of appropriate structural


& material modeling
for case-1
selection
of appropriate structural
& material modeling for case-N

Chapter 3,4,5,6

behavioral simulation output


behavioral simulation output
structural responses
structural responses
and material states for case-1
and material states for case-N

No

performance
case-1 satisfied?

Chapter 3,4,5,6
Chapter 7

performance
case-2 satisfied?

Yes

Yes

performance
case-N satisfied?

No

Yes

END

Figure 10.1: Flowchart of structural design and performance assessment.

fatigue and durability requirements under service loads have to be examined after deciding
design details by the strut-tie model.
The performance assessment shown in the lower part of Figure 10.1 is conducted for each set
of structural requirements and factored designed loads. Here, the decision making process of
shape, dimension, detailing and material arrangement has nothing to do with the following
behavioural simulation of the structure. Designers must select the most appropriate structural
(Chapter 4 and Chapter 5) and material modelling (Chapter 3 and Chapter 6) for computing
responses based on which the performance requirement is to be verified.
After computing the design responses of the structure and constituent concrete and steel,
judgment on acceptability of the computationally simulated behaviours has to be made. Here,
the following issues have to be decided in practice:
i) What sort of indicators are mechanically appropriate for limit states or design criterion
in order to quantitatively express safety and functionality requirements?

332

10 Concluding remarks

ii) What is the appropriate limit value corresponding to the level of required
performance?
iii) How reliable is the computed indicator and the limit value for performance
verification? In other words, what is the reasonable safety factor of nonlinear analysis?
For items (i) and (ii), we can use the same set of indicators and limit values as that used in
current design schemes based on linear and nonlinear analyses, that is, sectional forces and
capacities, stress and strength, etc. In this case, nonlinear analysis can be used for computing
the sectional forces as well as section capacities that cannot be evaluated only by on-going
codified equations and formula in practice (Chapter 1). Here, the capacity computation by 2D
and 3D nonlinear analyses (Chapters 4 and 5) can contribute to decisions. As multidimensional nonlinear analyses convey more detailed information on internal damage and
material states, as well as displacements and section capacities, that that obtained by more
conventional methods of analysis, more mechanics friendly and consistent indicators may be
used in making decisions when adopting a non-linear modelling approach. This is discussed
further in Section 10.3.
For item (iii), one must quantitatively evaluate the reliability of computed responses obtained
from nonlinear analysis in terms of each performance indicator. More specifically, the safety
factor, which is applied to the indicator derived from the utilized nonlinear modelling, has to
be specified by conducting a systematic experimental verification. Evaluation of the overall
(global) reliability of the computer simulation is crucial and high-quality database of
experiments (Chapter 7) is of great importance. In public works, these highly technological
issues are the tasks of in-house engineers who are responsible for public financing and
project/risk management, or code writers who produce statements on the use of nonlinear
analyses.

10.3 Benefits of non-linear modelling and analyses


Rational design may realize both reasonable cost and performances of concrete structures.
When some technology is advanced and can be properly used, the total cost may be generally
reduced if the targeted performance is specified. For conventional beams, slabs and columns
that occur in the majority of buildings and bridges, room of improving cost performance is
minimal since these widely used members are more or less reasonably efficiently designed in
practice. However, there exist opportunities of rational design of structures where there is a
shortage of past experiences and there exists a strong nonlinearity with the surrounding
media. Figure 10.2 shows a cost benefit analysis of one of the worlds largest underground
LNG storage tanks (JSCE, 1999). It consists of thin reinforced concrete side walls to bear
horizontal static/dynamic loads of soil and underground water pressures and a large circular
reinforced concrete tank base subjected to high static shear forces induced by underground
water pressure and its dead weight.
After the Hyogo-ken Nanbu earthquake (Kobe, Japan) in 1995, the design seismic loads were
re-evaluated and increased. If conventional design procedures are applied without revision,
the increased loading would lead to about a 25% increase in the cost of construction. This
situation led to an engineering motivation to drastically change the design scheme from a
capacity method by simplified elastic analysis to a performance based one with full 3D
nonlinear coupled soil-RC analysis. This makes for a substantial cost reduction while safety
and durability performances remain unchanged. A more than 20% cost reduction was realized

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

333

S 4/0+(.8-'

K125L
L2H

120

X 140-30-8+YQ+
0-, F41)*+Y)11(-1/

K120L
L2H

A -/(7.+/-(/, (3+*4)8
Z F71)8-8+)90-1+$[[#
\41+C%] +3*)//

110

K100L

80

conventional design

100

90

V99-30+49
M -8W3-8+
/0(99.-//

L2L
L2L

X 140-30-8+YQ+
0-, F41)*+
Y)11(-1/
M -8W3-8+
/0(99.-//

O )9-0Q+9)3041+
Z F71)8-8+941
O P-)1+1-(.9"

