Cervenka Consulting Ltd.

Na Hrebenkach 55
150 00 Prague
Czech Republic
Phone: +420 220 610 018
E-mail: cervenka@cervenka.cz
Web: http://www.cervenka.cz

ATENA Program Documentation

Part 1

Theory

Written by
Vladimír Červenka, Libor Jendele,
and Jan Červenka

Prague, August 24, 2009

 iii

CONTENTS

1 CONTINUUM GOVERNING EQUATIONS 9

1.1 Introduction 9

1.2 General Problem Formulation 10

1.3 Stress Tensors 12

1.3.1 Cauchy Stress Tensor 12

1.3.2 2nd Piola-Kirchhoff Stress Tensor 12

1.4 Strain Tensors 14

1.4.1 Engineering Strain 14

1.4.2 Green-Lagrange Strain 14

1.5 Constitutive tensor 14

1.6 The Principle of Virtual Displacements 15

1.7 The Work Done by the External Forces 17

1.8 Problem Discretisation Using Finite Element Method 17

1.9 Stress and strain smoothing 19

1.9.1 Extrapolation of stress and strain to element nodes 19

1.10 Simple, complex supports and master-slave boundary conditions. 21

1.11 References 22
2 CONSTITUTIVE MODELS 23
2.1 Constitutive Model SBETA (CCSbetaMaterial) 23

2.1.1 Basic Assumptions 23

2.1.2 Stress-Strain Relations for Concrete 26

2.1.3 Localization Limiters 32

2.1.4 Fracture Process, Crack Width 33

2.1.5 Biaxial Stress Failure Criterion of Concrete 33

2.1.6 Two Models of Smeared Cracks 35

iv

2.1.7 Shear Stress and Stiffness in Cracked Concrete 37

2.1.8 Compressive Strength of Cracked Concrete 37

2.1.9 Tension Stiffening in Cracked Concrete 38

2.1.10 Summary of Stresses in SBETA Constitutive Model 38

2.1.11 Material Stiffness Matrices 39

2.1.12 Analysis of Stresses 41

2.1.13 Parameters of Constitutive Model 41

2.2 Fracture–Plastic Constitutive Model (CC3DCementitious,

CC3DNonLinCementitious, CC3DNonLinCementitious2,
CC3DNonLinCementitious2User, CC3DNonLinCementitious2Variable,
CC3DNonLinCementitious2SHCC, CC3DNonLinCementitious3) 43

2.2.1 Introduction 43

2.2.2 Material Model Formulation 44

2.2.3 Rankine-Fracturing Model for Concrete Cracking 44

2.2.4 Plasticity Model for Concrete Crushing 46

2.2.5 Combination of Plasticity and Fracture model 50

2.2.6 Variants of the fracture plastic model 53

2.2.7 Tension stiffening 56

2.2.8 Crack spacing 56

2.2.9 Fatigue 57

2.2.10 Strain Hardening Cementitious Composite (SHCC, HPFRCC) material 60

2.2.11 Confinement-sensitive constitutive model 62

2.3 Von Mises Plasticity Model 66

2.4 Drucker-Prager Plasticity Model 69

2.5 User Material Model 70

2.6 Interface Material Model 70

2.7 Reinforcement Stress-Strain Laws 74

2.7.1 Introduction 74
 v

2.7.2 Bilinear Law 74

2.7.3 Multi-line Law 75

2.7.4 No Compression Reinforcement 76

2.7.5 Cycling Reinforcement Model 76

2.8 Reinforcement bond models 77

2.8.1 CEB-FIP 1990 Model Code 78

2.8.2 Bond Model by Bigaj 79

2.9 Microplane material model (CCMicroplane4) 81

2.9.1 Equivalent localization element 81

2.10 References 85
3 FINITE ELEMENTS 89
3.1 Introduction 89

3.2 Truss 2D and 3D Element 91

3.3 Plane Quadrilateral Elements 95

3.4 Plane Triangular Elements 102

3.5 3D Solid Elements 103

3.6 Spring Element 114

3.7 Quadrilateral Element Q10 116

3.7.1 Element Stiffness Matrix 116

3.7.2 Evaluation of Stresses and Resisting Forces 119

3.8 External Cable 120

3.9 Reinforcement Bars with Prescribed Bond 121

3.10 Interface Element 123

3.11 Truss Axi-Symmetric Elements. 126

3.12 Ahmad Shell Element 128

3.12.1 Coordinate systems. 130

3.12.2 Geometry approximation 135

vi

3.12.3 Displacement field approximation. 136

3.12.4 Strain and stresses definition. 137

3.12.5 Serendipity, Lagrangian and Heterosis variant of degenerated shell element. 138

3.12.6 Smeared Reinforcement 144

3.12.7 Transformation of the original DOFs to displacements at the top and bottom of the
element nodal coordinate system 144

3.12.8 Shell Ahmad elements implemented in ATENA 148

3.13 Curvilinear nonlinear 3D beam element. 149

3.13.1 Geometry and displacements and rotations fields. 149

3.13.2 Strain and stress definition 153

3.13.3 Matrices used in the beam element formulation 154

3.13.4 The element integration 160

3.14 Global and local coordinate systems for element load 162

3.15 References 165

4 SOLUTION OF NONLINEAR EQUATIONS 166
4.1 Linear Solvers 166

4.1.1 Direct Solver 167

4.1.2 Direct Sparse Solver 168

4.1.3 Iterative Solver 168

4.2 Full Newton-Raphson Method 172

4.3 Modified Newton-Raphson Method 173

4.4 Arc-Length Method 174

4.4.1 Normal Update Method 177

4.4.2 Consistently Linearized Method 178

4.4.3 Explicit Orthogonal Method 179

4.4.4 The Crisfield Method. 180

4.4.5 Arc Length Step 181

 vii

4.5 Line Search Method 181

4.6 Parameter β 182

4.7 References 183

5 CREEP AND SHRINKAGE ANALYSIS 184
5.1 Implementation of creep and shrinkage analysis in ATENA 184

5.1.1 Basic theoretical assumptions 184

5.2 Approximation of compliance functions Φ (t , t ') by Dirichlet series. 186

5.3 Step by step method 187

5.4 Integration and retardation times 188

5.5 Creep and shrinkage constitutive model 190

5.6 References 195

6 TRANSPORT ANALYSIS 197
6.1 Numerical solution of the transport problem – spatial discretisation 201

6.2 Numerical solution of the transport problem – temporal discretisation 208

6.2.1 θ -parameter Crank Nicholson scheme 209

6.2.2 Adams-Bashforth integration scheme 209

6.2.3 Reduction of oscillations and convergence improvement 210

6.3 Material constitutive model 211

6.4 Fire element boundary load 214

6.4.1 Hydrocarbon fires 214

6.4.2 Fire exposed boundary 215

6.4.3 Implementation of fire exposed boundary in ATENA 216

6.5 References 217

7 DYNAMIC ANALYSIS 219
7.1 Structural damping 222
8 EIGENVALUES AND EIGENVECTORS ANALYSIS 224
8.1 Inverse subspace iteration 224
viii

8.1.1 Rayleigh-Ritz method 225

8.1.2 Jacobi method 225

8.1.3 Inverse iteration method 227

8.1.4 Algorithm of Inverse subspace iteration 227

8.1.5 Sturm sequence property check 230

8.2 References 230

9 GENERAL FORM OF DIRICHLET BOUNDARY CONDITIONS 231
9.1 Theory behind the implementation 231

9.1.1 Single CBC 232

9.1.2 Multiple CBCs 234

9.2 Application of Complex Boundary Conditions 238

9.2.1 Refinement of a finite element mesh 238

9.2.2 Mesh generation using sub-regions 239

9.2.3 Discrete reinforcement embedded in solid elements 240

9.2.4 Curvilinear nonlinear beam and shell elements 241

9.3 References 242

INDEX 243
 9

1 CONTINUUM GOVERNING EQUATIONS

1.1 Introduction
This chapter presents the general governing continuum equations for non-linear analysis. In
general, there exist many variants of non-linear analysis depending on how many non-linear
effects are accounted for. Hence, this chapter first introduces some basic terms and entities
commonly used for structural non-linear analyses, and then it concentrates on formulations that
are implemented in ATENA.
It is important to realize that the whole structure does not have to be analyzed using a full non-
linear formulation. However, a simplified (or even linear) formulation can be used in many
cases. It is a matter of engineering knowledge and practice to assess, whether the inaccuracies
due to a simplified formulation are acceptable, or not.
The simplest formulation, i.e. linear formulation, is characterized by the following assumptions:
The constitutive equation is linear, i.e. the generalized form of Hook's law is used.
The geometric equation is linear that is, the quadratic terms are neglected. It means that during
analysis we neglect change of shape and position of the structure.
Both loading and boundary conditions are conservative, i.e. they are constant throughout the
whole analysis irrespective of the structural deformation, time etc.
Generally linear constitutive equations can be employed for a material, which is far from its
failure point, usually up to 50% of its maximum strength. Of course, this depends on the type of
material, e.g. rubber needs to be considered as a non-linear material earlier. But for usual civil
engineering materials the previous assumption is satisfactory.
Geometric equations can be considered linear, if deflections of a structure are much smaller than
its dimensions. This must be satisfied not only for the whole structure but also for its parts. Then
the geometric equations for the loaded structure can then be written using the original (unloaded)
geometry.
It is also important to realize that a linear solution is permissible only in the case of small strains.
This is closely related to material property because if strains are high, the stresses are usually,
although not necessarily, high as well.
Despite the fact that for the vast majority of structures linear simplifications are quite acceptable,
there are structures when it is necessary to take in account some non-linear behavior. The
resulting governing equations are then much more complicated, and normally they do not have a
closed form solution. Consequently some non-linear iterative solution scheme must be used (see
Chapter Solution of Nonlinear Equations further in this document).
Non-linear analysis can be classified according to a type of non-linear behavior:
Non-linear material behavior only needs to be accounted for. This is the most common case for
ordinary reinforced concrete structures. Because of serviceability limitations, deformations
are relatively small. However, the very low tensile strength of concrete needs to be accounted
for.
Deformations (either displacements only or both displacements and rotations) are large enough
such that the equilibrium equations must use the deformed shape of the structure. However
the relative deformations (strains) are still small. The complete form of the geometric
10

equations, including quadratic terms, has to be employed but constitutive equations are
linear. This group of non-linear analysis includes most stability problems.

The last group uses non-linear both material and geometric equations. In addition, it is usually
not possible to suddenly apply the total value of load but it is necessary to integrate in time
increments (or loading increments). This is the most accurate and general approach but
unfortunately is also the most complicated.
There are two basic possibilities for formulating the general structural behavior based on its
deformed shape:

Lagrange formulation:
In this case we are interested in the behavior of infinitesimal particles of volume dV . Their
volume will vary dependent on a loading level applied and, consequently, on the amount of
current deformations. This method is usually used to calculate civil engineering structures.

Euler formulation:
The essential idea of Euler's formulation is to study the "flow" of the structural material through
infinitesimal and fixed volumes of the structure. This is the favorite formulation for fluid
analysis, analysis of gas flow, tribulation etc. where large material flows exist.
For structural analysis, however, Lagrangian formulation is better, and therefore attention will be
restricted to this. Two forms of the Lagrangian formulation are possible. The governing
equations can either be written with respect to the undeformed original configuration at time t =
0 or with respect to the most recent deformed configuration at time t. The former case is called
Total Lagrangian formulation (TL) while the latter one is called the Updated Lagrangian
formulation (UL).
It is difficult to say which formulation is better because both have their advantages and
drawbacks. Usually it depends on a particular structure being analyzed and which one to use is a
matter of engineering judgement. Generally, provided the constitutive equations are adequate,
the results for both methods are identical.
ATENA currently uses Updated Lagrangian formulation, (which is described later in this
chapter) and supports the highest, i.e. 3rd level of non-linear behavior. Soon, it should also
support Total Lagrangian formulation.

1.2 General Problem Formulation

A general analysis of a structure usually consists of application of many small load increments.
At each of those increments an iterative solution procedure has to be executed to obtain a
structural response at the end of the increment. Hence, denoting start and end of the load
increment by t and t + Δt , at each step, we know structural state at time t (from the previous
steps) and solve for the state at time t + Δt . This procedure is repeated as many times as
necessary to reach the final (total) level of loading.
This process is depicted in Fig. 1-1. At time t = 0 the volume of structure is 0V , the surface area
is 0 S and any arbitrary point M has coordinates 0 X 1 , 0 X 2 , 0 X 3 . Similarly at the time t the same
 11

structure has a volume tV , surface area t S and coordinates of point M are t X 1 , t X 2 , t X 3 .

Similar definition applies for time t + Δt by replacing index t by t + Δt .

[0X3, tX3, t+ΔtX3]

M0 M1 M2
Configuration t+Δt
Configuration 0 Configuration t

0 t t+Δt
[0X2, tX2, t+ΔtX2]
[ X1, X1, X1]

Fig. 1-1 The movement of body of structure in Cartesian coordinate system.

For the derivations of nonlinear equations it is important to use clear and simple notation. The
same system of notation will be used throughout this document:
Displacements u are defined in a similar manner to that adopted for coordinates, hence t ui is the
i -th element of the displacement vector at time t ,
ui = t +Δt X i − t X i is i -th element of vector of displacement increments at time t ,
The left superscript denotes the time corresponding to the value of the entity, the left subscript
denotes the configuration with respect to which the value is measured and subscripts on the
right identify the relationships to the coordinate axis. Thus for example t +Δ0tτ ij denotes
element i , j of stress tensor τ at time t + Δt with respect to the original (undeformed)
configuration.
For derivatives the abbreviated notation will be used, i.e. all right subscripts that appear after a
comma declare derivatives. For example:
t +Δt ∂ t +Δt
u =
0 i, j ui (1.2)
∂X j
The general governing equations can be derived in the form of a set of partial differential
equations (for example using the displacement method) or an energy approach can be used. The
final results are the same.
One of the most general methods of establishing the governing equations is to apply the principle
of virtual work. There are three basic variants of this:

The principle of virtual displacements,

The principle of virtual forces,
12

The Clapeyron divergent theorem.

Using the virtual work theorems it is possible to derive several different variation principles
(Lagrange principle, Clapeyron principle, Hellinger-Reissner principle, Hu-Washizu principle
etc.). There are popular especially in linear analysis. They can be used to establish equilibrium
equations, to study possible deformation modes in finite element discretization etc.
Unfortunately in nonlinear analysis they do not always work.
In this document all the following derivations will be presented in their displacement form and
consequently the principle of virtual displacements will be used throughout.
The following section deals with the definition of the stress and strain tensors, which are usually
used, in nonlinear analysis. All of them are symmetric.

1.3 Stress Tensors

1.3.1 Cauchy Stress Tensor

This tensor is well known from linear mechanics. It express the forces that act on infinitesimal
small areas of the deformed body at time t. Sometimes these are also called a "engineering"
stress. The Cauchy stress tensor is the main entity for checking ultimate stress values in
materials. In the following text it will be denoted by τ . It is energetically conjugated with
Engineering strain tensor described later.

1.3.2 2nd Piola-Kirchhoff Stress Tensor

The 2nd Piola-Kirchhoff tensor is a fictitious entity, having no physical representation of it as in
the case of the Cauchy tensor. It expresses the forces, which act on infinitesimal areas of body in
the undeformed configuration. Hence it relates forces to the shape of the structure which no
longer exists.
The mathematical definition is given by:
0
ρ 0
Sij = t X i , m τ mn t X j , n
t t 0
(1.3)
ρ
0 t

where
0
ρ
is the ratio of density of the material at time 0 and t ,
t
ρ
t
τ mn is the Cauchy stress tensor at time t ,
0
t X i , m is the derivative of coordinates, ref. (1.5).

Using inverse transformation, we can express Cauchy stress tensor in terms of the 2nd Piola-
Kirchhoff stress tensor, i.e.:
t
ρ t
τ mn =
t t t
0 X m ,i 0 S mn 0 X n , j (1.4)
0
ρ
The elements 0t X i ,m are usually collected in the so-called Deformation gradient matrix:

X = ( 0∇ t X T )
t T
0 (1.5)
 13

where:

T
⎡ ∂ ∂ ∂ ⎤
0∇ = ⎢ 0
T
, 0 , 0 ⎥
⎣ ∂ X1 ∂ X 2 ∂ X 3 ⎦
t
X T = ⎡⎣ t X 1 , t X 2 , t X 3 ⎤⎦
0
ρ
The ratio can be computed using:
t
ρ
0
ρ =t ρ det( 0t X ) (1.6)
Expression (1.6) is based on the assumption that the weight of an infinitesimal particle is
constant during the loading process.
Some important properties can be deduced from definition of 2nd Piola-Kirchhoff tensor (1.3):
at time 0, i.e. the undeformed configuration, there is no distinction between 2nd Piola-Kirchhoff
0
ρ
and Cauchy stress tensors because 00 X = E , i.e. unity matrix and the density ratio = 1.,
t
ρ
2nd Piola-Kirchhoff tensor is an objective entity in the sense that it is independent of any
movement of the body provided the loading conditions are frozen. This is a very important
property. The Cauchy stress tensor does not satisfy this because it is sensitive to the rotation
of the body. It is energetically conjugated with Green-Lagrange tensor described later.
They’re some other stress tensor commonly used for structural nonlinear analysis, e.g. Jaumann
stress rate tensor (describes stress rate rather than its final values) etc, however they are not used
in ATENA and therefore not described in this document.
14

1.4 Strain Tensors

1.4.1 Engineering Strain

It is the most commonly used strain tensor, comprising strains that are called Engineering strains.
Its main importance is that it is used in linear mechanics as a counterpart to the Cauchy stress
tensor.
1 ⎛ ∂um ∂u ⎞
t emn = ⎜ t + t n ⎟ (1.7)
2 ⎝ ∂ Xn ∂ Xm ⎠

1.4.2 Green-Lagrange Strain

This is energy conjugate of the 2nd Piola-Kirchhoff tensor and its properties are similar (i.e.
objective etc.). It is defined as:
1 t
ε =
t
0 ij
2
( 0 ui , j + 0 u j ,i + 0 uk ,i 0 uk , j )
t t t
(1.8)

If we calculate the length of an infinitesimal fibber prior and after deformation in the original
coordinates, we get exactly the terms of the Green-Lagrange tensor.
The following equation gives relation between variation of Green-Lagrange and Engineering
strain tensors:
∂t X m ∂t X n
∂ ( t0 ε ij ) = ∂ ( t emn ) (1.9)
∂0 X i ∂0 X j
These are the strain tensors used in ATENA. From the other strain tensors commonly used in
non-linear analysis we can mention Almansi strain tensor, co-rotated logarithmic strain, strain
rate tensor etc.

1.5 Constitutive tensor

Although the whole chapter later in this document is dedicated to the problem of constitutive
equations and to material failure criteria, assume for the moment that stress-strain relation can
be written in the following form:
t
0 Sij = 0t Cijrs 0t ε rs (1.10)

where t0 Cijrs is the constitutive tensor.

This form is applicable for linear materials or in its incremental form it can be used also for
nonlinear materials. The following important relations apply for transformation from coordinates
to time 0 to coordinates at time t :
t
ρ t
t
Cmnpq = t t t t
0 xm ,i 0 xn , j 0Cijrs 0 x p , r 0 xq , s (1.11)
ρ
t 0

or in the other direction

 15

0
ρ0
t
Cijrs = 0 t 0 0
t xi , m t x j , n t Cmnpq t xr , p t xs , q (1.12)
0 t
ρ
Using constitutive tensor (1.11) and Almansi strains tt ε , we can write for Cauchy stresses (with
respect to coordinates at time t ):
t
τ ij = tt Cijrs ttε rs (1.13)

Almansi strains are defined (related to Green-Lagrange strains t0 ε ij by

t
t ε mn =0t xi ,m 0t x j ,n 0t ε ij (1.14)

or can be calculated directly:

1 t
t
ε =
t ij
2
( t ui, j +tt u j,i − tt uk ,i tt uk , j ) (1.15)

The equation (1.13) is equivalent to the equation (1.10) that was written for original
configuration of the structure. It is very important to know, with respect to which coordinate
system the stress, strain and constitutive tensors are defined, as the actual value can significantly
differ. ATENA currently assumes that all these tensors are defined at coordinates at time t .

1.6 The Principle of Virtual Displacements

This section presents how the principle of virtual displacement can be applied to the analysis of
a structure. For completeness both the Lagrangian Total and Updated formulations will be
discussed. In all derivations it is assumed that the response of the structure up to time t is known.
Now, at time t + Δt we apply load increment and using the principle of virtual displacement will
solve for state of the structure at t + Δt .
Virtual work of the structure yields following. For Total formulation:

∫( )
Sij δ ( t0+ dt ε ij ) dV = 0t + dt R
t + dt
0 (1.16)
0
V

for Updated formulation:

∫( )
Sij δ ( tt + dt ε ij ) dV = tt + dt R
t + dt
t (1.17)
t
V

where 0V , t V denotes the structure volume corresponding to time 0 and t and t + dt R is the total
virtual work of the external forces. The symbol δ denotes variation of the entity. Since energy
must be invariant with respect to the reference coordinate system, (1.16) and (1.17) must lead to
identical results.
Substituting expressions for strain and stress tensors the final governing equation for structure
can be derived. They are summarized in (1.18) through (1.29). Note that the relationships are
expressed with respect to configurations at an arbitrary time t and an iteration (i ) . Typically, the
time t may by 0 , in which case we have Total Lagrangian formulation or t + Δt (i − 1) , in which
case we have Updated Lagrangian formulation, where some terms can be omitted. ATENA also
support “semi” Updated Lagrangian formulation, when t conforms to time at the beginning of
time increment, i.e. the beginning of load step. The following table compares the above-
mentioned formulations:
16

Table 1.6-1 Comparison of different Lagrangian formulation.

Transform each Transform each load Transfor Calculate

Lagrangian iteration increment m stress t + Δt ( i −1)
u for
formulation and strain t i, j

IP state Material IP state Material e

for output t ij
variable properti variable properties
s es s
Total No No No No Yes Yes
Updated Yes Yes Yes Yes No No
“Semi”- No No Yes Yes No Yes
Updated
Governing equations:

∫
t +Δt
t Sij( i ) ∂ ( t +Δttε ij( i ) ) t dV = t +Δt R (1.18)
t
V
nd
where 2 Piola-Kirchhoff stress and Green Lagrange strain tensor are:
t
ρ
t +Δt
Sij( i ) = t ( i ) t +Δt ( i )
x
t +Δt i ,m τ
t
t mn t +Δt x (ji,n) (1.19)
t t +Δt
ρ
1 t +Δt ( i )
δ t +Δttε ij( i ) = δ
2
( t ui, j + t +Δt ( i )
u
t i, j + t +Δt ( i ) t +Δt ( i )
u
t k ,i u
t k, j ) (1.20)

The stress and strain increments:

t +Δt
t Sij( i ) = t +Δtt Sij( i −1) + t Sij( i ) (1.21)
t +Δt ( i )
t ij ε = ε
t +Δt ( i −1)
t ij + t ε ij( i )
(1.22)
t ε ij( i ) = t eij( i ) + tηij( i )
where linear part of the strain increment is calculated by:
1
e(i )
t ij =
2
( t ui , j + t ui , j +
(i ) (i ) t +Δt ( i −1)
t uk ,i t uk , j +
(i )
t uk , j t uk ,i )
t +Δt ( i −1) (i )
(1.23)

and nonlinear part by:

1 (i ) (i )
t ηij( i ) =
2
( t uk ,i t uk , j ) (1.24)

Using constitutive equations in form:

t +Δt
t Sij( i ) = t Cijrs t ε rs( i ) (1.25)
(i )
where t Cijrs is tangent material tensors and noting that δ ( t +Δttε ij(i ) ) = δ ( t ε ij(i ) ) , an incremental form of
(1.18) can be derived:
 17

(e + tηij( i ) ) δ ( t eij( i ) + tηij( i ) ) dV + Sij( i ) δ ( t eij( i ) + tηij( i ) ) dV = t +Δt R

t t
∫ ∫
(i ) t +Δt
t Cijrs t ij t (1.26)
t
t t
V V
nd
After linearisation, i.e. neglecting 2 order terms in (1.26):

(e + tηij( i ) ) δ ( t eij( i ) + tηij( i ) ) dV ≈ Cijrs t t eij( i ) δ ( t eij( i ) ) dV

t t
∫ ∫
(i )
t Cijrs t ij t (1.27)
t
t t
V V

we arrive to the final form of the governing equations:

Cijrs t ers( i ) δ ( t eij( i ) ) dV + Sij( i −1) δ ( tηij( i ) ) dV =

t t
∫ ∫
t +Δt
t t
t t
V V
(1.28)
Sij( i −1) δ ( t eij( i ) ) dV
t
∫
t +Δt t +Δt
R− t
t
V

Note that the term δ ( t eij( i ) ) = δ ( t eij ) is constant, i.e. independent of t ui( i ) , hence it is on RHS of
(1.28).

1.7 The Work Done by the External Forces

So far only the incremental virtual internal work has been considered. This work has to be
balanced by the work done by the external forces. It is calculated as follows:
∂ 2 t +Δtt ui( i −1)
t +Δt
R= ∫
t +Δt
fbi δ ( t u ) dV + ∫
i t +Δt
fsi δ ( t u ) dS + ∫ i t
ρ dV (1.29)
t
V t
S t
V
∂t 2

where fbi and fsi are body and surface forces, t S and t V denotes integration with respect to the
surface with the prescribed boundary forces and volume of the structure (at time and t ).
The 1st integral in (1.29) accounts for external work on surface (e.g. external forces), the second
one for work done by body forces (e.g. weight) and the last one accounts for work done by
inertia forces, which are applicable only for dynamic analysis problems).
At this point, all the relationships for incremental analysis have been presented. In order to
proceed further, the problem must be discretized and solved by iterations (described in Chapter
Solution of Nonlinear Equations).

1.8 Problem Discretisation Using Finite Element Method

Spatial discretisation consists of discretising primary variable, (i.e. deformation in case of
ATENA) over domain of the structure. It is done in ATENA by Finite Element Method. The
domain is decomposed into many finite elements and at each of these elements the deformation
field is approximated by
t
ui = h j t uij (1.30)

where
j is index for finite element node, j = 1...n ,
n is number of element nodes,
h j are interpolation function usually grouped in matrix H j = [ h1 ( r, s, t ), h2 ( r, s, t ).....hn ( r, s, t )] ,
r, s, t are local element coordinates.
18

The interpolation functions h j are usually created in the way that h j = 1 at node j and h j = 0 at
any other element nodes.
Combining (1.30) and equation for strain definition (1.8) it can be derived:
t +Δt
t ε ( i ) = ( t BL 0 + t +Δtt BL( i1−1) + t +Δtt BNL ) U
( i −1) t +Δt ( i )
(1.31)

where
t +Δt
t ε ( i ) is vector of Green-Lagrange strains,
t +Δt
U ( i ) is vector of displacements,
t +Δt ( i −1) t +Δt ( i −1) st
t BL 0 , t BL1 , t BNL are linear strain-displacements transformation matrices (the 1 two of
them) and nonlinear strain-displacements transformation matrix (the last one).
Similar equation can be written also for stress tensor.
t +Δt
t S (i ) = t +Δt
t C (i ) t +Δt
t ε (i ) (1.32)
where:
t +Δt ( i )
tS is vector of 2nd Piola-Kirchhoff stress tensor and
t +Δt
t C ( i ) is incremental stress-strain material properties matrix.
Applying the above discretisation for each finite element of the structure and assembling the
results the continuum based governing equations in (1.28) can be re-written in the following
form:
∂ t +Δt
t
M U ( i ) + ( t K L + t +Δtt K NL
( i −1)
) Δ t +ΔtU ( i ) =t +Δt R − t +Δt F ( i −1) (1.33)
∂t 2
where

t K L is the linear strain incremental stiffness matrix,

t +Δt ( i −1)
t K NL is the nonlinear strain incremental stiffness matrix,
t
M is the structural mass matrix,
Δ t +ΔtU ( i ) is the vector of nodal point displacements increments at time t + Δt , iteration i ;
∂ t +Δt ( i )
t+ Δt

∂t 2
( U ) is the vector of nodal accelerations,
t +Δt
R , t +Δt F ( i −1) is the vector of applied external forces and internal forces,
(i ) ( i−1)
, superscripts indicate iteration numbers.
Note that (1.33) contains also inertial term needed only for dynamic analysis. Finite element
matrices in (1.33) and corresponding analytical expressions are summarized:
 19

⎛ ⎞
Cijrs t ers( i ) δ ( t eij( i ) ) dV
t
K L ΔU ( i ) = ⎜ ∫ t BLT t C t BL dV ⎟ ΔU ( i ) ≈ ∫
t ⎜t ⎟ t
⎝V ⎠ t
V

⎛ ⎞
Sij( i −1) δ ( tηij( i ) ) dV
t
t +Δt ( i −1)
K NL ΔU ( i ) = ⎜ ∫ t +Δt ( i −1) T
BNL t +Δt
Sij( i −1) t +Δt ( i −1)
BNL dV ⎟ ΔU ( i ) ≈ ∫ t +Δt
t ⎜t t t t ⎟ t
⎝V ⎠ t
V

Sij( i −1) δ ( t eij( i ) ) dV

t
t +Δt
F ( i −1) = ∫ t +Δt
t Sij( i −1) dV ≈ ∫ t +Δt
t (1.34)
t t
V V

∫H ∫H
t +Δt T t +Δt T t +Δt
R= t f A dV dA + t f B dV ≈ t +Δt R
t t
A V

∂ ⎛ ⎞ ∂ ∂ t +Δt ui( i ) t ⎛ ∂ t +Δt ui( i ) ⎞

t
M U ( i ) = ⎜ ∫ H T t ρ H dV ⎟ 2
t +Δt
U (i ) ≈ ∫
t +Δt
ρ δ ⎜ ⎟ dV
∂t 2 ⎜t ⎟ ∂t ∂ t 2
⎝ ∂ t 2
⎠
⎝V ⎠ t
V

1.9 Stress and strain smoothing

All derivations and solution procedures in ATENA software are based on deformational form of
finite element method. Any structure is solved using weak (or integral) form of equilibrium
equations. The whole structure is divided into many finite elements and displacement u at each
particular element (at any location) is approximated by approximation functions hi and element
displacements u i as follows: u = ∑ hi u i , ( i is index of an element node). It is important to note
i
that in order not to loose any internal energy of the structure, the displacements over the whole
structure must be continuous. The continuity within finite elements is trivial. Use of continuous
approximation functions h j ensures this requirement. A bit more complicated situation is on
boundaries between adjacent elements, however, if the adjacent elements are of the same type,
their displacements are also continuous. Note that there exist are some techniques that alleviate
the continuity requirement but in ATENA they are not used.
Unlike displacements, stress and strain field is typically discontinuous. Moreover, a structure is
investigated within so-called material (or integral) points, which are points located somewhere
within each element. Their position is derived from requirement to minimize the approximation
error. In other words, standard finite element method provides stress and strain values only at
those material points and these values must be later somehow extrapolated into element nodal
points. Often, some sort of smoothing is required in order to remove the mentioned stress and
strain discontinuity. This section describes, how this goal is done in ATENA.
There are two steps in the process of stress and strain smoothing: 1/ extrapolation of stress and
strain from material points to element nodes and 2/ averaging of stress in global node. The whole
technique is described briefly. All details and derivations can be found e.g. (ZIENKIEWICZ,
TAYLOR 1989) and ČERVENKA et. al. 1993.

1.9.1 Extrapolation of stress and strain to element nodes

The extrapolation is done as follows (for each component of structural stress σ and strain ε ).
20

Let us define a vector of stresses σxx at element nodes i such as σxx = {σxx ,1 , σxx ,2 ,....σxx ,n } ,
T

where the 2nd index indicates element node number. Let us also define a vector
Pxx = { Pxx ,1 , Pxx ,2 ,....Pxx , n } , whose component are calculated
T

Pxx ,i = ∫ hi σ xx d Ωe (1.35)
Ωe

The nodal value σxx (with values of σ xx at nodes i =1..n ) is then calculated as follows:

σxx = [ M ] Pxx
inv
(1.36)

where:
M ij = ∫ hi h j d Ω e (1.37)
Ωe

In the above σ xx is an extrapolated field of stress of σ xx calculated by FEM. It is typically

discontinuous. n is number of element nods, Ωe is volume of the investigated finite element.
The same strategy is used also for remaining stress and strain components.
This smoothing technique is called variational as it is base on averaging energy over the element.
In addition to that ATENA supports also another way of extrapolating vales from integration
points to element nodes. In this case (1.37) is assumed to be a “lumped” diagonal matrix, in
order to eliminate the need for solving a system of linear equations. The process of lumping is
characterizes as follows:

M ij = ∫ hi ∑ hk δ ij d Ωe (1.38)
Ωe
k =1, n

As most element space approximations satisfy ∑h

k =1, n
k = 1 , the above equation is simplified to:

M ij = ∫ hi δ ij d Ω (1.39)
Ωe

where δ ij is Kronecker delta. This “lumped” formulation ATENA uses by default.

The above values are output as nodal element stress/strain values. It follows to calculate
   
averaged stress/stain value σ i = {σ xx , σ yy ,.....σ xz }i in a global node i that is participated by all
elements k with incidence at the global node i .


∑ σ Ω i ek
σi = k
(1.40)
∑Ω k
ek

where is vector of stresses σi = {σxx , σ yy ,.....σxz } at a node i , Ω ek is volume of element k that
i

has incidence of global node i . It should be noted that in ATENA the same extrapolation
 21

techniques is used for other integration point quantities as well such as: fracturing strains, plastic
strains and others.

1.10 Simple, complex supports and master-slave boundary

conditions.
Simple support and complex support boundary conditions represent boundary conditions of
Dirichlet types, i.e. boundary conditions that prescribe displacements. On the other hand, Simple
load boundary conditions is an example of von Neumann type boundary conditions, when forces
are prescribed.
Let K is structural stiffness matrix, u is vector of nodal displacements and R is a vector of
nodal forces. Further let u is subdivided into vector of free degrees of freedom uN (with von
Neumann boundary conditions) and constrained degrees of freedom uD (with Dirichlet boundary
conditions):
⎡u ⎤
u =⎢ N⎥ (1.41)
⎣ uD ⎦
The problem governing equations can then be written:
⎡ K NN K ND ⎤ ⎡ uN ⎤ ⎡ RN ⎤
⎢K = (1.42)
⎣ DN K DD ⎥⎦ ⎢⎣ uD ⎥⎦ ⎢⎣ RD ⎥⎦

ATENA software supports that any constrained degree of freedom can be a linear combination
of other degrees of freedom plus some constant term:
uDi = uDi ,0 + ∑ α k uNk (1.43)
k

where uDi ,0 is the constant term and α k are coefficients of the linear combination. Of course, the
equation (1.43) can include also the term ∑α u
l
l
l
D , however it is transformed into the constant

term.
The free degree of freedom are then solved by
uN = ( K NN )
−1
(R N − K ND RD ) (1.44)

and the dependent RD are solved by

RD = K DN uN + K DD uD (1.45)
The ATENA simple support boundary conditions mean that the boundary conditions use only
constant terms are uDi ,0 , (i.e. α k = 0 ). The complex support boundary conditions use the full
form of (1.43).
The boundary conditions as described above allow to specify for one degree of freedom either
Dirichlet, or von Neumann boundary condition, but not both of them the same time. It comes
from the nature of finite element method. However, ATENA can deal also this case of more
complex boundary conditions by introducing Lagrange multipliers. The derivation of theory
behind this kind of boundary conditions is beyond the scope of this manual. Details can be found
elsewhere, e.g. in (Bathe 1982). To apply this type of boundary conditions in ATENA, specify
22

for those degree of freedom both simple load and complex support boundary condition, the latter
one with the keyword “RELAX” keyword in its definition.
Nice feature about ATENA is that at any time it stores in RAM only K NN and all the elimination
with the remaining blocks of K is done at element level at the process of assembling the
structural stiffness matrix.
A special type of complex boundary conditions of Dirichlet type are so-called master-slave
boundary conditions. Such a boundary condition specifies that all (available) degrees of one
finite node, (i.e. slave node) are equal to degrees of freedom of another node (i.e. master node).
If more master nodes are specified, then these master nodes are assumed to form a finite element
and degrees of freedom of the slave node is assumed to be a node within that element. Its (slave)
degrees of freedom are approximated by element nodal (i.e. master) degrees of freedom in the
same way as displacements approximation within a finite element. The coefficients α k in (1.43)
are thus calculated automatically. This type of boundary conditions is used for example for
fixing discrete reinforcement bars to the surounding solid element .

1.11 References
BATHE, K.J. (1982), Finite Element Procedures In Engineering Analysis, Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey 07632, ISBN 0-13-317305-4.

ČERVENKA, J., KEATING, S.C., AND FELIPPA, C.A. (1993), “Comparison of strain
recovery techniques for the mixed iterative method”, Communications in Numerical Methods in
Engineering, Vol. 9, 925-932.
ZIENKIEWICZ, O.C., TAYLOR, R.L., (1989), The Finite Element Method, Volume 1,
McGraw-Hill Book Company, ISBN 0-07-084174-8.
 23

2 CONSTITUTIVE MODELS

2.1 Constitutive Model SBETA (CCSbetaMaterial)

2.1.1 Basic Assumptions

2.1.1.1 Stress, Strain, Material Stiffness

The formulation of constitutive relations is considered in the plane stress state. A smeared
approach is used to model the material properties, such as cracks or distributed reinforcement.
This means that material properties defined for a material point are valid within a certain
material volume, which is in this case associated with the entire finite element. The constitutive
model is based on the stiffness and is described by the equation of equilibrium in a material
point:

s = De, s = {σ x , σ y ,τ xy } , e = {ε x , ε y , γ xy }
T T
(2.1)

where s, D and e are a stress vector, a material stiffness matrix and a strain vector, respectively.
The stress and strain vectors are composed of the stress components of the plane stress state
σ x , σ y , τ xy , Fig. 2-1, and the strain components ε x , ε y , γ xy , Fig. 2-2, where γ xy is the engineering
shear strain. The strains are common for all materials. The stress vector s and the material matrix
D can be decomposed into the material components due to concrete and reinforcement as:
s = s c + s s , D = Dc + D s (2.2)
The stress vector s and both component stress vectors s c , s s are related to the total cross section
area. The concrete stress s c is acting on the material area of concrete A c , which is approximately
set equal to the cross section of the composite material A c ≈ A (the area of concrete occupied by
reinforcement is not subtracted).
The matrix D has a form of the Hooke's law for either isotropic or orthotropic material, as will be
shown in Section 2.1.11.

Fig. 2-1 Components of plane stress state.

24

Fig. 2-2 Components of strain state.

The reinforcement stress vector ss is the sum of stresses of all the smeared reinforcement
components:
n
s s = ∑ s si (2.3)
i =1

where n is the number of the smeared reinforcement components. For the ith reinforcement, the
global component reinforcement stress ssi is related to the local reinforcement stress σ si, by the
transformation:
s si = Tσ pi σ si, (2.4)
Asi
where pi is the reinforcing ratio pi = , Asi is the reinforcement cross section area. The local
Ac
reinforcement stress σ si, is acting on the reinforcement area Asi
The stress and strain vectors are transformed according to the equations bellow in the plane
stress state. New axes u, v are rotated from the global x, y axes by the angle α. The angle α is
positive in the counterclockwise direction, as shown in Fig. 2-3.

Fig. 2-3 Rotation of reference coordinate axes.

The transformation of the stresses:

s (u ) = Tσ s( x ) (2.5)
 25

⎡ cos(α ) 2 sin(α ) 2 2 cos(α ) sin(α ) ⎤

⎢ ⎥
Tσ = ⎢ sin(α ) 2
cos(α ) 2
−2 cos(α ) sin(α ) ⎥ (2.6)
⎢ − cos(α ) sin(α ) cos(α ) sin(α ) cos(α )2 − sin(α )2 ⎥
⎣ ⎦

s ( u ) = {σ u , σ v ,τ uv } , s ( x ) = {σ x , σ y , τ xy }
T T

The transformation of the strains:

e(u ) = Tε e( x ) (2.7)

⎡ cos(α ) 2 sin(α ) 2 cos(α ) sin(α ) ⎤

⎢ ⎥
Tε = ⎢ sin(α ) 2 cos(α ) 2 − cos(α ) sin(α ) ⎥ (2.8)
⎢ −2 cos(α ) sin(α ) 2 cos(α ) sin(α ) cos(α ) 2 − sin(α ) 2 ⎥
⎣ ⎦

e(u ) = {ε u , ε v , γ uv } , e( x ) = {ε x , ε y , γ xy } .
T T

The angles of principal axes of the stresses and strains, Fig. 2-1, Fig. 2-2, are found from the
equations:
2τ xy γ xy
tan(2ϑσ ) = , tan(2ϑε ) = (2.9)
σ x −σ y εx −εy
where ϑσ is the angle of the first principal stress axis and ϑε is the angle of the first principal
strain axis.
In case of isotropic material (uncracked concrete) the principal directions of the stress and strains
are identical, in case of anisotropic material (cracked concrete) they can be different. The sign
convention for the normal stresses, employed within this program, uses the positive values for
the tensile stress (strain) and negative values for the compressive stress (strain). The shear stress
(strain) is positive if acting upwards on the right face of a unit element.

2.1.1.2 Concept of Material Model SBETA

The material model SBETA includes the following effects of concrete behavior:
non-linear behavior in compression including hardening and softening,
fracture of concrete in tension based on the nonlinear fracture mechanics,
biaxial strength failure criterion,
reduction of compressive strength after cracking,
tension stiffening effect,
reduction of the shear stiffness after cracking (variable shear retention),
two crack models: fixed crack direction and rotated crack direction.
Perfect bond between concrete and reinforcement is assumed within the smeared concept. No
bond slip can be directly modeled except for the one included inherently in the tension stiffening.
However, on a macro-level a relative slip displacement of reinforcement with respect to concrete
over a certain distance can arise, if concrete is cracked or crushed. This corresponds to a real
mechanism of bond failure in case of the bars with ribs.
The reinforcement in both forms, smeared and discrete, is in the uniaxial stress state and its
constitutive law is a multi-linear stress-strain diagram.
26

The material matrix is derived using the nonlinear elastic approach. In this approach the elastic
constants are derived from a stress-strain function called here the equivalent uniaxial law. This
approach is similar to the nonlinear hypoelastic constitutive model, except that different laws are
used here for loading and unloading, causing the dissipation of energy exhausted for the damage
of material. The detailed treatment of the theoretical background of this subject can be found, for
example, in the book CHEN (1982). This approach can be also regarded as an isotropic damage
model, with the unloading modulus (see next section) representing the damage modulus.
The name SBETA comes from the former program, in which this material model was first used.
It means the abbreviation for the analysis of reinforced concrete in German language -
StahlBETonAnalyse.

2.1.2 Stress-Strain Relations for Concrete

2.1.2.1 Equivalent Uniaxial Law

The nonlinear behavior of concrete in the biaxial stress state is described by means of the so-
called effective stress σ c , and the equivalent uniaxial strain ε eq . The effective stress is in most
ef

cases a principal stress.

The equivalent uniaxial strain is introduced in order to eliminate the Poisson’s effect in the plane
stress state.
σ ci
ε eq = (2.10)
Eci
The equivalent uniaxial strain can be considered as the strain, that would be produced by the
stress σ ci in a uniaxial test with modulus Eci associated with the direction i. Within this
assumption, the nonlinearity representing a damage is caused only by the governing stress σ ci .
The details can be found in CHEN (1982).
The complete equivalent uniaxial stress-strain diagram for concrete is shown in Fig. 2-4.

Fig. 2-4 Uniaxial stress-strain law for concrete.

The numbers of the diagram parts in Fig. 2-4 (material state numbers) are used in the results of
the analysis to indicate the state of damage of concrete.
 27

Unloading is a linear function to the origin. An example of the unloading point U is shown in
Fig. 2-4. Thus, the relation between stress σ c and strain ε eq is not unique and depends on a load
ef

history. A change from loading to unloading occurs, when the increment of the effective strain
changes the sign. If subsequent reloading occurs the linear unloading path is followed until the
last loading point U is reached again. Then, the loading function is resumed.
The peak values of stress in compression f’cef and in tension f’tef are calculated according to the
biaxial stress state as will be shown in Sec.2.1.5. Thus, the equivalent uniaxial stress-strain law
reflects the biaxial stress state.
The above defined stress-strain relation is used to calculate the elastic modulus for the material
stiffness matrices, Sect. 2.1.11. The secant modulus is calculated as
s σc
Ec = (2.11)
ε eq
It is used in the constitutive equation to calculate stresses for the given strain state, Sect. 2.1.12.
The tangent modulus Ect is used in the material matrix Dc for construction of an element stiffness
matrix for the iterative solution. The tangent modulus is the slope of the stress-strain curve at a
given strain. It is always positive. In cases when the slope of the curve is less then the minimum
value Emint the value of the tangent modulus is set Ect = Emint. This occurs in the softening ranges
and near the compressive peak.
Detail description of the stress-strain law is given in the following subsections.

2.1.2.2 Tension before Cracking

The behavior of concrete in tension without cracks is assumed linear elastic. Ec is the initial
elastic modulus of concrete, ft ' ef is the effective tensile strength derived from the biaxial failure
function, Section 2.1.5.2.
σ c ef = Ec ε eq , 0 ≤ σ c ≤ ft ' ef (2.12)

2.1.2.3 Tension after Cracking

Two types of formulations are used for the crack opening:
A fictitious crack model based on a crack-opening law and fracture energy. This formulation is
suitable for modeling of crack propagation in concrete. It is used in combination with the
crack band, see Sect.2.1.3.
A stress-strain relation in a material point. This formulation is not suitable for normal cases of
crack propagation in concrete and should be used only in some special cases.
In following subsections are described five softening models included in SBETA material
model.
28

(1) Exponential Crack Opening Law

Fig. 2-5 Exponential crack opening law.

This function of crack opening was derived experimentally by HORDIJK (1991).

σ ⎧⎪ ⎛ w ⎞3 ⎫⎪ ⎛ w⎞ w
= ⎨1 + ⎜ c1 ⎟ ⎬ exp ⎜ −c2 ⎟ − (1 + c13 ) exp ( −c2 ) , (2.13)
f t ' ef ⎪⎩ ⎝ wc ⎠ ⎪⎭ ⎝ wc ⎠ wc

Gf
wc = 5.14
ft ' ef
where w is the crack opening, wc is the crack opening at the complete release of stress, σ is the
normal stress in the crack (crack cohesion). Values of the constants are, c1 =3, c2 =6.93. Gf is the
fracture energy needed to create a unit area of stress-free crack, ft 'ef is the effective tensile
strength derived from a failure function, Eq.(2.22). The crack opening displacement w is derived
from strains according to the crack band theory in Eq.(2.18).
(2) Linear Crack Opening Law

Fig. 2-6 Linear crack opening law.

σ c ef ft ' 2G
= ( wc − w) , wc = ' f (2.14)
f t ' ef wc ft
 29

(3) Linear Softening Based on Local Strain

Fig. 2-7 Linear softening based on strain.

The descending branch of the stress-strain diagram is defined by the strain c3 corresponding to
zero stress (complete release of stress).
(4) SFRC Based on Fracture Energy

Fig. 2-8 Steel fiber reinforced concrete based on fracture energy.

f1 f2 2G f
Parameters: c1 = , c2 = , wc =
ft ' ef
ft ' ef
f1 + f 2
(5) SFRC Based on Strain

Fig. 2-9 Steel fiber reinforced concrete based on strain.