O P-)1+3)F)3(0Q
\41, W*)+4.+/(N->-99-30

By K. Miyamoto, TEPCO

M )0(4.)*(N-8+
O P-)1+3)F)3(0Q+
R
M S T/4(*+.4.*(.-)1(0Q
U+?A +).)*Q/(/

$%#^!!!, ? 3)F)3(0Q+(.0"+8(), -0-1_+`!, ++


P-(7P0+_+@!, abc )**+0P(3:.-//
%, +941+/(8-+c )**++d>$!, +Y4004, +/*)Y

Figure 10.2: Cost performance analysis of nonlinear analysis based design for underground LNG storage tanks
(JSCE 1999).

even though the required safety performance against earthquake was very much raised, as
shown in Figure 10.2. For designing the sidewalls, nonlinearity of 3D in-plane shear and
transverse flexure (Chapter 5) of cracked RC shell elements was explicitly taken into account
and interaction with surrounding soil foundation was also analyzed using 3D FEM. This
nonlinear simulation resulted in reduced internal forces of the main body and a relatively
slight increase in the reinforcement even though a significantly high earthquake design load
was imposed. A key factor of the base slab design was the new 3D size-effect modelling of
punching shear failure because of its thickness (8-10m) with the disc span of approximately
100m. Thus, it was crucial to rationally estimate the transverse shear capacity for deciding on
the quantity and arrangement of shear reinforcement (Chapter 1).
As the total construction cost of an underground tank is massive, just a small cost reduction,
relative to the overall cost of the project, more that offsets the increased expense of the works
undertaken in the design office for a full 3D nonlinear analysis of the structure. In-house
engineers in charge of financing and energy development conducted experiments for verifying
their modelling of both the element and the structure. Finally, the global safety factor of overall
structural analysis was discussed and authorized by an independent academic sector in terms of
seismic performance limit state enabling re-use after the near and far field great earthquakes of
1000-years return period (JSCE, 1999). The expense of the large scale experiments was easily
covered within the actualized cost saving obtained from results of the nonlinear analyses.

334

10 Concluding remarks

10.4 Code provisions


As stated in Chapter 3, practitioners are required to assess the accuracy of their nonlinear
analyses and modelling as a design tool using a proper safety factor. Reliability or accuracy of
nonlinear analysis has to be examined as a set of material and structural models used because
of their nonlinearity. In other words, accuracy of individual modelling does not always stand
for the accuracy of entire computational system. It should also be emphasized that functioning
of computer programs or nonlinear analysis methods should be checked for each outcome.
For example, the appropriate safety factor for computed capacity is not necessarily the same
as that for ductility (deformability). Thus, the overall evaluation of program applicability is
one of the main concerns for practice and experimental verification of systematically arranged
parameters is essential.
This guideline can not present quantitative assessment of a set of nonlinear models for
particular design items of limit states but an outline is provided. Some design codes or
recommendations include clauses on nonlinear structural and material models. At this point in
time, conservative modelling is advocated due to shortage of experience. This guideline is
also presented so as to facilitate in further codification of non-linear modelling.

10.5 Specification of design loads


In some cases, the design loads are specified in consideration of the characteristics of the
structural analysis and the ease of the design process and do not, necessarily, represent the
realistic loading conditions. Thus, when practitioners apply load-path dependent nonlinear
analysis for performance assessment, careful checking of specified loads must be made as
code specified design loads are generally defined on the basis of load-path independent
structural analysis, such as for linear elasticity.
Figure 10.3 shows the bending moment diagram computed by the nonlinear smeared crack
analysis of a prestressed concrete viaduct. For the examination of the ultimate limit state, the
heavy loads are placed solely on the centre span of the bridge deck. If these specified loads
are just put on the centre span monotonically, the moment at the section of the span centre is
computed to be small, because the right and the left wing spans have greatly higher stiffness.
But, this loading history is unrealistic. As the design load of bridges represents the traffic, the
heavy live loads cannot reach the centre span without passing through the wing elements. If
we apply the specified ultimate limit state load after operating the serviceable loads over the
span, the computed flexural moment becomes larger due to the reduced stiffness of the wing
spans with bending cracks caused by the service pre-loads. It should be noted that the linear
elastic analysis results in reasonably better moment profiles because the sectional stiffness
profile over the whole span is assumed uniform and is closer to eventual reality than the case
of a nonlinear analysis with an unrealistic load history, no matter how unrealistic is the linear
elastic analysis.

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

335

specified patterns of design loads

Moment diagram: The specified Load is monotonically applied


just on the middle span.

Moment diagram: The specified Load is finally applied


after passing through the whole span of the bridge.