30

f1 f2
Parameters: c1 = ' ef
, c2 =
ft f t ' ef
Parameters c1 and c2 are relative positions of stress levels, and c3 is the end strain.

2.1.2.4 Compression before Peak Stress

The formula recommended by CEB-FIP Model Code 90 has been adopted for the ascending
branch of the concrete stress-strain law in compression, Fig. 2-10. This formula enables wide
range of curve forms, from linear to curved, and is appropriate for normal as well as high
strength concrete.
kx − x 2 ε E
σ c ef = f c ' ef ,x = ,k = o (2.15)
1 + (k − 2) x εc Ec

Fig. 2-10 Compressive stress-strain diagram.

Meaning of the symbols in the above formula in:

σcef - concrete compressive stress,
f c'ef - concrete effective compressive strength (See Section 2.1.5.1)
x - normalized strain,
ε - strain,
εc - strain at the peak stress f’cef ,
k - shape parameter,
Eo - initial elastic modulus,
f ' ef
Ec - secant elastic modulus at the peak stress, Ec = c .
εc
Parameter k may have any positive value greater than or equal 1. Examples: k=1. linear, k=2. -
parabola.
As a consequence of the above assumption, distributed damage is considered before the peak
stress is reached. Contrary to the localized damage, which is considered after the peak.
 31

2.1.2.5 Compression after Peak Stress

The softening law in compression is linearly descending. There are two models of strain
softening in compression, one based on dissipated energy, and other based on local strain
softening.

2.1.2.5.1 Fictitious Compression Plane Model

The fictitious compression plane model is based on the assumption, that compression failure is
localized in a plane normal to the direction of compressive principal stress. All post-peak
compressive displacements and energy dissipation are localized in this plane. It is assumed that
this displacement is independent on the size of the structure. This hypothesis is supported by
experiments conducted by Van MIER (1986).
This assumption is analogous to the Fictitious Crack Theory for tension, where the shape of the
crack-opening law and the fracture energy are defined and are considered as material properties.

Fig. 2-11 Softening displacement law in compression.

In case of compression, the end point of the softening curve is defined by means of the plastic
displacement wd. In this way, the energy needed for generation of a unit area of the failure plane
is indirectly defined. From the experiments of Van MIER (1986), the value of wd =0.5mm for
normal concrete. This value is used as default for the definition of the softening in compression.
The softening law is transformed from a fictitious failure plane, Fig. 2-11, to the stress-strain
relation valid for the corresponding volume of continuous material, Fig. 2-10. The slope of the
softening part of the stress-strain diagram is defined by two points: a peak of the diagram at the
maximal stress and a limit compressive strain εd at the zero stress. This strain is calculated from
a plastic displacement wd and a band size L'd (see Section 2.1.3) according to the following
expression:
wd
εd = εc + (2.16)
L'd
The advantage of this formulation is reduced dependency on finite element mesh.

2.1.2.5.2 Compression Strain Softening Law Based on Strain.

A slope of the softening law is defined by means of the softening modulus Ed . This formulation
is dependent on the size of the finite element mesh.
32

2.1.3 Localization Limiters

So-called localization limiter controls localization of deformations in the failure state. It is a
region (band) of material, which represents a discrete failure plane in the finite element analysis.
In tension it is a crack, in compression it is a plane of crushing. In reality these failure regions
have some dimension. However, since according to the experiments, the dimensions of the
failure regions are independent on the structural size, they are assumed as fictitious planes. In
case of tensile cracks, this approach is known as rack the “crack band theory“, BAZANT, OH
(1983). Here is the same concept used also for the compression failure. The purpose of the
failure band is to eliminate two deficiencies, which occur in connection with the application of
the finite element model: element size effect and element orientation effect.
4 noded element
y
crack
direction
θ2
θ1

Lc
Lt x

Fig. 2-12 Definition of localization bands.

2.1.3.1 Element size effect.

The direction of the failure planes is assumed to be normal to the principal stresses in tension
and compression, respectively. The failure bands (for tension Lt and for compression Ld) are
defined as projections of the finite element dimensions on the failure planes as shown in Fig.
2-12.

2.1.3.2 Element orientation effect.

The element orientation effect is reduced, by further increasing of the failure band for skew
meshes, by the following formula (proposed by CERVENKA et al. 1995).
Lt = γ Lt , Ld = γ Ld
' '

θ
γ = 1 + (γ max − 1) , θ ∈ 〈0; 45〉 (2.17)
45
 33

An angle θ is the minimal angle ( min (θ1 , θ 2 ) ) between the direction of the normal to the failure
plane and element sides. In case of a general quadrilateral element the element sides directions
are calculated as average side directions for the two opposite edges. The above formula is a
linear interpolation between the factor γ=1.0 for the direction parallel with element sides, and
γ= γ max , for the direction inclined at 45o. The recommended (and default) value of γ max =1.5.

2.1.4 Fracture Process, Crack Width

The process of crack formation can be divided into three stages, Fig. 2-13. The uncracked stage
is before a tensile strength is reached. The crack formation takes place in the process zone of a
potential crack with decreasing tensile stress on a crack face due to a bridging effect. Finally,
after a complete release of the stress, the crack opening continues without the stress.
The crack width w is calculated as a total crack opening displacement within the crack band.
w = ε cr L't (2.18)
where ε cr is the crack opening strain, which is equal to the strain normal to the crack direction in
the cracked state after the complete stress release.

Fig. 2-13 Stages of crack opening.

It has been shown, that the smeared model based on the refined crack band theory can
successfully describe the discrete crack propagation in plain, as well as reinforced concrete
(CERVENKA et al. 1991, 1992, and 1995).
It is also possible, that the second stress, parallel to the crack direction, exceeds the tensile
strength. Then the second crack, in the direction orthogonal to the first one, is formed using the
same softening model as the first crack. (Note: The second crack may not be shown in a
graphical post-processing. It can be identified by the concrete state number in the second
direction at the numerical output.)

2.1.5 Biaxial Stress Failure Criterion of Concrete

2.1.5.1 Compressive Failure

A biaxial stress failure criterion according to KUPFER et al. (1969) is used as shown in Fig.
2-14. In the compression-compression stress state the failure function is
34

Fig. 2-14 Biaxial failure function for concrete.

1 + 3.65a ' σ
f c ' ef = f c , a = c1 (2.19)
(1 + a) 2
σ c2
where σ c1 , σ c 2 are the principal stresses in concrete and f’c is the uniaxial cylinder strength. In
the biaxial stress state, the strength of concrete is predicted under the assumption of a
proportional stress path.
In the tension-compression state, the failure function continues linearly from the point
σ c1 = 0 , σ c 2 = f c' into the tension-compression region with the linearly decreasing strength:
σ c1
f c ' ef = f c' rec , rec = (1 + 5.3278 ), 1.0 ≥ rec ≥ 0.9 (2.20)
f c'
where rec is the reduction factor of the compressive strength in the principal direction 2 due to
the tensile stress in the principal direction 1.

2.1.5.2 Tensile Failure

In the tension-tension state, the tensile strength is constant and equal to the uniaxial tensile
strength f’t. In the tension-compression state, the tensile strength is reduced by the relation:
ft ' ef = ft ' ret (2.21)
where ret is the reduction factor of the tensile strength in the direction 1 due to the compressive
stress in the direction 2. The reduction function has one of the following forms, Fig. 2-15.
σ c2
ret = 1 − 0.8 (2.22)
f c'
A + ( A − 1) B σ
ret = , B = Kx + A, x = c '2 (2.23)
AB fc
The relation in Eq.(2.22) is the linear decrease of the tensile strength and (2.23) is the hyperbolic
decrease.
 35

Two predefined shapes of the hyperbola are given by the position of an intermediate point r, x.
Constants K and A define the shape of the hyperbola. The values of the constants for the two
positions of the intermediate point are given in the following table.

type point parameters

r x A K
a 0.5 0.4 0.75 1.125
b 0.5 0.2 1.0625 6.0208

Fig. 2-15 Tension-compression failure function for concrete.

2.1.6 Two Models of Smeared Cracks

The smeared crack approach for modeling of the cracks is adopted in the model SBETA. Within
the smeared concept two options are available for crack models: the fixed crack model and the
rotated crack model. In both models the crack is formed when the principal stress exceeds the
tensile strength. It is assumed that the cracks are uniformly distributed within the material
volume. This is reflected in the constitutive model by an introduction of orthotropy.

2.1.6.1 Fixed Crack Model

In the fixed crack model (CERVENKA 1985, DARWIN 1974) the crack direction is given by
the principal stress direction at the moment of the crack initiation. During further loading this
direction is fixed and represents the material axis of the orthotropy.
36

Fig. 2-16 Fixed crack model. Stress and strain state.

The principal stress and strain directions coincide in the uncracked concrete, because of the
assumption of isotropy in the concrete component. After cracking the orthotropy is introduced.
The weak material axis m1 is normal to the crack direction, the strong axis m2 is parallel with the
cracks.
In a general case the principal strain axes ε1 and ε2 rotate and need not to coincide with the axes
of the orthotropy m1 and m2. This produces a shear stress on the crack face as shown in Fig.
2-16. The stress components σc1 and σc2 denote, respectively, the stresses normal and parallel to
the crack plane and, due to shear stress, they are not the principal stresses. The shear stress and
stiffness in the cracked concrete is described in Section 2.1.7.

2.1.6.2 Rotated Crack Model

In the rotated crack model (VECCHIO 1986, CRISFIELD 1989), the direction of the principal
stress coincides with the direction of the principal strain. Thus, no shear strain occurs on the
crack plane and only two normal stress components must be defined, as shown in Fig. 2-17.

Fig. 2-17 Rotated crack model. Stress and strain state.

If the principal strain axes rotate during the loading the direction of the cracks rotate, too. In
order to ensure the co-axiality of the principal strain axes with the material axes the tangent shear
modulus Gt is calculated according to CRISFIELD 1989 as
σ c1 − σ c 2
Gt = (2.24)
2(ε1 − ε 2 )
 37

2.1.7 Shear Stress and Stiffness in Cracked Concrete

In case of the fixed crack model, the shear modulus is reduced according to the law derived by
KOLMAR (1986) after cracking. The shear modulus is reduced with growing strain normal to
the crack, Fig. 2-18 and this represents a reduction of the shear stiffness due to the crack
opening.

Fig. 2-18 Shear retention factor.

⎛ 1000ε u ⎞
− ln ⎜ ⎟
⎝ c1 ⎠
G = rg Gc , rg = c3 (2.25)
c2
c1 = 7 + 333( p − 0.005),c2 = 10 − 167( p − 0.005),0 ≤ p ≤ 0.02
where rg is the shear retention factor, G is the reduced shear modulus and Gc is the initial
concrete shear modulus
Ec
Gc = (2.26)
2(1 + ν )
where Ec is the initial elastic modulus and ν is the Poisson's ratio. The strain ευ is normal to the
crack direction (the crack opening strain), c1 and c2 are parameters depending on the reinforcing
crossing the crack direction, p is the transformed reinforcing ratio (all reinforcement is
transformed on the crack plane) and c3 is the user’s scaling factor. By default c3=1. In ATENA
the effect of reinforcement ratio is not considered, and p is assumed to be 0.0.
There is an additional constraint imposed on the shear modulus. The shear stress on the crack
plane τ uv = Gγ is limited by the tensile strength f’t. The secant and tangent shear moduli of
cracked concrete are equal.

2.1.8 Compressive Strength of Cracked Concrete

A reduction of the compressive strength after cracking in the direction parallel to the cracks is
done by a similar way as found from experiments of VECCHIO and COLLINS 1982 and
formulated in the Compression Field Theory. However, a different function is used for the
reduction of concrete strength here, in order to allow for user's adjustment of this effect. This
function has the form of the Gauss's function, Fig. 2-19. The parameters of the function were
38

derived from the experimental data published by KOLLEGER et al. 1988, which included also
data of Collins and Vecchio (VECCHIO at al.1982)
2
f c ' ef = rc f c' , rc = c + (1 − c) e− (128εu ) (2.27)
For the zero normal strain, ευ, there is no strength reduction, and for the large strains, the
strength is asymptotically approaching to the minimum value f c ' ef = cf c' .

Fig. 2-19 Compressive strength reduction of cracked concrete.

The constant c represents the maximal strength reduction under the large transverse strain. From
the experiments by KOLLEGGER et all. 1988, the value c = 0.45 was derived for the concrete
reinforced with the fine mesh. The other researchers (DYNGELAND 1989) found the reductions
not less than c=0.8. The value of c can be adjusted by input data according to the actual type of
reinforcing.
However, the reduction of compressive strength of the cracked concrete does not have to be
effected only by the reinforcing. In the plain concrete, when the strain localizes in one main
crack, the compressive concrete struts can cross this crack, causing so-called "bridging effect".
The compressive strength reduction of these bridges is also captured by the above model.

2.1.9 Tension Stiffening in Cracked Concrete

The tension stiffening effect can be described as a contribution of cracked concrete to the tensile
stiffness of reinforcing bars. This stiffness is provided by the uncracked concrete or not fully
opened cracks and is generated by the strain localization process. It was verified by simulation
experiments of HARTL, G., 1977 and published in the paper (MARGOLDOVA et.al. 1998).
Including an explicit tension stiffening factor would result in an overestimation of this effect.
Therefore, in the ATENA versions up to1.2.0 no explicit tension stiffening factor is possible in
the input.

2.1.10 Summary of Stresses in SBETA Constitutive Model

In the case of uncracked concrete the stress symbols have the following meaning:
σ c1 - maximal principal stress
σ c 2 - minimal principal stress
(tension positive, compression negative)
 39

In the case of cracked concrete, Fig. 2-16 stresses are defined on the crack plane:
σ c1 - normal stress normal to the cracks
σ c 2 - normal stress parallel to the cracks
τ c - shear stress on the crack plane

2.1.11 Material Stiffness Matrices

2.1.11.1 Uncracked Concrete

The material stiffness matrix for the uncracked concrete has the form of an elastic matrix of the
isotropic material. It is written in the global coordinate system x and y.
⎡ ⎤
⎢1 ν 0 ⎥
E ⎢ ⎥
Dc = 2 ⎢
ν 1 0 ⎥ (2.28)
1 −ν ⎢
1 −ν ⎥
⎢0 0 ⎥
⎣ 2 ⎦
In the above E is the concrete elastic modulus derived from the equivalent uniaxial law. The
Poisson's ratio ν is constant.

2.1.11.2 Cracked Concrete

For the cracked concrete the matrix has the form of the elastic matrix for the orthotropic
material. The matrix is formulated in a coordinate system m1, m2, Fig. 2-16 and Fig. 2-17, which
is coincident with the crack direction. This local coordinate system is referred to the superscript
L later. The direction 1 is normal to the crack and the direction 2 is parallel with the crack. The
definition of the elastic constants for the orthotropic material in the plane stress state follows
from the flexibility relation:
⎡ 1 ν ⎤
⎢ E − 21 0 ⎥
E2
⎧ ε1 ⎫ ⎢ 1 ⎥ ⎧σ 1 ⎫
⎪ ⎪ ⎢ ν 12 1 ⎥⎪ ⎪
⎨ε 2 ⎬ = ⎢ − 0 ⎥ ⎨σ 2 ⎬ (2.29)
⎪ γ ⎪ ⎢ E1 E2 ⎪
⎥ ⎩τ ⎭ ⎪
⎩ ⎭
⎢ 1⎥
⎢ 0 0
G ⎥⎦
⎣
First we eliminate the othotropic Poisson’s ratios for the cracked concrete, because they are
commonly not known. For this we use the symmetry relation ν 12 E2 = ν 21 E1 . Therefore, in (2.29)
there are only three independent elastic constants E1 , E2 ,ν 21 . Assuming that ν 21 = ν is the
Poisson's ratio of the uncracked concrete and using the symmetry relation, we obtain
E1
ν 12 = ν (2.30)
E2
L
The stiffness matrix Dc is found as the inverse of the flexibility matrix in (2.30):
40

⎡ ξ νξ 0 ⎤
D c = H ⎢⎢νξ 1 0 ⎥⎥,
L

⎢⎣ 0 0 G ⎥⎦ (2.31)
E
ξ = 1 ,H = E1 (1 − ξν 2 )
E2
In the above relation E2 must be nonzero. If E2 is zero and E1 is nonzero, then an alternative
1 E
formulation is used with the inverse parameter = 2 . In case that both elastic moduli are
ξ E1
L
zero, the matrix Dc is set equal to the null matrix.
L
The matrix Dc is transformed into the global coordinate system using the transformation matrix
Tε from (2.8).
D c = TεT D cL Tε (2.32)

The angle α is between the global axis x and the 1st material axis m1, which is normal to the
crack, Fig. 2-16.

2.1.11.3 Smeared Reinforcement

The material stiffness matrix of the ith smeared reinforcement is
⎡ cos( β i ) 4 cos( β i )2 sin( β i )2 cos( β i )3 sin( β i ) ⎤
⎢ ⎥
Dsi = pi Esi ⎢cos( β i )2 sin( β i )2 sin( β i ) 4 cos( β i ) sin( β i )3 ⎥ (2.33)
⎢ cos( β i )3 sin( β i ) cos( β i ) sin( β i )3 cos( β i )2 sin( β i )2 ⎥⎦
⎣
The angle β is between the global axis x and the ith reinforcement direction, and Esi is the elastic
modulus of reinforcement. The reinforcing ratio pi =As/Ac.

2.1.11.4 Material Stiffness of Composite Material

The total material stiffness of the reinforced concrete is the sum of material stiffness of concrete
and smeared reinforcement:
n
D = Dc + ∑ Dsi (2.34)
i =1

The summation is over n smeared reinforcing components. In ATENA the smeared

reinforcement is not added on the constitutive level, but it is modeled by a separate layers of
elements whose nodes are connected to those of the concrete elements. This corresponds to the
assumption of perfect bond between the smeared reinforcement and concrete.

2.1.11.5 Secant and Tangent Material Stiffness

The material stiffness matrices in the above Subsections 2.1.11.1, 2.1.11.2, 2.1.11.3, 2.1.11.4 are
either secant or tangent, depending on the type of elastic modulus used.
The secant material stiffness matrix is used to calculate the stresses for the given strains, as
shown in Section 2.1.12.
The tangent material stiffness matrix is used to construct the element stiffness matrix.
 41

2.1.12 Analysis of Stresses

The stresses in concrete are obtained using the actual secant component material stiffness matrix
s
s c = Dc e (2.35)
s
where Dc is the secant material stiffness matrix from Section 2.1.11 for the uncracked or
cracked concrete depending on the material state. The stress components are calculated in the
global as well as in the local material coordinates (the principal stresses in the uncracked
concrete and the stresses on the crack planes).
The stress in reinforcement and the associated tension stiffening stress is calculated directly from
the strain in the reinforcement direction.

2.1.13 Parameters of Constitutive Model

Default formulas of material parameters:

Parameter: Formula:
Cylinder strength f c' = −0.85 f cu'
Tensile strength 2
ft ' = 0.24 f cu' 3
Initial elastic modulus Ec = (6000 − 15.5 f cu' ) f cu'

Poisson's ratio ν = 0.2

Softening compression wd = −0.0005mm
Type of tension softening 1 – exponential, based on GF
Compressive strength in cracked concrete c = 0.8
Tension stiffening stress σ st = 0.
Shear retention factor variable (Sect.2.1.7)
Tension-compression function type linear
Fracture energy Gf according to VOS 1983 GF = 0.000025 ft ' ef [MN/m]
Orientation factor for strain localization γ max = 1.5 (Sect.2.1.3)

The SBETA constitutive model of concrete includes 20 material parameters. These parameters
are specified for the problem under consideration by user. In case of the parameters are not
known automatic generation can be done using the default formulas given in the table above. In
such a case, only the cube strength of concrete f’cu (nominal strength) is specified and the
remaining parameters are calculated as functions of the cube strength. The formulas for these
functions are taken from the CEB-FIP Model Code 90 and other research sources.
Used units are MPa.
The parameters not listed in the table have zero default value.
42

The values of the material parameters can be also influenced by safety considerations. This is
particularly important in cases of a design, where a proper safety margin should be met. For that
reason the choice of material properties depends on the purpose of analysis and the filed of an
application. The typical examples of the application are the design, the simulation of failure and
the research.
In case of the design application, according to most current standards, the material properties for
calculation of structural resistance (failure load) are considered by minimal values with applied
partial safety factors. The resulting maximum load can be directly compared with the design
loads.
According to some researchers, more appropriate approach would be to consider the average
material properties in nonlinear analysis and to apply a safety factor on the resulting integral
response variable (force, moment). However, this safety format is not yet fully established.
In cases of the simulation of real behavior, the parameters should be chosen as close as possible
to the properties of real materials. The best way is to determine these properties from mechanical
tests on material sample specimens.
 43

2.2 Fracture–Plastic Constitutive Model (CC3DCementitious,

CC3DNonLinCementitious, CC3DNonLinCementitious2,
CC3DNonLinCementitious2User,
CC3DNonLinCementitious2Variable,
CC3DNonLinCementitious2SHCC,
CC3DNonLinCementitious3)

2.2.1 Introduction
Fracture-plastic model combines constitutive models for tensile (fracturing) and compressive
(plastic) behavior. The fracture model is based on the classical orthotropic smeared crack
formulation and crack band model. It employs Rankine failure criterion, exponential softening,
and it can be used as rotated or fixed crack model. The hardening/softening plasticity model is
based on Menétrey-Willam failure surface. The model uses return mapping algorithm for the
integration of constitutive equations. Special attention is given to the development of an
algorithm for the combination of the two models. The combined algorithm is based on a
recursive substitution, and it allows for the two models to be developed and formulated
separately. The algorithm can handle cases when failure surfaces of both models are active, but
also when physical changes such as crack closure occur. The model can be used to simulate
concrete cracking, crushing under high confinement, and crack closure due to crushing in other
material directions.
Although many papers have been published on plasticity models for concrete (for instance,
PRAMONO, WILLAM 1989, MENETREY et al 1997, FEENSTRA 1993, 1998 ETSE 1992) or
smeared crack models (RASHID 1968, CERVENKA and GERSTLE 1971, BAZANT and OH
1983, DE BORST 1986, ROTS 1989), there are not many descriptions of their successful
combination in the literature. OWEN et al. (1983) presented a combination of cracking and
visco-plasticity. Comprehensive treatise of the problem was provided also by de BORST (1986),
and recently several works have been published on the combination of damage and plasticity
(SIMO and JU 1987, MESCHKE et al. (1998). The presented model differs from the above
formulations by ability to handle also physical changes like for instance crack closure, and it is
not restricted to any particular shape of hardening/softening laws. Also within the proposed
approach it is possible to formulate the two models (i.e. plastic and fracture) entirely separately,
and their combination can be provided in a different algorithm or model. From programming
point of view such approach is well suited for object oriented programming.
The method of strain decomposition, as introduced by DE BORST (1986), is used to combine
fracture and plasticity models together. Both models are developed within the framework of
return mapping algorithm by WILKINS (1964). This approach guarantees the solution for all
magnitudes of strain increment. From an algorithmic point of view the problem is then
transformed into finding an optimal return point on the failure surface.
The combined algorithm must determine the separation of strains into plastic and fracturing
components, while it must preserve the stress equivalence in both models. The proposed
algorithm is based on a recursive iterative scheme. It can be shown that such a recursive
algorithm cannot reach convergence in certain cases such as, for instance, softening and dilating
materials. For this reason the recursive algorithm is extended by a variation of the relaxation
method to stabilize convergence.
44

2.2.2 Material Model Formulation

The material model formulation is based on the strain decomposition into elastic ε ije , plastic ε ijp
and fracturing ε ijf components (DE BORST 1986).

ε ij = ε ije + ε ijp + ε ijf (2.36)

The new stress state is then computed by the formula:

σ ijn = σ ijn −1 + Eijkl (Δε kl − Δε klp − Δε klf ) (2.37)

where the increments of plastic strain Δε ijp and fracturing strain Δε ijf must be evaluated based on
the used material models.

2.2.3 Rankine-Fracturing Model for Concrete Cracking

Rankine criterion is used for concrete cracking

Fi f = σ ii′ t − f t′i ≤ 0 (2.38)

It is assumed that strains and stresses are converted into the material directions, which in case of
rotated crack model correspond to the principal directions, and in case of fixed crack model, are
given by the principal directions at the onset of cracking. Therefore, σ ii′ t identifies the trial
stress and f t′i tensile strength in the material direction i . Prime symbol denotes quantities in the
material directions. The trial stress state is computed by the elastic predictor.
σ ij′ t = σ ij′ n −1 + Eijkl Δε kl′ (2.39)

If the trial stress does not satisfy (2.38), the increment of fracturing strain in direction i can be
computed using the assumption that the final stress state must satisfy (2.40).
Fi f = σ ii′ n − f t′i = σ ii′ t − Eiikl Δε kl′ f − f t′i = 0 (2.40)
This equation can be further simplified under the assumption that the increment of fracturing
strain is normal to the failure surface, and that always only one failure surface is being checked.
For failure surface k , the fracturing strain increment has the following form.
∂ Fk f
Δε ij′ f = Δλ = Δλ δ ik (2.41)
∂ σ ij
After substitution into (2.40) a formula for the increment of the fracturing multiplier λ is
recovered.
σ kk′ t − f t′k σ kk′ t − f t′( wkmax )
Δλ = = ′ f + Δλ )
and wkmax = Lt (εˆ kk (2.42)
E kkkk E kkkk
This equation must be solved by iterations since for softening materials the value of current
tensile strength f t′( wkmax ) is a function of the crack opening w , and is based on Hordijk’s formula
(defined in SBETA model).
′ f in direction k ,
The crack opening w is computed from the total value of fracturing strain εˆ kk
plus the current increment of fracturing strain Δλ , and this sum is multiplied by the
 45

characteristic length Lt . The characteristic length as a crack band size was introduced by
BAZANT and OH. Various methods were proposed for the crack band size calculation in the
framework of finite element method. FEENSTRA (1993) suggested a method based on
integration point volume, which is not well suited for distorted elements. A consistent and rather
complex approach was proposed by OLIVIER. In the presented work the crack band size Lt is
calculated as a size of the element projected into the crack direction, Fig. 2-20. CERVENKA V.
et al. (1995) showed that this approach is satisfactory for low order linear elements, which are
used throughout this study. They also proposed a modification, which accounts for cracks that
are not aligned with element edges.

Fig. 2-20 Tensile softening and characteristic length

Equation (2.42) can be solved by recursive substitutions. It is possible to show by expanding

f t′( wkmax ) into a Taylor series that this iteration scheme converges as long as:

∂f t′( wkmax ) E kkkk

− < (2.43)
∂w Lt
Equation (2.43) is violated for softening materials only when snap back is observed in the stress-
strain relationship, which can occur if large finite elements are used. In the standard
displacement based finite element method, the strain increment is given, therefore, a snap back
on the constitutive level cannot be captured. This means that the critical region, with snap back
on the softening curve, will be skipped in a real calculation, which physically means, that the
energy dissipated by the system will be over estimated. This is of course undesirable, and finite
E ∂f t′(0) ∂f ′(0)
elements smaller then L < kkkk should be used, where t denotes the initial slope
∂w ∂w
of the crack softening curve.

It is important to distinguish between total fracturing strain εˆij′ f , which corresponds to the
maximal fracturing strain reached during the loading process, and current fracturing strain ε ij′ f ,
which can be smaller due to crack closure, and is computed using (2.44) derived by ROTS and
BLAUWENDRAAD.

ε kl′ f = ( Eijkl + Eijkl

′ cr ) −1 E klmn ε mn
′ , and Eijlk
′cr is defined by σ ij′ = Eijkl
′ cr ε kl′ f (2.44)
46

′cr represents the cracking stiffness in the local material directions.

The fourth order crack tensor Eijkl
In the current formulation, it is assumed, that there is no interaction between normal and shear
components. Thus, the crack tensor is given by the following formulas.
′cr = 0 for i ≠ k and j ≠ l
Eijkl (2.45)

Mode I crack stiffness equals

f t′( wimax )
′ cr =
Eiiii , (no summation of indices) (2.46)
εˆii′ f
and mode II and III crack stiffness is assumed as:
′cr = sF min ( Eiiii
Eijij ´cr
, E ´jjjj
cr
) , (no summation of indices) (2.47)

where i ≠ j , and sF is a shear factor coefficient that defines a relationship between the normal
and shear crack stiffness. The default value of sF is 20.
Shear strength of a cracked concrete is calculated using the Modified Compression Field Theory
of VECHIO and COLLINS (1986).
0.18 f c′
σ ij ≤ , i≠ j (2.48)
24 w
0.31 +
ag + 16

Where f c′ is the compressive strength in MPa, a g is the maximum aggregate size in mm and w
is the maximum crack width in mm at the given location. This model is activated by specifying
the maximum aggregate size a g otherwise the default behavior is used where the shear stress on
a crack surface cannot axceed the tensile strength.
The secant constitutive matrix in the material direction was formulated by ROTS and
BLAUWENDRAAD in the matrix format.
E′ s = E - E(E′ cr + E) -1 E (2.49)

Strain vector transformation matrix Tε (i.e. global to local strain transformation matrix) can be
used to transform the local secant stiffness matrix to the global coordinate system.
T
E s = Tε E′s Tε (2.50)
It is necessary to handle the special cases before the onset of cracking, when the crack stiffness
approaches infinity. Large penalty numbers are used for crack stiffness in these cases.

2.2.4 Plasticity Model for Concrete Crushing

New stress state in the plastic model is computed using the predictor-corrector formula.
σ ijn = σ ijn −1 + E ijkl ( Δε kl − Δε klp ) = σ ijt − E ijkl Δε klp = σ ijt − σ ijp (2.51)

The plastic corrector σ ijp is computed directly from the yield function by return mapping
algorithm.
F p (σ ijt − σ ijp ) = F p (σ ijt − Δλ lij ) = 0 (2.52)
 47

The crucial aspect is the definition of the return direction lij , which can be defined as

∂ G p (σ klt ) ∂ G p (σ ijt )
lij = E ijkl then Δε ijp = Δλ (2.53)
∂ σ kl ∂σ ij
where G (σ ij ) is the plastic potential function, whose derivative is evaluated at the predictor stress
state σ ijt to determine the return direction.

The failure surface of MENETREY, WILLAM is used in the current version of the material
model.

⎡ ρ⎤
2
⎡ ρ ξ ⎤
3P . ' ⎥ + m⎢
F = ⎢ 15
p
' r (θ , e ) + ⎥−c = 0 (2.54)
⎣ fc ⎦ ⎢⎣ 6 f c 3 f c' ⎥⎦

where
f c'2 − ft '2 e 4(1 − e 2 ) cos2 θ + (2e − 1) 2
m=3 , r (θ , e) = 1
f c' ft ' e + 1
[
2(1 − e ) cos θ + (2e − 1) 4(1 − e ) cos θ + 5e − 4e
2 2 2 2
] 2

In the above equations (ξ , ρ ,θ ) are Heigh-Vestergaard coordinates, f c′ and f t′ is compressive

strength and tensile strength respectively. Parameter e ∈ 0.5,10 . defines the roundness of the
failure surface. The failure surface has sharp corners if e = 0.5 , and is fully circular around the
hydrostatic axis if e = 10
. .
The position of failure surfaces is not fixed but it can move depending on the value of strain
hardening/softening parameter. The strain hardening is based on the equivalent plastic strain,
which is calculated according to the following formula.
Δε eqp = min(Δε ijp ) (2.55)

For Menétrey-Willam surface the hardening/softening is controlled by the parameter c ∈ 0,1 ,

which evolves during the yielding/crushing process by the following relationship:
2
⎛ f c′(ε eqp ) ⎞
c=⎜ ⎟ (2.56)
⎜ f′ ⎟
⎝ c ⎠
In the above two formulas the expression f c′(ε eqp ) indicates the hardening/softening law, which is
based on the uniaxial compressive test. The law is shown in Fig. 2-21, where the softening curve
is linear and the elliptical ascending part is given by the following formula:
2
⎛ ε c − ε eqp ⎞
σ = f co + ( f c − f co ) 1− ⎜
⎜ ε ⎟⎟ (2.57)
⎝ c ⎠
48

f'c

f’c0= 2f’t
εeq
p

εc =f'c/E
p

Fig. 2-21. Compressive hardening/softening and compressive characteristic length. Based on

experimental observations by VAN MIER.

The law on the ascending branch is based on strains, while the descending branch is based on
displacements to introduce mesh objectivity into the finite element solution, and its shape is
based on the work of VAN MIER. The onset of nonlinear behavior f c'0 is an input parameter as
well as the value of plastic strain at compressive strength ε cp . The Fig. 2-21 shows typical values
of these parameters. Especially the choice of the parameter f c'0 should be selected with care,
since it is important to ensure that the fracture and plastic surfaces intersect each other in all
material stages. On the descending curve the equivalent plastic strain is transformed into
displacements through the length scale parameter Lc . This parameter is defined by analogy to
the crack band parameter in the fracture model in Sec. 2.2.3, and it corresponds to the projection
of element size into the direction of minimal principal stresses. The square in (2.56) is due to the
quadratic nature of the Menétry-Willam surface.
Return direction is given by the following plastic potential
1
G p (σ ij ) = β I1 + 2 J 2 (2.58)
3
where β determines the return direction. If β < 0 material is being compacted during crushing,
if β = 0 material volume is preserved, and if β > 0 material is dilating. In general the plastic
model is non-associated, since the plastic flow is not perpendicular to the failure surface
The return mapping algorithm for the plastic model is based on predictor-corrector approach as
is shown in Fig. 2-22. During the corrector phase of the algorithm the failure surface moves
along the hydrostatic axis to simulate hardening and softening. The final failure surface has the
apex located at the origin of the Haigh-Vestergaard coordinate system. Secant method based
Algorithm 1 is used to determine the stress on the surface, which satisfies the yield condition and
also the hardening/softening law.
 49

Fig. 2-22 Plastic predictor-corrector algorithm.

Fig. 2-23. Schematic description of the iterative process (2.73). For clarity shown in two dimensions.
50

n −1
Algorithm 1: (Input is σ ijn −1 , ε ijp , Δε ijn )

Elastic predictor: σ ijt = σ ijn −1 + Eijkl Δε kln (2.59)

n −1
Evaluate failure criterion: f Ap = F p (σ ijt , ε ijp ) , Δλ A = 0 (2.60)

If failure criterion is violated i.e. f Ap > 0

∂ G p (σ ijt )
Evaluate return direction: mij = (2.61)
∂σ ij
n −1
Return mapping: F p (σ ijt − Δλ B E mij , ε ijp ) = 0 ⇒ Δλ B (2.62)
n −1
Evaluate failure criterion: f Bp = F p (σ ijt − Δλ B E mij , ε ijp + Δλ B mij ) (2.63)

Secant iterations (i ) as long as ΔλA − ΔλB > e (2.64)

ΔλB − Δλ A
New plastic multiplier increment: Δλ = Δλ A − f Ap (2.65)
f Bp − f Ap

∂ G p (σ ijt − Δλ E mij( i −1) )

New return direction: m (i )
ij = (2.66)
∂σ ij

Evaluate failure criterion: f p

= F p (σ ijt − Δλ E mij(i ) , ε ijp + Δλ mij(i ) ) (2.67)

New initial values for secant iterations:

f Bp < 0 ⇒ f Bp = f p , ΔλB = Δλ (2.68)

f Bp ≥ 0 ⇒ f Ap = f Bp , Δλ A = ΔλB , f Bp = f p , ΔλB = Δλ (2.69)

End of secant iteration loop

End of algorithm update stress and plastic strains.
n n −1
ε ijp = ε ijp + Δλ B mij( i ) , σ ijn = σ ijt − Δλ B E mij(i ) (2.70)

2.2.5 Combination of Plasticity and Fracture model

The objective is to combine the above models into a single model such that plasticity is used for
concrete crushing and the Rankine fracture model for cracking. This problem can be generally
stated as a simultaneous solution of the two following inequalities.
F p (σ ijn −1 + Eijkl (Δε kl − Δε klf − Δε klp )) ≤ 0 solve for Δε klp (2.71)

F f (σ ijn −1 + Eijkl (Δε kl − Δε klp − Δε klf )) ≤ 0 solve for Δε klf (2.72)

Each inequality depends on the output from the other one, therefore the following iterative
scheme is developed.
 51

Algorithm 2:
( i −1) ( i −1) (i ) (i )
Step 1: F p (σ ijn −1 + Eijkl (Δε kl − Δε klf + bΔε klcor − Δε klp )) ≤ 0 solve for Δε klp
(i ) (i ) (i )
Step 2: F f (σ ijn −1 + Eijkl (Δε kl − Δε klp − Δε klf )) ≤ 0 solve for Δε klf
(i ) (i ) ( i −1)
Step 3: Δε ijcor = Δε ijf − Δε ijf (2.73)

Iterative correction of the strain norm between two subsequent iterations can be expressed as
(i ) ( i −1)
Δε ijcor = (1 − b) α f α p Δε ijcor (2.74)

(i ) ( i −1)
Δεijf − Δεijf Δεijp
(i )
− Δεijp
( i −1)

where α = f
,α =p

Δε p (i )
ij − Δε p ( i −1)
ij
Δεijcor

and b is an iteration correction or relaxation factor, which is introduced in order to guarantee

convergence. It is to be determined based on the run-time analysis of α f and α p , such that the
convergence of the iterative scheme can be assured. The parameters α f and α p characterize the
mapping properties of each model (i.e. plastic and fracture). It is possible to consider each model
as an operator, which maps strain increment on the input into a fracture or plastic strain
increment on the output. The product of the two mappings must be contractive in order to obtain
a convergence. The necessary condition for the convergence is:
(1 − b)α f α p < 1 (2.75)

If b equals 0 , an iterative algorithm based on recursive substitution is obtained. The

convergence can be guaranteed only in two cases:
One of the models is not activated (i.e. implies α f or α p = 0 ),
There is no softening in either of the two models and dilating material is not used in the plastic
part, which for the plastic potential in this work means β ≤ 0 , (2.58). This is a sufficient but
not necessary condition to ensure that α f and α p < 1 .
It can be shown that the values of α f and α p are directly proportional to the softening rate in
each model. Since the softening model remains usually constant for a material model and finite
element, their values do not change significantly between iterations. It is possible to select the
scalar b such that the inequality (2.75) is satisfied always at the end of each iteration based on
the current values of α f and α p . There are three possible scenarios, which must be handled, for
the appropriate calculation of b :
α fα p ≤ χ , where χ is related to the requested convergence rate. For linear rate it can be
set to χ = 1 / 2 . In this case the convergence is satisfactory and b = 0 .

χ < α fα p < 1 , then the convergence would be too slow. In this case b can be estimated
α fα p
as b = 1 − , in order to increase the convergence rate.
χ
52

1 ≤ α f α p , then the algorithm is diverging. In this case b should be calculated as

χ
b = 1− to stabilize the iterations.
α αp
f

This approach guarantees convergence as long as the parameters α p , α f do not change

drastically between the iterations, which should be satisfied for smooth and correctly formulated
models. The rate of convergence depends on material brittleness, dilating parameter β and finite
element size. It is advantageous to further stabilize the algorithm by smoothing the parameter b
during the iterative process:
b = (b ( i ) + b ( i −1) ) / 2 (2.76)
where the superscript i denotes values from two subsequent iterations. This will eliminate
problems due to the oscillation of the correction parameter b . Important condition for the
convergence of the above Algorithm 2 is that the failure surfaces of the two models are
intersecting each other in all possible positions even during the hardening or softening.
Additional constraints are used in the iterative algorithm. If the stress state at the end of the first
step violates the Rankine criterion, the order of the first two steps in Algorithm 2 is reversed.
Also in reality concrete crushing in one direction has an effect on the cracking in other
directions. It is assumed that after the plasticity yield criterion is violated, the tensile strength in
all material directions is set to zero.
On the structural level secant matrix is used in order to achieve a robust convergence during the
strain localization process.
The proposed algorithm for the combination of plastic and fracture models is graphically shown
in Fig. 2-23. When both surfaces are activated, the behavior is quite similar to the multi-surface
plasticity (SIMO et al. 1988). Contrary to the multi-surface plasticity algorithm the proposed
method is more general in the sense that it covers all loading regimes including physical changes
such as for instance crack closure. Currently, it is developed only for two interacting models, and
its extension to multiple models is not straightforward.
There are additional interactions between the two models that need to be considered in order to
properly describe the behavior of a concrete material:
(a) After concrete crushing the tensile strength should decrease as well
(b) According to the research work of Collins (VECHIO and COLLINS (1986)) and
coworkers it was established the also compressive strength should decrease when
cracking occurs in the perpendicular direction. This theory is called compression field
theory and it is used to explain the shear failure of concrete beams and walls.
The interaction (a) is resolved by adding the equivalent plastic strain to the maximal fracturing
strain in the fracture model to automatically increase the tensile damage based on the
compressive damage such that the fracturing strains satisfies the following condition:

ft′ p
εˆkk′ f ≥ ε eq (2.77)
f c′
The compressive strength reduction (b) is based on the following formula based proposed by
Collins:

σ c = rc f c′
 53

1
rc = , rclim ≤ rc ≤ 1.0 (2.78)
0.8 + 170 ε1
Where ε1 is the tensile strain in the crack. In ATENA the largest maximal fracturing strain is
used for ε1 and the compressive strength reduction is limited by rclim . If rclim is not specified then
no compression reduction is considered.

2.2.6 Variants of the fracture plastic model

The several ATENA material models are based on the above theories: CC3DCementitious,
CC3DNonLinCementitious, CC3DNonLinCementitious2, CC3DNonLinCementitious2Variable,
CC3DNonLinCementitious2Fatigue (described in section 2.2.9),
CC3DNonLinCementitious2User, CC3DNONLINCEMENTITIOUS2SHCC (described in section
2.2.10), and CC3DNonLinCementitious3 (described in section 2.2.11), with the following
differences:
CC3DCementitious assumes linear response up to the point when the failure envelope is reached
both in tension and compression. This means that there is no hardening regime in Fig. 2-21. The
material CC3DNonLinCementitious on the contrary assumes a hardening regime before the
compressive strength is reached. The material CC3DNonLinCementitious2 is equivalent to
CC3DNonLinCementitious but purely incremental formulation is used (in
CC3DNonLinCementitious a total formulation is used for the fracturing part of the model),
therefore this material can be used in creep calculations or when it is necessary to change
material properties during the analysis. The material CC3DNonLinCementitious2Variable is
based on the material CC3DNonLinCementitious2 and it allows to define history evolution laws
for selected material parameters. The following material parameters can be defined using an
arbitrary evolution laws: young modulus E , tensile strength ft ' , compressive strength f c' and
f c'0 . It is the responsibility of the user to define the parameters in a meaningful way. It means
that at any time:
ft ' < 1 2 f c'0 (2.79)

f c'0 > f c'0 , f c'0 ≤ 0, f c' ≤ 0 (2.80)

The material CC3DNonLinCementitious2User allows for user defined laws for selected material
laws such as: diagrams for tensile and softening behavior (see Fig. 2-24 and Fig. 2-25), shear
retention factor (Fig. 2-26) and the effect of lateral compression on tensile strength (Fig. 2-27).
54

σt/f’t

1.0
t
(εf1 − εt ) Lt/Lch
loc

εt loc ε~f1
Fig. 2-24. An example of a user defined tensile behavior for CC3DNonLinCementitious2User
material.

σc/f’c

1.0
p c
(εeq − εc ) Lc/Lch
loc

εcloc
ε~ peq<0
Fig. 2-25. An example of a user defined compressive behavior for CC3DNonLinCementitious2User
material.

G/Gc

1.0
f sh t
(ε1 − ε ) Lt/Lch
loc

~
εsh
loc εf1
Fig. 2-26. An example of a user defined shear retention factor for shear stiffness degradation after
cracking.
 55

In the user defined material mode II and III crack stiffness are evaluated with the help of the
shear retention factor rg as:

rg G
′cr =
Eijij (2.81)
1 − rg

where i ≠ j , rg = min(rgi , rgj ) is the minimum of shear retention factors on cracks in directions
i , j , and G is the elastic shear modulus. Shear retention factor on a crack in direction i is
evaluated from the user specified diagram as shown in Fig. 2-26.
In the above diagrams Lt and Lc represents the crack band size and crush band size respectively
as it is defined Section 2.1.3. Ltch and Lcch represents a size for which the tensile and compression
diagram respectively is valid. For instance it represents the measuring base that was used in an
experiment to determine the strain values in the diagrams above. ε loc represents the strain value,
after which strain localization can be expected. Usually, this is the strain after which the diagram
is entering into the softening regime. For instance, the strain value that is used to determine the
tensile strength is calculated based on the following assumptions:
if ε1f < ε loc
f

ε1f = ε1f
else

ε1f = ε locf + (ε1f − ε locf ) Lt (2.82)

Ltch
The calculation of the strain value for graphs in Fig. 2-25 and Fig. 2-26 is analogical to Eq.
(2.82) but the appropriate values of ε loc , L and Lch should be used. It should be noted that the
strain ε1f is the strain that is calculated from the strain tensor at the finite element integration
points, while the strain ε1f is used to determine the current tensile strength from the provided
stress-strain diagram (see Fig. 2-24). The equation (2.82) then represents a scaling that takes into
account the difference between the experimental size and the size of the integration point. This
approach guarantees that the same amount of energy is dissipated when using large and small
finite elements.
It is also possible to define a material law for the shear strength of a cracked concrete and for the
compressive strength reduction after cracking.

Compressive strength of cracked concrete σ c = rc (ε1f ) f c′ (2.83)

Shear strength of cracked concrete σ ij ≤ f sh (ε1f ) ft′ (2.84)

It should be realized that the compressive strength of the cracked concrete i.e. (2.83) is a
function of the maximal fracturing strain, i.e. maximal tensile damage at the given point. The
shear strength should be a function of the crack opening. Because of that the shear strength is
specified as a function of the fracturing strain ε1f after the localization transformation (2.82).
The shear strength law is specified as a value relative to f t′ . The compressive strength reduction
is specified as a function relative to f c′ .
56

σt/f’t

1.0

1.0 σ3/f’c
Fig. 2-27. An example of a user defined tensile strength degradation law due to lateral compressive
stress.

2.2.7 Tension stiffening

In heavily reinforced concrete structures the cracks cannot fully developed and concrete
contributes to the steel stiffness. This effect is called tension stiffening and in
CC3DNonLinCementitious2 material it can be simulated by specifying a tension stiffening
factor cts . This factor represents the relative limiting value of tensile strength in the tension
softening diagram. The tensile stress cannot drop below the value given by the product of cts ft
(see Fig. 2-28). The recommended default value for cts is 0.4 as recommended by CEB-FIP
Model Code 1990.

σ
ft

c ts ft

ε
Fig. 2-28: Tension stiffening.

2.2.8 Crack spacing

In heavily reinforced concrete structures, or structures with large finite elements, when many
reinforcement bars are crossing each finite element, the crack band approach described in
 57

Section 2.1.3 will provide too conservative results, and the calculated crack widths may be
overestimated. This is the consequence of the fact that the crack band approach assumes that the
crack spacing is larger than a finite element size. In heavily reinforced structures, or if large
finite elements are used, it may occur that the crack spacing will be smaller than finite element
size. This is especially true if shell/plate elements are used. In this case, typically large finite
elements can be used, and they usually contain significant reinforcement. In these cases, it is
useful to provide the crack spacing manually, since otherwise the program will overestimate the
cracking and due to that also larger deflections may be calculated. The program ATENA allows
the user to manually define the crack spacing. This user defined spacing is used as crack band
size Lt in cases when the user defined crack spacing is smaller than the Lt that would be
calculated by formulas presented in Section 2.1.3.

2.2.9 Fatigue
For modelling fatigue behavior of concrete (CEB 1988 and SAE AE-4) under tensile load, a new
material has been implemented in Atena. The new material
(CC3DNonLinCementitious2Fatigue) is based on the existing three-dimensional fracture plastic
material (CC3DNonLinCementitious2) and uses a stress based model (2.2.9.1). It has an
additional parameter, β fatigue , and additional data attributes for σ base , N , and ε fatigue , used in the
damage calculation as described in section 2.2.9.2. For details and validation against tests
conducted by KESSLER-KRAMER (2002) see ČERVENKA, PRYL (2007).

2.2.9.1 Stress based models

In this approach the fatigue is represented by the so called S-N curves relating the applied stress,
S, and the number of cycles, N, to failure. Such curves must be determined by tests, see Fig.
2-29.
For steel reinforcement bars the performance can be normally expressed as a simple power law
by BASQUIN (1910).
Δσ rm N = C (2.85)
where Δσ r is the stress range, N is the number of cycles to failure and m and C are constants.
This means a linear relationship between Δσ and N in a full logarithmic diagram. The equation
(2.85) is generally valid for the high-cycle range.
For plain concrete the performance can normally be expressed as a straight line in a semi-
logarithmic diagram of the form:
σ max
= 1 − β (1 − R ) log N (2.86)
f
σ min
where σ max is the maximum stress, f is static concrete strength, R = , σ min is minimum
σ max
stress and β is a material constant. The equation (2.86) holds for both compressive and tensile
stresses, however, the value of β is not neccessarily the same for tensile and compressive
behavior of a material. The value should be determined from experiments. For example,
β =0.052 was used based on the experimental results for load levels 0.7 and 0.9 Fstat when
modelling the test on a probe sealed during curing with a notch from section 3.5.2.4 of
KESSLER-KRAMER (2002) for validation.
58

Fig. 2-29: Typical S-N line for concrete in compression (KLAUSEN (1978))

The S-N relations mentioned above are mainly obtained by constant amplitude tests. However,
in real structures the stresses are varying. One method which can be of help in this context is the
well-known Palmgren-Miner hypothesis PALMGREN (1924), MINER (1945).
k
ni
∑N
i =1
=1 (2.87)
i

where ni is the number of constant amplitude cycles at stress level i , N i is the number of cycles
to failure at stress level i , and k is the number of stress levels. As a rough tool this hypothesis is
useful, especially concerning steel. It can also be used for concrete although some investigations
have suggested that a value lower than 1 should be used.

2.2.9.2 Fatigue Damage Calculation

In the implemented model, fatigue damage consists of a contribution based on cycling stress
(2.2.9.2.1), and an additional contribution from crack opening and closing in each cycle
(2.2.9.2.2). The former is dominant before cracking occurs, the latter in already cracked regions.

2.2.9.2.1 Stress Based Contribution

The number of cycles to failure N is determined from a simple stress based model, so called S-
N or Wöhler curve as described in the previous section 2.2.9.1.
 59

⎛ σ upper ⎞
⎜ 1− ⎟
⎜ f ⎟
⎜ β fatigue (1− R ) ⎟
σ upper ⎜ ⎟
= 1 − β fatigue (1 − R ) log N , i.e., N = 10 ⎝ ⎠
, where σ upper stands for the maximum
f
σ base
tensile or compressive stress and f for the corresponding strength, ft or f c , R = .
σ upper
Then, the damage due to fatigue after n cycles is calculated as an increase of the maximum
fracturing strain εˆij′ f (see section 2.2.3). The maximum fracturing strain in each principal
direction is adjusted by adding
w fatigue n
ε fatigue = , where w fatigue = w fail and the failing displacement for the given stress
ElemSize N
w fail = invert _ soft _ law(σ upper ) (see Fig. 2-30).

Fig. 2-30: Softening law and fatigue damage.

In ATENA 4.0, a single value of β fatigue is used to calculate fatigue damage caused by both
tensile and compressive stresses. So far, there is also no special provision implemented for loads
crossing zero, i.e., changing from tension to compression and back in each cycle, which lead to
faster damage according to experimental results presented in CEB 1988 and SAE AE-4. In that
situation, the damage is calculated separately for cycling from 0 to max. compression and from 0
to max. tension, and then the worse of the both damage values is considered. It should be also
noted that the damage is only introduced in form of maximum fracturing strain, which has no
direct impact on compressive material properties, i.e., the fatigue damage effectively only has
influence on tensile behaviour of the material.
60

2.2.9.2.2 Crack Opening Based Contribution

The damage due to cracks that open and close during the cycling is determined as
w
ε fatigueCOD = fatigueCOD , where w fatigueCOD = n ξ fatigue c fatigueCODload ΔCOD , and ΔCOD denotes the
ElemSize
difference between the maximum and minimum crack opening during a cycle. The resulting
ε fatigueCOD is added to ε fatigue before the fatigue damage is introduced into the material.

2.2.9.3 Bringing in Fatigue Damage

It is recommended to introduce the fatigue induced damage into the unloaded structure (i.e., at
the lower stress level). Several other approaches of introducing the damage into the model were
also tested, i.e., introducing the damage at the upper load level or during reloading, but they
usually bring more convergence problems, especially during unloading.

2.2.10 Strain Hardening Cementitious Composite (SHCC, HPFRCC) material

The CC3DNONLINCEMENTITIOUS2SHCC is suitable for fibre reinforced concrete, such as
SHCC (Strain Hardening Cementitious Composites) and HPFRCC or UHPFRC (high and ultra-
high performance fiber reinforced concrete) materials. The theory of this material model is
identical to those described in Sections 2.2.1 - 2.2.6. The tensile softening regime (Fig. 2-32) and
the shear retention factor (Eq. (2.94)) are modified based on the model, proposed in KABELE, P.
(2002). This model is based on a notion of a representative volume element (RVE), which
contains distributed multiple cracks (hardening) as well as localized cracks (softening) – see Fig.
2-31.

a) Multiple cracking regime b) Localized cracking regime

η η
seco
nda

ξ ξ
ry c
rack

t
y crack se
set

r
prima

Fig. 2-31: Representative volume element with cracks.

2.2.10.1 Basic assumptions

a) multiple cracking regime (hardening)
• A set of parallel planar multiple cracks forms when maximum principal stress σmax = σfc (first
crack strength).
• Crack planes are perpendicular to the direction of σmax (ξ-axis).
• The direction of a crack set is fixed.
• Secondary crack set may form in direction perpendicular to primary set if the maximum
 61

normal stress in the corresponding direction (η-axis) exceeds σfc.

• Cracks may slide if the direction of principal stress changes.
• Crack opening and sliding are resisted by fiber bridging.
• Crack opening and sliding displacements are averaged over the RVE as cracking strains
mc ,
ij , mc
ij
,
(notation: lower indices – components of tensor or vector, upper indices –
multiple or localized crack mc, lc and association with primary or secondary crack direction ξ,
η)
b) localized cracking regime (softening)
• A localized crack forms within a set of multiple cracks if the corresponding normal cracking
strain exceeds the level of εmcmb (cracking strain capacity, a material constant).
• Opening and sliding displacements of the i , i localized cracks are treated by the crack
lc, lc,
band model (i.e. they are transformed into cracking strains ij , ij by dividing them with
corresponding band width wξc or wηc).
The overall strain of the RVE is then obtained as a sum of strain of material between cracks
(which may possibly contain nonlinear plastic strain due to compressive yielding), cracking
strains due to multiple cracks, and cracking strains due to localized cracks:
s mc , mc , lc, lc ,
ij ij ij ij ij ij (2.88)
where ε ijs represents the strain of the continuous material between cracks.

2.2.10.2 Crack opening model

The crack-normal stress components are related to cracking strains corresponding to opening of
multiple and localized cracks by piecewise linear relations depicted in Fig. 2-32 [although linear
hardening and softening are shown, a user should be allowed to input piecewise linear curves].
Note that for multiple cracks, it is assumed that they do not close unless exposed to crack-normal
compression (plasticity-like unloading) while a localized crack is assumed to close so that
normal stress decreases linearly to reach zero at zero COD [these assumptions may need to be
revised in the future to some combination of plasticity and damage-like closure]. See also section
2.2.3.

multiple cracking regime localized cracking regime

σξξ σξξ
crack
loading
opening

unloading/ unloading/
reloading reloading
εmcmb Δ0

cracking strain εmcξξ cracking strain εmcξξ COD Δξ

Fig. 2-32: Stress vs. cracking strain relations in crack-normal direction.

62

2.2.10.3 Crack sliding model

The model for crack sliding phenomena is implemented by means of a variable shear retention
factor β. The shear retention factor is defined as a ratio of the material post-cracking shear
stiffness Gc to its elastic shear stiffness G,
Gc
β= . (2.89)
G
Let us determine stiffness Gc, while considering the most general 2-D case of an element, which
contains two perpendicular sets of multiple cracks and two perpendicular localized cracks. If the
problem is defined in plane ξ−η, then the total engineering shear strain has only one non-zero
component, which is obtained as:
s mc , mc , lc , lc ,
2 2 2 2 2 , (2.90)
which can be rewritten with use of the shear bridging model (Kabele, 2000) as:
1 1 1 1 1 1
G M mc ,
M mc ,
wc L wc L Gc (2.91)

Functions M and L are defined by

Vf k Gf
M (ε ) = (2.92)
2ε

⎛1 Δ ⎞ Vf k Gf
L(Δ) = ⎜ − ⎟ , for Δ ≤ Δ 0
⎝ 2 2Δ 0 ⎠ ⎡ 4 k G f ⎛ Δ ⎤
2
⎞
Δ ⎢1 + ⎜ ⎟⎟ ⎥
⎢ 3 E f ⎝⎜ d f ⎠ ⎥
⎣ ⎦
L(Δ) = 0 , for Δ > Δ 0 (2.93)
Here Vf is the fiber volume fraction, Gf is the fiber shear modulus, Ef is the fiber Young’s
modulus, df is the fiber diameter, and k is the fiber cross-section shape correction factor. These
parameters are to be supplied by the user. The shear retention factor is then expressed as
1
β= (2.94)
⎡ 1 1 1 1 ⎤
1+ G ⎢ + + ξ + η η ⎥
⎣⎢ M (ε ξξ ) M (εηη ) wc L(Δξ ) wc L(Δη ) ⎦⎥
mc ,ξ mc ,η ξ

Note that for an element containing only multiple cracks (before localization) 0 and
mc , mc ,
1/L terms approach zero. For an uncracked element, 0 and 1/M and 1/L
approach zero, giving β=1.

2.2.11 Confinement-sensitive constitutive model

The CC3DNonLinCementitious3 fracture-plastic constitutive model is an advanced version of
the CC3DNonLinCementitious2 material that can handle the increased deformation capacity of
concrete under triaxial compression. It is suitable for problems including confinement effects
such as confined reinforced concrete members (columns, bridge piers), nuclear vessels and
triaxial compression tests of plain concrete. A detailed description of the model formulation is
presented in PAPANIKOLAOU and KAPPOS (2007). In this section, only the main differences
 63

between the CC3DNonLinCementitious3 and the CC3DNonLinCementitious2 model are

described, which are mainly focused on the plasticity part of the model (section 2.2.4).

2.2.11.1 Hardening and softening function

The position of failure surface can expand and move along the hydrostatic axis (simulating the
hardening and softening stages), based on the value of the hardening/softening parameter (κ). In
the present model, this parameter identifies with the volumetric plastic strain (GRASSL et al.,
2002) :
dκ = dε pv = dε1p + dε 2p + dε 3p (2.95)
The instantaneous shape and location of the loading surface during hardening is defined by a
hardening function (k), which depends on the hardening/softening parameter (κ). This function is
directly incorporated in the Menétrey-Willam failure surface equations (2.54), operating as a
scaling factor on the compressive concrete strength (fc ). It has the same elliptic form with
CC3DNonLinCementitious2 (2.57), but herein in terms of the plastic volumetric strain :
2
⎛ εp − εp ⎞
k(κ) = k(ε ) = k o + (1 − k o ) ⋅ 1 − ⎜ v,t p v ⎟
p
(2.96)
v ⎜ ε ⎟
⎝ v,t ⎠
where ε pv,t is the plastic volumetric strain at uniaxial concrete strength (onset of softening) and ko
is the value that defines the initial yield surface that bounds the initial elastic regime (onset of
plasticity). At the end of the hardening process, the hardening function retains a constant value
of unity and the material enters the softening regime, which is controlled by the softening
function (c). This function simulates the material decohesion by shifting the loading surface
along the negative hydrostatic axis. It is assumed that it follows the softening function originally
proposed by VAN GYSEL and TAERWE (1996) for uniaxial compression:
2
⎛ ⎞
⎜ 1 ⎟
⎜ 2 ⎟
c(κ) = c(ε v ) = ⎜ ⎛ n1 − 1 ⎞ ⎟
p
(2.97)
1+
⎜ ⎜⎝ n 2 − 1 ⎟⎠ ⎟
⎜ ⎟
⎝ ⎠
where :
n1 = ε pv / ε pv,t (2.98)

n 2 = (ε pv,t + t) / ε pv,t (2.99)

Parameter t in equation (2.99) controls the slope of the softening function and the outmost square
is necessary due to the quadratic nature of the loading surface. The softening function value
starts from unity and complete material decohesion is attained at c = 0. The evolution of both
hardening and softening functions with respect to the hardening/softening parameter is
schematically shown in Fig. 2-33.
64

k(κ) / c(κ)
c k
1.0

k
0.8

0.6
c

0.4

0.2 ko

0.0
p
ε v,t κ = εpv

Fig. 2-33: Evolution of hardening (k) and softening (c) functions with respect to the plastic
volumetric strain.

2.2.11.2 Plastic potential function

The present plasticity model incorporates a non-associated flow rule using a polynomial plastic
potential function (g), with Lode angle (θ) dependency and adjustable order (n) :
n
⎛ ρ ⎞ ⎡ 1 ⎤ ρ ξ
g = A ⋅⎜ ⎟ + ⎢C + (B − C)(1 − cos3θ) ⎥ ⋅ + −a (2.100)
⎝ k ⋅ c ⋅ f c ⎠ ⎣ 2 ⎦ k ⋅ c ⋅ f c k ⋅ c ⋅ f c

Parameters A, B and C define the shape of the plastic potential function in stress space and their
calibration is based on the assumption that the inclination (ψ) of the incremental plastic strain
vector identifies with the inclination of the total plastic strain vector at three distinct stress states,
namely the uniaxial, equibiaxial and triaxial compressive concrete strength (Fig. 2-34). The
attraction constant (a) is included for mathematical clarity and is not a user parameter, due to
plastic potential function differentiation in the flow rule.
 65

Fig. 2-34: Direction (ψ) of the incremental (a) and total (b) plastic strain vectors.

2.2.11.3 Suggested model parameters

A detailed calibration scheme for the plasticity model parameters, based on and extensive
experimental database can be found in PAPANIKOLAOU and KAPPOS (2007) and suggested
values (including the fracture model parameters) for various uniaxial compressive concrete
strengths (fc) are shown in the following table (see Atena Input File Format document for the
material definition details):
Table 2.2-1 Suggested parameters for the fracture and plasticity models

fc (ΜPa) 20 30 40 50 60 70
Εc (MPa) 24377 27530 30011 32089 33893 35497
ν 0.2 0.2 0.2 0.2 0.2 0.2
ft (MPa) 1.917 2.446 2.906 3.323 3.707 4.066
λt 1.043 1.227 1.376 1.505 1.619 1.722
e 0.5281 0.5232 0.5198 0.5172 0.5151 0.5133
fco (MPa) -4.32 -9.16 -15.62 -23.63 -33.14 -44.11
ε pv,t 4.92·10-4 6.54·10-4 8.00·10-4 9.35·10-4 1.06·10-3 1.18·10-3
t 1.33·10-3 2.00·10-3 2.67·10-3 3.33·10-3 4.00·10-3 4.67·10-3
A 7.342177 5.436344 4.371435 3.971437 3.674375 3.43856
B -8.032485 -6.563421 -5.73549 -5.430334 -5.202794 -5.021407
C -3.726514 -3.25626 -3.055953 -2.903173 -2.797059 -2.719067
n 3 3 3 3 3 3
Gf (MN/m) 4.87·10-5 6.47·10-5 7.92·10-5 9.26·10-5 1.05·10-4 1.17·10-4

fc (ΜPa) 80 90 100 110 120

66

Εc (MPa) 36948 38277 39506 40652 41727

ν 0.2 0.2 0.2 0.2 0.2
ft (MPa) 4.405 4.728 5.036 5.333 5.618
λt 1.816 1.904 1.986 2.063 2.136
e 0.5117 0.5104 0.5092 0.5081 0.5071
fco (MPa) -56.50 -70.30 -85.48 -102.01 -114.00
ε pv,t 1.30·10-3 1.41·10-3 1.52·10-3 1.62·10-3 1.73·10-3
t 5.33·10-3 6.00·10-3 6.67·10-3 7.33·10-3 8.00·10-3
A 3.245006 3.082129 2.942391 2.820644 2.713227
B -4.871993 -4.745867 -4.637358 -4.542587 -4.458782
C -2.659098 -2.611426 -2.572571 -2.540158 -2.512681
n 3 3 3 3 3
Gf (MN/m) 1.29·10-4 1.40·10-4 1.50·10-4 1.61·10-4 1.71·10-4

2.3 Von Mises Plasticity Model

Von Mises plasticity model called also as J2 plasticity is based only on one parameter k. The
yield function is defined as:
F p (σ ij ) = J 2 − k ( ε eqp ) = 0 (2.101)

where J2 denotes the second invariant of stress deviator tensor. The parameter
k ( ε eqp ) = 1
3 σ y ( ε eqp ) is the maximal shear stress and σ y is the uniaxial yield stress. This
parameter controls the isotropic hardening of the yield criterion.
N inc
σ y ( ε eqp ) = σ y + H ε eqp , ε eqp = ∑ 2
3 ( Δε p
: Δε p ) (2.102)
i =1

σ y is the yield stress, H the hardening modulus and ε eqp is the equivalent plastic strain
calculated as a summation of equivalent plastic strains during the loading history.
In case of von Mises plasticity the plastic potential function is identical with the yield function:
G p (σ ij ) = F P (σ ij ) (2.103)

The associated flow rule is assumed. The background information can be found in (CHEN,
SALEEB 1982, Sec.5.4.2).
The Von Mises model could be used to model cyclic steel behavior including Bauschinger
effect. In this case the yield function is modified as:
1
2 ( σ′ − X ) : ( σ′ − X ) − k (ε eqp ) − (r − 1)k0 = 0 (2.104)

where σ′ is the deviatoric stress, k0 is an initial value of k (ε eqp ) according to (2.102), X is the so
called back stress controlling the kinematic hardening:
 67

ΔX = 2 3 k1 Δε p − k2 X Δε eqp (2.105)

In equations (2.104) and (2.105) quantities r , k1 , k2 are material parameters for the cycling
response. If r is non-zero the cyclic model is activated and it controls the radius of the Von
Mises surface. If r = 1 the yielding will start exactly when σ y is reached. For lower values the
non-linear behavior starts earlier and the slope of the response is mainly affected by parameter
k1 (larger value – higher slope). Parameter k2 on the other hand affects the memory of the cyclic
response. Some examples of various parameter combinations are shown at.
68

H=0.0, r=0.7, k1=74000, k2=1000

250
200
150
100
Stress [MPa]

50
0
-50
-100
-150
-200
-250
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014
Strain [-]

H=2000 MPa, r=0.7, k1=74000, k2=1000

400

300

200

100
Stress [MPa]

-100

-200

-300

-400
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014
Strain [-]

Fig. 2-35: Effect of material parameter choice on cyclic response for E=210 GPa and σ y = 200 MPa.
 69

2.4 Drucker-Prager Plasticity Model

Drucker-Prager plasticity model is based on a general plasticity formulation that is described in
Section 2.2.4. The yield function is defined as:
p
FDP (σ ij ) = α I 1 + J 2 − k = 0 (2.106)

Where α and k are parameters defining the shape of the failure surface. They can be estimated
by matching with the Mohr-Coulomb surface. If the two surface are to agree along the
compressive meridian, i.e. θ = 00 , the formulas are:
2sin φ 6 c cos φ
α= , k= (2.107)
3 ( 3 − sin φ ) 3 ( 3 − sin φ )
This corresponds to a outer cone to the Mohr-Coulomb surface. The inner cone, which passes
through the tensile meridian where θ = 600 has the constants given by the following expressions:
2sin φ 6 c cos φ
α= , k= (2.108)
3 ( 3 + sin φ ) 3 ( 3 + sin φ )
The position of failure surfaces is not fixed but it can move depending on the value of strain
hardening/softening parameter. The strain hardening is based on the equivalent plastic strain,
which is calculated according to the following formula.
Δε eqp = min(Δε ijp ) (2.109)

Hardening/softening in the Drucker-Prager model is controlled by the parameter k . This

parameter is selected such that the surface at the peak passes through the uniaxial compressive
strength, and it changes according to the following expression.
f c′(ε eqp )
k = k0 (2.110)
f c′

In the above two formulas the expression f c′(ε eqp ) indicates the hardening/softening law, which is
based on the uniaxial compressive test. The law is shown in Fig. 2-36.

Fig. 2-36. Linear softening in the Drucker-Prager material model

Return direction is given by the following plastic potential:

1
G p (σ ij ) = β I1 + 2 J 2 (2.111)
3
70

where β determines the return direction. If β < 0 material is being compacted during crushing,
if β = 0 material volume is preserved, and if β > 0 material is dilating. In general the plastic
model is non-associated, since the plastic flow is not perpendicular to the failure surface
The return mapping algorithm for the plastic model is based on predictor-corrector approach as
is shown in Fig. 2-22. During the corrector phase of the algorithm the failure surface moves
along the hydrostatic axis to simulate hardening and softening. The final failure surface has the
apex located at the origin of the Haigh-Vestergaard coordinate system. Secant method based
Algorithm 1 is used to determine the stress on the surface, which satisfies the yield condition and
also the hardening/softening law.

2.5 User Material Model

In some situations, none of the standard material models available in ATENA can describe the
behavior sufficiently. Many such cases can be handled by defining user laws in the fracture-
plastic material model (see CC3DNonLinCementitious2User described in section 2.2.6), in the
others the user can provide a dynamic link library implementing his own material model. The
user material is based on the elastic isotropic material, adding new material parameters and state
variables (both limited to floating point values). See the User Material DLL Manual for
description and reference, and the CCUserMaterialExampleDLL directory in Atena Science
Examples for an example project including the source code in C and a windows help file version
of the manual, AtenaV4_UserMaterialDLL.chm. Please note that the behavior of the user model
may have influence on convergence of the analysis.

2.6 Interface Material Model

The interface material model can be used to simulate contact between two materials such as for
instance a construction joint between two concrete segments or a contact between foundation
and concrete structure. The interface material is based on Mohr-Coulomb criterion with tension
cut off. The constitutive relation for a general three-dimensional case is given in terms of
tractions on interface planes and relative sliding and opening displacements.
⎧τ 1 ⎫ ⎡ K tt 0 0 ⎤ ⎧ Δv1 ⎫
⎪ ⎪ ⎢ ⎪ ⎪
⎨τ 2 ⎬ = ⎢ 0 K tt 0 ⎥⎥ ⎨ Δv2 ⎬ (2.112)
⎪σ ⎪ ⎢ 0 0 K nn ⎦⎥ ⎪⎩ Δu ⎪⎭
⎩ ⎭ ⎣
For two-dimensional problems second row and column are omitted.
The initial failure surface corresponds to Mohr-Couloumb condition (2.113) with tension cut-off.
After stresses violate this condition, this surface collapses to a residual surface which
corresponds to dry friction.
τ ≤ c − σ ⋅ φ , σ ≤ ft (2.113)

τ = 0 , σ > ft
 71

Trial stress

φ
1 c Initial surface
Final stress
Residual surface

ft σ
Fig. 2-37: Failure surface for interface elements.

In general three-dimensional case τ in Fig. 2-37 and equation (2.113) is calculated as:

τ = τ 12 + τ 22 (2.114)
72

τ
c−φσ
min
Ktt
1
−φσ
Ktt

1
(a) Δv

σ
ft

min
Knn Knn
1
1

(b) Δu
Fig. 2-38: Typical interface model behavior in shear (a) and tension (b)

The K nn , K tt denote the initial elastic normal and shear stiffness respectively. Typically for zero
thickness interfaces, the value of these stiffnesses correspond to a high penalty number. It is
recommended not to use extremely high values as this may result in numerical instabilities. It is
recommended to estimate the stiffness value using the following formulas
E G
K nn = , K tt = (2.115)
t t
where E and G is minimal elastic modulus and shear modulus respectively of the surrounding
material. t is the width of the interface zone. Its value can be selected either on the basis of the
reality. For instance for mortar between masonry bricks the value is typically 10-20 mm.
Alternatively, it can be estimated as a dimension, which can be considered negligible with
respect to the structural size. For instance in case of a dam analysis, where the dam dimensions
are typically in the order of 100 meters, the width of the interface zone can be estimated to be 0.5
meters. It is suitable due to numerical reasons if stiffness is about 10 times of the stiffness of
adjacent finite elements.
 73

There are two additional stiffness values that need to be specified in the ATENA input. They are
denoted in Fig. 2-38 as K nnmin and K ttmin . They are used only for numerical purposes after the
failure of the element in order to preserve the positive definiteness of the global system of
equations. Theoretically, after the interface failure the interface stiffness should be zero, which
would mean that the global stiffness will become indefinite. These minimal stiffnesses should be
about 0.001 times of the initial ones.
It is possible to define evolution laws for tensile as well as shear softening by arbitrary
multilinear laws. Examples of such laws are shown in Fig. 2-39. The figure describes bi-linear
softening laws. The break point of this law can be determined for instance by the formula
proposed by Bruehwiler and Wittman (1990).

ft GF
s1 = , v1 = 0.75 (2.116)
4 ft

σ/ft c/c 0

I II
GF GF
s1σ s1c

Δu1σ Δu Δv 1c Δv
Fig. 2-39: Example of a softening law for tension and cohesion.
74

3.0
2.5
2.0
1.5
Shear stress [MPa]

1.0
0.5
0.0
-0.5
-1.0
-1.5
-2.0
-1.0 -0.5 0.0 0.5 1.0 1.5 2.0
Sliding [mm]

Fig. 2-40: Example of a cyclic response of the model in shear under constant normal pre-stress.

2.7 Reinforcement Stress-Strain Laws

2.7.1 Introduction
Reinforcement can be modeled in two distinct forms: discrete and smeared. Discrete
reinforcement is in form of reinforcing bars and is modeled by truss elements. The smeared
reinforcement is a component of composite material and can be considered either as a single
(only one-constituent) material in the element under consideration or as one of the more such
constituents. The former case can be a special mesh element (layer), while the later can be an
element with concrete containing one or more reinforcements. In both cases the state of uniaxial
stress is assumed and the same formulation of stress-strain law is used in all types of
reinforcement.

2.7.2 Bilinear Law

The bilinear law, elastic-perfectly plastic, is assumed as shown in Fig. 2-41
 75

Fig. 2-41 The bilinear stress-strain law for reinforcement.

The initial elastic part has the elastic modulus of steel Es. The second line represents the
plasticity of the steel with hardening and its slope is the hardening modulus Esh. In case of
perfect plasticity Esh =0. Limit strain εL represents limited ductility of steel.

2.7.3 Multi-line Law

The multi-linear law consists of four lines as shown in Fig. 2-42. This law allows to model all
four stages of steel behavior: elastic state, yield plateau, hardening and fracture. The multi-line is
defined by four points, which can be specified by input.

Fig. 2-42 The multi-linear stress-strain law for reinforcement.

The above described stress-strain laws can be used for the discrete as well as the smeared
reinforcement. The smeared reinforcement requires two additional parameters: the reinforcing
ratio p (see Section 2.1.1.1) and the direction angle β as shown in Fig. 2-43.
76

Fig. 2-43 Smeared reinforcement.

The spacing s of the smeared reinforcement is assumed infinitely small. The stress in the
smeared reinforcement is evaluated in the cracks, therefore it should include also a part of stress
due to tension stiffening (which is acting in concrete between the cracks, section 2.1.9).
σ scr ' = σ s ' + σ ts (2.117)
where σ s is the steel stress between the cracks (the steel stress in smeared reinforcement), σ scr
' '

is the steel stress in a crack. If no tension stiffening is specified σ ts =0 and σ scr = σ s . In case of
' '

the discrete reinforcement the steel stress is always σ s .

'

2.7.4 No Compression Reinforcement

Normally all reinforcement material models in ATENA exhibit the same behavior in tension as
well as in compression. The material types CCReinforcement and CCSmearedReinforcement
include the capability to deactivate the compressive response of the reinforcement. This is
sometimes useful, if this material model is used to simulate the behavior of reinforcement
elements that have a very low bending stiffness, so it can be assumed that when the
reinforcement is loaded by compressive forces, buckling occurs and the strength of the elements
in compression is negligible. This is controlled by the command COMPRESSION 0 or 1, which
deactivates and activates the compressive response respectively (for more details see ATENA
Input File Format).

2.7.5 Cycling Reinforcement Model

The reinforcing steel stress-strain behavior can be described by the nonlinear model of
Menegotto and Pinto (1973). In ATENA this model is extended to account of the isotropic
hardening due to an arbitrary hardening law that can be specified for reinforcement (see Sections
2.7.2, 2.7.3). The stress in the cycling model is calculated according to the following expression.
σ = (σ 0 − σ r ) σ * + σ r (2.118)
where
 77

σ = bε +
* * (1 − b ) ε * , ε* =
ε − εr
, R = R0 −
c1ξ
(2.119)
(1 + ε ) ε0 − εr c2 + ξ
1
*R
R

where R0 , c1 and c2 are experimentally determined parameters. The Fig. 2-44 shows the
meaning of strain values ε r , ε 0 , ξ and stress values σ r and σ 0 . These values changes for each
cycle. The values with the subscript r indicate the point where the cycle started, and the
subscript 0 indicates the theorethical yield point that would be reached during the unloading if
the response would not have been modified by the hysteretic behavior. During the calculation of
this point the material stress-strain law is considered (see Sections 2.7.2, 2.7.3)
Nincr .
σ * = f R ( ε eq ) , ε eq = ∑ Δε i
eq (2.120)
i =1

Fig. 2-44: Cycling reinforcement model based on Menegotto and Pinto (1973).

2.8 Reinforcement bond models

The basic property of the reinforcement bond model is the bond-slip relationship. This
relationship defines the bond strength depending on the value of current slip between
reinforcement and surrounding concrete. ATENA contains three bond-slip models: according to
the CEB-FIB model code 1990, slip law by Bigaj and the user defined law. In the first two
models, the laws are generated based on the concrete compressive strength, reinforcement
diameter and reinforcement type. The important parameters are also the confinement conditions
and the quality of concrete casting.
78

2.8.1 CEB-FIP 1990 Model Code

Fig. 2-45: Bond-slip law by CEB-FIP model code 1990.

α
⎛s⎞
τ =τ ⎜ ⎟
max ⎜ s ⎟
, 0 ≤ s ≤ s1 (2.121)
⎝ 1⎠

τ =τ max
, s1 < s ≤ s 2 (2.122)

⎛ s−s ⎞
τ =τ − (τ max − τ f )⎜ ⎟, s 2 < s ≤ s3
2
(2.123)
⎝s −s ⎠
max

3 2

τ =τ f
, s3 < s (2.124)
 79

Table 2.8-1: Parameters for defining the mean bond stress-slip relationship for ribbed bars.

2 3 4 5
Value Unconfined concrete* Confined concrete**
Bond conditions Bond conditions
Good All other cases Good All other cases
S1 0.6 mm 0.6 mm 1.0 mm
S2 0.6 mm 0.6 mm 3.0 mm
S3 1.0 mm 2.5 mm clear rib spacing
α 0.4 0.4
τmax
2.0 f 1.0 f 2.5 f 1.25 f
C C C C

τf
0.15 τ max 0.40 τ max
* Failure by splitting of the concrete
**Failure by shearing of the concrete between the ribs

Table 2.8-2: Parameters for defining the bond stress-slip relationship for smooth bars.

Values Cold drawn wire Hot rolled bars

Bond conditions Bond conditions
Good All other Good All other cases
cases
0.01 mm 0.1 mm
s1 = s 2 = s3
α 0.5 0.5

τ max = τ f 0.1 f 0.05 f 0.3 f 0.15 f

C C C C

2.8.2 Bond Model by Bigaj

The second pre-defined bond model available in ATENA is based on the work by BIGAJ 1999.
'
This model depends on the bond quality, concrete cubic compressive strength cu and f
reinforcement bar radius D . The slip law for this model is shown in Fig. 2-46.
80

3 4

Fig. 2-46: Bond law by BIGAJ 1999

The ascending part of the stress-slip law i.e. part a is modeled by a bi-linear curve. The
coordinates of the four points defining this stress-slip relationship are listed in the table below.

Table 2.8-3: Parameters for defining the bond stress-slip relationship for smooth bars.

Concrete Bond Point 1 Point 2 Point 3 Point 4

Type quality
Excelent s/D 0.000 0.020 0.044 0.480

τ b / 0.8 f cu' 0.500 3.000 0.700 0.000

'
f c < 60 Good
s/D 0.000 0.030 0.047 0.480

τ b / 0.8 f cu' 0.500 2.000 0.700 0.000

Bad s/D 0.000 0.040 0.047 0.480

τ b / 0.8 f cu' 0.500 1.000 0.700 0.000

Excelent s/D 0.000 0.012 0.030 0.340

τ b / 0.88 f cu' 0.600 2.500 0.900 0.000

'
f c > 60 Good
s/D 0.000 0.020 0.030 0.340

τ b / 0.88 f cu' 0.600 1.900 0.900 0.000

Bad s/D 0.000 0.025 0.030 0.340

τ b / 0.88 f cu' 0.600 1.100 0.900 0.000

 81

2.9 Microplane material model (CCMicroplane4)

The basic idea of the microplane model is to abandon constitutive modelling in terms of tensors
and their invariants and formulate the stress-strain relation in terms of stress and strain vectors
on planes of various orientations in the material, now generally called the microplanes. This
idea arose in G.I. Taylor’s (TAYLOR 1938) pioneering study of hardening plasticity of
polycrystalline metals. Proposing the first version of the microplane model, BAZANT 1984, in
order to model strain softening, extended or modified Taylor’s model in several ways (in detail
see BAZANT et al. 2000), among which the main one was the kinematic constraint between the
strain tensor and the microplane strain vectors. Since 1984, there have been numerous
improvements and variations of the microplane approach. A detailed overview of the history of
the microplane model is included in BAZANT et al 2000 and CANER and BAZANT 2000. In
what follows, we briefly review the derivation of the microplane model that is used in this work.
In the microplane model, the constitutive equations are formulated on a plane, called
microplane, having an arbitrary orientation characterized by its unit normal ni . The kinematic
constraint means that the normal strain ε N and shear strains ε M , ε L on the microplane are
calculated as the projections of the macroscopic strain tensor ε ij :

1 1
ε N = ni n j ε ij , ε M =
2
( mi n j + m j ni ) ε ij , ε L = ( li n j + l j ni ) ε ij
2
(2.125)

where mi and li are chosen orthogonal vectors lying in the microplane and defining the shear
strain components. The constitutive relations for the microplane strains and stresses can be
generally stated as:
σ N (t ) = Fτt=0 [ε N (τ ), ε L (τ ), ε M (τ ) ]
σ M (t ) = Gτt =0 [ε N (τ ), ε L (τ ), ε M (τ ) ] (2.126)
σ L (t ) = Gτt =0 [ε N (τ ), ε L (τ ), ε M (τ ) ]
where F and G are functionals of the history of the microplane strains in time t. For a detailed
derivation of these functionals a reader is referred to BAZANT et al 2000 and CANER and
BAZANT 2000. The macroscopic stress tensor is obtained by the principle of virtual work that is
applied to a unit hemisphere Ω. After the integration, the following expression for the
macroscopic stress tensor is recovered (BAZANT 1984):
3 Nm
σ σ
σ ij = ∫ s d Ω ≈ 6 ∑ wμ sij( μ ) , where sij = σ N ni n j + M ( mi n j + m j ni ) + L ( li n j + l j ni ) (2.127)
2π Ω
ij
μ =1 2 2
where the integral is approximated by an optimal Gaussian integration formula for a spherical
surface; numbers μ label the points of the integration formula and wμ are the corresponding
optimal weights.

2.9.1 Equivalent localization element

The objective of the equivalent localization element is to achieve equivalence with the crack
band model. This basic idea is that the material properties and parameters of the softening
material model are not modified to account for the differences in the finite element size, but
rather the softening crack band is coupled in series with an elastically behaving layer, in order to
obtain equivalence. For brevity, this layer will henceforth be called the `spring’. For large finite
elements, the effective length of this added elastic spring, representing the thickness of the added
82

elastic layer having the elastic properties of the material, will be much larger than the size (or
thickness) of the localization zone (crack band). Thus, after the crack initiation, the energy stored
in the elastic spring can be readily transferred to the localization zone and dissipated in the
softening (i.e., fracturing) process.
Inside each finite element at each integration point, an equivalent localization element is
assumed. The localization element is a serial arrangement of the localization zone, which is
loading, and an elastic zone (spring), which is unloading. The total length of the element is
equivalent to the crack band size L (width), and can be determined using the same methods as
described in Section 2.1.3 (see Fig. 2-12). The width of the localization zone is given either by
the characteristic length of the material or by the size of the test specimen for which the adopted
material model has been calibrated.
The three-dimensional equivalent element is constructed by three serial arrangements of the
elastic zone (spring) and localization band. The spring-band systems are perpendicular to each
other, and they are arranged parallel to the principal strain directions (Fig. 2-47). The simplified
two-dimensional version is shown in Fig. 2-48. In this arrangement of spring-band systems it is
possible to identify the following unknown stresses and strains:
σ ijb , 1σ iju , 2σ iju , 3σ iju and ε ijb , 1ε iju , 2ε iju , 3ε iju
m
where superscript b denotes the quantities in the localization band and the symbol xu with
superscripts u and m defines the quantities in the elastic spring in the direction m .

3
εiju,3σiju

εijb,σijb
1 u1
εij ,σiju
3
L
2 u2
3
h εij ,σiju

1
h 2
h
1
L 2
L
Fig. 2-47: The arrangement of the three-dimensional equivalent localization element.
 83

Finite element Localization

element
ε1

Elastic
springs
1
εij
u

ε2
2
1
h εijb εiju

2
Localization h
2 band
L 2
L
Fig. 2-48: The simplified two-dimensional view of the spring-band arrangement.

Ideally, the chosen directions should be perpendicular to the planes of failure propagation. In
ATENA, it is assumed for them to be aligned with the principal axes of the total macroscopic
strain tensor, which in most cases should approximately correspond to the above requirement.
Altogether there are 48 unknown variables. In the subsequent derivations, it is assumed that
these stresses and strains are defined in the principal frame of the total macroscopic strain tensor.
The set of equations available for determining these variables starts with the constitutive
formulae for the band and the elastic springs:
σ ijb = F (ε ijb ) (2.128)
m
σ iju = Dijlk mε klu for m = 1...3 (2.129)

The first formula (2.128) represents the evaluation of the non-linear material model, which in our
case is the microplane model for concrete. The second equation (2.129) is a set of three elastic
constitutive formulations for the three linear zones (springs) that are involved in the arrangement
at Fig. 2-47. This provides the first 24 equations, which can be used for the calculation of
unknown strains and stresses.
The second set of equations is provided by the kinematic constrains on the strain tensors.
84

1 ⎡ b1 1 u 1
ε11 =
L⎣1
ε11 h + ε11 L − 1h ⎤⎦( )
1
(
ε 22 = 2 ⎡⎣ε 22b 2h + 2ε 22u 2 L − 2h ⎤⎦
L
)
1 ⎡ b 3 3 u 3
ε 33 =
L⎣
ε 33 h + ε 33 L − 3h ⎤⎦
3 ( )
1⎧ 1 1 ⎫ (2.130)
2⎩ L
( L
)
ε12 = ⎨ 1 ⎡⎣ε12b 1h + 1ε12u 1L − 1h ⎤⎦ + 2 ⎡⎣ε12b 2h + 2ε12u ( 2
)
L − 2h ⎤⎦ ⎬
⎭
1⎧ 1 1 ⎫
ε 23 = ⎨ 2 ⎡⎣ε 23b 2h + 2ε 23u ( 2 L − 2h ) ⎤⎦ + 3 ⎡⎣ε 23b 3h + 3ε 23u ( 3 L − 3h ) ⎤⎦ ⎬
2⎩ L L ⎭
1⎧ 1 1 ⎫
ε13 = ⎨ 1 ⎡⎣ε13b 1h + 1ε13u ( 1L − 1h ) ⎤⎦ + 3 ⎡⎣ε13b 3h + 3ε13u ( 3 L − 3h ) ⎤⎦ ⎬
2⎩ L L ⎭
These 6 additional equations can be written symbolically as:
1⎧ 1 1 ⎫
ε ij = ⎨ i ⎣⎡ε ijb i h + iε iju ( i L − i h ) ⎦⎤ + j ⎣⎡ε ijb j h + jε iju ( j L − j h ) ⎦⎤ ⎬ (2.131)
2⎩ L L ⎭
The next set of equations is obtained by enforcing equilibrium in each direction between the
corresponding stress components in the elastic zone and in the localization band. For each
direction m , the following condition must be satisfied:
σ ijb m e j = mσ iju m e j for m = 1...3 (2.132)

where m e j denotes coordinates of a unit direction vector for principal strain direction m . Since
the principal frame of the total macroscopic strain tensor is used the unit vectors have the
following coordinates:
1
e j = (1, 0, 0 ) , 2 e j = ( 0,1, 0 ) , 3e j = ( 0, 0,1) (2.133)

The remaining equations are obtained by enforcing equilibrium between tractions on the other
surfaces of the band and the elastic zone (layer) imagined as a spring:
σ ijb m e j = nσ iju m e j where m = 1..3, n = 1...3, m ≠ n (2.134)

The equation (2.134) is equivalent to a static constraint on the remaining stress and strain
components of the elastic springs. Formulas (2.132) and (2.134) together with the assumption of
stress tensor symmetry represent the remaining 18 equations that are needed for the solution of
the three-dimensional equivalent localization element. These 18 equations can be written as:
σ ijb = mσ iju for m = 1...3 (2.135)
This means that the macroscopic stress must be equal to σ ijb , i.e., the stress in the localization
element, and that the stresses in all the three elastic zones must be equal to each other and to the
microplane stress σ ijb . This implies also the equivalence of all the three elastic strain tensors.

Based on the foregoing derivations, it is possible to formulate an algorithm for the calculation of
unknown quantities in the three-dimensional equivalent localization element.
Input: ε ij , Δε ij , ε ijb , ε iju (2.136)

Initialization: Δε ijb = Δε iju = Δε ij (2.137)

 85

(i )
i
L jh + j L ih
Step 1: d ε iju = i j
Cijkl rkl( i −1) (2.138)
2L L
(i ) ( i −1) (i )
Step 2: Δε iju = Δε iju + d ε iju (2.139)

(i ) 2iL jL 2 i L j L − i L jh − j L ih u
Step 3: Δε ijb = Δε ij − Δε ij (2.140)
i
L j h + j L ih i
L jh + j L ih
(i ) (i )
Step 4: rij( i ) = σ ijb − σ iju (2.141)

where Cijlk is the compliance tensor. The above iterative process is controlled by the following
convergence criteria;
d ε iju (i ) rij( i ) rij( i ) T d ε iju ( i )
<e , <e , <e (2.142)
Δε ij Δσ ijb Δσ ijb Δε ij
The macroscopic stress is then equal to the stress in the localization band σ ijb . More details about
the derivations of the above algorithm as well as various examples of application can be obtained
from the original reference CERVENKA et al. 2004. It should be noted that the described
equivalent localization element is used only if the calculated crack band size L (see Section
2.1.3) in each principal strain direction is larger than the prescribed localization band size h . For
smaller element sizes the equivalent localization approach is not used and mesh-dependent
results may be obtained.