Figure 10.3: Loading pattern and path-dependency on nonlinear analysis based design.

10.6 Maintenance
As introduced in Chapter 1, nonlinear analyses of collapsed real structures have greatly
contributed to investigation of the causes of failure. In reality, lessons from bitter experiences
have been driving forces for further improvement of design methods. Currently, nonlinear
analysis is expected to play a role of maintenance issues for real structures in use. Owing to
recent advances in non-destructive testing and sensing technology, much information related
to mechanical damage of constituent materials becomes available (e.g., crack location, crack
depth, crack width, corrosion level of steel inside concrete, yield hysteresis of steel,
permanent displacement, elastic wave velocity, etc.). Here, engineers are expected to answer
questions such as what potential capacity remains and how long the damaged structure may
continue to function for planning of maintenance?
This guideline does not methodically handle the performance assessment of damaged real
structures under natural or artificial environments, although some knowledge included in this
guideline may assist the practitioner in evaluating this issue. For example, damaged
reinforced concrete beams subjected to flexure-shear are shown in Figure 10.4. Some part of
main reinforcement was computationally decayed by volumetric expansion of corroded
substances. This, induced, corrosion created pre-cracking as well as influencing the formation
of subsequent diagonal shear cracks. This is introduced here merely to demonstrate the future
potential of nonlinear analysis in the field of maintenance engineering.
Guidelines on structural and service-life designs are thought to be an important development
issue of the future. The issue of analysis for maintenance assessment is a target of some
importance that involves the coupling of structural and materials engineering. Equally,
performance assessment is needed in providing more rational tools for the planning of
maintenance, as well as design of newly constructed structures. At present, nonlinear
structural analysis and modelling technology are under development for meeting these
challenges.

336

10 Concluding remarks

No corrosion damage: after shear failure

corrosion induced at the span center: after shear failure

corrosion induced in shear span: just before failure


Initiation of shear crack from pre-horizontal cracks

corrosion induced in shear span: after shear failure

Experiment Failure Crack Pattern

Analysis Failure Crack Pattern

Light corrosion beam


Figure 10.4: Partially corroded RC beams in shear and cracking pattern (Maekawa et al. 2003).

10.7 References
JSCE (1999), Recommendation for Structural Performance Verification of LNG
Uuderground Storage Tanks, Japan Society of Civil Engineers, Concrete Library, 98.
JSCE (2002), Standard Specification of Concrete Structures Structural Performance
Verificaiton , Japan Society of Civil Engineers.
Maekawa, K., Ishida, T. and Kishi, T. (2003), Multi-scale modeling of concrete performance
Integrated material and structural mechanics -, Journal of Advanced Concrete Technology,
JCI, 2(1).

!
fib Bulletin 45: Practitioners guide to finite element modelling
of reinforced concrete structures

337

fib Bulletins published since 1998


N
1

Title
Structural Concrete Textbook on Behaviour, Design and Performance;
Vol. 1: Introduction - Design Process Materials
Manual - textbook (244 pages, ISBN 978-2-88394-041-3, July 1999)

Structural Concrete Textbook on Behaviour, Design and Performance


Vol. 2: Basis of Design
Manual - textbook (324 pages, ISBN 978-2-88394-042-0, July 1999)

Structural Concrete Textbook on Behaviour, Design and Performance


Vol. 3: Durability - Design for Fire Resistance - Member Design - Maintenance, Assessment
and Repair - Practical aspects
Manual - textbook (292 pages, ISBN 978-2-88394-043-7, December 1999)

Lightweight aggregate concrete: Extracts from codes and standards


State-of-the-art report (46 pages, ISBN 978-2-88394-044-4, August 1999)

Protective systems against hazards: Nature and extent of the problem


Technical report (64 pages, ISBN 978-2-88394-045-1, October 1999)

Special design considerations for precast prestressed hollow core floors


Guide to good practice (180 pages, ISBN 978-2-88394-046-8, January 2000)

Corrugated plastic ducts for internal bonded post-tensioning


Technical report (50 pages, ISBN 978-2-88394-047-5, January 2000)

Lightweight aggregate concrete: Part 1 (guide) Recommended extensions to Model Code 90;
Part 2 (technical report) Identification of research needs; Part 3 (state-of-art report) Application of
lightweight aggregate concrete (118 pages, ISBN 978-2-88394-048-2, May 2000)
9 Guidance for good bridge design: Part 1 Introduction, Part 2 Design and construction
aspects. Guide to good practice (190 pages, ISBN 978-2-88394-049-9, July 2000)
10 Bond of reinforcement in concrete

State-of-art report (434 pages, ISBN 978-2-88394-050-5, August 2000)