2.10 References
BASQUIN, H.O. (1910), The exponential law of endurance tests, Proc. ASTM, 10 (II).
BAZANT, Z.P, OH, B.H (1983) - Crack Band Theory for Fracture of Concrete, Materials and
Structures, RILEM, Vol. 16, 155-177.
BAŽANT, Z.P., (1984), ‘Microplane model for strain controlled inelastic behavior’, Chapter 3 in
Mechanics of Engineering Materials (Proc., Conf. held at U. of Arizona, Tucson, Jan. 1984),
C.S. Desai and R.H. Gallagher, eds., J. Willey, London, 45-59.
BAŽANT, Z.P., CANER, F.C., CAROL, I., ADLEY, M.D., AND AKERS, S.A., (2000),
‘Microplane Model M4 for Concrete: I. Formulation with Work-Conjugate Deviatoric Stress’, J.
of Engrg. Mechanics ASCE, 126 (9), 944-961.
BIGAL, A.J (1999) - Structural Dependence of Rotation Capacity of Plastic Hinges in RC
Beams and Slabs, PhD Thesis, Delft University of Technogy, ISBN 90-407-1926-8.
BRUEHWILER, E., and WITTMAN, F.H. (1990), “The Wedge Splitting Test, A New Method
of Performing Stable Fracture-Mechanics Tests”, Engineering Fracture Mechanics, Vol. 35, No.
1-3, pp. 117-125.
CANER, F.C., AND BAŽANT, Z.P., (2000) ‘Microplane Model M4 for Concrete: II. algorithm
and calibration.", J. of Engrg. Mechanics ASCE, 129 (9), 954-961.
CEB-FIP Model Code 1990, First Draft, Comittee Euro-International du Beton, Bulletin
d'information No. 195,196, Mars.
CEB 1988, Bulletin D’Information No 188, Fatigue of concrete structures, State of the art report.
86

CERVENKA, V., GERSTLE, K. (1972) - Inelastic Analysis of Reinforced Concrete Panels: (1)
Theory, (2) Experimental Verification and application, Publications IABSE, Zürich, V.31-00,
1971, pp.32-45, and V.32-II,1972, pp.26-39.
CERVENKA, V. (1985) - Constitutive Model for Cracked Reinforced Concrete, Journal ACI,
Proc. V.82, Nov-Dec., No.6,pp.877-882.
CERVENKA, V., PUKL, R., ELIGEHAUSEN, R. (1991) - Fracture Analysis of Concrete Plane
Stress Pull-out Tests, Proceedings, Fracture process in Brittle Disordered Materials, Noordwijk,
Holland, June 19-21.
CERVENKA, V., PUKL, R., OZBOLT, J., ELIGEHAUSEN, R. (1995), Mesh Sensitivity
Effects in Smeared Finite Element Analysis of Concrete Structures, Proc. FRAMCOS 2, 1995,
pp 1387-1396.
CERVENKA, V., PUKL, R. (1992) - Computer Models of Concrete Structures, Structural
Engineering International, Vol.2, No.2, May 1992. IABSE Zürich, Switzerland, ISSN 1016-
8664, pp.103-107.
CERVENKA, V., PUKL, R., OZBOLT, J., ELIGEHAUSEN, R. (1995) - Mesh Sensitivity
Effects in Smeared Finite Element Analysis of Concrete Fracture, Proceedings of FRAMCOS2,
Zurich, Aedificatio.
CERVENKA, V., CERVENKA, J. (1996) - Computer Simulation as a Design Tool for Concrete
Structures, ICCE-96, proceedings of The second International Conference in Civil Engineering
on Computer Applications Research and Practice, 6-8 April, Bahrain.
CERVENKA, J, CERVENKA, V., ELIGEHAUSEN, R. (1998), Fracture-Plastic Material Model
for Concrete, Application to Analysis of Powder Actuated Anchors, Proc. FRAMCOS 3, 1998,
pp 1107-1116.
ČERVENKA, J., BAŽANT Z.P., WIERER, M., (2004), `Equivalent Localization Element for
Crack Band Approach to Mesh Sensitivity in Microplane Model’, submitted for publication, Int.
J. for Num. Methods in Engineering.
ČERVENKA, J., PRYL, D., (2007), `Fatigue Modelling of Crack Growth by Finite Element
Method and Smeared Crack Approach’, Internal Report 2007-08-03-2002-DP, Cervenka
Consulting.
CRISFIELD, M.A., WILLS, J. (1989)- The Analysis of Reinforced Concrete Panels Using
Different Concrete Models, Jour. of Engng. Mech., ASCE, Vol 115, No 3, March, pp.578-597.
CRISFIELD, M.A. (1983) - An Arc-Length Method Including Line Search and Accelerations,
International Journal for Numerical Methods in Engineering, Vol.19,pp.1269-1289.
CHEN, W.F, SALEEB, A.F. (1982) - Constitutive Equations For Engineering Materials, John
Willey \& Sons, ISBN 0-471-09149-9.
DARWIN, D., PECKNOLD, D.A.W. (1974) - Inelastic Model for Cyclic Biaxial Loading of
Reinforced Concrete, Civil Engineering Studies, University of Illinois, July.
DE BORST, R. (1986), Non-linear analysis of frictional materials, Ph.D. Thesis, Delft
University of Technology, 1986.
DRUCKER, D.C., PRAGER, W., Soil Mechanics and Plastic Analysis or Limit Design, Q.
Appl. Math., 1952, 10(2), pp 157-165.
DYNGELAND, T. (1989) - Behavior of Reinforced Concrete Panels, Dissertation, Trondheim
University, Norway, BK-report 1989:1
 87

FEENSTRA, P.H., Computational Aspects of Bi-axial Stress in Plain and Reinforced Concrete.
Ph.D. Thesis, Delft University of Technology, 1993.
FEENSTRA, P.H., ROTS, J.G., AMESEN, A., TEIGEN, J.G., HOISETH, K.V., A 3D
Constitutive Model for Concrete Based on Co-rotational concept. Proc. EURO-C 1998, 1, pp.
13-22.
ETSE, G., Theoretische und numerische Untersuchung zum diffusen und lokalisierten Versagen
in Beton, Ph.D. Thesis, University of Karlsruhe 1992.
FELIPPA, C. (1966) - Refined Finite Element Analysis of Linear and Nonlinear Two-
Dimensional Structures, Ph.D. Dissertation, University of California, Engineering, pp.41-50.
GRASSL, P., LUNDGREN, K., and GYLLTOFT, K. (2002) “Concrete in compression : A
plasticity theory with a novel hardening law”, International Journal of Solids and Structures,
39(20), 5205-5223.
VAN GYSEL, A., and TAERWE, L. (1996) “Analytical formulation of the complete stress-
strain curve for high strength concrete”, Materials and Structures, RILEM, 29(193), 529-533.
HARTL, G. (1977) “Die Arbeitlinie Eingebetete Staehle bei erst und kurz=Belastung”,
Dissrtation, Univbersitaet Innsbruck
HORDIJK, D.A. (1991) - Local Approach to Fatigue of Concrete, Doctor dissertation, Delft
University of Technology, The Netherlands, ISBN 90/9004519-8.
KABELE, P. (2002) - Equivalent Continuum Model of Multiple Cracking, Engineering
Mechanics 2002, 9 (1/2), pp.75-90, Assoc.for Engineering Mechanics, Czech Republic
KESSLER-KRAMER, CH., (2002) “Zugverhalten von Beton unter Ermüdungsbeanspruchung”,
Schriftenreihe des Instituts für Massivbau und Baustofftechnologie, Heft 49, Karlsruhe.
KLAUSEN, D. (1978), Festigkeit und Schadigung von Beton bei haufig wiederholter
Beanschpruchung, PhD Thesis, University of Technology Darmstadt, 85 pp.
KOLLEGGER, J. - MEHLHORN, G. (1988) - Experimentelle und Analytische Untersuchungen
zur Aufstellung eines Materialmodels für Gerissene Stahbetonscheiben, Nr.6 Forschungsbericht,
Massivbau, Gesamthochschule Kassel.
KOLMAR, W. (1986) - Beschreibung der Kraftuebertragung über Risse in nichtlinearen Finite-
Element-Berechnungen von Stahlbetontragwerken", Dissertation, T.H. Darmstadt, p. 94.
KUPFER, H., HILSDORF, H.K., RÜSCH, H. (1969) - Behavior of Concrete under Biaxial
Stress, Journal ACI, Proc. V.66,No.8, Aug., pp.656-666.
MARGOLDOVA, J., CERVENKA V., PUKL R. (1998), Applied Brittle Analysis, Concrete
Eng. International, November/December 1998.
MENETREY, P., WILLAM, K.J. (1995), Triaxial failure criterion for concrete and its
generalization. ACI, Structural Journal, 1995, 92(3), pp 311-318.
MENETREY, Ph., WALTHER, R., ZIMMERMAN, Th., WILLAM, K.J., REGAN, P.E.
Simulation of punching failure in reinforced concrete structures. Journal of Structural
Engineering, 1997, 123(5), pp 652-659.
MIER J.G.M van (1986) - Multiaxial Strain-softening of Concrete, Part I: fracture, Materials
and Structures, RILEM, Vol. 19, No.111.
MINER M.A. (1945), Cumulative damage in fatigue. Transactions of the American Society of
Mechanical Engineering, 67:A159-A164.
88

OLIVIER, J., A Consistent Characteristic Length For Smeared Cracking Models, Int. J. Num.
Meth. Eng., 1989, 28, pp 461-474.
OWEN, J.M., FIGUEIRAS, J.A., DAMJANIC, F., Finite Element Analysis of Reinforced and
Pre-stressed concrete structures including thermal loading, Comp. Meth. Appl. Mech. Eng., 1983,
41, pp 323-366.
PALMGREN, A. (1924), Die Lebensdauer von Kugellagern. Zeitschrift Verein Deutscher
Ingenieure, 68(14):339-341.
PAPANIKOLAOU, V.K., and KAPPOS, A.J. (2007) “Confinement-sensitive plasticity
constitutive model for concrete in triaxial compression”, International Journal of Solids and
Structures, 44(21), 7021-7048.
PRAMONO, E, WILLAM, K.J., Fracture Energy-Based Plasticity Formulation of Plain
Concrete, ASCE-JEM, 1989, 115, pp 1183-1204.
RAMM, E. (1981) - Strategies for Tracing Non- linear Responses Near Limit Points, Non- linear
Finite Element Analysis in Structural Mechanics, (Eds. W.Wunderlich,E.Stein, K.J.Bathe)
RASHID, Y.R. (1968), Ultimate Strength Analysis of Pre-stressed Concrete Pressure Vessels,
Nuclear Engineering and Design,1968, 7, pp 334-344.
ROTS, J.G., BLAAUWENDRAAD, J., Crack models for concrete: discrete or smeared? Fixed,
multi-directional or rotating? HERON 1989, 34(1).
SAE, AE-4, Fatigue Design Handbook
SIMO, J.C., JU, J.W., Strain and Stress-based Continuum Damage Models-I. Formulations, II-
Computational Aspects, Int. J. Solids Structures, 1987, 23(7), pp 821-869.
SIMO, J.C., KENNEDY, J.G., GOVINDJEE, S., (1988), Non-smooth Multi-surface Plasticity
and Visco-plasticity. Loading/unloading Conditions and Numerical Algorithms, Int. J. Num.
Meth. Eng., 26, pp 2161–2185.
TAYLOR, G.I., (1938), ‘Plastic strain in metal’, J. Inst. Metals, 62, 307-324.
VAN MIER J.G.M. (1986), Multi-axial Strain-softening of Concrete, Part I: fracture, Materials
and Structures, RILEM, 1986, 19(111).
VECCHIO, F.J., COLLINS, M.P (1986)- Modified Compression-Field Theory for Reinforced
Concrete Beams Subjected to Shear, ACI Journal, Proc. V.83, No.2, Mar-Apr., pp 219-231.
VOS, E. (1983) - Influence of Loading Rate and Radial Pressure on Bond in Reinforced
Concrete, Dissertation, Delft University, pp.219-220.
WILKINS, M.L., Calculation of Elastic-Plastic Flow, Methods of Computational Physics, 3,
Academic Press, New York, 1964.
 89

3 FINITE ELEMENTS

3.1 Introduction
The preceding chapters dealt with the general formulation of the problem, geometric and
constitutive equations. All expressions were derived independently of the structural shape, the
finite elements used etc. Here, an information about finite elements currently implemented in
ATENA is given.

t
h2
3
7 6 h1

4 9 2

5 s
8
r 1

t t

3 3 h6
h8 7 6
7 6
2 4 9 2
4 9
5 5 s
8 s 8

1 r 1
r

t
h9
3
7 6

4 9 2

5 s
8
r 1

Fig. 3-1 Examples of interpolation function for plane quadrilateral elements.

The available elements can be divided into three groups: plane elements for 2D, 3D and axi-
symmetric analysis, solid 3D elements and special elements, which comprises elements for
modeling external cable, springs, gaps etc.
With few exceptions all elements implemented in ATENA are constructed using isoparametric
formulation with linear and/or quadratic interpolation functions. The isoparametric formulation
90

of one, two and three dimensional elements belongs to the "classic" element formulations. This
is not because of its superior properties, but due to the fact that it is a versatile and general
approach with no hidden difficulties and, also very important, these elements are easy to
understand. This is very important particularly in nonlinear analysis. For example it is highly
undesirable to add element-related problems to problems related to e.g. material modeling.
Big advantage of ATENA isoparametric elements is that their interpolation functions hi ( r, s, t )
are constructed in hierarchical manner. Take an example of plane quadrilateral element. Some of
its interpolation functions are depicted in Fig. 3-1. The 1st four functions, i.e. functions h1 ( r , s, t )
to h4 ( r , s, t ) has to be always present in the interpolation set, (to ensure bilinear approximation).
Then, any additional function h6 ( r, s, t ) through h9 ( r, s, t ) can be added independently. This
would involve adding the new function itself and also amendments to the already present
interpolation functions. This approach (and use of C++ templates) makes possible that one
element formulation generates quadrilateral elements with nodes (1,2,3,4), (1,2,3,4,5),
(1,2,3,4,6), ... (1,2,3,4,8), (1,2,3,4,9), (1,2,3,4,5,6), (1,2,3,4,5, 7), ... (1,2,3,8,9), ...
(1,2,3,4,5,6,7,8,9). Additional mid-side points are particularly useful for changing mesh density,
(i.e. element size), see Fig. 3-2, as they allow change of mesh density without need triangular
elements.
Although the concept of hierarchical elements was described for plane quadrilateral elements, in
ATENA it applies for plane triangular elements, 3D bricks, tetrahedral and wedge elements, too.
Always there is a set of basic interpolation function that can be extended by any “higher”
interpolation function.
Apart of interpolation functions finite element properties depend strongly on numerical
integration scheme used to integrate element stiffness matrix, element nodal forces etc. In Atena,
majority of elements are integrated by Gauss integration scheme that ensure n( n − 1) order
accuracy, where n is degree of the polynomial used to approximate the integrated function.

9 10 11 9 10 11

6 7 8 7
6 8

3 4 5 3 4 5

1 2 1 2

Solution with Standard

hierarchical solution
elements

Fig. 3-2 Change of finite element mesh density.

 91

3.2 Truss 2D and 3D Element

2D and 3D truss elements in ATENA are coded in group of elements CCIsoTruss<xx> ...
CCIsoTruss<xxx>. The string in < > describes present element nodes, (see Atena Input File
Format document for more information). These are isoparametric elements integrated by Gauss
integration at 1 or 2 integration points for the case of linear or quadratic interpolation, i.e. for
elements with 2 or 3 element nodes, respectively. They are suitable for plane 2D as well as 3D
analysis problems. Geometry, interpolation functions and integration points of the elements are
given in Fig. 3-3, Table 3.2-1 to Table 3.2-3.

s
y
2 r
3 CCIsoTruss<xx>
1
CCIsoTruxx<xxx>
x

Fig. 3-3 Geometry of CCIsoTruss<...> elements.

Table 3.2-1 Interpolation functions of CCIsoTruss<...> elements.

Node Function Include only if node 3 is

i defined
hi

1 1 1
(1 − r ) − h3
2 2
2 1 1
(1 + r ) − h3
2 2
3 (1 − r 2 )

Table 3.2-2 Sample points for Gauss integration of 1 node CCIsoTruss<xx> element.

Integration point Coordinate r Weight

1 0. 2.
92

Table 3.2-3 Sample points for Gauss integration of 2 and 3 nodes CCIsotruss<xxx> elements.

Integrati Coordinate r Weight

on point

1 0.577350269189626 1.
2 -0.577350269189626 1.

The element vectors and matrices for Total Lagrangian formulation , configuration at time t
and iteration (i) are as follows. Note that they are equally applicable for Updated Lagrangian
formulation upon applying changes related to the element reference coordinate system
(undeformed vs. deformed element axis.). The formulation is present for 3-nodes element option.
The 2-nodes variant is obtained by simply neglecting the terms for the element mid-point.
An arbitrary point on the truss element has at reference time t coordinates t X = [t x1 , t x1 , t x1 ] :
t
x1 = t x11 h1 + t x12 h2 + t x13 h3

t
x2 = t x21 h1 + t x22 h2 + t x23 h3 (3.1)

t
x3 = t x31 h1 + t x32 h2 + t x33 h3

At time t + Δt ( i −1) the same point has coordinates t +Δt

X ( i −1) :

t +Δt
x1( i −1) = ( t x11 + t u11( i −1) ) h1 + ( t x12 + t u12( i −1) )h2 + ( t x13 + t u13( i −1) )h3

t +Δt
x2( i −1) = ( t x21 + t u21( i −1) )h1 + ( t x22 + t u22( i −1) ) h2 + ( t x23 + t u23( i −1) )h3 (3.2)

t +Δt
x3( i −1) = ( t x31 + t u31( i −1) )h1 + ( t x32 + t u32( i −1) ) h2 + ( t x33 + t u33( i −1) )h3
t +Δt
and at time t + Δt ( i ) coordinates X (i )
t +Δt
x1( i ) = ( t x11 + t u11( i ) ) h1 + ( t x12 + t u12( i ) )h2 + ( t x13 + t u13( i ) )h3

t +Δt
x2( i ) = ( t x21 + t u21( i ) )h1 + ( t x22 + t u22( i ) ) h2 + ( t x23 + t u23( i ) )h3 (3.3)

t +Δt
x3( i ) = ( t x31 + t u31( i ) )h1 + ( t x32 + t u32( i ) ) h2 + ( t x33 + t u33( i ) )h3
t +Δt ( i ) t +Δt
Increment of Green Lagrange strain t ε11( i ) = ε
t 11 − t +Δttε11( i −1) (at time t + Δt , iteration (i )
with
to configuration at time t ) is calculated:
 93

⎛ ⎛ ∂ t +Δt l ( i ) ⎞2 ⎛ ∂ t +Δt l ( i −1) ⎞2 ⎞

⎜⎜ ⎟⎟ − ⎜ ⎟ ⎟⎟
1 ⎜ ⎝⎜ ∂r ⎠ ⎝ ∂r ⎠
ε(i )
t 11 = ⎜ 2 ⎟ (3.4)
2⎜ ⎛ ∂tl ⎞ ⎟
⎜⎜ ⎜ ∂r ⎟ ⎟⎟
⎝ ⎝ ⎠ ⎠
where truss length differentials are
2 2 2 2
⎛ ∂ t l ⎞ ⎛ ∂ t x1 ⎞ ⎛ ∂ t x2 ⎞ ⎛ ∂ t x3 ⎞
⎜ ⎟ =⎜ ⎟ +⎜ ⎟ +⎜ ⎟
⎝ ∂r ⎠ ⎝ ∂r ⎠ ⎝ ∂r ⎠ ⎝ ∂r ⎠

2 2 2 2
⎛ ∂ t +Δt l ( i −1) ⎞ ⎛ ∂ t +Δt x1( i −1) ⎞ ⎛ ∂ t +Δt x2( i −1) ⎞ ⎛ ∂ t +Δt x3( i −1) ⎞
⎜ ⎟ =⎜ ⎟ +⎜ ⎟ +⎜ ⎟ (3.5)
⎝ ∂r ⎠ ⎝ ∂r ⎠ ⎝ ∂r ⎠ ⎝ ∂r ⎠

2 2 2 2
⎛ ∂ t +Δt l ( i ) ⎞ ⎛ ∂ t +Δt x1( i ) ⎞ ⎛ ∂ t +Δt x2( i ) ⎞ ⎛ ∂ t +Δt x3( i ) ⎞
⎜ ⎟ =⎜ ⎟ +⎜ ⎟ +⎜ ⎟
⎝ ∂r ⎠ ⎝ ∂r ⎠ ⎝ ∂r ⎠ ⎝ ∂r ⎠

Substituting (3.5), (3.3) into (3.4) after some math manipulation it can be derived:
⎡ ∂h1 ∂h1 t 1 ∂h1 ∂h2 ∂h1 ∂h3 t 3 ⎤
⎢ ∂r ∂r x1 + ∂r ∂r x12 +
t
x1 ⎥
∂r ∂r
⎢ ⎥
⎢ ∂h2 ∂h1 t x1 + ∂h2 ∂h2 ∂h ∂h
x1 + 2 3 t x13 ⎥
t 2
⎢ ∂r ∂r 1 ∂r ∂r ∂r ∂r ⎥
⎢ ∂h ∂h ∂h ∂h ∂h ∂h ⎥
⎢ 3 1 t x11 + 3 2 x1 + 3 3 t x13 ⎥
t 2

⎢ ∂r ∂r ∂r ∂r ∂r ∂r ⎥
⎢ ∂h1 ∂h1 t 1 ∂h1 ∂h2 ∂h ∂h ⎥
⎢ x2 + x2 + 1 3 t x23 ⎥
t 2

⎢ ∂r ∂r ∂r ∂r ∂r ∂r ⎥
t +Δt 1 ⎢ ∂h2 ∂h1 t 1 ∂h2 ∂h2 ∂h2 ∂h3 t 3 ⎥
t BL 0 = x2 + x2 +
t 2
2 ⎢ x2 ⎥
⎛ ∂ l ⎞ ⎢ ∂r ∂r
t ∂r ∂r ∂r ∂r
⎥
⎜ ∂r ⎟ ⎢ ∂h3 ∂h1 t 1 ∂h3 ∂h2 ∂h3 ∂h3 t 3 ⎥
⎝ ⎠ x + x2 +
t 2
x2
⎢ ∂r ∂r 2 ∂r ∂r ∂r ∂r ⎥
⎢ ∂h ∂h ∂h ∂h ∂h1 ∂h3 t 3 ⎥
⎢ 1 1 t x31 + 1 2 x3 +
t 2
x3 ⎥
⎢ ∂r ∂r ∂r ∂r ∂r ∂r ⎥
⎢ ∂h2 ∂h1 t 1 ∂h2 ∂h2 ∂h ∂h ⎥
⎢ x3 + x3 + 2 3 t x33 ⎥
t 2

⎢ ∂r ∂r ∂r ∂r ∂r ∂r ⎥
⎢ ∂h3 ∂h1 t 1 ∂h3 ∂h2 ∂h3 ∂h3 t 3 ⎥
⎢⎣⎢ ∂r ∂r x3 + ∂r ∂r x3 + (3.6)
t 2
x3 ⎥
∂r ∂r ⎦⎥
94

⎡ ∂h1 ∂h1 t +Δt 1( i −1) ∂h1 ∂h2 t +Δt ∂h1 ∂h3 t +Δt 3( i −1) ⎤
⎢ ∂r ∂r u1 + u12( i −1) + u1 ⎥
∂r ∂r ∂r ∂r
⎢ ⎥
⎢ 2 1 t +Δt u1( i −1) + h2 ∂h2
∂h ∂h ∂ t +Δt 2( i −1)
u1 +
∂h2 ∂h3 t +Δt 3( i −1) ⎥
u1
⎢ ∂r ∂r 1
∂r ∂r ∂r ∂r ⎥
⎢ ∂h ∂h ∂ h ∂h ∂h ∂h ⎥
⎢ 3 1 t +Δt u11( i −1) + 3 2 t +Δt 2( i −1)
u1 + 3 3 t +Δt u13( i −1) ⎥
⎢ ∂r ∂r ∂r ∂r ∂r ∂r ⎥
⎢ ∂h1 ∂h1 t +Δt 1( i −1) ∂h1 ∂h2 t +Δt 2( i −1) ∂h ∂h ⎥
⎢ u2 + u2 + 1 3 t +Δt u23( i −1) ⎥
⎢ ∂r ∂ r ∂r ∂r ∂r ∂r ⎥
t +Δt 1 ⎢ ∂h2 ∂h1 t +Δt 1( i −1) ∂h2 ∂h2 ∂ h2 ∂h3 t +Δt 3( i −1) ⎥
t BL( i1−1) = 2 ⎢ u2 + t +Δt 2( i −1)
u2 + u2 ⎥
⎛ ∂ t l ⎞ ⎢ ∂r ∂r ∂r ∂r ∂r ∂r
⎥
⎜ ∂r ⎟ ⎢ ∂h3 ∂h1 t +Δt 1( i −1) ∂h3 ∂h2 ∂h3 ∂h3 t +Δt 3( i −1) ⎥
⎝ ⎠ u2 + t +Δt 2( i −1)
u2 + u2
⎢ ∂r ∂r ∂r ∂r ∂r ∂r ⎥
⎢ ∂h ∂h ∂h ∂h ∂h ∂h ⎥
⎢ 1 1 t +Δt u31( i −1) + 1 2 t +Δt 2( i −1)
u3 + 1 3 t +Δt u33( i −1) ⎥
⎢ ∂r ∂r ∂r ∂r ∂r ∂r ⎥
⎢ ∂h2 ∂h1 t +Δt 1( i −1) ∂h2 ∂h2 t +Δt 2( i −1) ∂h ∂h ⎥
⎢ u3 + u3 + 2 3 t +Δt u33( i −1) ⎥
⎢ ∂r ∂r ∂r ∂r ∂r ∂r ⎥
⎢ ∂h3 ∂h1 t +Δt 1( i −1) ∂h3 ∂h2 t +Δt 2( i −1) ∂h3 ∂h3 t +Δt 3( i −1) ⎥
u3 + u3 + u3 ⎥ (3.7)
⎢⎢⎣ ∂r ∂r ∂r ∂r ∂r ∂r ⎥⎦
and
⎡ ∂h1 ∂h2 ∂h3 ⎤
⎢ ∂r 0 0 0 0 0 0 ⎥
∂r ∂r
⎢ ⎥
1 ⎢ ∂h1 ∂h2 ∂h3
t +Δt ( n −1)
BNL = t 0 0 0 0 0 0 ⎥ (3.8)
⎛∂ l⎞⎢ ⎥
t
∂r ∂r ∂r
⎜ ∂r ⎟ ⎢ ∂h1 ∂h2 ∂h3 ⎥
⎝ ⎠⎢ 0 0 0 0 0 0 ⎥
⎣⎢ ∂r ∂r ∂r ⎦⎥

The 2nd Piola-Kirchhoff stress matrix and tensor are:

⎡ t +Δtt S11( i −1) 0 0 ⎤
t +Δt ⎢ ⎥
t S ( i −1) =⎢ 0 t +Δt ( i −1)
t S11 0 ⎥,
t +Δt
t S ( i −1) = [ t +Δtt S11( i −1) ] (3.9)
⎢ 0 0 t +Δt ( i −1)
t S11
⎥
⎣ ⎦
t +Δt (i )
The formulation is completed by relationship for element deformation gradient t X 1,1 , which
yields:
⎛ ∂ t +Δt l ( i ) ⎞
⎜ ∂r ⎟
t +Δt
t
(i )
X 1,1 =⎝ ⎠ (3.10)
⎛∂ l⎞t

⎜ ∂r ⎟
⎝ ⎠
Note that 2-nodes truss element has constant strains along its length and thus the increment of
Green Lagrange strain can be calculated directly, (i.e. not using differentials truss length as it
was the case of (3.4) ):
 95

1⎜( l ) −( l ) ⎟
⎛ t +Δt ( i ) 2 t +Δt ( i −1) 2 ⎞
t ε11( i ) = (3.11)
2⎜ t 2
l ⎟
⎝ ⎠
This yields a bit simpler element formulation (with the same results). However, for the sake of
preserving unified approach to all truss elements, ATENA uses even in this case the equation
(3.4).

3.3 Plane Quadrilateral Elements

Plane quadrilateral elements in ATENA are coded in group of elements CCIsoQuad<xxxx> ...
CCIsoQuad<xxxxxxxxx>. The string in < > describes present element nodes (see Atena Input
File Format document for more information). These are isoparametric elements integrated by
Gauss integration at 4 or 9 integration points for the case of bilinear or bi-quadratic interpolation,
i.e. for elements with 4 or 5 and more element nodes, respectively. They are suitable for plane
2D, axisymmetric and 3D problems.
CCIsoQuad2_5<...> elements present a simplified 3D formulation of the CCIsoQuad<...>
elements. Their higher execution performance is achieved at cost of omitting some nonlinear
terms, see below.
Geometry, interpolation functions and integration points of the elements are given in Fig. 3-4,
Table 3.3-1, Table 3.3-2, and Table 3.3-3.

y 2 s
5
1 CCIsoQuad<xxxx>
6
CCIsoQuad<xxxxx>
9 CCIsoQuad<xxxx_x>
8 r ....
3 CCIsoQuad<xxxx_x_x_>
7 ....
4 x CCIsoQuad<xxxxxxxxx>

Fig. 3-4 Geometry of CCIsoQuad<...> elements.

96

Table 3.3-1: Interpolation functions of CCIsoQuad<...> elements.

Node Function hi Include only if node i is defined

i

i=5 I=6 i=7 i=8 i=9

1 1 1 1 1
(1 + r )(1 + s ) − h5 − h8 h9
4 2 2 4
2 1 1 1 1
(1 − r )(1 + s ) − h5 − h6 h9
4 2 2 4
3 1 1 1 1
(1 − r )(1 − s ) − h6 − h7 h9
4 2 2 4
4 1 1 1 1
(1 + r )(1 − s ) − h7 − h8 h9
4 2 2 4
5 1 1
(1 − r 2 )(1 + s ) − h9
2 2
6 1 1
(1 − s 2 )(1 − r ) − h9
2 2
7 1 1
(1 − r 2 )(1 − s ) − h9
2 2
8 1 1
(1 + r )(1 − s 2 ) − h9
2 2
9 1
(1 − r 2 )(1 − s 2 )
2
 97

Table 3.3-2: Sample points for Gauss integration of 4 nodes CCIsoQuad<...> element.

Integrati Coordinate r Coordinate s Weight

on point

1 0.577350269189626 0.577350269189626 1.
2 0.577350269189626 - 1.
0.577350269189626
3 - 0.577350269189626 1.
0.577350269189626
4 - - 1.
0.577350269189626 0.577350269189626

Table 3.3-3: Sample points for Gauss integration 5 to 9 nodes CCIsoQuad<...> elements.

Integrati Coordinate r Coordinate s Weight

on point

1 0.774596669241483 0.774596669241483 0.3086419753

2 0.774596669241483 0. 0.4938271605
3 0.774596669241483 - 0.3086419753
0.774596669241483
4 0. 0.774596669241483 0.4938271605
5 0. 0. 0.7901234568
6 0. - 0.4938271605
0.774596669241483
7 - 0.774596669241483 0.3086419753
0.774596669241483
8 - 0. 0.4938271605
0.774596669241483
9 - - 0.3086419753
0.774596669241483 0.774596669241483

Equations (3.12) through (3.21) present CCIsoQuad<...> axisymmetric element formulation. 2D

element formulation is simply obtained by removing terms associated with circumferential
strains and stresses t +Δttε 33
( i ) t +Δt ( i )
, t S33 .
98

Incremental strains:

ε
(i )
t 11 = t u1,1
(i )
+ t +Δtt u1,1
( i −1)
t u1,1 +
(i ) t +Δt ( i −1)
t u2,1 t u2,1 +
(i ) 1
2
(( u
t 1,1) +( u ) )
(i ) 2
t
(i ) 2
2,1

t ε 22( i ) = t u2,2
(i )
+ t +Δtt u1,2
( i −1)
t u1,2 +
(i ) t +Δt ( i −1)
t u2,2 t u2,2 +
(i ) 1
2
(( u
t 1,2 ) +( u ) )
(i ) 2
t
(i ) 2
2,2

1
ε
(i )
t 12 =
2
( t u1,2(i ) + t u2,1(i ) ) +
1 t +Δt ( i −1) ( i ) t +Δt ( i −1) ( i ) t +Δt ( i −1) ( i ) t +Δt ( i −1) ( i )
2
( t u1,1 t u1,2 + t u2,1 t u2,2 + t u1,2 t u1,1 + t u2,2 t u2,1 ) +
1
2
( t u1,1(i ) t u1,2(i ) + t u2,1(i ) t u2,2(i ) )
2
u1( i ) t +Δt
u1( i ) u1( i ) 1 ⎛ u1( i ) ⎞
ε (i )
= t + + ⎜t ⎟
( x)
t 33 2
x1 t 2 ⎝ x1 ⎠ (3.12)
1

Displacement derivatives:
∂ ( t +Δt ui( i ) − t +Δt ui( i −1) )
t ui , j =
(i )

∂t x j
(3.13)
t +Δt ( i −1)
t +Δt ( i −1) ∂ u
u
t i, j = i

∂ xj
t

Strains and matrices to calculate them:

t ε (i ) = t +Δt
t BL( i −1) ΔU ( i )

t ε ( i ) = ⎡⎣ t ε11( i ) , t ε 22( i ) , 2 t ε12( i ) , t ε 33( i ) ⎤⎦ (3.14)

ΔU ( i ) = t +Δt U ( i ) − t +Δt U ( i −1) = ⎡⎣u11( i ) , u1(2 i ) , u12( i ) , u22( i ) ,..... u1n ( i ) , u2n ( i ) ⎤⎦

Linear strain-displacement matrix:

t +Δt
t BL( i −1) = t +Δt
t BL 0 + t +Δtt BL( i1−1) (3.15)
Linear strain-displacement matrix – constant part:
⎡ t h1,1 0 t h2,1 0 ... 0 0 ⎤
⎢ 0 ⎥
⎢ t h1,2 0 t h2,2 ... 0 t hn ,2 ⎥
t +Δt
B
t L0 = ⎢ t h1,2 t h1,1 t h2,2 t h2,1 ... t hn ,2 t hn ,1 ⎥ (3.16)
⎢ ⎥
⎢ h1 0
h2
0
h
... t n 0 ⎥
ti ti
⎣⎢ x1 x1 xi1 ⎦⎥
 99

where

∂hi
t hi , j =
∂t x j

(3.17)
ui( i ) = t +Δt ui( i ) −t +Δt ui( i −1)

n
t
xi1 = ∑ hk t x1k
k =1

Linear strain-displacement matrix – non-constant part:

⎡ t +Δt ( i −1)
l11 t h1,1 t +Δt ( i −1)
l
21 t 1,1 h t +Δt ( i −1)
l
11 t 2,1 h t +Δt ( i −1)
l h
21 t 2,1
⎢ t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1)
⎢ l11 t h1,2 l
21 t 1,2 h l
11 t 2,2 h l h
21 t 2,2
t +Δt ( i −1)
t B
L1 = ⎢ l11 t h1,2 + t +Δt l11( i −1) t h1,1
t +Δt ( i −1) t +Δt ( i −1)
l
21 t 1,2 h + t +Δt ( i −1)
21 t 1,1 l h t +Δt ( i −1)
l
11 t 2,2 h t +Δt ( i −1)
+ l
11 t 2,1 h t +Δt ( i −1)
l h
21 t 2,2 + t +Δt ( i −1)
l h
21 t 2,1
⎢
⎢ t +Δt ( i −1) h1 t +Δt ( i −1) h
l33 t 0 l 2
0
⎢⎣ xi1 xi
33 t
1

... t +Δt ( i −1)

l11 t n ,1 h t +Δt ( i −1)
l
21 t n ,1 h ⎤
t +Δt ( i −1) t +Δt ( i −1) ⎥
... l11 t n ,2 h l
21 t n ,2 h ⎥
... t +Δt ( i −1)
l h t +Δt ( i −1)
+ l h t +Δt ( i −1)
l h t +Δt ( i −1)
+ l h ⎥ (3.18)
11 t n ,2 11 t n ,1 21 t n ,2 21 t n ,1
⎥
t +Δt ( i −1) h ⎥
... l n

xi ⎥⎦
33 t
1

where
n
t +Δt ( i −1)
11l = ∑ t hk ,1 t +Δtt u1k ( i −1)
k =1

n
t +Δt ( i −1)
12l = ∑ t hk ,2 t +Δt k ( i −1)
u
t 1
k =1

n
t +Δt ( i −1)
21l = ∑ t hk ,1 t +Δtt u2k ( i −1) (3.19)
k =1

n
t +Δt ( i −1)
22l = ∑ t hk ,2 t +Δtt u2k ( i −1)
k =1

1 n
ti ∑
t +Δt ( i −1) t +Δt k ( i −1)
33l = hk t 1 u
x1 k =1
100

Nonlinear strain-displacement matrix

⎡ t h1,1 0 t h2,1 0 ... hn ,1t 0 ⎤
⎢ h 0 0 ⎥⎥
⎢ t 1,2 t h2,2 0 ... t hn ,2

t +Δt ( i −1)
⎢ 0 t h1,1 0 t h2,1 ... 0 t hn ,1
⎥
t BNL =⎢ ⎥ (3.20)
⎢ 0 t h1,2 0 t h2,2 ... 0 t hn ,2 ⎥
⎢ h1 h2 h ⎥
⎢ ti 0 t i
0 ... t n 0 ⎥
⎣⎢ x1 x1 xi1 ⎦⎥
2nd Piola-Kirchhoff stress tensor and vector
⎡ t +Δtt S11( i −1) t +Δt ( i −1)
t 12 S 0 0 0 ⎤
⎢ t +Δt ( i −1) t +Δt ( i −1) ⎥
⎢ t S 21 t 22 S 0 0 0 ⎥
t S (i −1) =⎢ 0 0 t +Δt ( i −1)
t S11
t +Δt ( i −1)
t S12 0 ⎥
⎢ t +Δt ( i −1) t +Δt ( i −1) ⎥
⎢ 0 0 t S 21 t S 22 0 ⎥
⎢ 0 0 0 0 t +Δt ( i −1) ⎥
⎣ t S33 ⎦

(3.21)
t S (i ) = ⎡⎣ t +Δtt S11(i −1) t +Δt
t
( i −1)
S 22 t +Δt
t
( i −1)
S 21 t +Δt
t S33(i −1) ⎤⎦

In case of the simplified 3D analysis, i.e. elements CCIsoQuad2_5<...>, the equations are further
extended as follows:
All element matrices and vectors are computed with respect to element local coordinate
system xlocal ,1 , xlocal ,2 using equations in (3.12) through (3.21). They are transformed into
3D global coordinate system by means of simple transformation:
M global = T M local T T , vglobal = T vlocal (3.22)
where
M global , M local , vglobal , vlocal are global and local finite element matrices and vectors,

T is transformation matrix from local to global coordinate system:

⎡ cos( xlocal ,1 , xglobal ,1 ), cos( xlocal ,2 , xglobal ,1 ) ⎤
⎢ ⎥
T = ⎢ cos( xlocal ,1 , xglobal ,2 ), cos( xlocal ,2 , xglobal ,2 ) ⎥ (3.23)
⎢ cos( xlocal .1 , x global ,3 ), cos( xlocal ,2 , x global ,3 ) ⎥
⎣ ⎦
where:
xlocal ,i , x global ,i are local and global coordinates (in 2D and 3D space).

The local element coordinate system (see Fig. 3-5) is defined by local xlocal ,1 , xlocal ,2 , xlocal ,3
coordinates. All of them pass through origin of the global (reference) coordinate system. The
axes xlocal ,1 and xlocal ,2 constitute a local coordinates element plane that is parallel to the element..
The axis xlocal ,3 is perpendicular to the element and the axis xlocal ,1 is defined as a projection of
global x1 axis to the local coordinate element plane. An exception to that is, when the element is
normal to the global x1 . In this case the local xlocal ,1 coincides with the global x2 axis.
 101

The present definition of local element coordinate system depends on plane of the finite element
but it does not depend on its shape itself. This is very important property, as ATENA supports
use of local (instead of global) nodal degrees of freedom and, (of course) these degrees of
freedom must refer to a coordinate system common to all elements of the plane, in which they
lie.

x3
3

xl o c a l , 3

4 O

X
X’ 2
xl o c a l , 2

1 x2
x1 xl o c a l , 1

Fig. 3-5 Local plane element coordinate system.

Full 3D formulation of the CCIsoQuad<...> elements is much the same as that for simplified 3D
elements CCIsoQuad2_5<...>. The only difference is that the matrix t0 BNL will include also terms
related to the „out-of-element-plane“ direction:
⎡ t h1,1 0 0 t h2,1 0 0 t h3,1 0 0 t hN ,1 0 0 ⎤
⎢ h 0 0 h2,2 0 0 h3,2 0 0 hN ,2 0 0 ⎥
⎢ t 1,2 t t t ⎥
t +Δt ( i −1)
⎢ 0 h 0 0 h2,1 0 0 h3,1 0 0 hN ,1 0 ⎥
t BNL =⎢ t 1,1 t t
...... t
⎥ (3.24)
⎢ 0 t h1,2
0 0 t h2,2
0 0 t h3,2
0 0 t hN ,2
0 ⎥
⎢ 0 0 t h1,1 0 0 t h2,1 0 0 t h3,1 0 0 t hN ,1
⎥
⎢ ⎥
⎢⎣ 0 0 h
t 1,2 0 0 t h2,2 0 0 t h3,2 0 0 t hN ,2 ⎥
⎦
102

3.4 Plane Triangular Elements

Plane triangular elements in ATENA are coded in group of elements CCIsoTriangle<xxx> ...
CCIsoTriangle<xxxxxx>. The string in < > describes present element nodes (see Atena Input
File Format document for more information). These are isoparametric elements integrated by
Gauss integration at 1 or 3 integration points for the case of bilinear or bi-quadratic interpolation,
i.e. for elements with 3 or 4 and more element nodes, respectively. They are suitable for plane
2D, axisymmetric and 3D problems. Geometry, interpolation functions and integration points of
the elements are given in Fig. 3-6, Table 3.4-1, Table 3.4-2, and Table 3.4-3.

3 s
CCIsoTriangle<xxx>
...
6 CCIsoTriangle<xxxxxx>
y
5

1
4 r
2

x
Fig. 3-6: Geometry of CCIsoTriangle<...> elements.

Table 3.4-1: Interpolation functions of CCIsoTriangle<...> elements.

Node Function hi Include only if node i is

i defined

i=4 i=5 i=6

1 1− r − s 1 1
− h4 − h6
2 2
2 r 1 1
− h4 − h5
2 2
3 s 1 1
− h5 − h6
2 2
4 4r (1 − r − s )
5 4rs
6 4 s (1 − r − s )
 103

Table 3.4-2: Sample point for Gauss integration of 3 nodes CCIsoTriangle<...> elements.

Integration point Coordinate r Coordinate s Weight

1 1/3 1/3 1/2

Table 3.4-3: Sample points for Gauss integration of 3 to 6 nodes CCIsoTriangle<...> elements.

Integration Coordinate r Coordinate s Weight

point

1 1/6 1/6 1/6

2 2/3 1/6 1/6
3 1/6 2/3 1/6
All the above expressions for the formulation for plane quadrilateral elements remain valid also
for the triangular elements, including the extension from 2D to simplified and full 3D analysis.
The expressions only use different approximation functions hi ( r, s, t ) and different integration
points [ r, s, t ] , see Table 3.4-1, Table 3.4-2, and Table 3.4-3.

3.5 3D Solid Elements

ATENA finite element library includes the following group of 3D solid elements:
tetrahedral elements CCIsoTetra<xxxx> ... CCIsoTetra<xxxxxxxxxx> with 4 to 10 nodes,
see Fig. 3-7,
brick elements CCIsoBrick<xxxxxxxx> ... CCIsoBrick<xxxxxxxxxxxxxxxxxxxx> with 8 up
to 20 nodes see Fig. 3-8 and
wedge elements CCIsoWedge<xxxxxx> ... CCIsoWedge<xxxxxxxxxxxxxxx> with 6 to 15
nodes, see Fig. 3-9.
The string in < > describes present element nodes (see Atena Input File Format document for
more information). These are isoparametric elements integrated by Gauss integration at
integration points given in the following tables. Interpolation functions for all variants of the
elements are also given in the tables below.
104

4
CCIsoTetra<xxxx>
9 .............
10
3 CCIsoTetra<xxxxxxxxxx>
8

7 6
1
5
2

Fig. 3-7 Geometry of CCIsoTetra<...> elements.

t
3
11 10
4 2
12 9 CCIsoBrick<xxxxxxxx>
1
..................
z20 19
CCIsoBrick<xxxxxxxxx...x>
7 18
15 s

8 17 14
6
r 16
13
5 y
x

Fig. 3-8 Geometry of CCIsoBrick<...> elements.

1 9 3

7 8
CCIsoWedge<xxxxxx>
2 .....
15
13 CCISoWedge<xxxxxx...x>

12 14
4 6

10 11
5

Fig. 3-9 Geometry of CCIsoWedge<...> elements.

 105

Table 3.5-1 Interpolation functions of CCIsoTetra<...> elements.

Node Function hi Include only if node i is defined

i

i=5 i=6 i=7 i=8 i=9 i = 10

1 1− r − s − t 1 1 1
− h5 − h7 − h10
2 2 2
2 r 1 1 1
− h5 − h6 − h8
2 2 2
3 s 1 1 1
− h6 − h7 − h9
2 2 2
4 t 1 1 1
− h8 − h9 − h10
2 2 2
5 4r (1 − r − s − t )
6 4rs(1 − t )
7 4 s (1 − r − s − t )
8 4rt (1 − s )
9 4 s t (1- r )
10 4t (1- r - s - t )

Table 3.5-2 Sample point for Gauss integration of 4 nodes CCIsoTetra<...> element.

Integration point Coordinate r Coordinate s Coordinate t Weight

1 1/4 1/4 1/4 1/6

106

Table 3.5-3 Sample points for Gauss integration of 5 to 10 nodes CCIsoTetra<...> elements.

Integration point Coordinate r Coordinate s Coordinate t Weight

1 0.13819660 0.13819660 0.13819660 1/24

2 0.13819660 0.13819660 0.58541020 1/24
3 0.58541020 0.13819660 0.13819660 1/24
4 0.13819660 0.58541020 0.13819660 1/24

Table 3.5-4 Interpolation functions of CCIsoBrick<...> elements.

Function hi Include only if node i is defined

Node i

i=9 i = 10 i = 11 i = 12 i = 13 i = 14 i = 15 i = 16 i = 17 i = 18 i = 19 i = 20

1 1 1
(1 + r )(1 + s )(1 + t ) − h 9
1
− h12
1
− h17
8 2 2 2

2 1
(1 − r )(1 + s )(1 + t )
1 1
− h9 − h10
1
− h18
8 2 2 2

3 1
(1 − r )(1 − s )(1 + t )
1 1
− h10 − h11
1
− h19
8 2 2 2

4 1
(1 + r )(1 − s )(1 + t )
1 1
− h11 − h12
1
− h20
8 2 2 2

5 1
(1 + r )(1 + s )(1 − t )
1
− h13
1 1
− h16 − h17
8 2 2 2

6 1
(1 − r )(1 + s )(1 − t )
1 1
− h13 − h14
1
− h18
8 2 2 2

7 1
(1 − r )(1 − s )(1 − t )
1 1
− h14 − h15
1
− h19
8 2 2 2

8 1
(1 + r )(1 − s )(1 − t )
1 1
− h15 − h16
1
− h20
8 2 2 2

9 1
(1 − r 2 )(1 + s )(1 + t )
4
 107

10 1
(1 − r )(1 − s 2 )(1 + t )
4

11 1
(1 − r 2 )(1 − s )(1 + t )
4

12 1
(1 + r )(1 − s 2 )(1 + t )
4

13 1
(1 − r 2 )(1 + s )(1 − t )
4

14 1
(1 − r )(1 − s 2 )(1 − t )
4

15 1
(1 − r 2 )(1 − s )(1 − t )
4

16 1
(1 + r )(1 − s 2 )(1 − t )
4

17 1
(1 + r )(1 + s )(1 − t 2 )
4

18 1
(1 − r )(1 + s )(1 − t 2 )
8

19 1
(1 − r )(1 − s )(1 − t 2 )
4

20 1
(1 + r )(1 − s )(1 − t 2 )
4
108

Table 3.5-5 Sample points for Gauss integration of 8 nodes CCIsoBrick<...> element.

Integrati Coordinate r Coordinate s Coordinate t Weight

on point

1 0.5773502691896 0.5773502691896 0.577350269189626 1.

26 26
2 0.5773502691896 0.5773502691896 - 1.
26 26 0.577350269189626
3 0.5773502691896 - 0.577350269189626 1.
26 0.5773502691896
26
4 0.5773502691896 - - 1.
26 0.5773502691896 0.577350269189626
26
5 - 0.5773502691896 0.577350269189626 1.
0.5773502691896 26
26
6 - 0.5773502691896 - 1.
0.5773502691896 26 0.577350269189626
26
7 - - 0.577350269189626 1.
0.5773502691896 0.5773502691896
26 26
8 - - - 1.
0.5773502691896 0.5773502691896 0.577350269189626
26 26

Table 3.5-6 Sample points for Gauss integration of 9 to 20 nodes CCIsoBrick<...> element.

Integrati Coordinate r Coordinate s Coordinate t Weight

on point

1 0.7745966692414 0.7745966692414 0.774596669241483 0.1714677641

83 83
2 0.7745966692414 0.7745966692414 0. 0.2743484225
83 83
3 0.7745966692414 0.7745966692414 - 0.1714677641
83 83 0.774596669241483
 109

4 0.7745966692414 0. 0.774596669241483 0.2743484225

83
5 0.7745966692414 0. 0. 0.4389574760
83
6 0.7745966692414 0. - 0.2743484225
83 0.774596669241483
7 0.7745966692414 - 0.774596669241483 0.1714677641
83 0.7745966692414
83
8 0.7745966692414 - 0. 0.2743484225
83 0.7745966692414
83
10 0. 0.7745966692414 0.774596669241483 0.2743484225
83
11 0. 0.7745966692414 0. 0.4389574760
83
12 0. 0.7745966692414 - 0.2743484225
83 0.774596669241483
13 0. 0. 0.774596669241483 0.4389574760
14 0. 0. 0. 0.7023319616
15 0. 0. - 0.4389574760
0.774596669241483
16 0. - 0.774596669241483 0.2743484225
0.7745966692414
83
17 0. - 0. 0.4389574760
0.7745966692414
83
18 0. - - 0.2743484225
0.7745966692414 0.774596669241483
83
19 - 0.7745966692414 0.774596669241483 0.1714677641
0.7745966692414 83
83
20 - 0.7745966692414 0. 0.2743484225
0.7745966692414 83
83
21 - 0.7745966692414 - 0.1714677641
0.7745966692414 83 0.774596669241483
83
110

22 - 0. 0.774596669241483 0.2743484225
0.7745966692414
83
23 - 0. 0. 0.4389574760
0.7745966692414
83
24 - 0. - 0.2743484225
0.7745966692414 0.774596669241483
83
25 - - 0.774596669241483 0.1714677641
0.7745966692414 0.7745966692414
83 83
26 - - 0. 0.2743484225
0.7745966692414 0.7745966692414
83 83
27 - - - 0.1714677641
0.7745966692414 0.7745966692414 0.774596669241483
83 83

Table 3.5-7 Interpolation functions of CCIsoWedge<...> elements.

hh1 = (1 − r − s )
hh2 = r
hh3 = s
hh4 = 4r (1 − r − s )
hh5 = 4rs
hh6 = 4s (1 − r − s )
1+ t
hv1 =
2
1− t
hv2 =
2
hv3 = (1 − t 2 )
Node I

Function Include only if node i is defined

hi i=7 i=8 i=9 i = 10 i = 11 i = 12 i = 13 i = 14 i = 15

1 hh1 hv1 1
− h7
1
− h9
1
− h13
2 2 2

2 hh2 hv1 1
− h7
1
− h8
1
− h14
2 2 2
3 hh3 hv1 1
− h8
1
− h9
1
− h15
2 2 2
 111

4 hh1 hv2 1
− h10
1
− h12
1
− h13
2 2 2

5 hh2 hv2 1
− h10
1
− h11
1
− h14
2 2 2

6 hh3 hv2 1
− h11
1
− h12
1
− h15
2 2 2

7 hh4 hv1

8 hh5 hv1

9 hh6 hv1

10 hh4 hv2

11 hh5 hv2

12 hh6 hv2

13 hh1 hv3

14 hh2 hv3

15 hh3 hv3

Table 3.5-8 Sample points for Gauss integration of 6 nodes CCIsoWedge<...> element.

Integration point Coordinate r Coordinate s Coordinate t Weight

1 1/6 1/6 0.577350269189626 1/6

2 2/3 1/6 0.577350269189626 1/6
3 1/6 2/3 0.577350269189626 1/6
4 1/6 1/6 -0.577350269189626 1/6
5 2/3 1/6 -0.577350269189626 1/6
6 1/6 2/3 -0.577350269189626 1/6

Table 3.5-9 Sample points for Gauss integration of 7 to 15 nodes CCIsoWedge<...> element.

Integration point Coordinate r Coordinate s Coordinate t Weight

1 1/6 1/6 0.774596669241483 0.0925925926

2 2/3 1/6 0.774596669241483 0.0925925926
3 1/6 2/3 0.774596669241483 0.0925925926
4 1/6 1/6 0. 0.1481448148
112

5 2/3 1/6 0. 0.1481448148

6 1/6 2/3 0. 0.1481448148
7 1/6 1/6 -0.774596669241483 0.0925925926
8 2/3 1/6 -0.774596669241483 0.0925925926
9 1/6 2/3 -0.774596669241483 0.0925925926

Formulation of 3D solid elements is given in the following equations:

Incremental strains:
1 (i ) 1 t +Δt ( i −1) ( i ) t +Δt ( i −1) ( i ) 1 ( i ) ( i )
t ε ij( i ) =
2
( t ui , j + t u j ,i ) +
(i )

2
( t uk ,i t uk , j + t uk , j t uk ,i ) + 2 ( t uk ,i t uk , j ) (3.25)

where indices i, j, k ∈< 1...3 >

Displacement derivatives:
∂ ( t +Δt ui( i ) − t +Δt ui( i −1) )
t u (i )
i, j =
∂t x j
(3.26)
t +Δt ( i −1) ∂ t +Δt ui( i −1)
t i, j u =
∂t x j
Strains and matrices to calculate them:

t ε ( i ) = t +Δtt BL(i −1) ΔU (i )

t ε ( i ) = ⎡⎣ t ε11(i ) t ε 22(i ) t ε 33(i ) 2 t ε12(i ) 2 t ε 23( i ) 2 t ε13(i ) ⎤⎦

(3.27)

ΔU (i ) =t +Δt U (i ) −t +Δt U (i −1) =

⎡⎣u11(i ) u1(2 i ) u31(i ) u12( i ) u22(i ) u32(i ) ... u1n ( i ) u2n (i ) u3n (i ) ⎤⎦

Linear strain-displacement matrix:

t +Δt
t BL( i −1) = t +Δt
t BL 0 + t +Δtt BL( i1−1) (3.28)
Linear strain-displacement matrix – constant part:
⎡ t h1,1 0 0 t h2,1 0 0 ... t hn ,1 0 0 ⎤
⎢ 0 h 0 0 h2,2 0 ... 0 hn ,2 0 ⎥
⎢ t 1,2 t t ⎥
t +Δt
⎢ 0 0 h
t 1,3 0 0 t h2,3 ... 0 0 t n ,3 ⎥
h
t BL 0 = ⎢ ⎥ (3.29)
⎢ t h1,2 h
t 1,1 0 t h2,2 t h2,1 0 ... t hn ,2 t hn ,1 0 ⎥
⎢ 0 h
t 1,3 h
t 1,2 0 t h2,3 t h2,2 ... 0 t hn ,3 t hn ,2
⎥
⎢ ⎥
⎢⎣ t h1,3 0 h
t 1,1 t h2,3 0 t h2,1 ... t hn ,3 0 t hn ,1 ⎥
⎦
where
 113

∂hi
t hi , j =
∂t x j (3.30)

ui( i ) = t +Δt ui( i ) − t +Δt ui( i −1)

Linear strain-displacement matrix – non-constant part:

⎡ t +Δt ( i −1)
l11 t h1,1 t +Δt ( i −1)
l
21 t 1,1 h t +Δt ( i −1)
l h
31 t 1,1
t +Δt ( i −1)
l h
11 t 2,1
⎢ t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1)
⎢ l12 t h1,2 l
22 t 1,2 h l h
32 t 1,2 l h
12 t 2,2

t +Δt
⎢ t +Δt ( i −1)
l t h1,3 t +Δt ( i −1)
l h t +Δt ( i −1)
l h t +Δt ( i −1)
l h
t BL( i1−1) = ⎢ t +Δt ( i −1) 13 t +Δ t ( i −1) t +Δt ( i −1)
23 t 1,3
t +Δt ( i −1) t +Δt ( i −1)
33 t 1,3
t +Δt ( i −1) t +Δt ( i −1)
13 t 2,3
t +Δt ( i −1)
⎢ l11 t h1,2 + l11 t h1,1 l 21 t 1,2 h + l
21 t 1,1 h l
31 t 1,2 h + l h
31 t 1,1 l h +
11 t 2,2 l h
12 t 2,1
⎢ t +Δt l12( i −1) t h1,3 + t +Δt l13( i −1) h1,2 t +Δt ( i −1)
l 22 t 1,3 h + t +Δt ( i −1)
l
23 t 1,2 h t +Δt ( i −1)
l
32 t 1,3 h + t +Δt ( i −1)
l h
33 t 1,2
t +Δt ( i −1)
l h +
12 t 2,3
t +Δt ( i −1)
l h
13 t 2,2
⎢ t +Δt ( i −1) t +Δt ( i −1)
t
t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1)
⎢⎣ l11 t 1,3 h + l13 t h1,1 l21 t 1,3 h + l
23 t 1,1 h l
31 t 1,3 h + l h
33 t 1,1 l h +
11 t 2,3 l h
13 t 2,1

(3.31)

... t +Δt ( i −1)

l
31 t n ,1 h ⎤
t +Δt ( i −1) ⎥
... l
31 t n ,2 h ⎥
... t +Δt ( i −1)
l
33 t n ,3 h ⎥
t +Δt ( i −1) t +Δt ( i −1) ⎥
... l
31 t n ,2 h + 32 t n ,1 l h ⎥
... t +Δt ( i −1)
l
32 t n ,3 h + t +Δt ( i −1)
33 t n ,2 l h ⎥
t +Δt ( i −1) t +Δt ( i −1) ⎥
... l
31 t n ,3 h + 33 t n ,1 l h ⎥⎦

where
n
t +Δt ( i −1)
ijl = ∑ t hk , j t +Δt k ( i −1)
t i u (3.32)
k =1

Nonlinear strain-displacement matrix

⎡ t h1,1 0 0 t h2,1 ... t hn ,1 0 0 ⎤
⎢ h 0 0 h2,2 ... hn ,2 0 0 ⎥
⎢ t 1,2 t t
⎥
⎢ t h1,3 0 0 t h2,3 ... t hn ,3 0 0 ⎥
⎢ ⎥
⎢ 0 h
t 1,1 0 0 ... 0 t hn ,1 0 ⎥
t +Δt
t B( i −1)
NL =⎢ 0 h
t 1,2 0 0 ... 0 t hn ,2 0 ⎥ (3.33)
⎢ ⎥
⎢ 0 h
t 1,3 0 0 ... 0 t hn ,3 0 ⎥
⎢ 0 0 h
t 1,1 0 ... 0 0 t hn ,1
⎥
⎢ ⎥
⎢ 0 0 h
t 1,2 0 ... 0 0 t hn ,2 ⎥
⎢⎢ 0 0 h 0 ... 0 0 ⎥
t hn ,3 ⎥
⎣ t 1,3 ⎦
114

2nd Piola-Kirchhoff stress tensor and vector

⎡ t +Δtt S11( i −1) t +Δt ( i −1)
t 12S t +Δt ( i −1)
t 13 S 0 0 0 0 0 0 ⎤
⎢ t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1) ⎥
⎢ t S21 t 22S t 23 S 0 0 0 0 0 0 ⎥
⎢ t +Δtt S31( i −1) t +Δt ( i −1)
t 32S t +Δt ( i −1)
t 33 S 0 0 0 0 0 0 ⎥
⎢ t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1) ⎥
⎢ 0 0 0 S
t 11 t 12S S
t 13 0 0 0 ⎥
t S ( i −1)
=⎢ 0 0 0 t +Δt ( i −1)
S
t 21
t +Δt ( i −1)
t 22S t +Δt ( i −1)
S
t 23 0 0 0 ⎥
⎢ t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1) ⎥
⎢ 0 0 0 S
t 31 t 32S S
t 33 0 0 0 ⎥
⎢ 0 0 0 0 0 0 t +Δt ( i −1)
S t +Δt ( i −1)
S t +Δt ( i −1) ⎥
t S13
⎢ t 11
t +Δt ( i −1)
t 12
t +Δt ( i −1) t +Δt ( i −1)
⎥
⎢ 0 0 0 0 0 0 S
t 21 S
t 22 t S 23 ⎥
⎢⎢ 0 t +Δt ( i −1) t +Δt ( i −1) t +Δt ( i −1) ⎥
⎥
⎣ 0 0 0 0 0 S
t 31 S
t 32 t S33 ⎦

S ( i ) = ⎡⎣ t +Δtt S11( i −1) t +Δt

t
( i −1)
S22 t +Δt
t S33( i −1) t +Δt ( i −1)
S
t 12
t +Δt
t
( i −1)
S23 t +Δt ( i −1)
S
t 13 ⎤⎦
(3.34)

3.6 Spring Element

Spring elements in ATENA are used to model spring-like boundary conditions, i.e. situation
where external forces acting on boundary of the structure are linearly proportional to the
associated displacements. Three elements of this type are available, see also Fig. 3-10, Fig. 3-11:
CCSpring – 2D and 3D element to model spring-like boundary conditions at a point,
CCLineSpring – 2D element to model spring-like boundary conditions along a line
CCPlaneSpring – 3D element to model spring-like boundary conditions along a triangular area.
All these elements are derived from 2D or 3D formulation of the CCIsoTruss<xx> element
described earlier in this chapter. For example, CCSpring element consists of one
CCIsoTruss<xx> element. The 1st node of each CCIsoTruss<xx> coincides with one node of the
CCSpring element, whereas the 2nd node of the CCIsoTruss<xx> is set by direction vector, see
Fig. 4-4. Note that as the analysis is nonlinear, length of the direction does matter. This vector is
specified in ATENA &SPRING_GEOMETRY_SPEC command and is common for all spring
elements that use this geometry.
CCLineSpring and CCPlaneSpring elements were created to enable convenient definition of
„uniform“ spring-like conditions along the boundaries. The boundary force at a node i of the
spring element is calculated:
ui kA
Ri = (3.35)
n direction
where
k is spring material stiffness parameter set by &MATERIAL SPRING command,
(parameter k has character of multi-linear Young modulus),
ui is displacement at spring element node i ,
 115

A is the area of CCPlaneSpring element or length of CCLineSpring multiplied by

thickness (which defaults to 1 if not specified in element geometry) or the area defined in
element geometry for CCSpring (similarly, with a default of 1 if not specified) for the
respective element,
n is number element nodes, i.e. 1, 2 or 3 for CCSpring, CCLineSpring or CCPlaneSpring
element respectively,
direction is Euclidean norm (i.e. length) of the direction vector, see above.

CCSpring CCLineSpring

Fig. 3-10 Geometry of 2D CCSpring and CCIsoLineSpring.

CCPlaneSpring

CCSpring

area A

Fig. 3-11 Geometry of 3D CCSpring and CCPlaneSpring.

116

3.7 Quadrilateral Element Q10

3.7.1 Element Stiffness Matrix

The quadrilateral finite element Q-10 is derived from a six-node triangle (CCQ10<xxxx>,
CCQ10Sbeta<xxxx>). The derivation of the stiffness matrix is taken from FELIPPA 1966. The
position of any internal point P in the element is defined by the triangular coordinates ζι (called
also natural coordinates). These coordinates are expressed by means of areas within the triangle
as shown in Fig. 3-12. Sub-areas Ai are subtended by the point P and two corners. A is the area of
triangular element.

Fig. 3-12 Coordinate systems of the six-node triangular element.

A1 A A
ζ1 = , ζ2 = 2 , ζ3 = 3 (3.36)
A A A
ζ1 + ζ 2 + ζ 3 = 1
Using the quadratic interpolation function, the displacement components u( ζ i ), v( ζ i ) is written
in the terms of triangular coordinates ζ i and nodal displacement vectors :

u (ζ i ) = F(ζ i )T u, v(ζ i ) = F(ζ i )T v (3.37)

The displacement vectors u, v contain six components of the nodal displacements and the vector
F (ζ i ) contains the quadratic interpolation functions in triangular coordinates:

u = { u1 u2 u3 u4 u5 u6 } , v = { v1 v2 v3 v4 v5 v6 }
T T
(3.38)

F(ζ i ) = {ζ 1 (2ζ 1 −1) ζ 2 (2ζ 2 − 1) ζ 3 (2ζ 3 − 1) 4ζ 1ζ 2 4ζ 3ζ 1}

T
4ζ 2ζ 3 (3.39)
A general procedure to construct the element stiffness matrix is described by the set of following
equations:
(a) The constitutive equation:
s = De (3.40)
 117

(b) The strain-displacement equations in the Cartesian coordinates:

∂u ( x, y ) ∂v ( x, y ) ∂u ( x, y ) ∂v ( x, y )
εx = , εy = , γ = + (3.41)
∂x ∂y ∂y ∂x
which is written in terms of the natural coordinates ζ i and the nodal displacements vectors u, v:

⎧u ⎫
ε (ζ i ) = F∂ T ⎨ ⎬ (3.42)
⎩v⎭
The stiffness matrix:
K = ∫ F∂ DF∂ dV
T
(3.43)
V

The matrix F∂ contains partial derivatives of the interpolation function F and the integral in the
last equation is made over the element volume V. The details of the derivation can be found in
FELIPPA 1966 and here only the final matrix equations are presented.

Fig. 3-13 Quadrilateral element (b) composed from two triangular elements (a).

The quadrilateral finite element is composed from two 4-node triangular elements, as shown in
Fig. 3-13. Two degrees of freedom in a node are the horizontal and vertical displacements. The
triangular element is derived from the 6-node triangle by imposing kinematic constraints on two
mid-side nodes. The resulting strain-displacement matrix relation for the 4-node triangle is:
⎧e x ⎫ ⎡ U O ⎤
e = Bd
⎪ ⎪ ⎢ ⎥ ⎧u ⎫
⎨e y ⎬ = ⎢O V ⎥ ⎨ ⎬ (3.44)
⎪ g ⎪ ⎢V U ⎥ ⎩v ⎭
⎩ ⎭ ⎣ ⎦
where ex, ey are the normal strain vectors, g is the shear strain vector (engineering type) and O is
the null matrix. The strain and displacement vectors contain nodal components:

e x = {ε x1 ε x 2 ε x 3 } , e y = {ε y1 ε y 2 ε y 3 } , g = {γ x1 γ x 2 γ x 3 }
T T T
(3.45)

u = { u1 u2 u3 u4 } , v = { v1 v2 v3 v4 }
T T
(3.46)

The strain interpolation function in the element is linear and is uniquely specified by three nodal
values in the corners of the triangular element, while the displacement interpolation function is
quadratic and is specified by three corners and one mid-side nodal displacement. The
118

components ui, vi are the horizontal and vertical displacements, respectively, in the node i. The
indexes 1, 2 and 3 denote the corner nodes of a sub-triangle and the index 4 is for the mid-side
node, see Fig. 3-13 (a). The strain-displacement sub-matrices in (3.44) are
⎡3b1 + 2b3 −b2 b3 4b2 ⎤
1 ⎢
U= −b1 3b2 + b3 b3 4b1 ⎥⎥
2S ⎢
⎢⎣ b1 b2 b3 . ⎥⎦

⎡3a1 + 2a3 − a2 a3 4a2 ⎤

1 ⎢
V= −a1 3a2 + a3 a3 4a1 ⎥⎥ (3.47)
2S ⎢
⎢⎣ a1 a2 a3 . ⎥⎦

a1 = x3 − x2 b1 = y2 − y3
a2 = x1 − x3 b2 = y3 − y1
a3 = x2 − x1 b3 = y1 − y2
2S = a3 b2 − a2 b3
where xi, yi are the global Cartesian coordinates of the node i in a sub-triangle, S is the area of the
sub-triangle.
The element stiffness matrix for the 4-node sub-triangle is
⎡K K uv ⎤
K = ⎢ uu (3.48)
⎣ K vu K vv ⎥⎦

The stiffness matrix K has an order 8 and is so partitioned that the upper four rows correspond to
the horizontal displacement components (index u) and the lower four rows correspond to the
vertical displacement components (index v). The integration of the stiffness coefficients is made
exactly and the resulting sub-matrices are:
K uu = St ⎡⎣ d11 A + d13 (H + HT ) + d33 C ⎤⎦

K vv = St ⎡⎣ d 22 C + d 23 (H + HT ) + d33 A ⎤⎦

K uv = St ⎡⎣ d12 H + d13 A + d 23 C + d33 HT ⎤⎦ (3.49)

where t is the thickness of the element, dij are the coefficients of the material stiffness matrix D,
(3.40). The integration in (3.43) is done explicitly by the following matrix multiplication:
A = UT QU, H = UT QV, C = VT QV (3.50)
Where the area integration matrix Q is:
⎡2 1 1⎤
1 ⎢
Q= ⎢ 1 2 1 ⎥⎥ (3.51)
12
⎢⎣1 1 2 ⎥⎦

The element stiffness matrix of the 5-node quadrilateral, Fig. 3-13(b), is composed of the two 4-
node sub-triangles by summing the stiffness coefficients of the appropriate nodes. The resulting
matrix of the 5-node quadrilateral K10 has the order 10. The coefficients of the matrix can be
rearranged according to the external (index e) and internal (index i) degrees of freedom:
 119

⎡K K ei ⎤
K10 = ⎢ ee (3.52)
⎣ K ie K ii ⎥⎦

The sub-matrices corresponding to two internal degrees of freedom are eliminated by

condensation procedure and the final element stiffness matrix K of the order 8 is obtained:
−1
K = K ee − K ei K ii K ie (3.53)

Fig. 3-14 Subdivision of quadrilateral element.

The subdivision of the quadrilateral element into the triangular elements must be done in an
optimal way and it is preformed automatically by the program. The examples of the subdivisions
are illustrated by Fig. 3-14. Due to this method of the subdivision, a concave form of the
quadrilateral element is acceptable. This element form could not be achieved by an isoparametric
element.

3.7.2 Evaluation of Stresses and Resisting Forces

For the given displacement field the strains and stresses are evaluated in the center of the
quadrilateral element. The stresses at this point are obtained from material laws as functions of
strains according to Section 2.1.12. Also the constitutive law for the element and the matrix D
are calculated from the stresses and strains at the center of the element. These stresses and strains
are written in the output file as a part of the results.
The calculation of resisting nodal forces of the sub-triangle for a current displacement field and a
constitutive law is done by the following equation:
R = t BT Q 9 s 9 (3.54)
where R is the vector of nodal forces (same arrangement and numbering as in the vector d in
(3.44)). The matrix Q9 contains three integration matrices Q in the diagonal. The stress vector s9
(same numbering as the vector e, (3.40), is calculated from the current strains and secant material
matrix, Section 2.1.12.
There are two variations of this element in program ATENA: CCQ10<xxxx> and
CCQ10Sbeta<xxxx>. The main difference between these two elements lies in the way how the
resisting forces are calculated. In case CCQ10<xxxx>, they are computed as described by
Equation (3.54). In the second case, however, the material law is evaluated only at the element
centroid. Based on the current state of damage a secant constitutive matrix is calculated and it is
used to determine the integration point stresses and resulting resisting forces. This element type
is almost identical to the element that was implemented in the program SBETA, i.e. the former
version of this program. Due to this approach there are some limitations for usage of this element
with respect to some material models. It can be only used with material models that are able to
calculate and exact secant constitutive matrix. This means that only the following material
models can be used with the element CCQ10Sbeta<xxxx>: CCElastIsotropic and
CCSbetaMaterial.
120

3.8 External Cable

External pre-stressing cables are reinforcing bars, which are not connected with the most of the
concrete body, except of limited number of points, so called deviators, as shown in Fig. 3-15.
This element type is denoted in ATENA as CCExternalCable.

Fig. 3-15 External cable model.

Each cable has two ends provided with anchors. The anchor, where the pre-stressing force is
applied is denoted as the active anchor, the anchor on the other side is the passive anchor. The
points between the anchors are called deviators (or links). After applying pre-stressing the cable
is fixed at anchors. In the deviators, cable can slide while its movements and the forces are
governed by the law of dry friction. The slips of the cable in the deviators (the relative
displacement of the cable ends with respect to the deviators) are denoted as δ1, δ2, … They are
introduced as variables to be determined by the analysis.

Fig. 3-16 Forces at the deviator.

The forces, F1 and F2 acting on a deviator i are the cable forces at the adjacent cable sections,
Fig. 3-16. Their difference Pi = F1 -F2, (F1> F2) is the loss of the pre-stressing force due to
friction in the deviator i. The relation between these forces according to the law of friction is
expressed as:
 121

F2 = F1 exp(−ϕα i ) − Q (3.55)
In this expression αi is the angular change of the cable direction at the deviator i and ϕ is the
friction coefficient. In the above equation the constant part of the friction Q = k R αi , where k is
the cohesion (a constant part of the friction ) of the cable of a unit length and R is the radius of
the deviator. The product Rαi is the length on which is the cohesion k acting. If the constant part
of friction is neglected the term Q is zero.

Fig. 3-17 Forces and displacements in the cable element (cable section).

A section of the cable between the deviators is considered as the uniaxial bar element, Fig. 3-17.
The force F in the cable element depends on: the pre-stressing force P, the displacements of ends
u1, u2 due to structural deformation and the cable slips δ1, δ2 in the deviators. The slip is
introduced as an additional variable for the external cables. The equilibrium equation of the
cable section is:
F = P + k (u2 − u1 + δ 2 − δ1 ) (3.56)
The element stiffness k = Es A/L, where A, L are the cable cross section and the length,
respectively, and Es is the actual secant or tangent modulus derived in the same way as in case of
other reinforcement using bilinear or multi-linear law.
The cable forces F1, F2,…, are determined by applying the above equations in all cable deviators
and by simultaneous solution for slips δi .
Introduction of pre-stressing is accomplished by applying an initial slip (cable pull-out) at the
anchor end until a prescribed pre-stressing force is reached. This procedure reflects a real
process of pre-stessing and takes into account the loss of pre-stressing due to friction deviators
and deformation of the structure.

3.9 Reinforcement Bars with Prescribed Bond

Reinforcement bars with prescribed bonds are deviation of the external cables described in the
previous section. The main difference is that they are fully connected to the surrounding concrete
body, however this connection has only limited strength. This “bond” strength is specified by so
called cohesion stress.
This type of element is denoted as CCBarWithBond in ATENA. Typical reinforcement bar of
this type is depicted in the figure below. The detail shows undeformed and deformed shape of a
segment of the bar. The original length l0 will change to l due to displacement u of the
surrounding body and bar slips δ .
122

σc σc
1 2 i-1 i m

1 2 i-1 i i+1 m+1

undeformed truss i

deformed truss i

u δ δ
i i i+1
u
i+1
lo
l

Fig. 3-18 Reinforcement bar with slips.

Normal stress at element i is calculated by:

(ui +i + δ i +1 − ui − δ i )
σi = E (3.57)
li
Its derivative is compared with the cohesion stress. If the cohesion stress between the bar and the
surrounding concrete is becoming too high, the bar will slip to reduce this stress. Otherwise, the
slips δ will remain unchanged (or initially equal to zero), which correspond to the case of
perfect bond. The cohesion stress can be constant or it can be defined as a function of δ .
σ c = σ c (δ ) (3.58)
The equilibrium condition for reinforcement bar with prescribed bond yields:
∂σ p
≤ σc (3.59)
∂x A
where σ is stress in the bar, x is local coordinate axis in direction of the bar and p, A means
perimeter and cross sectional area of the bar.
Discretized form of (3.59) for node i reads (the bars are of constant strain type):
li + li −1
A(σ i − σ i −1 ) ≤ p σc (3.60)
2
If the above equation is written for all nodes on the bar, we obtain a set of inequalities. It has to
be solved in iterative manner (within each iteration of the main solution loop).
In order to obtain more realistic shape, the resulting cohesion stresses are prior their output
smoothed. The smoothing operation for node i is expressed as follows:
 123

σ i +1li +1 + σ i li
σi right =
li +1 + li

σ i li + σ i −1li −1
σi left = (3.61)
li + li −1

(σi right − σi left ) A

σc =
p li
The equation (3.57) together with (3.60) completes the element description. The element can be
used to model realistically cohesion between reinforcement bar and concrete. Such a model is
needed for analysis of pullout tests etc. Although the adopted solution is fairly simple, it
provides reasonable results accuracy at low computation cost. A more elaborate model of
cohesion between reinforcement bar and surrounding concrete can be achieved by using special
interface elements that is described in the next section.

3.10 Interface Element

The interface elements are used to model a contact between two surfaces. Currently, the
following element types are available: CCIsoCCIsoGap<xxxx> and CCIsoGap<xxxxxx>,
CCIsoGap<xxxxxxxx> for 2D and 3D analysis, respectively. These elements use linear
approximation of geometry. For the case of nonlinear geometry, use element type
CCIsoGap<xxxxxx> for 2D and CCIsoGap<xxxxxxxxxxxx> or
CCIsoGap<xxxxxxxxxxxxxxxx> for 3D. The string in < > describes present element nodes, (see
Atena Input File Format document for more information). The elements are derived from the
corresponding isoparametric elements (described in sections 3.3 and 3.4), i.e. they use the same
geometry and nodal ids etc. Geometry of the supported gap elements is depicted in Fig. 3-19.
124

Linear geometry Nonlinear geometry

v 1
2D v 1
r,u(r) u(r)
r,
2 2 5
4 4
6
3 3
CCIsoGap<xxxx> CCIsoGap<xxxxx_x>

w 3D 10 w

1 1
2 7
2 9
r,u(r,s) r,u(r,s) 8
4 3 4
3
5 5 12
11
6 6

s,v(r,s) s,v(r,s)
CCIsoGap<xxxxxx> CCIsoGap<xxxxxxxxxxxx>

13 w
w 1
1 9
2 2
12 16
5 5
4 r,u(r,s) 6 10 4
r,u(r,s) 6 11
3 3
14
8 8
7 15
7
s,v(r,s) s,v(r,s)
CCIsoGap<xxxxxxxx>
CCIsoGap<xxxxxxxxxxxxxxxx>

Fig. 3-19 CCIsoGap elements

The interface is defined by a pair of lines, (or surfaces in 3D) each located on the opposite side
of interface. In the original (i.e. undeformed) geometry, the interface lines/surfaces can share the
same position, or they can be separated by a small distance. In this case we speak about the
interface with nonzero thickness.
In the following, the interface behavior is explained on a simple 2-dimensional case, see section
2.6 for a full description of the interface material.
The interface element has two states:
• Open state: There is no interaction of the contact sides.
• Closed state: There is full interaction of the contact sides. In addition, friction sliding of
the interface is possible in case of interface element with a friction model.

Penalty method is employed to model the above behavior of the interface. For this purpose we
define a constitutive matrix of the interface in the form:
⎧ F ⎫ ⎡K 0 ⎤ ⎧ Δu ⎫
F = ⎨ τ ⎬ = ⎢ tt ⎨ ⎬ = Du (3.62)
⎩ Fσ ⎭ ⎣ 0 K nn ⎦⎥ ⎩ Δv ⎭
 125

in which Δu, Δv are the relative displacements of the interface sides (sliding and opening
displacements of the interface) in the local coordinate system r , s and K tt , K nn are the shear and
normal stiffness, respectively. This coefficient can be regarded as stiffness of one material layer
(real, or fictious) having a finite thickness. It should be understood that the layer is only a
numerical tool to handle the gap opening and closing. Fτ , Fσ are forces at the interface, (again at
the local coordinate system).
The actual derivation of gap elements is now demonstrated for the case of linear 2D gap element
CCIsoGap<xxxx>, see Fig. 3-19. The other elements are constructed in a similar way.
The element has two degrees of freedom defined in the local coordinate system, which is aligned
with the gap direction. They are relative displacements Δv, Δu and are defined as follows:

1 1
h1 = (1 + r ), h2 = (1 − r )
2 2

⎡ Δu ⎤ ⎡ h2 Δu1,4 + h1Δu2,3 ⎤
Δu = ⎢ ⎥ = ⎢ ⎥
⎣ Δv ⎦ ⎣ h2 Δv1,4 + h1Δv2,3 ⎦
⎡ u1 ⎤
⎢v ⎥
⎢ 1⎥
⎢ u2 ⎥
⎢ ⎥
⎡h 0 h2 0 −h2 0 −h1 0 ⎤ ⎢ v2 ⎥
Δu = ⎢ 1 = Bu
⎣0 h1 0 h2 0 − h2 0 − h1 ⎦⎥ ⎢ u3 ⎥
⎢ ⎥
⎢ v3 ⎥ (3.63)
⎢u ⎥
⎢ 4⎥
⎣⎢ v4 ⎦⎥

The rest of the element derivation is the same as in case of any other elements, i.e. the stiffness
matrix K = ∫ ΒT DBdV , vector of internal forces Q = ∫ ΒT FdV etc. A numerical integration in
two Gauss points is used to integrate the interface element stiffness matrix. The matrix K a nd
the vector Q are in local coordinate system and therefore before they are assembled in the
problem governing equations they must be transformer in global coordinates.
The stiffness coefficients depend on the gap state. The interface is considered open, if the normal
force Fσ >Rti (Rti is the interface tensile strength force) and the corresponding constitutive law is
(stress free interface):
⎧ Fτ ⎫ ⎧0 ⎫
⎨ ⎬=⎨ ⎬ (3.64)
⎩ Fσ ⎭ ⎩0 ⎭

The stiffness coefficients are set to small, but nonzero values K ttop , K nnop .
The interface element is considered closed if Fσ ≤ Rti. The stiffness coefficients are set to large
values K ttcl , K nncl . It should be noted that the stiffness coefficients are defined only for the purpose
of the numerical iterative solution. (Hint: The values of coefficients in the closed state (the large
126

values) are based on thickness comparable to the size of neighbor quadrilateral elements. The
minimum values in the open state can be about 1000 times smaller. )
The interface thickness in the out-of plane direction is normally provided as an input parameter.
In the case of axi-symmetric analysis it is however calculated using the formula:
t = 2π x (3.65)
where x is the distance from the axis of symmetry.

3.11 Truss Axi-Symmetric Elements.

In the following a circumferential truss elements for axisymmetric analysis are described. The
elements call CCCircumferentialTruss and CCCircumferentialTruss2 and they are aimed mainly
for modeling structural circumferential reinforcement. For radial reinforcement refer to
CCIsoTruss<xx> and CCIsoTruss<xxx> elements.
The CCCircumferentialTruss has one node only, whereas the CCCircumferentialTruss2 has
nodes two. They behaves much the same, the difference being only in calculation of their “cross-
sectional area”. In case of the CCCircumferentialTruss element the area is entered directly from
input data. The CCCircumferentialTruss2 element calculate the area as its thickness (defined in
its geometry data) multiplied by its length. Unlike isoparametric elements thses elements are
derived and computed analytically.
Geometry, interpolation functions and integration points of the elements are given in
Fig. 3-20.

y
y

2
1
1

CCCircumferentialTruss CCCircumferentialTruss2

x x

Fig. 3-20 Geometry of CCCircumferentialTruss and CCCircumferentialTruss2 elements.

In the following structural vectors and matrices for the CCIsoTruss element are derived.
Development of the CCIsoTruss2 is much the same. In fact, it is CCIsoTruss acting at the centre-
point of the CCIsoTruss2 element with its cross-sectional area calculated as explained above.
The element vectors and matrices for Total Lagrangian formulation (TL), configuration at time t
and iteration (i) are as follows. Note that they are equally applicable for Updated Lagrangian
formulation (UL) upon applying changes related to the element reference co-ordinate system
(undeformed vs. deformed element axis.).
 127

The truss element center has at reference time t and t + Δt ( i −1) co-ordinates t X = [t x1 , t x1 ] and
t +Δt ( i −1) ( i −1) ( i −1)
X = [t +Δt x1 , t +Δt x1 ] , respectively. The element length (at respective time) is its length is
t
l = 2π t x1 and t +Δt ( i −1)
l = 2π t ( t x11 + t u11( i −1) ) .
t +Δt ( i ) t +Δt
Increment of Green Lagrange strain t ε11( i ) = ε
t 11 − t +Δttε11( i −1) (at time t + Δt , iteration (i )
with
to configuration at time t ) is calculated:

1⎜( l ) −( l ) ⎟
⎛ t +Δt ( i ) 2 t +Δt ( i −1) 2 ⎞
ε
(i )
= (3.66)
( )
t 11
2⎜ t
l
2
⎟
⎝ ⎠
where truss length t +Δt ( i )
l = 2π t ( t x11 + t u11( i −1) + t u11( i ) ) . Note that t u11( i ) is co-ordinate increment
(i ) ( i −1)
( t +Δt x1 − t +Δt x1 ) . Substituting expressions for element length into (3.66) yields:

ε(i )
=
4π 2 (( x + u
t 1
1
t 1( i −1)
1 + t u11( i ) ) − ( t x11 + t u11( i −1) )
2 2
)=
(x)
t 11
t 1 2
1
(3.67)
1( i −1) 1( i ) 2
u1( i )
u u t
1⎛ u ⎞ 1( i )
t 1
+ 1 t 1
+ ⎜ ⎟
t 1

( x1 ) t 1 2
t 1 t 1
x 1 2⎝ x ⎠ 1

Separating t u11( i ) from (3.67) and rearranging in matrix form we obtain:

t +Δt 1
t BL 0 = (3.68)
t
x11

t +Δt
t
u11( i −1)
BL( i1−1) = (3.69)
(x)
t
t 1 2
1

and
t +Δt ( n −1) 1
BNL = (3.70)
( x11 )
t 2
t

The 2nd Piola-Kirchhoff stress matrix and tensor are:

t +Δt
t S ( i −1) = t +Δt
t S ( i −1) = [ t +Δtt S11( i −1) ] (3.71)
t +Δt (i )
The formulation is completed by relationship for element deformation gradient t X 1,1 , which
yields:

( x11 + t u11( i ) )
t 2
t +Δt
t
(i )
X 1,1 = 1 + t e11( i ) = (3.72)
t
x11
128

where engineering strain t e11( i ) is calculated by

⎛ 2 t 1 2 ⎞ ⎛ ⎞
⎜ 4π ( x1 + u1 ) − 4π ( x1 ) ⎟ ⎜ ( x11 + t u11( i ) ) − t x11 ⎟
2 t 1 t 1( i ) 2 t 2
t +Δt ( i )
l
e(i )
t 11 = =⎝ ⎠=⎝ ⎠ (3.73)
4π 2 ( t x11 )
t t 1
l 2 x1

3.12 Ahmad Shell Element

This section describes Ahmad shell element implemented in ATENA, see Fig. 3-24. It can be
used to model thin as well as thick shell or plate structures. It accounts for both plane and
bending structural stiffness. The element features quadratic geometry and displacement
approximation and therefore, the element’s shape can be non-planar. It is possible to account for
structural curvatures. Big advantage of this element is that it is seamlessly connectible to true 3D
ATENA elements.
Three modifications of this element are supported and these are characterized by Lagrangian,
Serendipity and Heterosis variant of geometry and displacement field approximation. In order to
avoid or minimize membrane and shear locking of shell element it is further possible to use full
integration scheme, as well as reduced and/or selective integration. The problems concerned with
combination of particular displacement approximation and integration scheme with respect to
locking phenomena are discussed.
The element is derived in a way similar as the other finite elements, which are described in this
manual. Hence, in the present description will concentrate mainly on features that are specific for
this element. Following Total Lagrangian formulation of the problem, the principle of virtual
displacement is used to assemble incremental form of governing equations of structure.
The present Ahmad element belongs to group of shell element formulation that is based on 3D
elements’ concept. Nevertheless, it uses some assumptions and restrictions, so that the originally
3D element is transformed into 2D space only. It saves computational time and it also avoids
some formulation difficulties pertaining to 3D elements.
The element’s in-plane integration is carried out in usual way by Gauss integration scheme,
whilst in the 3rd dimension (i.e. perpendicular to mid surface of element) the integration can be
done in closed (analytical) form. However, in order to enable accounting for nonlinearity of
constitutive equations, the so called layer concept is used instead. Hence, in the 3rd dimension
simple quadrilateral integration is employed.
The present degenerate continuum element was originally proposed by Ahmad et al. (Ahmad,
Irons et al. 1970). Following general shell element theory concept, every node of element has
five degree of freedom, e.g. three displacements and two rotations in planes normal to mid-
surface of element. In order to facilitate a simple connection of this element with other true 3D
elements, the (original) five degrees of freedom are transformed into x,y,z displacement of a top
node and x,y displacement of a bottom node degrees of freedom. The two nodes are located on
the normal to mid-surface passing thru the original mid-surface element’s node, see Fig. 3-25.
The essential point in the element’s derivation is that displacements and rotations fields are
approximated "independently", (see e.g. (Jendele 1981), where similar approach is used for
plates). This means that they are handled separately. Unlike in true Mindlin theory our
formulation matches geometric equations automatically. However, a special technique is used to
improve the element’s shear behaviour (Hinton and Owen 1984).
The first formulation of this element proposed by Ahmad was linear but since that time many
improvements have been achieved. The most important is the application of reduced or selective
 129

integration scheme that reduces or totally removes locking of the element. Also, many authors
extended the original formulation to geometrically and later also materially nonlinear analysis.
One such an advanced form of the element is the formulation implemented in ATENA.
On input, the Ahmad element uses the same geometry as 20 nodes isoparametric brick element,
i.e. CCIsoBrick<xxxxxxxxxxxxxxxxxxxx>, see Fig. 3-24. This is needed, in order to be able to
use the same pre- and postprocessors’ support for the shell and native 3D brick (i.e. hexahedron)
elements. After the 1st step of the analysis, the input geometry will automatically change to the
external geometry from Fig. 3-24. As nodes 17 and 18 contain only so called bubble function,
the element is post-processed in the same way is it would be the element
CCIsoBrick<xxxxxxxxxxxxxxxx>. Internally, all element’s vectors and matrices are derived
based on the internal geometry as depicted also in Fig. 3-24.
With shell elements, the best connection at edges is to cut both at 45 degrees, or a different
corresponding angle if the thicknesses are not the same, or if connected at other than right angle,
see Fig. 3-21 (a). Another option is to use a volume brick element at the corner, which is the only
feasible way when more than two shells are connected, see Fig. 3-21 (b). The nodes on the
surface connected to the volume element have to be listed in the INTERFACE subcommand in
the shell geometry definition for correct behavior. Connecting like in Fig. 3-22 is not
recommended, as the master-slave relations induced by the fixed thickness of the shell may
cause numerical problems.

Shell1

Brick Shell2

Shell3
(a) (b)

Fig. 3-21: Shell - recommended connection (a) 2 shells (b) 3 shells

Fig. 3-22: Shell - not recommended connection

130

3.12.1 Coordinate systems.

The essential point in the element’s derivation is to understand coordinate systems that are used
within the derivation. These are as follows. Note that all vectors indicating coordinate systems’
axes are normalized. Thus, any directional cosines are simply computed as scalar products that
need not be divided by the vectors’ norm.
Global coordinate system.
It is used to define the whole FE model. Global coordinates are denoted by t x1 , t x2 , t x3 , where
the index t referrers to time. Note that we are using Modified Lagrangian formulation, in which
model configuration is updated after each time step, while within one step (for iterating) the
configuration from the step beginning is employed. Thus, 0 x1 , 0 x2 , 0 x3 are a point global
coordinates prior any load has been applied.
Nodal coordinate system
This coordinate system is defined at each point of element mid-plane surface, i.e. mid-nodes 1-9.
t t t
At a node k it is specified by vectors V1k , V2k , V3k , see Fig. 3-23.
t

s
t r
X 3L t
V3k

t
t
X 3G
L
X2

t
node k V2k
t
X 1L
t
V1k
t
X 2G
t
X 1G

Fig. 3-23 Ahmad element coordinate systems

t t
The vectors V1k , V2k , V3k are defined as follows: Firstly, two auxiliary vectors V1 , V3 are
t t t

t
calculated. Vector V3 at a point is defined as a line joining bottom and top coordinates at the
t
node k (prior any deformations, i.e. at reference configuration). The second vector V1 is normal
t
to V3 and is parallel to plane of global X 1G and
0 0
X 3G . Hence:

V3 = ⎡ V3 1 , V3 2 , V3 3 ⎤

t t t t

⎣ ⎦
(3.74)
V1 = ⎡ V1 1 , V1 2 , V1 3 ⎤ = ⎡ V3 , 0, − V3 ⎤
t t t t t t

⎣ ⎦ ⎣ 3 1⎦
 131

t t
If V3 is parallel to X 2G (i.e. V3 1 = V3 3 = 0 ), V1 is defined by
0 t t

V1 = ⎡ − V3 2 , 0, 0 ⎤
t t
(3.75)
⎣ ⎦
t t t t
After that, the coordinate system V1k , V2k , V3k itself is defined. The vector V3k is constructed in
t
the same way as was the vector V3 , however, current, i.e. deformed configuration is used. The
remaining two vectors are defined as vector product:
t
V2k = V3k ⊗ V1
t t
(3.76)
t t t
V1k = V2k ⊗t V3k (3.77)
t t t
The vectors V1k , V2k , V3k define local nodal shell coordinate system in which the shell rotations
are specified. As already mention, the original formulation of the element has 5 DOFs per nodes.
These are 3 displacements, expressed in the global coordinate systems and two rotations α , β .
t t t
They are rotations along the vectors V1k , V2k . It comes from definition that V3 need not be
normal to the element surface. It must only connect the top and bottom nodes of the shell.
t
Sometimes, it is advantageous to modify the nodal coordinate system so that V3k remains
t t t
unchanged but V1k and V2k are rotated (along V3k ) to a certain direction. Note however, that
t t t
mutual orthogonality of V1k , V2k , V3k must not be damaged.

Local coordinate system

t
Local coordinates are denoted by x1L , t x2L , t x3L . The system refers to coordinate axes
t t t
X 1L , X 2L , X 3L . It is used mainly at sampling (integration) points to calculate strains and
t t t
stresses. The vector axes X 1L , X 2L , X 3L are defined by:

⎡ ∂ t x1 ⎤ ⎡ ∂ t x1 ⎤
⎢ ⎥ ⎢ ⎥
⎢ ∂t r ⎥ ⎢ ∂t s ⎥
t ⎢∂ x ⎥ ⎢∂ x ⎥
X 3L = ⎢ 2 ⎥ ⊗ ⎢ 2 ⎥ (3.78)
∂r ∂s
⎢ t ⎥ ⎢ t ⎥
⎢ ∂ x3 ⎥ ⎢ ∂ x3 ⎥
⎢ ∂r ⎥ ⎢ ∂s ⎥
⎣ ⎦ ⎣ ⎦
t t t
X 2L = X 3L ⊗ V1k
t t t
X 1L = X 2L ⊗ X 3L (3.79)

t t t t
As the nodal coordinate system V1k , V2k , V3k can rotate along V3k , the local coordinate system
t t t t
would X 1L , X 2L , X 3L rotate simultaneously along X 3L . This definition allows for user defined
132

shell local coordinate system that is common for all shell elements, irrespective of their
t t
incidences. Note that unlike V3k the vector X 3L is always normal to the element mid-plane
surface.

Curvilinear coordinate system

This system is used to calculate derivatives and integration in element integration points. Its
coordinates are r , s for in-plane direction and t in direction of element thickness, see Fig. 3-23.
The in-plane displacements are approximated by Lagrange, Hetherosis or Serendipity
approximation similar 2D isoparametric elements. For the 3rd direction, i.e. through the depth of
the element. linear approximation is used within the frame of the shell layer concept.
 133

t3 Input geometry
11 10
4 2
12 9
1
20 19
7 18
15 s

8 17 14
6
r 16
13
5

t3
11 10 External geometry
4 2
12 9
117
7
15 s

8 14
18 6
r 16
13
5

t
Internal geometry
1
2 8
3 9 7
s
4 6
5
r

Fig. 3-24 Geometry and the element’s nodes

134

Fig. 3-25 Degenerate shell Ahmad element – coordinate systems and degree of freedoms
(DOFs)
 135

3.12.2 Geometry approximation

The coordinates of the top and bottom element surface is used to define the element geometry:
⎡ t x1 ⎤ ⎛ ⎡ t x1k ,top ⎤ ⎡ t x1k ,bot ⎤ ⎞
⎢t ⎥ t N
⎜ 1 + t ⎢ t k ,top ⎥ 1 − t ⎢ t k ,bot ⎥ ⎟
⎢ x2 ⎥ = x = ∑ hk ⎜ 2 ⎢ x2 ⎥ + 2 ⎢ x2 ⎥ ⎟ (3.80)
⎢ t x3 ⎥ k =1 ⎜ ⎢ t x3k ,top ⎥ ⎢ t x3k ,bot ⎥ ⎟
⎣ ⎦ ⎝ ⎣ ⎦ ⎣ ⎦⎠
where N=8 is number of nodes per element, (geometry is always interpolated by 8-nodes
Serendipity interpolation, irrespective of displacement interpolation), h(r,s) is k-th interpolation
⎡ t x1k ,top ⎤ ⎡ t x1k ,bot ⎤
⎢ ⎥ ⎢ ⎥
function, r,s,t are isoparametric coordinates (see Fig. 3-24), ⎢ t x2k ,top ⎥ and ⎢ t x2k ,bot ⎥ are vector of
⎢ t x3k ,top ⎥ ⎢ t x3k ,bot ⎥
⎣ ⎦ ⎣ ⎦
top and bottom coordinates of point k, see Fig. 3-26.

t
X k ,top = [ t x1k ,top , t x2k ,top , t x3k ,top ]

node k t
X k ,mid = [t x1k , mid , t x2k ,mid , t x3k ,mid ]

t
X k ,bot = [t x1k ,bot , t x2k ,bot , t x3k ,bot ]

Fig. 3-26 Approximation of the element geometry

Using the above the equation (3.80) can be rewritten in the following form:
⎛ t k ,mid ⎡ t V3k ⎤ ⎞
⎡ t x1 ⎤ ⎜ ⎡ x1 ⎤ ⎢ 1⎥ ⎟
⎢t ⎥ t N
⎜ ⎢ t k ,mid ⎥ t ⎢ t k ⎥ ⎟
⎢ x2 ⎥ = x = ∑ hk ⎜ ⎢ x2 ⎥ + 2 ⎢ V3 2 ⎥ [thick ]k ⎟ (3.81)
⎢ t x3 ⎥ k =1
⎜ ⎢⎣ t x3k ,mid ⎥⎦ ⎢t k ⎥ ⎟
⎣ ⎦ ⎜ V3 3 ⎥ ⎟
⎝ ⎢
⎣ ⎦ ⎠
where [thick ]k is element thickness in node k (i.e. distance between top and bottom points) and
136

⎡ t x1mid ⎤ ⎡ ⎡ t x1top ⎤ ⎡ t x1bot ⎤ ⎤

⎢ t mid ⎥ 1 ⎢ ⎢ t top ⎥ ⎢ t bot ⎥ ⎥
⎢ x2 ⎥ = 2 ⎢ ⎢ x2 ⎥ + ⎢ x2 ⎥ ⎥ (3.82)
⎢ t x3mid ⎥ ⎢ ⎢ t xtop ⎥ ⎢ t x bot ⎥ ⎥
⎣ ⎦k ⎣⎣ 3 ⎦k ⎣ 3 ⎦k ⎦
are coordinates of mid surface.