11 Factory applied corrosion protection of prestressing steel


State-of-art report (20 pages, ISBN 978-2-88394-051-2, January 2001)

12 Punching of structural concrete slabs


Technical report (314 pages, ISBN 978-2-88394-052-9, August 2001)

13 Nuclear containments State-of-art report (130 pages, 1 CD, ISBN 978-2-88394-053-6, September 2001)
14 Externally bonded FRP reinforcement for RC structures
Technical report (138 pages, ISBN 978-2-88394-054-3, October 2001)

15 Durability of post-tensioning tendons


Technical report (284 pages, ISBN 978-2-88394-055-0, November 2001)

16 Design Examples for the 1996 FIP recommendations Practical design of structural concrete
Technical report (198 pages, ISBN 978-2-88394-056-7, January 2002)

17 Management, maintenance and strengthening of concrete structures


Technical report (180 pages, ISBN 978-2-88394-057-4, April 2002)

18 Recycling of offshore concrete structures


State-of-art report (33 pages, ISBN 978-2-88394-058-1, April 2002)

19 Precast concrete in mixed construction


State-of-art report (68 pages, ISBN 978-2-88394-059-8, April 2002)

20 Grouting of tendons in prestressed concrete


Guide to good practice (52 pages, ISBN 978-2-88394-060-4, July 2002)

21 Environmental issues in prefabrication


State-of-art report (56 pages, ISBN 978-2-88394-061-1, March 2003)

Title

22 Monitoring and safety evaluation of existing concrete structures


State-of-art report (304 pages, ISBN 978-2-88394-062-8, May 2003)

23 Environmental effects of concrete State-of-art report (68 pages, ISBN 978-2-88394-063-5, June 2003)
24 Seismic assessment and retrofit of reinforced concrete buildings
State-of-art report (312 pages, ISBN 978-2-88394-064-2, August 2003)

25 Displacement-based seismic design of reinforced concrete buildings


State-of-art report (196 pages, ISBN 978-2-88394-065-9, August 2003)

26 Influence of material and processing on stress corrosion cracking of prestressing steel - case
studies. Technical report (44 pages, ISBN 978-2-88394-066-6, October 2003)
27 Seismic design of precast concrete building structures
State-of-art report (262 pages, ISBN 978-2-88394-067-3, January 2004)

28 Environmental design State-of-art report (86 pages, ISBN 978-2-88394-068-0, February 2004)
29 Precast concrete bridges State-of-art report (83 pages, ISBN 978-2-88394-069-7, November 2004)
30 Acceptance of stay cable systems using prestressing steels
Recommendation (80 pages, ISBN 978-2-88394-070-3, January 2005)

31 Post-tensioning in buildings Technical report (116 pages, ISBN 978-2-88394-071-0, February 2005)
32 Guidelines for the design of footbridges
Guide to good practice (160 pages, ISBN 978-2-88394-072-7, November 2005)

33 Durability of post-tensioning tendons


Recommendation (74 pages, ISBN 978-2-88394-073-4, December 2005)

34 Model Code for Service Life Design


Model Code (116 pages, ISBN 978-2-88394-074-1, February 2006)

35 Retrofitting of concrete structures by externally bonded FRPs.


Technical Report (224 pages, ISBN 978-2-88394-075-8, April 2006)

36 2006 fib Awards for Outstanding Concrete Structures


Bulletin (40 pages, ISBN 978-2-88394-076-5, May 2006)

37 Precast concrete railway track systems


State-of-art report (38 pages, ISBN 978-2-88394-077-2, September 2006)

38 Fire design of concrete structures materials, structures and modelling


State-of-art report (106 pages, ISBN 978-2-88394-078-9, April 2007)

39 Seismic bridge design and retrofit structural solutions


State-of-art report (300 pages, ISBN 978-2-88394-079-6, May 2007)

40 FRP reinforcement in RC structures


Technical report (160 pages, ISBN 978-2-88394-080-2, September 2007)

41 Treatment of imperfections in precast structural elements


State-of-art report (74 pages, ISBN 978-2-88394-081-9, November 2007)

42 Constitutive modelling of high strength / high performance concrete


State-of-art report (130 pages, ISBN 978-2-88394-082-6, January 2008)

43 Structural connections for precast concrete buildings


Guide to good practice (370 pages, ISBN 978-2-88394-083-3, February 2008)

44 Concrete structure management: Guide to ownership and good practice


Guide to good practice (208 pages, ISBN 978-2-88394-084-0, February 2008)

45 Practitioners guide to finite element modelling of reinforced concrete structures


State-of-art report (344 pages, ISBN 978-2-88394-085-7, June 2008)
Abstracts for fib Bulletins, lists of available CEB Bulletins and FIP Reports, and an order form are
given on the fib website at www.fib-international.org/publications.