3.12.3 Displacement field approximation.

The general concept of displacement approximation is very similar, (although not identical) to
geometry approximation. As already mentioned the original version of Ahmad element uses 5
T
degrees of freedom per node, see Fig. 3-25. These are ⎡⎣ t u1mid ,t u2mid ,t u3mid ,t α ,t β ⎤⎦ , where
u1 , u2 , u3 are displacements of the element’s node at the mid-surface and t α ,t β are
t mid t mid t mid

rotations with respect to vectors v1k and v 2k respectively. These degree of freedoms (DOFs) are
used throughout the whole element’s development. However, in order to improve compatibility
of the present shell element with other 3D elements implemented in ATENA, externally the
T
element uses ⎡⎣ t u1top ,t u2top ,t u3top ,t u1bot ,t u2bot ⎤⎦ DOFs, i.e. displacements at the top and bottom of the
element. The 6th displacement, i.e. u3bot is eliminated due to application of shell theory that
assumes σ 33 = 0 .
Approximation of the original three "displacement" and two rotation degrees of freedom is
independent. Nevertheless, the curvatures used in governing element equations use all of them in
the sense dictated by geometric equations. This approach enables to satisfy not only equilibrium
equations for membrane stresses and in-plane shear (in mid-surface) as it is the case of popular
Kirchhoff hypothesis, but also to satisfy equilibrium condition for transversal shears (normal to
mid-surface).
Note that in the following derivation of the element we will deal with the original set of
element’s DOFs , see (10). Every point thus has five degree of freedom,
T
⎡⎣ t u1mid ,t u2mid ,t u3mid ,t α ,t β ⎤⎦ . Displacement vector is calculated by:

⎛ t k , mid ⎡ − t V2k t
V1k 1 ⎤ ⎞
⎡ t u1k ⎤ ⎜ ⎡ u1 ⎤ ⎢ 1 ⎥ t k ⎟
⎢t k ⎥ t ⎜ ⎢ t k , mid ⎥ t t k ⎥⎡ α ⎤⎟
N
⎢− V k
⎢ u2 ⎥ = u = ∑ hk ⎜ ⎢ u2 ⎥ + 2 [thick ]k
t
V1 2 ⎢ t k ⎥ ⎟ (3.83)
⎢ 22 ⎥ β
⎢ t u3k ⎥ k =1
⎜ ⎢⎣ t u3k , mid ⎥⎦ ⎢ t k t k ⎥
⎣ ⎦⎟
⎣ ⎦ ⎜
⎝ ⎢⎣ − V2 3 V1 3 ⎥
⎦
⎟
⎠
T
The original displacement vector at point k has the form ⎡⎣ t u1mid ,t u2mid ,t u3mid ,t α ,t β ⎤⎦ . Unlike in
the case of geometry approximation, were N=8, displacements approximation accounts also for
displacement in the element mid-point, i.e. N=9. The ninth function h is so called bubble
function.
 137

u k ,mid = [ t u1k ,mid , t u2k , mid , t u3k , mid ]

t
V3k

node k

αk
t t
V2k
t
β k

t
V1k

Fig. 3-27 Displacement approximation

3.12.4 Strain and stresses definition.

The 2nd Piolla Kirchhoff tensor and Green Lagrange strain tensor is used. They are calculated
and printed in the local coordinate system t x '1 , t x ' 2 and t x '3 .
Green - Lagrange tensor.
The general definition for Green-Lagrange strain tensor has the form (see eq. (1.8)):
1 t
t
0 ijε =
2
( 0 ui , j + 0 u j ,i + 0 uk ,i 0 uk , j )
t t t
(3.84)

Using the above equation and applying the Von-Karman assumption, Eqn. (3.84) can be written
as:
⎡ ∂ t u1 ⎤
⎢ ⎥ ⎡ 2
⎤
⎢ ∂ x1
t
⎥ ⎢ 1 ⎛⎜ ∂ u3 ⎞⎟ ⎥
t

⎢ ∂ t u2 ⎥ ⎢ 2 ⎝ ∂ t x1 ⎠ ⎥
⎡ 0 ε11 ⎤ ⎢
t
⎥ ⎢ 2 ⎥
⎥ ⎢ ∂ x2
t
⎢ t ⎥ ⎢ 1 ⎛ ∂ t u3 ⎞ ⎥
⎢ 0 ε 22 ⎥ ⎢ ∂ t u ∂ t u ⎥ ⎢ ⎜ t ⎟ ⎥
⎢ 2 0tε12 ⎥ = ⎢ t 1 + t 2 ⎥ + ⎢ 2 ⎝ ∂ x2 ⎠ ⎥ = 0t ε L + 0t ε NL (3.85)
⎢ t ⎥ ⎢ ∂ x2 ∂ x1 ⎥ ⎢ ∂ t u ∂ t u ⎥
⎢ 2 0 ε13 ⎥ ⎢ t ⎥ ⎢ t 3 + t 3⎥
⎢ 2 tε ⎥ ⎢ ∂ u1 + ∂ u3 ⎥ ⎢ ∂ x2 ∂ x1 ⎥
t

⎣ 0 23 ⎦ ⎢ ∂ t x ∂ t x ⎥
3 1 ⎢ 0 ⎥
⎢ t ⎥ ⎢ ⎥
⎢ ∂ u2 + ∂ u3 ⎥ ⎣⎢
t
0 ⎥
⎦
⎢ ∂t x ∂t x ⎥
⎣ 3 2⎦

The Von-Karman assumptions simplify the calculation of strain by accepting that:

• All strains are relatively small,
• The deflection normal to mid surface of shell is of order of thickness,
• The both curvatures are much smaller than 1.,
• The in-plane displacements are much smaller than transverse displacement and thus their
derivatives in 2nd order terms can be neglected.
138

ε
t
0 L and 0t ε NL represents linear and nonlinear part of strain vector, respectively. More
information about their calculation is beyond the scope of this publication. It is available e.g. in
(Jendele 1992).
2nd Piolla Kirchhoff tensor.
Energetically conjugated with Green - Lagrange tensor is 2nd Piolla Kirchhoff tensor and this
tensor is used by the present shell element. Remind that we account for all stresses with
exclusion of normal stress which is perpendicular to shell mid surface (as it is usual practice in
shell analysis). This is the reason, why we introduced local coordinate system and all expression
are derived with respect to it.
Obviously the local coordinate system varies depending on element deformation and thus it is
necessary to re-compute (each iteration) the transformation matrix T (that relates local and
global coordinate systems).
To compute internal forces we will use 2nd Piolla Kirchhoff tensor in vector form (in a node k):
⎡⎣ 0t S ⎤⎦ = ⎡⎣ 0t S11 0
t
S 22 t
S
0 12
t
S
0 13 0
t
S 23 ⎤⎦ (3.86)
k k

Note that that it is possible to abbreviate full 3 by 3 element tensor to the above vector, because
of adopting Von Karmann simplifying assumption.

3.12.5 Serendipity, Lagrangian and Heterosis variant of degenerated shell

element.
Until now no information about interpolation function h and number of integration points were
given. The present shell element analysis uses Serendipity interpolation functions. Note that
bubble function h9 (used in displacement approximation only) represents relative departure of
approximated function with respect to the function value calculated by previous eight
approximation functions.
 139

The interpolation functions hi read:

1
h1 (r , s ) = (1- r )(1- s )(-r - s -1)
4
1
h2 (r , s ) = (1- s )(1- r 2 )
2
1
h3 (r , s ) = (1 + r )(1- s)(r - s -1)
4
1
h4 (r , s ) = (1 + r )(1- s 2 )
2
1
h5 (r , s ) = (1 + r )(1 + s )(r + s -1)
4
1
h6 (r , s ) = (1 + s )(1- r 2 )
2
1
h7 (r , s ) = (1 + r )(1- s )(r - s -1)
4
1
h8 (r , s ) = (1- r )(1- s 2 )
2
1
h9 (r , s ) = (1- r 2 )(1- s 2 ) (3.87)
2
The actual values in center point can be calculated by:
8
a9 = ∑ ai hi (r = 0, s = 0) + Δa9 (3.88)
i =1

where hi are values of interpolation function at point (0,0), ai are corresponding node values,
Δa9 is departure in the center (i.e. computed value corresponding to degree of freedom at center)
and a9 is total value in center.
Depending on combination how many nodes and integration points are used, we distinguish the
Serendipity, Lagrange and Heterosis degenerated element variants, see Fig. 3-28.
Serendipity element.
This element was used in the original Ahmad work. It comprises eight nodal points (center point
corresponding to bubble function is omitted).
Gauss integration scheme is used for integration . It can be integrated by full, reduced or
selective integration procedure. Using full integration ,i.e. at three by three sample points,
element exhibits shear locking for thin and even moderately thick element. If reduced integration
is used. the problem of locking is significantly improved without creating spurious energy modes
on structure level. However, in case of thin element there are two non communicable spurious
energy mode on element level.
It should be noted that there were reported some difficulties, if some unfavorable constraints are
applied. Nevertheless the element is popular. If reduced integration is used the provided results
are relatively good.
140

Fig. 3-28 Node notation for element variants of the Ahmad shell element

Nine node Lagrangian element.

The nine point Lagrangian element is still considered to be the best variant of the degenerated
element. This is especially because of its versatility. For full integration scheme there is no
problem with membrane and shear locking for very thin plate and shell application. If element is
moderately thick, shear locking can be improved by reduced integration scheme. However, in
that case the element exhibits rank deficiency.
Heterosis element.
The Heterosis element is very similar to Lagrangian element. The difference is that it assumes
first three DOFs at the element centre to be constrained, (i.e. only the rotations are retained)
The element exhibits better behavior compared with previous quadratic elements and especially
in combination with selective integration scheme no locking is produced. With reduced
integration the element provides results better than Lagrangian element. In that case there are
some spurious mechanisms, but for practical solution there are not probable.
 141

Fig. 3-29 Summary of locking and spurious energy modes

Problem of membrane and shear locking for linear analysis are summarized in Fig. 3-29. In the
case of nonlinearity, the situation is much more complicated and depends primarily on the
material state at the sampling points. For more information refer to (Jendele, Chan et al. 1992)
Element’s integration
In previous paragraphs we mentioned many time full, reduced and selective integration scheme.
The sense of these procedures is best to demonstrate in Fig. 3-30.
142

Fig. 3-30 Integration schemes and sampling point notation

The steps during selective integration of shear can been explained on example of integration
arbitrary function f (r , s ) :
1/ First we calculate the value of f at sampling points that corresponds to two by two integration
rule, i.e
f (-0.5773, -0.5773), f (-0.5773, 0.5773), f (0.5773, -0.5773), f (0.5773, 0.5773)
2/ Using bilinear approximation we calculate the values of f at points that correspond to three by
three integration rule. There are two possibility to that.
The first one is based on original approximation area and the main idea is that we calculate the
value of function f at "corners" of isoparametric element (i.e. r = ±1., s = ±1. ):
 143

4
f (−0.5773, −0.5773) = ∑ fi hi' (−0.5773, −0.5773)
k =1
4
f (−0.5773, 0.5773) = ∑ f i hi' (−0.5773, 0.5773)
k =1
4
(3.89)
f (0.5773, −0.5773) = ∑ f h (0.5773, −0.5773)
i i
'

k =1
4
f (0.5773, 0.5773) = ∑ fi hi' (0.5773, 0.5773)
k =1

where fi are element nodal values of function f and hi' are interpolation functions corresponding
to two by two interpolation and a node i.

Fig. 3-31 Extension of bilinear approximation function for arbitrary rectangular

The set of equation (3.89) can be solved for fi . Having these value we can bi-linearly
approximate function f and compute function value at any point, i.e. also at sampling points
corresponding to three by three integration rule.
The second and more elegant solution is direct approximation. The interpolation function hi are
presented for an square area of the size 2x2 units, but they can be extended to a rectangular of
any size, as shown in Fig. 3-31.
Since the functional values for the 2x2 sampling points in the corner of the square
lr = ls = 2 x0.5775 are available, the approximation functions hi' can be used directly to calculate
the values of the function f at sampling points corresponding to the 3x3 integration rule.
For integration in direction perpendicular to r - s plane, that is in t coordinate it is also possible to
use Gauss integration, but due to material nonlinearity there is more advantages to use direct
rectangular integration. This concept is called the Layered model, see Fig. 3-32.
The main idea of it is to divide the element along the thickness to layers whereby in particular
layer the values of strain and stresses are expected to be constant and equal to their value at
weight point of layer. Hence the integration in t direction is computed as a sum of integrated
expressions multiplicated by adequate area of layer for all layers from bottom to top of element.
It was found that to achieve good accuracy it is necessary to about six to ten layers.
This concept. i.e. layer model is advantageous because it enables us to create for example
reinforcing layers in element and also we can use finer division near top and bottom of shell,
where higher stress level can be expected.
144

Fig. 3-32 The layer model

3.12.6 Smeared Reinforcement

The ATENA implementation of the Ahmad shell element supports embedding of smeared
reinforcement layers. In this concept, reinforcement bars with the same coordinate z, (see Fig.
3-32), material and the same directions are replaced by a layer of smeared reinforcement. Such a
layer is placed at the same elevation z as the original reinforcement bars and its thickness is
calculated so that sum of cross sectional area of the bars and the replacing smeared
reinforcement layer is the same. The layer is usually superimposed over existing concrete layers
and it employs CCSmeardReinforcement material law, which makes possible to account for the
original reinforcement bars’ direction.
Because of the fact that each layer of the Ahmad shell can use a distinct material model, concrete
and smeared reinforcement layers are treated in similar way. (Constitutive equations, i.e.
material law are placed outside of ATENA finite elements’ code). Description of syntax of
related input commands is beyond scope of this document, but it can be found in the “ATENA
Input File Format” document.
Note that the support for smeared reinforcement does not exclude use of structural discrete
reinforcement. Both the type of reinforcement can be combined in one model to achieve the best
likeness of the the real structure with its numerical model.

3.12.7 Transformation of the original DOFs to displacements at the top and

bottom of the element nodal coordinate system

This section describes in detail the whole procedure of transforming Ahmad elements from its
original formulation to the new one used by ATENA SW. Just to remind you: The original
formulation (described in the previous sections) differs from the new one in selection of element
degree of freedom, see Section 3.12.3.
Let us start to work in nodal coordinate system first. The following equation states
transformation rules for transforming three global displacements and two nodal rotations at the
element mid-plane, (i.e. the original DOFs at a node k), to 6 displacements at nodal coordinate
 145

system, three at the top and three at the bottom surface of the shell. Note the right superscripts
“N” that indicate nodal coordinate system.
⎡t k t t 1 ⎤
⎢ V1 1 V1k 2 V1k 3 thick 0 ⎥
2
⎢ ⎥
⎡ t u1k , N ,top ⎤ ⎢ t V k 1
− thick ⎥ ⎡ t u k , mid ⎤
t t
V2k 2 V2k 3 0 ⎡ t u1k , mid ⎤
⎢ t k , N ,top ⎥ ⎢ 2 1 2 ⎥ 1
⎢ u2 ⎥ ⎢t k ⎥ ⎢ t u k , mid ⎥ ⎢ t k , mid ⎥
⎢ u2 ⎥
t t
⎢ t u3k , N ,top ⎥ ⎢ V3 1 V3k 2 V3k 3 0 0 ⎥ ⎢ t 2k , mid ⎥
⎢ t k , N ,bot ⎥ = ⎢ ⎥ ⎢ u3 ⎥ = T1 ⎢ t u3k , mid ⎥ (3.90)
⎢ u1 ⎥ ⎢ tV k t
V k t
V k 1
− thick 0 ⎥⎢ t k ⎥ ⎢ t k ⎥
⎢ u ⎥ ⎢ 11 1 2 1 3
⎥ ⎢ α ⎥ ⎢ α ⎥
t k , N , bot
2
⎢ t k , N ,bot ⎥ ⎢ t
2 ⎢
⎥⎣ β ⎦t k ⎥ ⎢ tβk ⎥
1 ⎣ ⎦
⎣⎢ u3 ⎦⎥ ⎢ V2k 1
t t
V2k 2 V2k 3 0 thick ⎥
⎢ 2 ⎥
⎢t k t t ⎥
⎢⎣ V3 1 V3k 2 V3k 3 0 0 ⎥⎦

Transformation from nodal to global coordinated system

The next step in the element’s derivation is to write transformation of the left-hand side vector of
(3.90) from nodal to global coordinate system. It reads:
⎡ tV k t
V2k 1
t
V3k 1 0 0 0 ⎤
⎢ 11 ⎥ t k , N ,top
⎡ u1 ⎤ ⎢ t k
t k ,top
⎡ u ⎤ ⎡ t u1k , N ,top ⎤
0 ⎥ ⎢ t 1k , N ,top ⎥
t t
⎢ t k ,top ⎥ ⎢ V1 2 V2k 2 V3k 2 0 0 ⎢ t k , N ,top ⎥
⎥ u
⎢ u2 ⎥ ⎢ t k t ⎥ ⎢ t 2k , N ,top ⎥ ⎢ u2 ⎥
⎢ t u3k ,top ⎥ ⎢ V1 3 V2k 3 V3k 3 0 0 0 ⎥⎢ u ⎥ ⎢ t k , N ,top ⎥
u
⎢ t k ,bot ⎥ = ⎢ ⎥ = T2 ⎢ t k , N ,bot ⎥
3 3
⎢
t k ⎥ t k , N ,bot
(3.91)
⎢ u1 ⎥ ⎢ 0 V3 1 ⎥ ⎢ u1 ⎢ u1
t t
0 0 V1k 1 V2k 1 ⎥ ⎥
⎢ t u k ,bot ⎥ ⎢ ⎢ t k , N ,bot ⎥ ⎢ t k , N ,bot ⎥
t k ⎥ u2 u
⎢ t 2k ,bot ⎥ ⎢ 0 0 0
t
V1k 2
t
V2k 2 V3 ⎥ ⎢ t k , N ,bot ⎥ ⎢ t 2k , N ,bot ⎥
⎣⎢ u3 ⎦⎥ ⎢ 2 ⎢ u
⎥⎣ 3 ⎦⎥ ⎣⎢ u3 ⎦⎥
t t k
⎢ 0 0 0 V1k 3 V2k 3 V3 3 ⎥
⎣ ⎦

Complete transformation of the original DOFs to the new element formulation DOFs

The final transformation from the original to the new element DOFs at a node k is obtain by
substituting (3.90) into (3.91). Thus we can write

⎡ t u1k ,top ⎤
⎢ t k ,top ⎥ ⎡ t u1k , mid ⎤ ⎡ t u1k , mid ⎤
⎢ u2 ⎥ ⎢ t k , mid ⎥
u2 ⎥
⎢ t k , mid ⎥
⎢ u3 ⎥
t k , top ⎢ ⎢ u2 ⎥
⎢ t k ,bot ⎥ = T2 T1 ⎢ u3 ⎥ = T ⎢ t u3k , mid ⎥
t k , mid
(3.92)
⎢ u1 ⎥ ⎢ t k ⎥ ⎢ t k ⎥
⎢ u ⎥
t k , bot ⎢ α ⎥ ⎢ α ⎥
⎢ t k ,bot ⎥
2 ⎢ β ⎥
t k ⎢ tβk ⎥
⎣ ⎦ ⎣ ⎦
⎢⎣ u3 ⎥⎦
146

where T
⎡
( ) ( ) ( ) thick k t k thick k t k ⎤
t k 2 t 2 t 2 t t t t t t t t t t t t
⎢ V1 1 + V2k 1 + V3k 1 V1k 1 V1k 2 + V2k 1 V2k 2 + V3k 1 V3k 3 V1k 1 V1k 3 + V2k 1 V2k 3 + V3k 1 V3k 3 V1 1 − V2 1 ⎥
⎢ 2 2 ⎥
⎢t k t k
( V ) +( V ) +( V ) thick t k ⎥
k k
t k t k t k t k t 2 t 2 t 2 t t t t t t thick t k
⎢ V1 1 V1 2 + V2 1 V2 2 + V3 1 V3 3 V1k 2 V1k 3 + V2k 2 V2k 3 + V3k 2 V3k 3 −
k k k
1 2 2 2 3 2 V1 2 V2 2 ⎥
⎢ 2 2 ⎥
⎢t k t k
( V ) +( V ) +( V ) thick t k ⎥
k k
t k t k t k t k t t t t t t t 2 t 2 t 2
thick t k
⎢ V1 1 V1 3 + V2 1 V2 3 + V3 1 V3 3 V1k 2 V1k 3 + V2k 2 V2k 3 + V3k 2 V3k 3 −
k k k
1 3 2 3 3 3 V1 3 V2 3 ⎥
⎢ 2 2 ⎥
⎢
( ) ( ) ( ) thick k t k thick k t k ⎥
t k 2 t 2 t 2 t t t t t t t t t t t t
⎢ V1 1 + V2k 1 + V3k 1 V1k 1 V1k 2 + V2k 1 V2k 2 + V3k 1 V3k 3 V1k 1 V1k 3 + V2k 1 V2k 3 + V3k 1 V3k 3 − V1 1 V2 1 ⎥
2 2
⎢ ⎥
( V ) +( V ) +( V )
k
⎢t k t k t k t k t k t k t 2 t 2 t 2 t t t t t t thick t k thick k t k ⎥
⎢ V1 1 V1 2 + V2 1 V2 2 + V3 1 V3 3 V1k 2 V1k 3 + V2k 2 V2k 3 + V3k 2 V3k 3 −
k k k
1 2 2 2 3 2 V1 2 V2 2 ⎥
2 2
⎢ ⎥
⎢t k t k
( V ) +( V ) +( V ) thick k t k thick k t k ⎥
t t t t t t t t t t t 2 t 2 t 2
V1 1 V1 3 + V2k 1 V2k 3 + V3k 1 V3k 3 V1k 2 V1k 3 + V2k 2 V2k 3 + V3k 2 V3k 3 k k k
− V1 3 V2 3 ⎥
⎣⎢
1 3 2 3 3 3
2 2 ⎦

In a very similar way we can define inverse transformation, i.e. from the new DOFs to to original
one. Without any derivation the matrix reads:
⎡ t u1k ,top ⎤
⎡ t k , mid
1 u ⎤ ⎢ t k ,top ⎥
⎢ t k , mid
u
⎥ ⎢ u2 ⎥
⎢ 2 ⎥ ⎢ t u k ,top ⎥
⎢ t k , mid
3 u ⎥ = T' ⎢ t 3k ,bot ⎥ (3.93)
⎢ ⎥ ⎢ u1 ⎥
⎢
t k
α ⎥ ⎢ t u k ,bot ⎥
⎢ t
βk ⎥ ⎢ t 2k ,bot ⎥
⎣ ⎦
⎢⎣ u3 ⎥⎦

⎡ T11 T12 T13 T11 T12 T13⎤

⎢ 2 2 2 2 2 2 ⎥
⎢ ⎥
⎢ T21 T22 T23 T21 T22 T23⎥
⎢ 2 2 2 2 2 2 ⎥
⎢ ⎥
⎢ T31 T32 T33 T31 T32 T33⎥
where T' = ⎢ 2 2 2 2 2 2 ⎥
⎢ t k t t t t t k ⎥
⎢ V1 1 V1k 2 V1k 3 − V1k 1 − V1k 2 − V1 3 ⎥
⎢ thick k thick k thick k thick k thick k thick k ⎥
⎢ t t t t t t k
⎥
⎢ − V2k − V2k 2 − V2k 3 V2k 1 V2k 2 V2 3 ⎥
⎢ 1
⎥
⎣ thick k thick k thick k thick k thick k thick k ⎦

Constraining the redundant DOF to comply with shell theory

As noted earlier, the original set of DOFs at a node comprises 5 DOFs, whilst the new one has
T
six DOFs. Consequently, one DOF from ⎡⎣ t u1k ,top ,t u2k ,top ,t u3k ,top ,t u1k ,bot ,t u2k ,bot ,t u3k ,bot ⎤⎦ must be
fixed. The presented work prefers to constrain t u3k ,bot but u1k ,bot or u2k ,bot are also good
candidates, if t u3k ,bot can not be fixed due some numerical problems, usually due to a special
position of the element with respect to global coordinate system.
Derivation of the constrain is now demonstrated on the case of t u3k ,bot . Using (3.92)
 147

t
u3k ,bot =t u3k ,top + (t u3k ,bot −t u3k ,top ) =
t
(
u3k ,top + (T16 − T13 ) u1k ,mid + (T26 − T23 ) u2k ,mid + .... (T56 − T53 ) β k =
t t t
) (3.94)
t t t t
t
u3k ,top − thick k V1k 3 α k + thick k V2k 3 β k

Now in (3.94) eliminate α k and t β k using (3.93). Thus we obtain one equation relating
T
⎡⎣ t u1k ,top ,t u2k ,top ,t u3k ,top ,t u1k ,bot ,t u2k ,bot ,t u3k ,bot ⎤⎦ , which is then used to constrain t u3k ,bot as a linear
combination of t u1k ,top ,t u2k ,top ,t u3k ,top ,t u1k ,bot ,t u2k ,bot :

u3k ,bot = c1top t u1k ,top + c2top t u2k ,top + c3top t u3k ,top + c1bot t u1k ,bot + c2bot t u2k ,bot (3.95)
where:

t t t t
V1k 3 V1k 1 + V2k 3 V2k 1
c top
=
( V ) + ( V ) −1
1 2 2
t k t k
1 3 2 3

t t t t
V1k 3 V1k 2 + V2k 3 V2k 2
c top
=
( V ) + ( V ) −1
2 2 2
t k t k
1 3 2 3

1 − ( V ) − ( V ) yy
t 2 t 2
k k 2
1 3 2 3 z
c top
=−
( V ) + ( V ) −1
3 2 2
t k t k
1 3 2 3

t t t t
V1k 3 V1k 1 + V2k 2 V2k 1
c bot
=−
( ) ( ) −1
1 2 2
t t
V1k 3 + V2k 3
t t t t
V1k 3 V1k 2 + V2k 3 V2k 2
c bot
=−
( ) ( ) −1
2 2 2
t t
V1k 3 + V2k 3
(3.96)
The DOFs u1k ,bot or u2k ,bot can be eliminated in the same way. During the execution of the
element, it is recommended to constrain one of t u1k ,bot ,t u2k ,bot ,t u3k ,bot based on which solution is
the most stable, (i.e. maximum denominator in (3.96)).

Constraining DOFs at the centre of Hetherosis element

A special attention needs to be paid to the 9th mid-plane node of Hetherosis element, when we
have to additionally constrain t u1k ,mid , t u2k ,mid , t u3k ,mid . Thus, of the 6 DOFs we need to constrain 4
of them.
148

t
For example, suppose we want to keep free u2k ,top and t
u3k ,top and we need to
fix t u2k ,top ,t u1k ,bot ,t u2k ,bot ,t u3k ,bot . Equation (3.94) from the previous paragraph needs to be added by
three more equations. These are:
⎡ t u1k ,top ⎤
⎢ t k ,top ⎥
u
⎡ t u1k , mid ⎤ ⎡T11' T12' T13' T14' T15' T16 ⎤ ⎢ t 2k ,top ⎥ ⎡ 0 ⎤
'

⎢ t k , mid ⎥ ⎢ ' ⎥⎢ u ⎥
⎢ u2 ⎥ = ⎢T21 T21 T23 T24 T25
' ' ' '
T26' ⎥ ⎢ t 3k ,bot ⎥ = ⎢⎢ 0 ⎥⎥ (3.97)
u
⎢ t u3k , mid ⎥ ⎢T31' T32' T33' T34' T35'
⎣ ⎦ ⎣ T36' ⎦⎥ ⎢ t 1k ,bot ⎥ ⎣⎢ 0 ⎦⎥
⎢ u ⎥
⎢ t 2k ,bot ⎥
⎣⎢ u3 ⎦⎥
Equations (3.94) and (3.97) are then solved for t u2k ,top ,t u1k ,bot ,t u2k ,bot ,t u3k ,bot as a linear combination
of t u2k ,top and t u3k ,top .

−1
⎡ t u2k ,top ⎤ ⎡ T11' T14' T15' T16' ⎤ ⎡ T12' t u2k ,top + T13' t u3k ,top ⎤
⎢ k ,bot ⎥ ⎢ ' ⎥ ⎢ ' t k ,top ' t k ,top ⎥
⎢ u1 ⎥ = − ⎢ T21 T24' T25' T26' ⎥ ⎢ T22 u2 + T23 u3 ⎥ (3.98)
⎢ u2k ,bot ⎥ ⎢ T31' T34' T35' T35' ⎥ ⎢ T32' t u2k ,top + T33' t u3k ,top ⎥
⎢ t k ,bot ⎥ ⎢ top ⎥ ⎢ top t k ,top top t k ,top ⎥
⎣⎢ u3 ⎦⎥ ⎣⎢c1 −1 c1bot c2bot ⎦⎥ ⎣⎢ c2 u2 + c3 u3 ⎦⎥

Again, there are several alternatives regarding of which of the 6 DOFs to keep and which to
eliminate. The best option is chosen the same way as described in Section 0.

3.12.8 Shell Ahmad elements implemented in ATENA

Several modifications the Ahmad shell elements are implemented in ATENA. They are listed
in the following table:
 149

Table 3.12-1 Ahmad shell elements.

Element name Type of Number of Number of Comment

approximation in-plane in-plane
integration integration
poins per axis poins per axis
direction for direction for
bending shear
CCAhmadElement33L9 Lagrange 3 3 No spurious
modes, locking
for this shells
CCAhmadElement32L9 Lagrange 3 2
CCAhmadElement33H9 Heterosis 3 3
CCAhmadElement32H9 Heterosis 3 2 Good
compromise
between locking
and spurious
energy modes
CCAhmadElement22S8 Serendipity 2 2 Fast, but
spurious modes

3.13 Curvilinear nonlinear 3D beam element.

A curvilinear 3D beam finite element implemented in ATENA, is described here. It is called
CCBeamNL. The element is based on a similar beam element from BATHE(1982). It is fully
nonlinear, in terms of its geometry and material response. It uses quadratic approximation of its
shape, so the it can be curvilinear, twisted, with variable dimensions of the cross-sections.
Moreover, beam’s cross-sections can be of any shape, optionally even with holes.
The element belongs to the group of isoparametric elements with Gauss integration along its axis
and trapezoidal (Newton-Cotes) quadrature within the cross-section. The integration (or
material) points are placed in a way similar to the layered concept applied to shell elements,
however, the “layers” are located in both “s,t” directions.

3.13.1 Geometry and displacements and rotations fields.

Geometry of the element is depicted in Fig. 3-33. The depicted brick nodes specification is
employed to ensure compatibility of the element with ATENA preprocessor. The beam 3D nodes
definition is used by ATENA postprocessor. The element response is computed within the 1D
beam geometry. Thus, on input the element has 20 nodes, while during the calculation it has only
15 nodes, i.e. 12 nodes for 3D beam shape definition and 3 nodes for the 1D beam geometry.
Any of the 15 nodes can be subject to a kinematic or static constraint. The 1D beam nodes have
6 degrees of freedom (dofs) – three displacements and three rotations with respect to global
coordinate axes. The 3D beam nodes allocate only the three displacement dofs per node. The
redundant brick nodes are ignored and they allocate no dofs.
The element uses three configuration. The reference configuration corresponds to shape of the
beam at the beginning of the step, i.e. prior any load in the current step is applied was employed.
150

It is used as a reference coordinate system for all calculation within a loading step t, with respect
to which all derivatives are computed. This configuration is denoted by a t superscript left to a
referred symbol, e.g. t x . The element shape after all previous iterations within the current step
and prior the current iteration is denoted by t + dt superscript, t + dt x . Increments within the
current iteration do not use any superscript, e.g. x.
 151

t
a2 Geom etry
t
b2

t 1
Vt

t
t
b3 a3

6
13
t
a1 1
Vs 14 5 Bric k nodes
7 16
t
s
b1 18 15 8

19
r 17 6
20 5 Beam 3D nodes
7
10 8
2
9
1 11
10
12 9
3 11 Beam 1D nodes
4 14
12

2
1 15 t

3
4
w
θz
13 z 2

2 s
2
y r
θx
x
θy v
u Isoparam etric shape
Global c oord. system and elem ent dofs
Fig. 3-33 CCBeamNL element
152

The beam’s geometry at the configuration t and t + dt is defined by:

⎛ t s ⎞
t
x = hi ⎜ t X i + t ai tVi tx + t bi tVi sx ⎟
⎝ 2 2 ⎠
⎛ t t s s ⎞
t
y = hi ⎜ t Yi + t ai tVi y + t bi tVi y ⎟ (3.99)
⎝ 2 2 ⎠
⎛ t s ⎞
t
z = hi ⎜ t Z i + t ai tVi tz + t bi tVi sz ⎟
⎝ 2 2 ⎠
In the above i refers to axial nodes, i.e. i = 1..3 for the nodes 13,14,15, see the 1D beam nodes.
T
hi = hi (r ) is i-th nodal interpolation function i described in Section 3.2. ⎡⎣ t X i ,t Yi ,t Z i ⎤⎦ are
T T
global coordinates of a node i at time t. The vectors ⎡⎣ t Vi tx ,t Vi y ,t Vi tz ⎤⎦ , ⎡⎣ t Vi sx ,t Vi y ,t Vi sz ⎤⎦ are the
t s

vectors t Vt , tVs depicted in Fig. 3-33, in a cross section i, at time t, which define local
coordinate axis s,t. The symbols t ai , t bi refers to dimensions of the cross section i, time t; see
the figure, too.
Geometry of the beam at time t + dt is defined in a similar way:

t + dt ⎛ t s ⎞
x = hi ⎜ t + dt X i + t ai t + dtVi tx + t bi t + dtVi sx ⎟
⎝ 2 2 ⎠
t + dt ⎛ t t s s ⎞
y = hi ⎜ t + dt Yi + t ai t + dtVi y + t bi t + dtVi y ⎟ (3.100)
⎝ 2 2 ⎠
t + dt ⎛ t s ⎞
z = hi ⎜ t + dt Z i + t ai t + dtVi tz + t bi t + dtVi sz ⎟
⎝ 2 2 ⎠

The element’s displacements at time t + dt is calculated as follows:

t + dt
u = t + dt x −t x
t + dt
v = t + dt y −t y (3.101)
t + dt t + dt
w= z− z
t
 153

and displacement increments within a iteration:

⎛ t s ⎞
u = hi ⎜ U i + t ai Vi tx + t bi Vi sx ⎟
⎝ 2 2 ⎠
⎛ t t s s ⎞
v = hi ⎜ Vi + t ai Vi y + t bi Vi y ⎟ (3.102)
⎝ 2 2 ⎠
⎛ t s ⎞
w = hi ⎜ Wi + t ai Vi tz + t bi Vi sz ⎟
⎝ 2 2 ⎠

In the above equation the vectors Vi t , Vi s are Vi t =t + dt Vi t −t Vi t and Vi s =t + dt Vi s −t Vi s and they are
approximated by

Vi tx = ( t + dt
Vi tz θiy − t + dt t
Vi y θiz )
t
Vi y = ( t + dt
Vi tx θiz − Vi tz θix
t + dt
)
Vi tz = ( t + dt t
Vi y θix − Vi tx θiy
t + dt
)
(3.103)
Vi sx = ( t + dt
Vi sz θiy − t + dt
Vi θiz
sy
)
s
Vi y = ( t + dt
Vi sx θiz − t + dt
Vi sz θ ix )
Vi sz = ( t + dt
Vi y θ ix −
s t + dt
Vi sx θiy )
The parametrs θix , θiy , θiz are rotations around the global axis, with respect to beginning of the
current load step. Note that (3.103) is valid only approximately.

3.13.2 Strain and stress definition

The element uses Green-Lagrange strain and Piola-Kirchhof stresses, see Section 1.4.2 and
Section 1.3.2. transformed to the local isoparametric r,s,t coordinate system. As the beam theory
implies, only normal strain component ε r and shear components γ rs , γ rt are considered. The
stress vector includes the corresponding σ rr ,τ rs ,τ rt entries, whereby the remaining strains have
to remain zero. The procedure of calculation stress-strain response is as follows:
1. Calculate all 6 components of Green-Lagrange strains (1.8) and their increments within
global coordinate systems. The increments are computed with respect to the beginning of
the current load step.
2. Transform the strains increments into local r,s,t coordinate system.
3. Zeroise components Δε ss , Δε tt , Δγ st .
4. Execute material law to compute corresponding stresses.
5. Transform the stresses to the global coordinate system.
The following expressions are used to calculate displacement derivatives needed for calculation
of the strains:
154

−1
⎡ df ⎤ ⎡ dx dy dz ⎤ ⎡ df ⎤ ⎡ df ⎤
⎢ dx ⎥ ⎢ dr dr dr ⎥ ⎢ dr ⎥ ⎢ dr ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ df ⎥ = ⎢ dx dy dz ⎥ ⎢ df ⎥ = J −1 ⎢ df ⎥ (3.104)
⎢ dy ⎥ ⎢ ds ds ds ⎥ ⎢ ds ⎥ ⎢ ds ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ df ⎥ ⎢ dx dy dz ⎥ ⎢ df ⎥ ⎢ df ⎥
⎢⎣ dz ⎥⎦ ⎢⎣ dt dt dt ⎥⎦ ⎢⎣ dt ⎥⎦ ⎢⎣ dt ⎥⎦
where f is a displacement function to be derivated.

3.13.3 Matrices used in the beam element formulation

Substituting equations (3.99) to (3.104) into the expressions for calculating element matrices
(1.31) to (1.34) all important matrices and vectors of the beam element can be calculated. Their
explicit presentation is beyond the scope of this document. Nevertheless, the most important
ones are now given:
The Jacobian matrix:
∂ t x ∂hi ⎛ t t t t t x s t t sx ⎞
J11 = = ⎜ X i + ai Vi + bi Vi ⎟
∂r ∂r ⎝ 2 2 ⎠
∂ y ∂hi ⎛ t
t
t t t ty s t t sy ⎞
J12 = = ⎜ Yi + ai Vi + bi Vi ⎟
∂r ∂r ⎝ 2 2 ⎠
∂ t z ∂hi ⎛ t t t t t z s t t sz ⎞
J13 = = ⎜ Z i + ai Vi + bi Vi ⎟
∂r ∂r ⎝ 2 2 ⎠
∂xt
⎛1 ⎞
J 21 = = hi ⎜ t bi tVi sx ⎟
∂s ⎝2 ⎠
∂t y ⎛1 s ⎞
J 22 = = hi ⎜ t bi tVi y ⎟
∂s ⎝2 ⎠
∂t y ⎛1 ⎞
J 23 = = hi ⎜ t bi tVi sz ⎟
∂t ⎝2 ⎠
∂x
t
⎛1 ⎞
J 31 = = hi ⎜ t ai tVi tx ⎟
∂t ⎝2 ⎠
∂t y ⎛1 t ⎞
J 32 = = hi ⎜ t ai tVi y ⎟
∂t ⎝2 ⎠
(3.105)
∂t z ⎛1 ⎞
J 33 = = hi ⎜ t ai tVi tz ⎟
∂t ⎝2 ⎠
 155

t + dt
The matrix t BNL . :
It is constructed in the way that

⎡U 1 ⎤
⎡ ∂u ⎤ ⎢V ⎥
⎢ ∂x ⎥ ⎢ 1⎥
⎢ ⎥ ⎢W1 ⎥
⎢ ∂u ⎥ ⎢ x⎥
⎢ ∂y ⎥ ⎢θ1 ⎥
⎢ ⎥ ⎢θ1y ⎥
⎢ ∂u ⎥ ⎢ z⎥
⎢ ∂z ⎥ ⎢θ1 ⎥
⎢ ⎥ ⎢U ⎥
⎢ ∂v ⎥ ⎢ 2⎥
⎢ ∂x ⎥ ⎢ V2 ⎥
⎢ ∂v ⎥ ⎢W ⎥
⎢ ⎥ = t + dt
t BNL
⎢ 2⎥ = t + dt
BNL U
⎢ ∂y ⎥ ⎢θ 2x ⎥ t

⎢ ∂v ⎥ ⎢ y⎥
⎢ ⎥ ⎢θ 2 ⎥
⎢ ∂z ⎥ ⎢θ 2z ⎥
⎢ ∂w ⎥ ⎢ ⎥
⎢ ⎥ ⎢U 3 ⎥
⎢ ∂x ⎥ ⎢V ⎥
⎢ ∂w ⎥ ⎢ 3⎥
⎢ ∂y ⎥ ⎢W3 ⎥
⎢ ⎥ ⎢θ x ⎥
⎢ ∂w ⎥ ⎢ 3y ⎥ (3.106)
⎢ ∂z ⎥ ⎢θ3 ⎥
⎣ ⎦ ⎢ z⎥
⎣θ3 ⎦
The detailed expressions for calculating t + dtt BNL are given in (3.109) and (3.110). The equations
are important because they present the way, how spatial derivatives of all the displacements are
calculated. The entries in t + dtt BNL are thus used to setup also the matrix t + dtt BL 0 and t + dtt BL1 .
These matrices are computed as follows:

t + dt t + dt
t BL 0(1,i ) = t BNL (1,i )
t + dt t + dt
t BL 0(2,i ) = t BNL (5,i )
t + dt t + dt
t BL 0(3,i ) = t BNL (9,i )
t + dt t + dt
(3.107)
t BL 0(4,i ) = t BNL (4,i ) + t + dtt BNL (2,i )
t + dt t + dt
t BL 0(5,i ) = t BNL (6,i ) + t + dtt BNL (8,i )
t + dt t + dt
t BL 0(6,i ) = t BNL (7,i ) + t + dtt BNL (3,i )
156

t + dt ∂ t + dt u t + dt ∂ t + dt v t + dt ∂ t + dt w t + dt
t BL1(1,i ) = B
t NL (1,i ) + B
t NL (4,i ) + t BNL (7,i )
∂x ∂x ∂x
t + dt ∂ t + dt u t + dt ∂ t + dt v t + dt ∂ t + dt w t + dt
t BL1(2,i ) = B
t NL (2,i ) + B
t NL (5,i ) + t BNL (8,i )
∂y ∂y ∂y
t + dt ∂ t + dt u t + dt ∂ t + dt v t + dt ∂ t + dt w t + dt
t BL1(3,i ) = B
t NL (3,i ) + B
t NL (6,i ) + t BNL (9 i )
∂z ∂z ∂z
t + dt ∂ t + dt u t + dt ∂ t + dt u t + dt ∂ t + dt v t + dt
t BL1(4,i ) = B
t NL (2,i ) + B
t NL (1,i ) + t BNL (5,i ) +
∂x ∂y ∂x
∂ t + dt v t + dt ∂ t + dt w t + dt ∂ t + dt w t + dt
t BNL (4,i ) + B
t NL (8,i ) + t BNL (7,i )
∂y ∂x ∂y
t + dt ∂ t + dt u t + dt ∂ t + dt u t + dt ∂ t + dt v t + dt
t BL1(5,i ) = t BNL (3,i ) + t BNL (2,i ) + t BNL (6,i ) +
∂y ∂z ∂y
∂ t + dt v t + dt ∂ t + dt w t + dt ∂ t + dt w t + dt
t BNL (5,i ) + t BNL (9,i ) + z t BNL (8,i )
∂z ∂y ∂y
t + dt ∂ t + dt u t + dt ∂ t + dt u t + dt ∂ t + dt v t + dt
t BL1(6,i ) = t BNL (1,i ) + t BNL (3,i ) + t BNL (4,i ) +
∂z ∂x ∂z
(3.108)
∂ t + dt v t + dt ∂ t + dt w t + dt ∂ t + dt w t + dt
B
t NL (6,i ) + B
t NL (7,i ) + t BNL (9,i )
∂x ∂z ∂x
 157

t + dt −1 ∂hi
t BNL (1,1) = J1,1
∂r
t + dt
t BNL (1,2) = 0
t + dt
t BNL (1,3) = 0
t + dt
t BNL (1,4) = 0
t + dt −1∂hi ⎛ t t t + dt tz s t t + dt sz ⎞ 1 −1 t t + dt sz 1 −1 ∂hi t t + dt tz
t BNL (1,5) = J1,1 ⎜ ai Vi + bi Vi ⎟ + J1,2 hi bi Vi + J1,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt −1 ∂hi ⎛ t t t + dt t y s t t + dt s y ⎞ 1 −1 t t + dt s y 1 −1 ∂hi t t + dt t y
t BNL (1,6) = J1,1 ⎜ − ai Vi − bi Vi ⎟ − J1,2 hi bi Vi + J1,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt −1 ∂hi
t BNL (2,1) = J 2,1
∂r
t + dt
B
t NL (2,2) = 0
t + dt
t BNL (2,3) = 0
t + dt
t BNL (2,4) = 0
t + dt −1∂hi ⎛ t t t + dt t z s t t + dt sz ⎞ 1 −1 t t + dt sz 1 −1 ∂hi t t + dt tz
t BNL (2,5) = J 2,1 ⎜ ai Vi + bi Vi ⎟ + J 2,2 hi bi Vi + J 2,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt −1 ∂hi ⎛ t t t + dt t y s t t + dt s y ⎞ 1 −1 t t + dt s y 1 −1 ∂hi t t + dt t y
t BNL (2,6) = J 2,1 ⎜ − ai Vi − bi Vi ⎟ − J 2,2 hi bi Vi + J 2,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt −1 ∂hi
t BNL (3,1) = J 3,1
∂r
t + dt
t BNL (3,2) = 0
t + dt
t BNL (3,3) = 0
t + dt
t BNL (3,4) = 0
t + dt −1 ∂hi ⎛ t t t + dt tz s t t + dt sz ⎞ 1 −1 t t + dt sz 1 −1 ∂hi t t + dt tz
t BNL (3,5) = J 3,1 ⎜ ai Vi + bi Vi ⎟ + J 3,2 hi bi Vi + J 3,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt −1 ∂hi ⎛ t t t + dt t y s t t + dt s y ⎞ 1 −1 t t + dt s y 1 −1 ∂hi t t + dt t y
t BNL (3,6) = J 3,1 ⎜ − ai Vi − bi Vi ⎟ − J 3,2 hi bi Vi + J 3,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt
t BNL (4,1) = 0
t + dt −1 ∂hi
t BNL (4,2) = J1,1
∂r
t + dt
t BNL (4,3) = 0
t + dt −1 ∂hi ⎛ t t t + dt tz s t t + dt sx ⎞ 1 −1 t t + dt sz 1 −1 ∂hi t t + dt tz
t BNL (4,4) = J1,1 ⎜ − ai Vi − bi Vi ⎟ − J1,2 hi bi Vi − J1,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt
t BNL (4,5) = 0
t + dt −1 ∂hi ⎛ t t t + dt tx s t t + dt sx ⎞ 1 −1 t t + dt sx 1 −1 ∂hi t t + dt tx
t BNL (4,6) = J1,1 ⎜ ai Vi + bi Vi ⎟ + J1,2 hi bi Vi + J1,3 ai Vi
∂r ⎝2 2 ⎠ 2 2 ∂r
t + dt
t BNL (5,1) = 0
t + dt −1 ∂hi
t BNL (5,2) = J 2,1
∂r
t + dt
t BNL (5,3) = 0
t + dt −1 ∂hi ⎛ t t t + dt tz s t t + dt sx ⎞ 1 −1 t t + dt sz 1 −1 ∂hi t t + dt tz
t BNL (5,4) = J 2,1 ⎜ − ai Vi − bi Vi ⎟ − J 2,2 hi bi Vi − J 2,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt
t BNL (5,5) = 0 (3.109)
t + dt −1 ∂hi ⎛ t t t + dt tx s t t + dt sx ⎞ 1 −1 t t + dt sx 1 −1 ∂hi t t + dt t x
t BNL (5,6) = J1,1 ⎜ ai Vi + bi Vi ⎟ + J1,2 hi bi Vi + J1,3 ai Vi
∂r ⎝2 2 ⎠ 2 2 ∂r
158

t + dt
t BNL (6,1) = 0
t + dt −1 ∂hi
t BNL (6,2) = J 3,1
∂r
t + dt
t BNL (6,3) = 0
t + dt −1 ∂hi ⎛ t t t + dt tz s t t + dt sx ⎞ 1 −1 t t + dt sz 1 −1 ∂hi t t + dt tz
t BNL (6,4) = J 3,1 ⎜ − ai Vi − bi Vi ⎟ − J 3,2 hi bi Vi − J 3,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt
t BNL (6,5) = 0
t + dt −1 ∂hi ⎛ t t t + dt tx s t t + dt sx ⎞ 1 −1 t t + dt sx 1 −1 ∂hi t t + dt tx
t BNL (6,6) = J 3,1 ⎜ ai Vi + bi Vi ⎟ + J 3,2 hi bi Vi + J 3,3 ai Vi
∂r ⎝2 2 ⎠ 2 2 ∂r
t + dt
t BNL (7,1) = 0
t + dt
t BNL (7,2) = 0
t + dt −1 ∂hi
t BNL (7,3) = J1,1
∂r
t + dt −1 ∂hi ⎛ t t t + dt t y s t t + dt s y ⎞ 1 −1 t t + dt s y 1 −1 ∂hi t t + dt t y
t BNL (7,4) = J1,1 ⎜ ai Vi + bi Vi ⎟ + J1,2 hi bi Vi + J1,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt −1 ∂hi ⎛ t t t + dt tx s t t + dt sx ⎞ 1 −1 t t + dt sx 1 −1 ∂hi t t + dt tx
t BNL (7,5) = J1,1 ⎜ − ai Vi − bi Vi ⎟ − J1,2 hi bi Vi − J1,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt
t BNL (7,6) = 0
t + dt
t BNL (8,1) = 0
t + dt
t BNL (8,2) = 0
t + dt −1 ∂hi
t BNL (8,3) = J1,1
∂r
t + dt −1 ∂hi ⎛ t t t + dt t y s t t + dt s y ⎞ 1 −1 t t + dt s y 1 −1 ∂hi t t + dt t y
t BNL (8,4) = J 2,1 ⎜ ai Vi + bi Vi ⎟ + J 2,2 hi bi Vi + J 2,3 ai Vi
∂r ⎝2 2 ⎠ 2 2 ∂r
t + dt −1 ∂hi ⎛ t t t + dt tx s t t + dt sx ⎞ 1 −1 t t + dt sx 1 −1 ∂hi t t + dt tx
t BNL (8,5) = J 2,1 ⎜ − ai Vi − bi Vi ⎟ − J 2,2 hi bi Vi − J 2,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt
t BNL (8,6) = 0
t + dt
t BNL (9,1) = 0
t + dt
t BNL (9,2) = 0
t + dt −1 ∂hi
t BNL (9,3) = J 3,1
∂r
t + dt −1 ∂hi ⎛ t t t + dt t y s t t + dt s y ⎞ 1 −1 t t + dt s y 1 −1 ∂hi t t + dt t y (3.110)
t BNL (9,4) = J 3,1 ⎜ ai Vi + bi Vi ⎟ + J 3,2 hi bi Vi + J 3,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt −1 ∂hi ⎛ t t t + dt tx s t t + dt sx ⎞ 1 −1 t t + dt sx 1 −1 ∂hi t t + dt tx
t BNL (9,5) = J 3,1 ⎜ − ai Vi − bi Vi ⎟ − J 3,2 hi bi Vi − J 3,3 ai Vi
∂r ⎝ 2 2 ⎠ 2 2 ∂r
t + dt
t BNL (9,6) = 0
 159

t +Δt
The stress matrix t Sij from (1.34) has he form:

⎡ t +Δttσ xx t +Δt
t xy τ t +Δt
t xz τ ⎤
⎢ ⎥
⎢
t +Δt
t yy σ t +Δt
t yz τ ⎥
⎢ t +Δt
t zz σ ⎥
⎢ ⎥
⎢
t +Δt
t σ xx t +Δt
τ
t xy τ
t +Δt
t xz ⎥
t +Δt
t Sij =
⎢ t +Δt
σ τ
t +Δt ⎥ (3.111)
⎢ t yy t yz
⎥
⎢ σ
t +Δt
t zz ⎥
⎢ σ xx τ ⎥
t τ xz ⎥
t +Δt t +Δt t +Δt
sym.
⎢ t t xy

⎢ t +Δt
t yy σ t +Δt
t τ yz ⎥
⎢ ⎥
t σ zz ⎦
t +Δt
⎣

As already mentioned, stress-strain relations are calculated in r,s,t coordinate system, hence we
need equations for their transformations from global x,y,z coordinate system to the isoparametric
system with r,s,t coordinates and vice versa.
Let us denote t + dt Tε , t + dtTσ transformation matrices for strain and stress transformation from
global to isoparametric coordinate system, so that:

⎡ t +Δttσ xx ⎤
⎢ t +Δt ⎥
⎡ t +Δttσ rr ⎤ ⎢ t σ yy ⎥
⎢ t +Δt ⎥ t + dt ⎢ t σ zz ⎥
t +Δt

⎢ t rs ⎥ τ = Tσ ⎢ ⎥
⎢ t τ rt ⎥
t +Δ t ⎢
t +Δt
t τ xy ⎥
⎣ ⎦ ⎢ t +Δtτ ⎥
⎢ t +Δtt yz ⎥
⎣⎢ t τ xz ⎦⎥
(3.112)
⎡ t +Δttε xx ⎤
⎢ t +Δt ⎥
⎡ t ε rr ⎤
t +Δt ⎢ t ε yy ⎥
⎢ t +Δt ⎥ t + dt ⎢ t ε zz ⎥
t +Δt

⎢ t γ rs ⎥ = Tε ⎢ t +Δtγ ⎥
⎢ t +Δttγ rt ⎥ ⎢ t xy ⎥
⎣ ⎦ ⎢ t +Δtγ ⎥
⎢ t +Δtt yz ⎥
⎢⎣ t γ xz ⎥⎦

Then the transformation matrices are calculated by:

(
⎡ t + dt V rx 2
) ( ) ( V rz ) ⎤
ry 2 2 ry ry t + dt
t + dt t + dt
V 2t + dt V rx t + dt
V 2t + dt V V rz 2t + dt V rx t + dt
V rz
⎢ ⎥
t + dt ⎢ ry t + dt sy s ry t + dt ry t + dt sy ⎥
Tσ = ⎢ t + dt V rx t + dtV sx t + dt
V V t + dt
V rz t + dt
V sz t + dt
V rx t + dt
V y + t + dt V V sx t + dt
V V sz + t + dt V rz t + dt
V t + dt rx t + dt s z
V V + t + dt V rz t + dtV sx ⎥
⎢ t + dt V rx t + dtV tx t + dt
V
ry t + dt
V
ty t + dt
V rz t + dt
V sz t + dt
V rx t + dt ty
V + t + dt V
ry t + dt
V tx t + dt
V
ry t + dt
V tz + t + dt V rz t + dt
V
ty
V + t + dt V rz t + dtV tx ⎥
t + dt rx t + dt t z
V
⎢ ⎥
⎣ ⎦
(3.113)
160

(
⎡ t + dt V rx 2
) ( ) ( V rz ) ⎤
ry 2 2 ry ry t + dt
t + dt t + dt t + dt t + dt t + dt t + dt t + dt
V V rx V V V rz V rx V rz
⎢ ⎥
t + dt ⎢ ry t + dt sy s ry t + dt ry t + dt sy ⎥
Tε = ⎢ 2t + dt V rx t + dtV sx 2t + dt V V 2t + dt V rz t + dt
V sz t + dt
V rx t + dt
V y + t + dt V V sx t + dt
V V sz +t + dt V rz t + dt
V t + dt rx t + dt sz
V V + t + dt V rz t + dtV sx ⎥
⎢ 2t + dt V rx t + dtV tx 2 t + dt
V
ry t + dt
V
ty
2 t + dt
V rz t + dt
V sz t + dt
V rx t + dt
V +
ty t + dt
V
ry t + dt
V tx t + dt
V
ry t + dt
V +
tz t + dt
V rz t + dt
V
ty t + dt rx t + dt t z
V V + Vt + dt rz t + dt t x ⎥
V
⎢ ⎥
⎣ ⎦
(3.114)
V s = ⎡⎣ t + dt V sx t + dt V y t + dt V sz ⎤⎦ , t + dt V t = ⎡⎣ t + dt V tx t + dt V y t + dt V tz ⎤⎦ are vectors
t + dt s t
where vectors
of unity length from Fig. 3-33. The remaining vector is calculated as a vector product of the
previous two vectors:
V r = ⎣⎡ t + dt V rx V rz ⎦⎤ =t + dt V s ⊗ t + dtV t
t + dt t + dt ry t + dt
V (3.115)

Inverse transformation matrices are calculated as:

t + dt
Tσ−1 =t + dt TεT
(3.116)
t + dt −1 t + dt
Tε = Tσ T

3.13.4 The element integration

The element is integrated numerically. Along its longitudinal axis the element is integrated by
standard two to six nodes Gaussian integration. The table below lists r coordinates and
associated weights for utilized integration points:
Table 3.13-1: Gaussian integration of the beam element along the longitudinal axis

Integrat Coordinate r Weight

Number of
ion
integ. points
point

1 0.577350269189626 1.
2
2 -0.577350269189626 1.
1 0.774596669241483 0.555555555555556
3 2 0. 0.888888888888889
3 -0.774596669241483 0.555555555555556
1 0.861136311594053 0.347854845137454
2 0.339981043584856 0.652145154862546
4
3 -0.339981043584856 0.652145154862546
4 0.861136311594053 0.347854845137454
5 1 0.906179845938664 0.236926885056189
2 0.538469310105683 0.478628670499366
 161

3 0. 0.568888888888889
4 -0.538469310105683 0.478628670499366
5 -0.906179845938664 0.236926885056189
1 0.932469514203152 0.171324492379170
2 0.661209386466265 0.360761573048139
3 0.238619186083197 0.467913934572691
6
4 -0.238619186083197 0.467913934572691
5 -0.661209386466265 0.360761573048139
6 -0.932469514203152 0.171324492379170

Although the 2-nodes integration may be sufficient, (with respect to the quadratic displacement
transformation), a higher order integration scheme will yield better result in a case of high
material nonlinearity and/or in a case of a very curved beam geometry.
As for integration within the cross-section, i.e. in s,t coordinates, trapezoidal quadrature is used.
The element cross-section is subdivided into ns , nt “strips” as depicted in the following figure.
t

dtnt
2 s

dt2
dt1
ds1 dsns ind ividual “weight“
ds2 2 and m aterial

Fig. 3-34 The beam cross section integration

The integration is then carried out by summing functional values in centre of the all
quadrilaterals multiplied by their area.
Note that the element is integrated within the isoparametric coordinate system, hence we have to
use dx dy dz = det( J ) dr ds dt , see (3.104).
Nice feature of the ATENA’s implementation of the beam is that each of the quadrilaterals in a
cross section adopts an artificial input weight factor. By default, such a “weight” is equal to one,
however, if we set its value to zero, essentially a hole is introduced. This mechanism, together
with possibility of defining an customized material law in each of the quadrilaterals facilitates to
analyse beams that have a arbitrary shape of cross-sections.
The present beam implementation supports also smeared reinforcement. This is done in the same
way as it was for the Ahmad elements described in the previous section.
162

3.14 Global and local coordinate systems for element load

Most element loads can be defined in global or local coordinate system. Global coordinate
system is always available, hence using it is usually the safest way to input a desired element
load. Nevertheless, some elements are internally defined in a local coordinate system and it can
be employed for an element load definition, too. Location of such a local system, (if it exists) has
been described together with description of the associated finite element. For example, local
coordinate systems are defined for plane 3D isoparametric elements, shell and beam elements
etc. On the other hand, elements such as tetrahedrons, bricks and others are defined in directly in
global coordinate system and therefore a local element load is treated as if it were input as a
global element load.
An exception to the above are truss elements. Although they are defined in global coordinate
system, they do support local element load. Their local coordinate system (for element loading
only) is defined as follows:
• local X axis points in direction of the truss element,
• local Y axis is normal to local X axis and lies in the global XY plane,
• its positive orientation is chosen so that the local X and local Y forms a right-hand (2D)
coordinate system in the plane defined by these local axes,
• local Z axis is vector product of the local X and local Y axes, (for 3D case only).
• if the truss is parallel to global z, then local X points in direction of global Z, local Y
coincides with global Y and local Z has opposite direction of the global X, (for 3D case
only).

2D 3D

ZG
YG N2
YG
YL
N1

N1 XL YL
YG
XG

XL N2
XG

Fig. 3-35 Local and global coordinate systems for truss element N1-N2, (e.g. loaded element edge)

Specification of a boundary load deserves slightly more attention. Firstly, it is applied only to an
element’s edge or an element’s surface, (see also the note below), as opposed to e.g. an element
body load that is for the whole element. Local coordinate system is thus defined by location of
the loaded edge or surface. Secondly, a boundary load definition must include a reference to a
selection, which contains nodes to be loaded. Their order in the list is irrelevant, as what really
 163

matters is the order in which they appear in the element incidences. When processing a boundary
load, ATENA loops thru all element’s surfaces and edges, (in the order specified in the table
below) and checks appropriate incidental nodes. If the tested node is present in the list of loaded
boundary nodes, it is picked up and put into incidences of a new planar or line element. This
element is later used to process the boundary load. It is its local coordinate system, that is
(possibly) used to deal with local/global load transformations.
The table below defines the orders, in which element surfaces and edges are tested for a surface
or edge element load. (It is assumed that element incidences are (n1 , n2 , ... nnum _ elem _ nodes ) ). It
describes linear elements but surfaces and edges of nonlinear elements are treated in the same
order.

Table 3.14-1: Order of element surface and nodes as they are tested within a boundary load
definition.
Element shape Type Surface/node incidences
Truss Edge (n1 , n2 )
Triangle Surface (n1 , n2 , n3 )
Edge (n1 , n2 ); (n2 , n3 ); (n3 , n1 )
Quad Surface (n1 , n2 , n3 , n4 )
Edge (n1 , n2 ); (n2 , n3 ); (n3 , n1 ); (n4 , n1 )
Hexahedron, Surface (n1 , n2 , n3 , n4 ); (n5 , n6 , n7 , n8 ); (n1 , n4 , n8 , n5 ); (n2 , n3 , n7 , n6 );
(brick) (n1 , n2 , n6 , n5 ); (n4 , n3 , n7 , n8 );
Edge (n1 , n2 ); (n2 , n3 ); (n3 , n4 ); ( n4 , n1 );
(n5 , n6 ); (n6 , n7 ); (n7 , n8 ); (n8 , n5 );
(n1 , n5 ); (n2 , n6 ); ( n3 , n7 ); (n4 , n8 )
Tetrahedron Surface (n1 , n2 , n3 ); (n1 , n2 , n4 ); (n1 , n3 , n4 ); (n2 , n3 , n4 )
Edge (n1 , n2 ); (n2 , n3 ); (n3 , n1 );
(n4 , n1 ); (n4 , n2 ); (n4 , n3 )
Wedge Surface (n1 , n2 , n3 ); (n4 , n5 , n6 );
(n1 , n2 , n5 , n4 ); (n6 , n5 , n2 , n3 ); (n4 , n6 , n3 , n1 )
Edge (n1 , n2 ); (n2 , n3 ); (n3 , n1 );
(n4 , n5 ); (n5 , n6 ); ( n6 , n4 );
(n1 , n4 ); ( n2 , n5 ); ( n3 , n6 );

Note that only one surface or one edge of each element can be loaded in a single boundary load
specification. If more element’s surfaces or edges are to be loaded, use more boundary load
164

definitions. Violation of this rule causes an error report and skipping of the offending boundary
load.

3D edge load
2D edge load planar element
planar element

n1 ZG n1 XL ZL
YG XL

YL
YL n2 n3
n2
n3
YG
XG
XG

3D edge load
solid element 3D surface load
solid element
n1 X L Z L n1
ZG ZL
ZG
YL
YL
n3 XL
n2 n2 n3

n6 n6
n5 YG YG
XG n5
XG

Fig. 3-36 Examples of positioning local coordinate system used by surface and element load for 2D
and 3D elements

Transport analysis does not distinguish between local and global element loads. Hence, a local
element “load” is treated as being a global load. The actual load value is always scalar, (unlike
vectors in statics) and it is assumed positive for flow out of the element.
 165

3.15 References
AHMAD, S., B. M. IRONS, ET AL. (1970). "Analysis of Thick and Thin Shell Structures by
Curved Finite Elements." International Journal of Numerical Methods in Engineering 2:
419-451.
BATHE, K.J.(1982), Finite Element Procedures In Engineering Analysis, Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey 07632, ISBN 0-13-317305-4.
CRISFIELD, M.A. (1983) - An Arc-Length Method Including Line Search and Accelerations,
International Journal for Numerical Methods in Engineering, Vol.19,pp.1269-1289.
FELIPPA, C. (1966) - Refined Finite Element Analysis of Linear and Nonlinear Two-
Dimensional Structures, Ph.D. Dissertation, University of California, Engineering, pp.41-50.
HINTON, E. AND D. R. J. OWEN (1984). Finite Element Software for Plates and Shells,
Peridge Press.
JENDELE, L. (1981). Thick Plate Finite Element based on Mindlin's Theory. Prague, student
research work.
JENDELE, L. (1992). Nonlinear Analysis of 2D and Shell Reinforced Concrete Structures
Including Creep and Shrinkage. Civil Engineering Department. Glasgow, University of
Glasgow: 393.
JENDELE, L., A. H. C. CHAN, ET AL. (1992). "On the Rank Deficiency of Ahmad's Shell
Element." Engineering Computations 9(6): 635-648.
RAMM, E. (1981) - Strategies for Tracing Non- linear Responses Near Limit Points, Non- linear
Finite Element Analysis in Structural Mechanics, (Eds. W.Wunderlich, E.Stein, K.J.Bathe)
166

4 SOLUTION OF NONLINEAR EQUATIONS

The main objective of this chapter is to review methods for solution of a set of nonlinear
equations. Several methods, which are implemented in ATENA are described later in this
Chapter. However, all of them need to solve a set of linear algebraic equation in the form
Ax = b (4.1)
where A, x, b stands for a global structural matrix and vectors of unknown variables and rhs of
the problem, respectively. Hence, this problem is discussed first.

4.1 Linear Solvers

Two types of the solvers are supported: direct and iterative, each of them having some pros and
cons. Without going into details, a direct solver is recommended for smaller problems or
problems that are ill-posed. On the other hand, iterative solvers are typically more efficient to
solve well-posed large 3D analyses.
The two approaches also differ in a way they store the structural matrix A . It comes from the
nature of FEM that the structural matrices have sparse character, with most of non-zero elements
located near the diagonal. The matrix has banded pattern and ATENA works with band of
variable width.
If a direct solver is used, then each column of matrix A stores all entries between the diagonal
element and the last non-zero element in the column. This structure is sometimes called sky-line
profile structure. The matrix A
⎡ a11 a12 a13 a15 ⎤
⎢a a22 a23 a24 a25 ⎥
⎢ 21 ⎥
⎢ a31 a32 a33 a34 a35 ⎥
⎢ ⎥
A=⎢ a42 a43 a44 a45 a46 ⎥ (4.2)
⎢ a51 a52 a53 a54 a55 a56 ⎥
⎢ ⎥
⎢ a64 a65 a66 a67 ⎥
⎢ a76 a77 ⎥⎦
⎣
is thus stored in three vectors d , u , l with actual data and one vector p with information about
matrix’s profile:
d = [ a11 a77 ]
T
a22 a33 a44 a55 a66
u = [ a12 a67 ]
T
a13 a23 a13 a24 a34 a15 a25 a35 a45 a46 a56
l = [ a21 a76 ]
T
a31 a32 a13 a43 a43 a51 a52 a53 a54 a64 a65 (4.3)

p = [ 0 1 3 5 9 11 12]
T

For each column i of the matrix A the vector p stores location of ai (i −1) within the array u ,
resp. l . If A is symmetric, then u = l and only l is stored. Note the a direct solver we have to
store all elements within the bandwidth, even though some of them may be equal to zero,
because that they can become non-zero in the process of solution, (i.e. matrix factorization).
 167

Iterative solver can store only true non-zero elements, irrespective of whether they are located
above or below the skyline. Suppose the matrix A from (4.2) that stores some zero elements
below the skyline
⎡ a11 0 a13 a15 ⎤
⎢0 a22 a23 a24 0 ⎥
⎢ ⎥
⎢ a31 a32 a33 a34 0 ⎥
⎢ ⎥
A=⎢ a42 a43 a44 0 a46 ⎥ (4.4)
⎢ a51 0 0 0 a55 0 ⎥
⎢ ⎥
⎢ a64 a65 a66 a67 ⎥
⎢ a76 a77 ⎥⎦
⎣
All iterative solvers would store the matrix A in three vectors. All the data are stored in a
vector a and location of the stored element is maintained in vectors r , c . The above matrix is
stored as follows:
a = [ a11 a31 a51 a22 a32 a42 a33 a13 a23 a43 ... a77 a67 ]
(4.5)
c =[ 1 3 5 2 3 4 3 1 2 3 ... 7 6 ]
r = [ 1 4 7 21 23 ]
The vector a stores for each column of A first diagonal element, followed by all non-zero
elements, from the top to the bottom of the column. The vector c stores row index of each entry
in the vector a . r stores location of all diagonal elements aii within a appended by an artificial
pointer to an +1n +1 , where n = dim( A) .

4.1.1 Direct Solver

The well known Cholesky decomposition is used to solve the problem. The matrix A is
decomposed into
A = LDU (4.6)
where L, U is lower and upper matrix and D is diagonal matrix. The method to compute the
decomposition is described elsewhere, e.g. (Bathe 1982). Equation (4.1) is then solved in two
steps:
v = L−1b
(4.7)
x = ( DU ) v
−1

Both of the above equations are computed easily, because the involved matrices have triangular
pattern. Hence, the solution of (4.7) represents back substitution only. If A is symmetric, (which
is usually the case), then
U = LT (4.8)
168

4.1.2 Direct Sparse Solver

Direct sparse solvers are similar to the above Direct solvers, however they should work more
economically both in terms of RAM and CPU requirements. They belong to a group of direct
(i.e. non-iterative) solution methods. They are based on matrix decomposition similar to (4.6).
The decomposition can be LU or LDU for nonsymmetric matrices and/or LLT or LDLT
decomposition for symmetric matrices.
The main difference between these solvers and those from Section 4.1.3 is that they run so called
pre-factorization procedure, before the actual factorization is executed. Such a pre-factorization
has two jobs:
1. Find out, what initially zero aij entries of the matrix A (that are stored below the skyline)
become non-zero due to factorization of A. Such entries are called fill-in.
2. Per mutate lines and columns of A so that the filling gets minimum.
Once a map of fill-in is known, it is added to the originally non-zero data of A and only these
data are to be stored and maintained in the next operations. Hence, as it is not necessary to store
and work upon all data below the skyline of A (as it is he case of solvers in Section 4.1.1); we
can use here a sparse matrix storage scheme. The incurred savings in both RAM and CPU
resources is significant and it pays well of for a computation overhead caused by the pre-
factorization phase and a bit more complicated storage scheme in use.
It is beyond the scope of this document to describe all details about implementation of this
solver. It is based on (Vondracek, 2006) and (Davis et. al, 1995). A number of optimization
techniques are used to speed up the solution procedure, such as the problem (4.6) can be solved
using a block structure. This apply to pre-factorization, factorization as well as for
backward/forward substitution phases. Typical size of such a block is 2x2 .. 6x6. The bigger
block size, the smaller overhead for pre-factorization and mapping of the matrix and the faster
the operation to actually factorize and solve the problem (4.6). Use of a bigger block, however,
results also in a higher waste of RAM because all non-zero data and fill-in are rounded into a
storage with block pattern.
Direct sparse solvers are a compromise between Direct Solvers and Sparse Solvers. They
typically need more RAM and CPU than Sparse solvers do, (and less than Direct Solvers),
however, they never diverge and bring uncertainties as what precoditioner to use etc. Therefore,
they are recommended for middle size (may-be ill conditioned) problems, the solution of which
would not fit into RAM subject a Direct Solver is used, and for which Sparse solvers are not
sufficiently robust.

4.1.3 Iterative Solver

The table below lists all solvers in ATENA that can solve the problem (4.1) iteratively. Although
the list is long, from practical point of view only a few of them are recommended, see the
column “Description”. In addition, only the methods DCG and ICCG are designed to take full
advantage of symmetry of A (if present). The remaining solvers would store only symmetric part
of A, however they will operate on it in the same way as it is not symmetric. Therefore, for
symmetric problems the solvers DCG and ICCG are preferable.
Each of the iterative solver typically consists of two routine, one for “preparation” of the solution
and the other for the solution itself, i.e. “execution” phase. The former routine is particularly
important for the case of preconditioned iterative solvers. This is where a preconditioning matrix
is created.
 169

The most efficient preconditioning routine are based on incomplete Cholesky decomposition
(Rektorys 1995). The preconditioning matrix A' is decomposed in the same way as (4.6), i.e.
A' = L'D'U' (4.9)
Comparing A and A' , it can be written
for aij ≠ 0 a 'ij = aij
(4.10)
for aij = 0 a 'ij ≠ aij
The incomplete Cholesky decomposition is carried out in the same way as complete Cholesky
decomposition (4.6), however, entries in A , which were originally zero and became nonzero
during the factorization are ignored, i.e. they stay zero even after the factorization. The incurred
inaccuracy is the penalty for memory savings due to usage of the iterative solvers’ storage
scheme. For symmetric problem, use ssics routine, for nonsymmetric problems the ssilus is
available to construct A' = L'D'(L')T or A' = L'D'U' .
Last, but not the least note that each solver needs some temporary memory. Such requirements
are included in the table below. Typically, the more advanced iterative solver, the more extra
memory it needs and the less number of iterations needed to achieve the same accuracy.

Table 4.1-1 SOLVER TYPES.

Type D/I Prep. Exec. Sym/N Temporary memory Description

phase phase on- required
sym
LU D --- --- S,NS ----- For smaller or ill-
posed probems
JAC I ssds sir S,NS 4*(11)+8*(1+4*n) Simple, not
recommended
GS I --- sir S,NS 4*(11+nel+n+1)+8*(1+3
*n+nel)
ILUR I ssilus sir S,NS 4*(13+4*n+nu+nl)+8*(1
+4*n+nu+nl)
DCG I ssds scg S 4*(11)+8*(1+5*n) For large symmetric
well-posed problems
ICCG I ssics scg S 4*(12+nel+n)+8*(1+5*n For large symmetric
+nel) problems,
recommended
DCGN I ssd2s scgn S,NS 4*(11)+8*(1+8*n) For large non-
symmetric well-
posed problems
LUCN I ssilus scgn S,NS 4*(13+4*n+nl+nl)+8*(1 For large non-
+8*n+nl+nu) symmetric problems,
recommended
DBCG I ssds sbcg S,NS 4*(11)+8*(1+8*n)
170

LUBC I ssilus sbcg S,NS 4*(13+4*n+nl+nu)+8*(1

+8*n+nu+nl)
DCGS I ssds scgs S,NS 4*(11)+8*(1+8*n)
LUCS I ssilus scgs S,NS 4*(13+4*n+nl+nu)+8*(1
+8*n+nu+nl)
DOMN I ssds somn S,NS 4*(11)+8*(1+4*n+nsave
+3*n*(nsave+1))
LUOM I ssilus somn S,NS 4*(13+4*n+nu+nl)+8*(1
+nl+nu+4*n+nsave+3*n
*(nsave+1))
DGMR I ssds sgmres S,NS 4*(31)+8*(2+n+n*(nsav
e+6)+nsave*(nsave+3))
LUGM I ssilus sgmres S,NS 4*(33+4*n+nl+nu)+8*(2
+n+nu+nl+n*(nsave+6)+
nsave*(nsave+3))
In the above:
n is number of degree of freedom of the problem. nel is the number of nonzeros in the lower
triangle of the problem matrix (including the diagonal). nl and nu is the number of nonzeros in
the lower resp. upper triangle of the matrix (excluding the diagonal).
Table 4.1-2: EXECUTION PHASES.

Phase name Description

sir Preconditioned Iterative Refinement sparse Ax = b solver. Routine to solve a
general linear system Ax = b using iterative refinement with a matrix
splitting.
scg Preconditioned Conjugate Gradient iterative Ax=b solver. Routine to solve a
symmetric positive definite linear system Ax = b using the Preconditioned
Conjugate Gradient method.
scgn Preconditioned CG Sparse Ax=b Solver for Normal Equations. Routine to
solve a general linear system Ax = b using the Preconditioned Conjugate
Gradient method applied to the normal equations AA'y = b, x=A'y.
sbcg Solve a Non-Symmetric system using Preconditioned BiConjugate Gradient.
scgs Preconditioned BiConjugate Gradient Sparse Ax=b solver. Routine to solve a
Non-Symmetric linear system Ax = b using the Preconditioned BiConjugate
Gradient method.
somn Preconditioned Orthomin Sparse Iterative Ax=b Solver. Routine to solve a
general linear system Ax = b using the Preconditioned Orthomin method.
sgmres Preconditioned GMRES iterative sparse Ax=b solver. This routine uses the
generalized minimum residual (GMRES) method with preconditioning to
solve non-symmetric linear systems of the form: A*x = b.
 171

Table 4.1-3: PREPARATION PHASES.

Phase name Description

ssds Diagonal Scaling Preconditioner SLAP Set Up. Routine to compute the
inverse of the diagonal of a matrix stored in the SLAP Column format.
ssilus Incomplete LU Decomposition Preconditioner SLAP Set Up.Routine to
generate the incomplete LDU decomposition of a matrix. The unit lower
triangular factor L is stored by rows and the unit upper triangular factor U is
stored by columns. The inverse of the diagonal matrix D is stored. No fill in
is allowed.
ssics Incompl Cholesky Decomposition Preconditioner SLAP Set Up. Routine to
generate the Incomplete Cholesky decomposition, L*D*L-trans, of a
symmetric positive definite matrix, A, which is stored in SLAP Column
format. The unit lower triangular matrix L is stored by rows, and the inverse
of the diagonal matrix D is stored.
ssd2s Diagonal Scaling Preconditioner SLAP Normal Eqns Set Up. Routine to
compute the inverse of the diagonal of the matrix A*A'. Where A is stored in
SLAP-Column format.

As for the solution procedure, i.e. the latter of the two solution phases, the most commonly used
method is Conjugate gradient method (with incomplete Cholesky preconditioner) (Rektorys
1995). The flow of execution is as follows:
r 1 = b − A x1
z 1 = M −1 r 1

r i zi
βi =
r i −1 z i −1
p i = z i + β i p i −1
(4.11)
r i zi
α = i
i

p Ap i ( )
x i +1 = x i + α i p i
r i +1 = r i − α i Ap i( )
z i +1 = M −1 r i +1
i = i +1
This solution procedure is implemented in scg routine.
The iterative solvers in ATENA are based on SLAP package (Seager and Greenbaum 1988) that
where modified to fit into ATENA framework. The authors of the package refer to (Hageman
and Young 1981), where all of the implemented solution techniques are fully described.
172

4.2 Full Newton-Raphson Method

Using the concept of incremental step by step analysis we obtain the following set of nonlinear
equations:
K ( p ) Δp = q − f ( p ) (4.12)

where:
q is the vector of total applied joint loads,

f ( p ) is the vector of internal joint forces,

Δp is the deformation increment due to loading increment,

p are the deformations of structure prior to load increment,

K ( p ) is the stiffness matrix, relating loading increments to deformation increments.

The R.H.S. of (4.12) represents out-of-balance forces during a load increment, i.e. the total load
level after applying the loading increment minus internal forces at the end of the previous load
step. Generally, the stiffness matrix is deformation dependent, i.e. a function of p , but this is
usually neglected within a load increment in order to preserve linearity. In this case the stiffness
matrix is calculated based on the value of p pertaining to the level prior to the load increment.

The set of equations (4.12) is nonlinear because of the non-linear properties of the internal
forces:
f ( kp ) ≠ kf ( p ) (4.13)

and non-linearity in the stiffness matrix

K ( p ) ≠ K ( p + Δp ) (4.14)

where k is an arbitrary constant.

The set of equations represents the mathematical description of structural behavior during one
step of the solution. Rewriting equations (4.12) for the i-th iteration within a distinct loading
increment we obtain:
K ( pi −1 ) Δpi = q − f ( pi −1 ) (4.15)

All the quantities for the (i-1)-th iteration have already been calculated during previous solution
steps. Now we solve for p i at load level q using:

pi = pi −1 + Δpi (4.16)

As pointed out earlier, equation(4.15) is nonlinear and therefore it is necessary to iterate until
some convergence criterion is satisfied. The following possibilities are supported in ATENA
( k marks k -th component of the specified vector):
 173

ΔpiT Δpi
≤ ε rel . disp
piT pi

( q − f ( pi −1 ))T ( q − f ( pi −1 ))
≤ ε rel . force
f ( pi )T f ( pi )
(4.17)
ΔpiT ( q − f ( pi −1 ))
≤ ε rel .energy
piT f ( pi )

max(( q k − f k ( pi −1 ))) max( ( q k − f k ( pi −1 )))

≤ ε abs. force
max( f k ( pi )) max( f k ( pi ))

The first one checks the norm of deformation changes during the last iteration whereas the
second one checks the norm of the out-of-balance forces. The third one checks out-of-balance
energy and the fourth conditions checks out-of-balanced forces in terms of maximum
components (rather then Euclid norms). The values of the convergence limits ε are set by
default to 0.01 or can be changed by input command SET.
The concept of solution nonlinear equation set by Full Newton-Raphson method is depicted in
Fig. 4-1:

Loading
q

Loading increment

p p p Deformation
0 1 2

Fig. 4-1 Full Newton-Raphson method.

4.3 Modified Newton-Raphson Method

The most time consuming part of solution (4.15) is the re-calculation of the stiffness matrix
K ( pi −1 ) at each iteration. In many cases this is not necessary and we can use matrix K ( p0 ) from
the first iteration of the step. This is the basic idea of the so-called Modified Newton-Raphson
method. It produces very significant time saving, but on the other hand, it also exhibits worse
convergence of the solution procedure.
174

The simplification adopted in the Modified Newton-Raphson method can be mathematically

expressed by:
K ( pi −1 )  K ( p0 ) (4.18)

The modified Newton-Raphson method is shown in Fig. 4-2. Comparing Fig. 4-1 and Fig. 4-2 it
is apparent that the Modified Newton-Raphson method converges more slowly than the original
Full Newton-Raphson method. On the other hand a single iteration costs less computing time,
because it is necessary to assemble and eliminate the stiffness matrix only once. In practice a
careful balance of the two methods is usually adopted in order to produce the best performance
for a particular case. Usually, it is recommended to start a solution with the original Newton-
Raphson method and later, i.e. near extreme points, switch to the modified procedure to avoid
divergence.

Loading
q

Loading increment

p0 p1 p2 Deformation
p3 p4

Fig. 4-2 Modified Newton-Raphson method

4.4 Arc-Length Method

Next to the Modified Newton-Raphson method, the most widely used method is the Arc-length
method. This method was first employed about fifteen years ago to solve geometrically non-
linear structures. Because of its excellent performance, it is now quite well established for
geometric non-linearity and for material non-linearity as well. Many workers have been
interested in using and improving Arc-length procedures.
The main reason for the popularity of this method is its robustness and computational efficiency
which assures good results even in cases where traditional Newton-Raphson methods fail. Using
an Arc-length method stability problems such as snap back and snap through phenomena can be
studied as well as materially non-linear problems with non-smooth or discontinuous stress-strain
diagrams. This is possible due to the changing load conditions during iterations within an
increment.
The main idea of this method is well explained by its name, arc-length. The primary task is to
observe complete load-displacement relationship rather then applying a constant loading
increment as it is in the Newton-Raphson method. Hence this method fixes not only the loading
but also the displacement conditions at the end of a step. There are many ways of fixing these,
but one of the most common is to establish the length of the loading vector and displacement
changes within the step.
From the mathematical point of view it means that we must introduce an additional degree of
freedom associated with the loading level (i.e. a problem has n displacement degrees of freedom
and one for loading) and in addition a constraint for the new unknown variable must be
 175

introduced. The new degree of freedom is usually named λ. There are many possibilities for
defining constraints on λ and those implemented in ATENA are briefly reviewed in the
following sections.
To derive the Arc-length method we rewrite the set of equations (4.12) in form of (4.19), where
λ defines the new loading factor:
K ( p ) Δp = λ q − f ( p ) (4.19)

Now re-writing (4.19) in a form suitable for iterative solution:

K ( pi −1 ) Δpi = λ q − f ( pi −1 ) = λ q − f i −1 (4.20)

pi = pi −1 + Δpi = pi −1 + ηi −1δ i −1 (4.21)

Δpi = Δpi −1 + ηi −1δ i −1 (4.22)

λi = λi −1 + Δλi −1 (4.23)
The notation is explained in Fig. 4-3. The matrix K can be recomputed for every iteration
(similar to Full Newton-Raphson method) or it can be fixed based on the 1st iteration for all
subsequent iterations (Modified Newton Raphson method). The vector q does not mean in this
case the total loading at the end of the step but only a reference loading "type". The actual
loading level is a multiple of this.
The scalar η is an additional variable introduced by the Line-search method, which will be
discussed later. The scalar η is used to accelerate solutions in cases of well-behaved load-
deformation relationships or to damp possible oscillations, if some convergence problems arose,
e.g. near bifurcation and extreme points.

q λ0

Δ λ0 q

q λ1
Δ λ1 q
q λ2 g
q λ3 1
R2
g0
Load increment

Loading R1

δT Δ λ0

δ 0
δ0
q λ start
Δp 1

Δp 2
Δp 3
η0δ0 η1δ1 η2 δ2

p0 p1 p2 Deformation

Fig. 4-3 The Arc-length method

176

Additional notation is defined as follows:

Out-of-balance forces in i-th iteration:
g ( pi ) = gi = f i − λi q = f − ( λi −1 + Δλi −1 ) qi (4.24)

R.H.S vector in i-th iteration:

RHS i = λi q − f i −1 = Δλi −1q − gi −1 (4.25)

Substituting (4.21) through (4.25) into (4.20), the deformation increment δ i −1 can be calculated
from:
Kδ i −1 = RHS i −1 = Δλi −1q − gi −1 (4.26)

Hence:
δ i −1 = δ i −1 + Δλi −1δ T (4.27)
where
δ i −1 = − K −1 gi −1
(4.28)
δT = K q −1

It remains only to set the additional constraint for Δλi −1 and ηi −1 and the whole algorithm is
defined. Thus compared to the Newton-Raphson methods in which we solve n dimensional non-
linear problem, the Arc-length method need to solve a (n + 2) dimensional problem, where the
first n unknowns correspond to deformations and the last two are Δλi −1 and ηi −1 .
If we set ηi −1 = 1 , then we deal with an (n + 1) dimensional problem that correspond to pure Arc-
length method, otherwise a combination of Arc-length and Line search must be employed. The
Line search method is discussed later in this chapter. Note that all vectors including δ i −1 , δ T are
of order (n + 1). Their (n + 1)-th coordinate corresponds to the loading dimension λ and it is set
to zero.
Now, introduce two new vectors ti −1 and ni −1 as shown in Fig. 4-4. There are defined by:
ti −1 = Δpi −1 + β ( λi −1 − λstart ) (4.29)

ni −1 = ηδ i −1 + βΔλi −1 (4.30)
where:
β is scalar that relates dimensions of λ to size of deformation space,
λi −1 is a (n + 1) dimensional vector with its firth n coordinates set to zero (deformation
space) and its (n + 1)-th coordinate equal to λi −1 .
λstart is a (n+1) dimensional vector similar to λi −1 , however its (n + 1)-th coordinate equal
to λstart .
 177

n1
βλ n2
n3

t1 t3
t2

p
Fig. 4-4 The vectors ti and ni and scalar β .

It is then obvious that

ti = ti −1 + nn −1 (4.31)
Defining the residual R :
Ri −1 = ti −1 ni −1 (4.32)

equations (4.20) through (4.32) lead to the final expression for the unknown Δλi −1 (noting that
ΔpiT−1 Δλi −1 = piT−1 λi −1 = 0 ):

Ri −1 − ΔpiT−1 δ i −1
Δλi −1 = (4.33)
ηΔpiT−1δ T + β 2 ( λi −1 − λstart )
To obtain Δλi −1 by (4.33) the residual Ri −1 must be defined. In fact, it also define type of Arc-
length constrain being used. The types supported in ATENA are described below.

4.4.1 Normal Update Method

Vector ti −1 and ni −1 are normals in this case, hence residual Ri −1 = 0 , see Fig. 4.4-3.
178

n1
βλ n2
n3

t1 t3
t2

p
Fig. 4.4-3 Normal update method.

The main advantage of this method is its simplicity. The Normal update plane is relatively
reliable, but it can fail if the l-p diagram suddenly changes its slope or turns back or down (snap
back and snap through). Nevertheless if these special conditions are treated by this method then a
very significant reduction in step length is unavoidable.

4.4.2 Consistently Linearized Method

The residual Ri −1 is defined in this case by

Ri −1 = ti −T1 ni −1 = ti −1 ni −1 cos(α ) = − ti −1 ( ti −1 − s ) (4.34)

The step length s and angle α are depicted in Fig. 4.3-4. The norm of the vector ti −1 is
calculated using (4.29):
2
ti −1 = ΔpiT−1Δpi −1 + β 2 ( λi −1 − λstart ) 2 (4.35)
 179

βλ
n
i-1
α
ti-1
current s
required
s
ti s = step length

p
Fig. 4.4-4 Consistently linearized method.

Substituting (4.34) and (4.35) in (4.33) we obtain the final expression for Δλi −1 . It should be
noted that the scalar s is set 'a priori' and governs the actual step length. Of course, the proper
choice of this parameter is essential for the solution and therefore it will be discussed later in
more detail.
This method is especially suitable for solutions that embrace λ − p diagrams with sudden breaks
and discontinuities, e.g. for materially nonlinear problems.

4.4.3 Explicit Orthogonal Method

The basic constraint for Δλi −1 in this case is that ti −1 = ti = s , where s is some distinct 'a priori'
set step length. Similar to the previous method we also have to evaluate the residual Ri −1 :

Ri −1 = ti −T1 ni −1 = ti −1 ni −1 cos(α ) = − ti −1 ri −1 (4.36)

Based on the similar triangles (see Fig. 4.4-), the following can be derived:
rl −1 tl −1
= (4.37)
t 'l −1 − s t 'l

− s 2 ( ti ' − s )
Ri −1 = (4.38)
ti '

ti ' = ti −1 + ni'−1 (4.39)

2 2 2
ti ' = ti −' 1 + β 2 Δλi2−1 + η 2 δ i −1 (4.40)

The vector ti −' 1 is calculated using (4.35). By substituting the above equations into (4.33) final
expression for Δλi −1 is obtained.
From the above derivation it is clear that in practice we at first employ Normal Update Method
(Chapter 4.4.1) to solve for ti ' and ni'−1 and thereafter we correct the Δλi −1 in order to satisfy
the constraint ti −1 = ti = s .
180

||ri-1|| n’i-1
βλ
α
|| r’i-1|| = ||t’i - s||
||ti-1||
ti-1 ni-1

ti
s
||t’i||

s = step length

p
Fig. 4.4-5 Explicit orthogonal method.

This method is usually utilized to analyze geometrically nonlinear structures, particularly

stability problems. Its main feature is robustness and compared with the "classical" Crisfield
cylinder method (see below) it avoids the problem of the choice of the proper Δλi −1 root (the
condition ti −1 = ti = s while expressing vector length analytically). As for convergence, the
method is comparable to the method 4.4.3, but has the advantage that it preserves the step length.

4.4.4 The Crisfield Method.

The Crisfield method is derived directly from the constraint of constant step length
ti −1 = ti = s The residual Ri −1 is not used in this case and we substitute equations (4.20)
through (4.31) straight into the above constraint. It leads to the following equation for Δλi −1 :

a1Δλi2−1 + a2 Δλi −1 + a3 = 0 (4.41)

where:
a1 = η 2δ TT δ T + β 2

a2 = 2 β 2 ( λi −1 − λstart ) + 2δ TT δ i −1η 2 (4.42)

a3 = β 2 ( λi −1 − λstart ) 2 + η 2δ i T−1δ i −1 − s 2
Equation (4.41) has generally two roots Δλi −1 and hence we must decide which of them to use.
There exist several strategies but ATENA chooses that root Δλi −1 , for which cos( ti −1 , ti ) ≥ 0 (or
higher of them), i.e. direction of new increment as close as possible to direction of the previous
increment (within the same step).
 181

4.4.5 Arc Length Step

The proper step length is of essential importance for good execution performance. It directly
influences the convergence radius on the one hand and the number of required steps on the other.
ATENA uses the following procedure to set (or optimize) s :
(1) Set loading vector q and thus define a reference loading level (within one load
increment).
(2) Structural response to this load in the 1st execution step, the 1st iteration defines step
length s1 in the 1st step. In the subsequent steps the step length is kept fixed or optimized
(based on SET ATENA input command, subcommand
&ARC_LENGTH_OPTIMISATION:
ni −1
si = si −1 (4.43)
n

ni −1
si = 4 si −1 (4.44)
n

n
si = si −1 (4.45)
ni −1
where
si and si −1 is Arc length step length in the current and the previous load increment,
respectively.
n and ni −1 is desired number of iterations and number of iterations in the previous step.
n is typically 5-6.

4.5 Line Search Method

The objective of this method is to calculate the parameter η that was already introduced in the
Chapter 4.4 Arc-Length Method The method can be used either independently or in combination
with Arc length method. The primary reason for introducing a new parameter (i.e. a new degree
of freedom to the set of equations) is to accelerate or to damp the speed of analysis of the load-
displacement relationship.
The basic idea behind η is to minimize work of current out-of-balance forces on displacement
increment.
Let us assume that we have already solved already two points p0 and p0 + η ' δ p and thus we
have also calculated out-of-balance forces g ( p0 ) and g ( p0 + η ' δ ) at these points. The aim of
this method is to set the parameter η so that the work being done by out-of- balance forces at
point p0 + ηδ is minimum.

The work of out-of-balance forces is:

p
Φ ( p ) = Φ ( p0 ) + ∫ g ( p )T dp = minimum (4.46)
po
182

Hence:
∂Φ ( p ) ∂ ⎛ p T ⎞
∂p ∂p
= 0+ ⎜ ∫p g ( p) ⎟ = g ( p )T =0 (4.47)
dη ∂p ⎝ 0 ⎠ ∂η ∂η

Interpolating linearly out-of-balance forces between points p0 and p0 + η ' δ

⎛ g ( p0 + η ' δ ) − g ( p0 ) ⎞ g ( p0 + η ' δ ) − g ( p0 )
g ( p0 + ηδ ) = g ( p0 ) + ⎜ ⎟ || p0 + ηδ − p0 ||= g ( p0 ) + η
⎜ || p0 + η ' δ − p0 || ⎟ η '
⎝ ⎠
(4.48)
and using :
p = p0 + ηδ
∂p (4.49)
=δ
∂η
The final expression for η ' can be derived:
g ( p0 )T δ
η =η ' (4.50)
g ( p0 )T δ − g ( p0 + η ' δ )T δ

Thus, the Line search method can be summarized:

Use any method to calculate displacement increment δ , (see Fig. 4-3 and (4.28)). The
parameter η ' can be set from the last load increment or simply to unity.
Calculate out-of-balance forces for both g ( p0 ) and g ( p0 + η ' δ ) .

Use (4.50) to calculate new value for η .

As all the above equations are nonlinear, the parameter η must be solved by iterations until
g ( p0 + ηδ
≤ a specified energy drop, typically < 0.6 – 0.8 >.
g ( p0 )

Practical experience suggests that the value of parameter η should be kept in interval < 0.1 – 5>.

4.6 Parameter β
The parameter β scales the deformation space p to the loading dimension λ . If β = 0 , the
solution for Δλi −1 is searched on an area of a cylindrical shape of radius equal to step length
s (Crisfield method) and the axis normal to the p (deformation) space. The solution is the point
of intersection of this area and the line, defined by the energy gradients of structure and by the
applied load at point p . If β > 0 , the solution is carried out in the same way on ellipsoidal or
spherical space.
The higher value of β , the higher "weight factor" for changes in loading space compared to
displacement increments.
ATENA currently supports the following formulae for setting and optimization of β (for current
step j ):
 183

Δ( j p )
bergan = (4.51)
Δ jλ
j

β = β ref ( j −1 bergan ) (4.52)

or
j −2 bergan
β = j −1 β (4.53)
j −1 bergan

If ratio of displacements changes Δ ( j p ) to load changes Δ( j λ ) in the last load step increased,
then the equation (4.52) or equation (4.53) will increase or reduce the current value of β in the
current step, thereby puts higher or lower „weight factor“ on loads compared to displacements,
respectively. Hence, the equation (4.52) tend to keep constant importance of loading space
irrespective of displacements, whilst the equation (4.53) tries to keep importance of both spaces
equal. Note that the equation (4.52) or (4.53) corresponds to
BETA_FORCES_DISPLS_RATIO_CONSTANT and BETA_BERGAN_RATIO_CONSTANT
switch in SET input command, respectively.

4.7 References
BATHE, K.J.(1982), Finite Element Procedures In Engineering Analysis, Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey 07632, ISBN 0-13-317305-4.
CRISFIELD, M.A. (1983) - An Arc-Length Method Including Line Search and Accelerations,
International Journal for Numerical Methods in Engineering, Vol.19, pp.1269-1289.
FELIPPA, C. (1966) - Refined Finite Element Analysis of Linear and Nonlinear
Two-Dimensional Structures, Ph.D. Dissertation, University of California, Engineering,
pp.41-50.
RAMM, E. (1981) - Strategies for Tracing Non- linear Responses Near Limit Points, Non- linear
Finite Element Analysis in Structural Mechanics, (Eds. W.Wunderlich,E.Stein, K.J.Bathe)
REKTORYS, K. (1995). Přehled užité matematiky. Prague, Prometheus.
SEAGER, M. K. and A. GREENBAUM (1988). A SLAP for the Masses, Lawrence Livermore
National Laboratory
VONDRACEK, R. (2006) - The Use Of The Sparse Direct Solver In The Egineering
Applications Of The Finite Element Method. Theses for Ph.D. Czech Technical University,
Prague.
DAVIS, T., AMESTOY, P., DUFF, I.S (1995) - An Aproximate Minimum Degree Ordering
Algorithm, Comp. and Information Science Dept., University of Florida, Tech. Report
TR-94-039.
184

5 CREEP AND SHRINKAGE ANALYSIS

Creep and shrinkage are undoubtedly features that have significant influence on concrete
behaviour. Although creep and shrinkage analysis can be neglected in design of most civil
structures, there exist cases, when these phenomena have to be accounted for. The Ref. (Bazant
and Baweja 1999) provides a five levels classification of structures that can serve as a simple
guidelines for making decision, when creep and shrinkage analysis is needed and when it is not
needed. The recognized levels of structures are as follows:
Level 1: Reinforced concrete beams, frames and slabs with span under 20m and heights of up to
30m, plain concrete footings, retaining walls.
Level 2. Prestressed beams or slabs of spans up to 20m, high-rise building frames up to 100m
high.
Level 3. Medium-span box girder, cable-stayed or arch bridges with spans of up to 80m,
ordinary tanks, silos, pavements.
Level 4. Long-span prestressed box-girder, cable-stayed or arched bridges; large bridges built
sequentially in stages by joining parts, large gravity, arch or buttress dams, cooling towers, large
roof shells, very tall buildings.
Level 5. Record span bridges, nuclear containments and vessels, large offshore structures, large
cooling towers, record-span thin roof shells, record-span slender arch bridges.
Full creep and shrinkage analysis is mandatory for design of structures level 4 and 5 and it is
recommended also for the level 3 structures.

5.1 Implementation of creep and shrinkage analysis in ATENA

ATENA software provides a powerful method for creep and shrinkage analysis for most
problems from engineering practice. It is based on so called cross-sectional approach, meaning
that the analysis builds upon creep and shrinkage behaviour of the whole cross section rather
than behaviour of individual material points only. The reason for choosing this method is that at
this moment, there is available numerous models for predicting creep and shrinkage behaviour of
a concrete cross section, whereas there is very low evidence about the same behaviour at
material point level. The second reason is that its accuracy suffices for most analyses from
engineering practice and it is much less expansive in terms of computational cost.

5.1.1 Basic theoretical assumptions

The implemented creep and shrinkage analysis is based on assumption of linear creep, which in
other words means that material compliance function Φ (t , t ') and accompanying function for
shrinkage ε 0 (t ) depends only on material composition, temperature, shape and time at
observation t and at loading t ' . It does not depend on stress-strain conditions. In spite of the
simplifications the provided analysis is in most practical cases sufficiently accurate and it is fast
and efficient. On the other hand, it is applicable only for structures, where stress value does not
exceed about 60% of ultimate strength of concrete. For higher load levels the material
nonlinearity becomes significant and a more elaborate solution has to be employed. The above
simplification applies to time dependent (i.e. long term) material behaviour only. For short-term
behaviour of the material model retains its nonlinearity, i.e. it accounts for phenomena such as
cracks, plasticity.
 185

The creep and shrinkage analysis is based on assumption of Stieltjes integral, which is is written
for the case of 1D analysis in the following form:
t ∂σ
ε (t ) = ∫ Φ(t , t ') dτ + ε 0 (t ) (5.1)
t' ∂τ
where:
t = observation time,
t ' = loading time,
σ (t ) =stress at time t ,
ε 0 (t ) = initial stress-independent strain such as concrete shrinkage,
Φ (t , t ') = compliance function of concrete.

Fig. 5-1 Decomposition of stress history into stress steps (left) or impulses (right).

The sense of Stieltjes integral is given in the above figure.

Equation (5.1) has to be modified for the case of 2 and 3D analyses for practical analyses. This
is done below. It is important to note that (5.1) applies for any stress and strain history and it is
defined in incremental form. It means that at a particular time t stress at t + Δt depends only on
∂σ
current material state at time t and stress increment at time t + Δt , i.e. Δσ = dτ .
∂τ
The final form of the above equations reads:
t ⎛ ∂σ ∂B(σ (τ )) ⎞
ε (t ) = ∫ Φ(t ,τ ) ⎜ B(σ (τ )) + σ (τ ) ⎟ dτ + ε 0 (t ) (5.2)
t'
⎝ ∂τ ∂τ ⎠
where:
σ (t ) = is stress vector at time t , (note the bar atop of a symbol indicates vector),
ε 0 (t ) = vector of initial strains, such as shrinkage,
B(σ (τ )) = matrix accounting for multiaxial stress-strain conditions, including all material
short-term nonlinearities.
Notice the way the equation (5.2) is written. Long-term and short-term material behaviour is
separated. The former is encapsulated in compliance function Φ (t , t ') , whereas the short-term
behaviour is comprised in the matrix B(σ (τ )) . This assumption brings significant simplification
of the creep and shrinkage analysis and it is believed that for most practical analysis the induced
inaccuracy is acceptable.
186

Substituting t = t '+ δ t , δ t → 0 into (5.2) and applying load increment Δσ (t ') = σ (t ') (i.e.
loading from zero level) at time t ' , it can be derived
ε (t '+ δ t ) = Φ (t '+ δ t , t ')B(σ (t '))σ (t ') + ε 0 (t '+ δ t ) (5.3)
Comparison of (5.3) with similar equations for constitutive relations for short-term loading
conditions, i.e. t '+ δ t  t ' , yields instantaneous secant material rigidity matrix:
D(t ') = ( B(σ (t '))Φ (t ', t ') )
−1
(5.4)

The matrix D(t ') corresponds to reciprocal value of the well known secant Young modulus
E (t ') in the case of 1D stress-strain conditions. In the case of plane stress conditions, the matrix
B(σ (τ )) reads (5.5), etc.

⎡ 1 −ν 0 ⎤
⎢
B=⎢ 1 0 ⎥⎥ (5.5)
⎢⎣ sym. 2(1 + ν ) ⎥⎦

5.2 Approximation of compliance functions Φ(t , t ') by Dirichlet

series.
Ref. (Bazant and Spencer 1973) and others show that significant improvement of computational
efficiency can be obtained, if the original material compliance function Φ(t , t ') is during the
creep solution approximated by Dirichlet series Φ '(t , t ') as follows:

n ⎛ ⎟⎞
⎛ t −t ' ⎞
⎜−
1 1 ⎜ ⎜ τμ ⎟
⎠⎟
Φ '(t , t ') = +∑ 1− e ⎝
(5.6)
E (t ') μ =1 Eμ (t ') ⎜ ⎟
⎝ ⎠
where :
τ μ = are so called retardation times,
n = number of approximation functions, i.e. this parameter is related to the input parameter
number of retardation times.
E (t ') = instant Young modulus at time t ' ,
Eμ (t ') =coefficients for the approximation functions.
 187

Fig. 5-2 Approximation of compliance (or retardation) function curve at age t’ at loading by a sum of
exponentials used as shape functions of Dirichlet series

The effect of use of Dirichlet series approximation is depicted in the above figure. A single
approximation exponential is drawn in sub-figure (a), while the whole process of decomposition
of compliance and retardation curves is depicted in the sub-figures (b), (c), respectively.

The incorporation of Dirichlet series Φ '(t , t ') brings the following benefits:
- Creep analysis is independent of material creep prediction model.
- Time integration is exact; hence, less temporal increments are necessary.
- Less demand of computer storage needed for storing data from the previous temporal
steps of the analysis. It suffices to store data from the previous analysis step only, rather
than the complete stresses-strain history of the analyzed structure.

5.3 Step by step method

Equation (5.2) (upon substitution (5.6) is solved numerically. The structure is discretized in
space by finite element method, (described elsewhere in this document). As for time, the solution
is carried out by Step-by-step method (SBS) (Bazant 1988). The structural behaviour is analysed
in several time steps, i.e. in time increments, as it corresponds to (5.2). After some mathematical
manipulations (Jendele and Phillips 1992) the final solution equations read:
i r −1/ 2 ( B )−1 (Δε - Δεi r )
Δσ r = E (5.7)
r -1/ 2 r

ε (tr ) = ε r = ε r −1 + Δε r (5.8)
σ (tr ) = σ r = σ r −1 + Δσ r (5.9)
n
1 1 1
= + ∑ (1 − λμ ,r ) (5.10)
i r −1/ 2
E Er −1/ 2 μ =1 Eμ ,r −1/ 2

⎛ Δt
− r
τ
⎞ τμ
λμr = ⎜1 − e μ ⎟ (5.11)
⎜ ⎟ Δtr
⎝ ⎠
188

⎡ n− i ⎤
Δt

Δεir = ∑ ⎢1 − e μ ⎥ ε μ*r −1 + Δε r0
τ
(5.12)
μ =1 ⎢ ⎥⎦
⎣
ε (tr ) = ε r = ε r −1 + Δε r (5.13)

εμ = e
*
−
Δt r
τμ
εμ +*
λμ E
r
(
i r −1/ 2 Δε − Δεi
r ) (5.14)
r r −1
Eμr −1/ 2

In the above the following notation is used:

r = identification of temporal increments, r ∈ 1..N , where N is number of time
increments for the analysis,
Δtr = tr − tr −1 = time increment,
Δσ r = σ r − σ r −1 = stress increment in time tr ,

ε μ*r = ε μ* (tr ) =internal variables at time tr ,

0 0
ε r = ε (tr ) = shrinkage at time tr ,
1
Er −1/ 2 = ( E (tr ) + E (tr −1 ) ) = constant average secant Young modulus at time incremenent
2
Δtr ,
1 1
(
Eμ r −1/ 2 =Eμ (tr ) + Eμ (tr −1 ) ) = ( Eμr + Eμr −1 ) = constant average value of Dirichlet
2 2
coefficient Eμ at Δtr ,

1
B r −1/ 2 = ( B(tr ) + B(tr −1 ) ) = average value of the matrix B at Δtr .
2
Equation (5.7) thru (5.14) defines all necessary relations to complete the creep and shrinkage
analysis in ATENA. Of course, they are supplemented by relations used by short-term material
constitutive model, i.e. equations for calculating the matrix B.
At each time increment, a typical short-term alike analysis is carried. Difference between the
short-term analysis and the described analysis of one step of the creep and shrinkage is that the
latter one uses especially adjusted Young modulus E i r −1/ 2 and initial strain increments Δεi r to
account for creep and shrinkage. After each step these have to be updated. It involves mainly
update of λμr and Δεi r . With these values a new E i r −1/ 2 is calculated and the next temporal
analysis step is carried out.

5.4 Integration and retardation times

Appropriate selection of retardation and integration times is of crucial importance for accurate
and efficient creep and shrinkage analysis. The choice of retardation times has direct impact on
accuracy of approximation an original compliance function by Dirichlet series, see Equation
(5.6) and Fig. 5-2, whilst the choice of integration times affects accuracy of the approximation of
loading function of the structure, see Equation (5.1) and Fig. 5-1. If number of the times is too
low, some important features of concrete behaviour can be disregarded. The opposite extrem, i.e.
 189

using too many of retardation or integration times results in worthless lengthy solution of the
problem.
The ATENA software respects recommendation in (Bazant and Whittman 1982). Retardation
times are spread uniformly in log(t ) space and they are automatically calculated as follows:
μ −1
τ μ = 10 m
τ1 , μ = 1, 2..n (5.15)

In the above m is number of retardation times per log(t ) unit, m ≥ 1 . By default this constant is
in ATENA set to 1. If required, a more detailed approximation is possible, i.e. any value m > 1
can be used. In the program this parameter is input as a number of retardation times per time unit
in logarithmic scale. For a typical concrete creep law a certain optimal value can be determined
and it is independent of a structure being analyzed. Note however, that the value depends on the
choice of time units.
Example: If the retardation times parameter is set to 2, the creep law will be approximated by
two approximation points for the time interval between 0 - 1 day, two points for the interval 1 -
10 days, then two points for 10 - 100 days, etc.
Therefore, the proper values will depend on the choice of time units. If the time unit is a day, the
recommended value is 1 - 2.
Start time τ 1 must be chosen sufficiently low, so that Dirichlet series can account for processes
in very young concrete, right after its loading has been applied. As a default, ATENA uses the
expression τ 1 = 0.1 t ' .
As for the upper limit for τ μ , it is required:

t
τn ≥ (5.16)
2
The above limits are applicable for the case, when the coefficients Eμ (t ') of Dirichlet series in
(5.6) are calculated by Least-square method (Jendele and Phillips 1992).
ATENA also supports alternative way of calculation of the coefficients Eμ (t ') of Dirichlet series
in (5.6). In this case, Inverse Laplace transformation (Bazant and Xi 1995) is used instead. This
method requires τ 1 → 0 , typically 1E-3 and
τn ≥ t (5.17)
Comparing the above two approaches, it can be said that Least-square method yields
approximation of the compliance function at discrete times, whereby Inverse transformation is
based on continuous approach. In some cases Least-square method results in better convergence
behaviour, however it sometimes suffers from numerical problems during calculation due to ill-
posed problem for solution of Eμ (t ') . It is left to experience and engineering judgment to decide,
which of the method is more appropriate for a particular solution.

Integration times or sample times tr are calculated in similar way. In this case, the times are
uniformly spread in log(t − t ') time scale. They are generated starting from the 1st loading time
t ' . Hence , we can write
r −1
tr = 10 l
( t1 − t ') (5.18)
190

where l ≥ 2 is number of time increments per unit of log(t − t ') and t1  t '+ 0.1 days. Each new
major load increment or decrement causes the generation procedure (5.18) must start again from
small time increments. This parameter defines the number of time steps, the program will use to
integrate the structural behavior. Creep or other nonlinear effects will cause a redistribution of
stresses inside the structure. In order to properly capture such processes a sufficiently small time
steps are needed. Its definition depends on the type of the analyzed structure as well as on the
choice of time units. For typical reinforced concrete structures and for the time unit being a day,
it is recommended to set this parameter to 2. This will mean that for each load interval longer
then 1 day, two sub-steps will be added. For a load that is interval longer then 10 days, 4 sub-
steps will be added. For an interval longer than 100 days, it will be 6 sub-steps, etc.

The creep and shrinkage analysis in ATENA requires that the user set number of retardation
times m and number of time increments l per unit of log time, (unless the default values are
OK). He/she also specifies time span, i.e. τ 1 and τ n . Then, retardation times are generated, i.e.
an appropriate command is issued. It follows to set stop time of the analysis. Usual input data
describing structural shape, material etc. are given thereafter, however, there are three important
differences from time-independent analysis:
1. Material model for concrete contains data for long-term as well as for short-term material
model.
2. Step data must include information about time, at which the step is applied.
3. It is recommended to input data for all intended load time steps prior the steps are
executed. It helps the generation of integration (intermediate) times
Intermediate time steps, i.e. times tr as well retardation times are generated automatically. The
analysis proceeds until the stop time is reached. If no stop time is specified, it is assumed to be
time of the last load step. If the time span for retardation times does not covered step load times,
the solution is aborted, giving an appropriate error message.

5.5 Creep and shrinkage constitutive model

In the above sections, it was silently assumed that long-term part of the material model, i.e.
compliance function Φ (t , t ') and shrinkage function ε r0 for concrete, is known and it was
shown, how it is utilized within creep and shrinkage analysis. It is primary intention of this
section to describe what long-term creep and shrinkage prediction models are implemented in
ATENA and how they should be used.
Generally speaking, ATENA applies no restriction on kind and shape of both Φ (t , t ') and ε r0 , as
it adopts SBS method solution algorithm, in which compliance function is approximated by
Dirichlet series. Hence, most widely recognized creep prediction models could be implemented.
The CCStructureCreep module currently supports the following models:

1. CCModelACI78 (ACI_Committee_209 1978), recommended by ACI, by now already

obsolete,
2. CCModelCEB_FIP78 (Beton 1984), recommended by CEB committee, by now already
obsolete,
 191

3. CCModelB3 (Bazant and Baweja 1999), developed by Bazant and Al Manaseer in 1996,
very efficient model recognized world-wide,
4. CCModelB3Improved, same as the above, improved to account for temperature history,
probably the best model available in ATENA,
5. CCModelCSN731202, model developed by CSN 731202 Code of practice in Czech
Republic,
6. CCModelBP1_DATA (Bazant and Panula 1978a; Bazant and Panula 1978d; Bazant and
Panula 1978b; Bazant and Panula 1978c), relatively efficient and complex model; now it
is superceeded by CCModelBP_KX or CCModelB3,
7. CCModelBP2_DATA (Bazant and Panula 1978e), simplified version of the above
model,
8. CCModelBP_KX (Bazant and Kim 1991a; Bazant and Kim 1991b; Bazant and Kim
1991c; Bazant and Kim 1991d), powerful model with accounts for humidity and
temperature history etc., for practical use it may-be too advanced,
9. CCModelGeneral general model into which experimentally obtained Φ (t , t ') and ε r0
function can be input.

The following data summarized input parameters for the supported models. Note that some
models allow improved prediction based on laboratory data. If it is the case, the model input the
corresponding experimentally measured values. Also, some model can account for material point
history of humidity h(t ) and temperature T (t ) . Again, a model supports this feature, if it can
input adequate data.
Table 5.5-1 : List of material parameters for creep and shrinkage prediction – definition and
description

Parameter name Description Units Default

Concrete. type Type of concrete according to ACI. Type 1 1
is Portland cement etc. Types 1,3 accepted
for static analysis, types 1-4 accepted for
transport analysis.
Thickness S / V Cross section thickness defined as ration of length 0.0767m
section’s surface to volume
Strength f cyl 28 Material cylindrical strength in compression stress 35.1MPa
at time 28 days
Young m. E28 Short-term material Young modulus at 28 stress F ( f cyl 28 )
days, i.e. inverse compliance at 28.01 days
loaded at 28 days
Ambient humid. h Ambient relative humidity. Accepted range 0.78
(0.4..1).
Ratio ac Total aggregate/cement weight ratio. 7.04

Ratio wc Water/cement weight ratio. 0.63

192

Ratio as Total aggregate/find sand weight ratio. 2.8

as = sa −1 .

Ratio sa Fine/total aggregate weight ratio. sa = as −1 0.4

Ratio g s Coarse gravel/fine aggregate weight ratio. 1.3

Ratio sc Fine aggregate/cement weight ratio. 1.8

Shape factor Cross section shape factor. It should be 1, 1.25

1.15, 1.25, 1.3, 1.55 for slab, cylinder,
square prism, sphere, cube, respectively.
Slump Result of material slump test. length 0.012m
Air content Material volumetric air content. % 5
Cement mass Weight of cement per volume of concrete mass/ 320kg/
length3 m3
Concr. density Material density used to evaluate strength mass/ 2125kg/
and Young modulus at 28 days.. length3 m3
Curing type Curing conditions. It can be either in water AIR
(i.e. WATER) or air under normal
temperature (i.e. WATER) or steamed
curing (i.e. STEAM).
End of curing Time at beginning of drying, i.e. end of days 7
curing.
Current time t
I/D

Load time t’
Tot.water loss w Total water loss (up to zero humidity and kg N/A
infinite time). It is measured in an oven in a
laboratory and it is used to enhance
prediction of shrinkage infinite tsh 2 ∞
(Bazant and Baweja 1999). This value is in
turn used to elaborate drying creep and
shrinkage prediction of the model. If it is
not specified, the model prediction
enhancement is not activated. It can be used,
if water loss w(t) are input as well.
Water loss w(t) Water losses at time t; measured at a kg N/A
Improvem.

laboratory. It is used to enhance

drying creep and shrinkage
prediction. See also description of
total water loss w.
 193

Shrink. ε 0 (t ) Measured shrinkage at time t. It is used to N/A

enhance drying creep and shrinkage
prediction. See also description of total
water loss w.
Compl. Φ(t , t ') Measured material compliance at time t. It is 1 N/A
used to improve overall creep and shrinkage /stress
prediction of the model.
Humidity h(t ) History of humidity in a material point. N/A
Value at time t. Some material models can
use these values to account for real temporal
humidity and temperature conditions.
Although the data can be input manually,
i.e. to group material points with similar
humidity and temperature history into a
group and dedicate a distinct material for
Hist.

that group, it is prepared for full automatic

processing being currently in development.
It will automatically link heat and humidity
transport analysis with the static analysis
using one of available creep and shrinkage
prediction model. Applicable range (0.4..1).
Temperat. T (t ) History of temperature in a material point. Celsia
See also description of h(t )
Compl. Φ(t , t ') Measured compliance at time t loaded at 1/
time t’. This and the next two parameters
stress
should be used, if known (measured)
compliance functions are to be employed
in ATENA creep and shrinkage analysis.
Hence, no prediction is done and the given
Direct

data are only used to calculate the

parameters of Dirichlet series
approximation.
Shrink. ε 0 (t ) Measured shrinkage at time t . See the N/A
parameter above.
Strength f cyl (t ) Measured shrinkage at time t . See the
parameter above

Table 5.5-2: Input parameters needed by individual creep and shrinkage prediction models

Model name B3 B3- BP- CEB ACI CSN BP1 BP2 Gen
impr KX eral
Model No. 3 4 8 2 1 5 6 7 9
Concrete. Type x x x x x x x
Thickness S / V x x x x x x x x
194

Strength f cyl 28 x x x x x x x x

Young m. E28 x x x x x

Ambient humid. h x x x x x x x x
Ratio ac x x x x x x

Ratio wc x x x x x x

Ratio as

Ratio sa x x

Ratio g s x x

Ratio sc x x

Shape factor x x x x x
Slump x
Air content x
Cement mass x
Concr. density x x x x
Curing type x x x x x x
End of curing x x x x x x x x
Current time t x x x x x x x x x
I/D

Load time t’ x x x x x x x x x
Tot.water loss w x x x
Water loss w(t) x x
Improvem.

Shrink. ε 0 (t ) x x x x x x x x

Compl. Φ(t , t ') x x x

Humidity h(t ) x x x
Hist.

Temperat. T (t ) x x x
Compl. Φ(t , t ') x
Direct

Shrink. ε 0 (t ) x

Strength f cyl (t ) x

The above parameter “Concrete type” actually referes to a cement type used in concrete subject
to creep analysis. The following table brings description of widely recognized cement types.
Note that only types 1,3 are supported in Atena static analysis. The transport analysis in Atena
recognizes types 1-4. The remaining types aredescribed just for information.
 195

Cement type Description

General purpose cements suitable for all uses where the special
I and Type IA 1
properties of other types are not required.
Type II cements contain no more than 8% tricalcium aluminate
1
II and Type IIA (C3A) for moderate sulfate resistance. Some Type II cements meet
the moderate heat of hydration option of ASTM C 150.
Chemically and physically similar to Type I cements except they
III and Type IIIA1
are ground finer to produce higher early strengths.
Used in massive concrete structures where the rate and amount of
IV heat generated from hydration must be minimized. It develops
strength slower than other cement types.
V Contains no more than 5% C3A for high sulfate resistance.
IS (X) 2 Portland blast furnace slag cement
IP (X)2 Portland-pozzolan cement.
GU 3
General use

HE3 High early strength

3
MS Moderate sulfate resistance
3
HS High sulfate resistance
MH3 Moderate heat of hydration
LH3 low heat of hydration

5.6 References

ACI_COMMITTEE_209 (1978). Prediction of Creep, Shrinkage and Temperature Effects in

Concrete Structures. Detroit, 2nd draft, ACI.
BATHE, K. J. (1982). Finite Element Procedures in Engineering Analysis. Englewood Cliffs,
New Jersey 07632, Prentice Hall, Inc.
BAZANT, Z. AND T. SPENCER (1973). "Dirichlet Series Creep Function for Aging Concrete."
ASCE Journal of Engineering and Mechanical Division: 367-387.

1
Air-entraining cements
2
Blended hydraulic cements produced by intimately and uniformly intergrinding or blending two or more types of
fine materials. The primary materials are portland cement, ground granulated blast furnace slag, fly ash, silica fume,
calcined clay, other pozzolans, hydrated lime, and pre-blended combinations of these materials. The letter “X”
stands for the percentage of supplementary cementitious material included in the blended cement. Type IS(X), can
include up to 95% ground granulated blast-furnace slag. Type IP(X) can include up to 40% pozzolans.
3
All portland and blended cements are hydraulic cements. "Hydraulic cement" is merely a broader term. ASTM C
1157, Performance Specification for Hydraulic Cements, is a performance specification that includes portland
cement, modified portland cement, and blended cements. ASTM C 1157 recognizes six types of hydraulic cements.
196

BAZANT, Z. P. (1988). Mathematical Modeling of Creep and Shrinkage of Concrete. New

York, John Wiley & Sons.
BAZANT, Z. P. AND S. BAWEJA, EDS. (1999). Creep and Shrinkage Predicition Model for
Analysis and design of Concrete Structures: Model B3. Creep and Shrinkage of Concrete, ACI
Special Publicatino.
BAZANT, Z. P. AND J. K. KIM (1991). "Improved Prediction Model for Time-Dependent
Deformation of Concrete: Part 1- Shrinkage." Materials and Structures 24: 327-345.
BAZANT, Z. P. AND J. K. KIM (1991). "Improved Prediction Model for Time-Dependent
Deformation of Concrete: Part 2- Basic Creep." Materials and Structures 24: 409-421.
BAZANT, Z. P. AND J. K. Kim (1991). "Improved Prediction Model for Time-Dependent
Deformation of Concrete: Part 3- Creep at Drying." Material and Structures 25: 21-28.
BAZANT, Z. P. AND J. K. KIM (1991). "Improved Prediction Model for Time-Dependent
Deformation of Concrete: Part 4- Temperature Effects." Material and Structures 25: 84-94.
BAZANT, Z. P. AND L. PANULA (1978). "Practical Prediction of Time-dependent
Deformations of Concrete; Part 1: Shrinkage." Material and Structures 11 (65): 301-316.
BAZANT, Z. P. AND L. PANULA (1978). "Practical Prediction of Time-dependent
Deformations of Concrete; Part 3: Drying Creep." Material and Structures 11 (65): 415-423.
BAZANT, Z. P. AND L. PANULA (1978). "Practical Prediction of Time-dependent
Deformations of Concrete; Part 4: Temperature Effect on Basic Creep." Material and Structures
11 (66): 424-434.
BAZANT, Z. P. AND L. PANULA (1978). "Practical Prediction of Time-dependent
Deformations of Time-dependent Deformation of Concrete; Part 2: Basic Creep." Material and
Structures 11 (65): 317-328.
BAZANT, Z. P. AND L. PANULA (1978). Simplified Prediction of Concrete Creep and
Shrinkage from Strength and Mix. Struct. Engng. Report No. 78-10/6405. Evanston, Illinoins,
Northwestern University, Dep. of Civ. Engng.
BAZANT, Z. P. AND F. H. WHITTMAN (1982). Creep and shrinkage in Concrete Structures.
New York, John Wiley & Sons.
BAZANT, Z. P. AND Y. XI (1995). "Continous Retardation Spectrum for Solidification Theory
of Concrete Creep." Journal of Engineering Mechanics 121(2): 281-287.
BETON, C. E.-I. D. (1984). CEB Design Manual on Structural Effects on Time Dependent
Behaviour of Concrete. Saint Saphorin, Switzerland, Georgi Publishing Company.
HAGEMAN, L. AND D. YOUNG (1981). Applied Iterative Methods. New York, Academic
Press.
JENDELE, L. AND D. V. PHILLIPS (1992). "Finite Element Software for Creep and Shrinkage
in Concrete." Computer and Structures 45 (1): 113-126.
REKTORYS, K. (1995). Přehled užité matematiky. Prague, Prometheus.
SEAGER, M. K. AND A. GREENBAUM (1988). A SLAP for the Masses, Lawrence Livermore
National Laboratory.
 197

6 TRANSPORT ANALYSIS

As pointed out in the previous section, creep material behaviour of concrete strongly depends on
moisture and temperature conditions. Some constitutive models for creep in ATENA, (see Table
5.5-2), can pay regards to these factors and based on previously computed moisture and
temperature histories within the structure they can predict concrete behaviour more accurately.
This section describes a module called CCStructuresTransport that is used to calculate the
histories. The more accurate creep analysis then typically consists of two steps: firstly execute
CCStructuresTransport module and calculate the moisture and humidity histories of the structure
and secondly execute CCStructuresCreep module to carry out the actual static analysis. Of
course, for both analyses we have to prepare an appropriate model. Export/Import of the results
between the modules is already done by ATENA automatically.
To be exact, both the transport and static analysis should be executed simultaneously but as
moisture and temperature transport does not depend significantly on structural deformations, i.e.
coupling of the analyses is low, the implemented “staggered” solution yields sufficiently
accurate results.

The governing equations for moisture transport read (for representative volume REV] :

∂w ∂ ( we + wn )
= = −div( J w ) (6.1)
∂t ∂t

where:
w is total water content defined as a ratio of weight of water at current time t to weight of
water and cement at time t0 = 0 in REV, [mass/mass], e.g. [kg/kg]
we , wn = stands for the amounts of free and fixed (i.e. bound) water contents, [mass/mass],
J w = moisture flux, [length*mass/ (time*mass)]. e.g. [m/day],
t =time, [time], e.g. [day].

The moisture flux is computed by

J w = − Dw∇we (6.2)
where
Dw is moisture diffusivity tensor of concrete [m2/day],
∇ is gradient operator.
Note that in (6.2) only diffusion of water vapor is considered. Moisture advection is negligible.
The equations (6.1) and (6.2) can be also written as being dependent on w or relative moisture h .
A relationship between h and w is given by
198

w = w(h) (6.3)

Using (6.3) Equation (6.2) can be written as follows

J w = − Dh ∇h (6.4)

A special attention must be paid to calculation of the above time derivatives and integration of
the governing equations. For example, in case of usual Gauss integration and use of exact time
derivatives the solution may suffer from mass losses. To remedy the problem the
CCStructuresTransport module integrates the structure, i.e. all the individual finite elements in
nodes and time derivatives are calculated numerically (Jendele 2001). This integration is similar
to use of finite volume method, which is also known to be robust against the mass losses.

Heat transfer is governed by similar equation

∂ ∂T
∂t
( CT (T − Tref ) ) = CT
∂t
= − div( J T ) (6.5)

where
CT is heat capacity [J/(K.m3)],
J T is heat flux [J/(day.m2)].
If hydration we want to add heat Qh (t ) , which expresses amount of hydration heat within unit
⎡ J ⎤
volume i.e Qh , ⎢ 3 ⎥ , Equation (6.5) changes to
⎣m ⎦
∂ ∂T ∂Qh
∂t
( CT (T − Tref ) + Qh ) = CT
∂t
+
∂t
= −div( J T ) (6.6)

⎡ J ⎤
Heat flux J T , ⎢ 2 ⎥ is calculated by
⎣m s⎦
J T = − KT grad (T ) (6.7)

and KT stands for heat conductivity, e.g. [J/(day.m.K)].

Note that Equation (6.5) accounts for heat transport due to conduction only. Heat advection is
negligible. In (6.5) we can also neglect hydration heat, because in large times its impact for heat
transfer is small. On the other hand, we cannot neglect concrete moisture consumption due to
hydration process. According to (Bazant and Thonguthai 1978; Bazant 1986) hydration water
content wh can be calculated by:
 199

1
⎛ t ⎞3
wn = wh ≈ 0.21 c ⎜ e ⎟ (6.8)
⎝ τ e + te ⎠

where
τ e = 23 days, te is equivalent hydration time in water at temperature 25 0C that corresponds to
the same degree of hydration subject to real age, moisture and temperature conditions of the
material. The parameter c relates to the amount of cement and is calculated by(6.53). If
temperature ranges from 0 to 100 0C , te is computed by

te = ∫ β h βT dt (6.9)

where dt is time increment after the mould has been removed and coefficients βT , β h are
calculated by
1
βh = (6.10)
1 + (3.5 − 3.5h) 4

⎡U ⎛ 1 1 ⎞ ⎤
βT = exp ⎢ h
⎜ − ⎟⎥ (6.11)
⎢⎣ ⎜⎝ Tl0 Tl ⎟⎠ ⎥⎦
R

Uh
In the fraction the symbol U h stands for activation energy of hydration and R is gas
R
Uh
constant. According to (Bazant 1986) = 2700 0 K . Tl, Tl0 are real and reference concrete
R
temperature expressed in 0 K . The reference temperature is given by

Tl0 = 273.15 + 25 (6.12)

The following figure depicts relationship between real t and equivalent time te for the case of
constant temperature T = 15 0C and moisture h = 0.8 . In practice, this relationship is rarely
linear, because with increase of time the amount of fixed water (due to hydration) wh is
increasing as well and it involves gradual decrease of relative moisture h .
200

Equivalent time for h=0.8, T=15 oC

Fig. 6-1 Equivalent vs. real time relationship

The amount of water that was needed for hydration of concrete according to Equation (6.8) for
the case of c = 300 kg is shown below:

wh for c = 300kg

te [days]

Fig. 6-2 Moisture consumed by hydration as a function of equivalent time

 201

6.1 Numerical solution of the transport problem – spatial

discretisation

The transport governing equations for a typical engineering problem are too complex for
analytical solution. Hence, similar to other ATENA engineering modules, finite element method
is used also for the CCStructuresTransport module. The transport problem gets spatially and
temporarily discretized and then the resulting set of nonlinear algebraic equations is solved by a
special iterative solver. This section is dedicated to detailed description of the former type of
discretisation.

The solution is based on Equations (6.1) thru(6.7). Note that the unknown variables are

h = h(t ); T = T (t ); w = w(h, T ); wh = wh (h, T , t ) (6.13)

and they are to be discretized. Let the left-hand side part of (6.1) and (6.4) is denoted
LHSh , LHST , respectively. The subscript h and T indicates moisture and temperature flux.
Similar subscripts are also used for right-hand-side of the equations, RHSh , RHST . Notice that
RHS expressions do not include the divergence operator!

∂
LHSh = ( w + wh ) (6.14)
∂t

∂T ∂Qh
LHST = CT + (6.15)
∂t ∂t

RHS h = − J w = − J h (6.16)

RHST = − J T (6.17)

The strip over an entity in the above equations means that the entity is vector. (Scalar entities do
not have the strip). The fluxes J w = J h are identical, i.e. the subscript w indicates also moisture
phase. Using the above notation Equations (6.1) and (6.5) can be written as follows

LHS h = div( RHS h )

(6.18)
LHST = div( RHST )
202

The LHSh includes time derivative of moisture. It is computed using the following expressions:

wh = wh (te )

∂te
= β h βT (6.19)
∂t

∂wh ∂wh ∂te ∂wh

= = β h βT
∂t ∂te ∂t ∂te

For the next derivation, let us write Equations(6.14), (6.15) in a general form:

∂h ∂w ∂T
LHS h = chh + chw + chT + ch 0
∂t ∂t ∂t
(6.20)
∂h ∂w ∂T
LHST = cTh + cTw + cTT + cT 0
∂t ∂t ∂t

and equations(6.16), (6.17)

RHSh = [ khh ] ∇h + [ khw ] ∇w + [ khT ] ∇T + kh 0

(6.21)
RHST = [ kTh ] ∇h + [ kTw ] ∇w + [ kTT ] ∇T + kT 0

where square bracket indicates that the enclosed entity is a matrix [ ].

Comparing (6.20) with (6.1) and (6.5) we find that

chh = chT = 0; chw = 1; ch 0 ≠ 0

(6.22)
∂Qh
cTh = cTw = 0; cTT ≠ 0; cT 0 = ≠0
∂t

The parameter cTT is in ATENA an input material parameter, ch 0 is computed from(6.19), i.e.
∂w ∂w ∂w
ch 0 = h β h βT . The solution also includes expressions ≠ 0; . Their values depend on a
∂te ∂h ∂T
 203

constitutive model being used in the solution. For more information please refer to Section
Material constitutive model.

For right-hand sides we can write in similar manner:

[ khw ] = [ khT ] = [0] ; [ khh ] ≠ 0; kh 0 = 0

(6.23)
[ kTh ] = [ kTw ] = [0] ; [ kTT ] ≠ 0; kT 0 = 0

The parameter [ kTT ] is a material input parameter, [ khh ] is calculated from a constitutive model,
see the next section.
For the next derivation, let us assume discretisation of the unknown variables as follows. Remind
that these are in the governing equations integrated in finite nodes, (Celia, Bouloutas et al. 1990;
Celia and Binning 1992).

T
h = NT h; ∇h = ⎡⎣∇N ⎤⎦ h

T
w = N T w; ∇w = ⎡⎣∇N ⎤⎦ w (6.24)

T
T = NT T ; ∇T = ⎡⎣∇N ⎤⎦ T

where
h , w, T stands for vectors of the corresponding entities. The vectors have dimension n equal
to number of finite nodes of the problem.

N is vector of interpolation, (i.e. shape) functions,

⎡ ∂N1 ∂N 2 ∂N n ⎤
⎢ ∂x ...
⎢ ∂x ∂x ⎥⎥
∂N ∂N 2 ∂N n ⎥
⎡⎣∇N ⎤⎦ = ⎢⎢ 1
T
...
∂y ∂y ∂y ⎥
⎢ ⎥
⎢ ∂N1 ∂N 2 ∂N n ⎥
⎢⎣ ∂z ...
∂z ∂z ⎥⎦

Using (6.24) Equations (6.20) and (6.21) can be written in the form
204

∂h ∂w ∂T
LHSh = chh N T + chw N T + chT N T + ch 0
∂t ∂t ∂t
(6.25)
∂h ∂w ∂T
LHST = cTh N T + cTw N T + cTT N T + cT 0
∂t ∂t ∂t

and

RHSh = [ khh ] ⎡⎣∇N ⎤⎦ h + [ khw ] ⎡⎣∇N ⎤⎦ w + [ khT ] ⎡⎣∇N ⎤⎦ T + kh 0

T T T

(6.26)
RHST = [ kTh ] ⎡⎣∇N ⎤⎦ h + [ kTw ] ⎡⎣∇N ⎤⎦ w + [ kTT ] ⎡⎣∇N ⎤⎦ T + kT 0
T T T

The resulting set of equations are solved iteratively using finite element method, see
(Zienkiewicz and Taylor 1989), (weak formulation, in which the shape functions N are used as
weight function):

∫ N ( LHS
V
h )
− div( RHS h ) dV = 0

(6.27)

∫ N ( LHS
V
T )
− div( RHST ) dV = 0

where V is volume of the analyzed structure. Each of the above equations represents a set of
equations with dimension equal to number of finite nodes n. Note that div( RHS h ) and
div( RHST ) are scalars !
In the next derivation the two parts of (6.27) are dealt with separately.

⎛ ∂h ∂w ∂T ⎞
∫ N ( LHS ) dV = ∫ N ⎜⎝ c
V
h
V
hh NT
∂t
+ chw N T
∂t
+ chT N T
∂t
+ ch 0 ⎟ dV =
⎠

∂h ∂w
∫c
V
hh NN T dV + ∫ chw NN T dV
∂t V ∂t
+ ... ∫c
V
h0 NdV = (6.28)

∂h ∂w
[cchh ] + [ cchw ] + ...cc h 0
∂t ∂t
 205

⎛ ∂h ∂w ∂T ⎞
∫ N ( LHS ) dV = ∫ N ⎜⎝ c
V
T
V
Th NT
∂t
+ cTw N T
∂t
+ cTT N T
∂t
+ cT 0 ⎟ dV =
⎠
(6.29)
∂h ∂w
[ ccTh ] + [ ccTw ] + ...ccT 0
∂t ∂t

and the matrices [ cc ] are calculated by

[cchh ] = ∫ chh NN T dV ; [ cchw ] = ∫ chw NN T dV ; ... cc h 0 = ∫ ch 0 NdV

V V V

(6.30)
[ccTh ] = ∫ cTh NN T dV ; [ ccTw ] = ∫ cTw NN T dV ; ... ccT 0 = ∫ cT 0 NdV
V V V

The second part of (6.27) are calculated using Green theorem (6.36):

∫ N ( −div( RHS ) ) dV = − v∫ N ( n )
RHS h dS + ∫ ⎡⎣∇N ⎤⎦ RHS h dV =
T
h s
V S V

= − v∫ N nsT
S
([k ] ⎡⎣∇N ⎤⎦ h + [k ] ⎡⎣∇N ⎤⎦ w + [k ] ⎡⎣∇N ⎤⎦ T + k ) dS +
hh
T
hw
T
hT
T
h0
(6.31)

+ ∫ ⎡⎣∇N ⎤⎦ ( [ k ] ⎡⎣∇N ⎤⎦ h + [ k ] ⎡⎣∇N ⎤⎦ w + [ k ] ⎡⎣∇N ⎤⎦ T + k )dV

T T T
hh hw hT h0
V

where S is structural surface (with possibly defined boundary conditions).

In case of heat transfer we can derive all the equations in a similar way. In analogy to (6.30) let
us introduce matrices [ kk ]
206

[ kkhh ] = ∫ ⎡⎣∇N ⎤⎦ [ khh ] ⎡⎣∇N ⎤⎦

T
dV
V

[ kkhw ] = ∫ ⎡⎣∇N ⎤⎦ [ khw ] ⎡⎣∇N ⎤⎦

T
dV
V

kk h 0 = ∫ ⎡⎣∇N ⎤⎦ kh 0 dV
V (6.32)
...

[ kkTT ] = ∫ ⎡⎣∇N ⎤⎦ [ kTT ] ⎡⎣∇N ⎤⎦

T
dV
V

kk T 0 = ∫ ⎡⎣∇N ⎤⎦ kT 0 dV
V

and also

[ J hh ] = v∫ N nsT [ khh ] ⎡⎣∇N ⎤⎦ dS

T

[ J hw ] = v∫ N nsT [ khw ] ⎡⎣∇N ⎤⎦ dS

T

...

[ JTT ] = v∫ N nsT [ kTT ] ⎡⎣∇N ⎤⎦ dS

T

J h 0 = v∫ N nsT kh 0 dS
S

J T 0 = v∫ N nsT kT 0 dS
S
(6.33)

Using (6.28) to (6.33) the original governing equations (6.27) can be written as follows:
 207

∂h ∂w ∂T
[cchh ] + [ cchw ] + [ cchT ] + cc h 0 + [ kkhh ] h + [ kkhw ] w + [ kkhT ] T + kk h 0 =
∂t ∂t ∂t
= [ J hh ] h + [ J hw ] w + [ J hT ] T + J h 0
(6.34)
∂h ∂w ∂T
[ccTh ] + [ ccTw ] + [ ccTT ] + ccT 0 + [ kkTh ] h + [ kkTw ] w + [ kkTT ] T + kk T 0 =
∂t ∂t ∂t
= [ J Th ] h + [ J Tw ] w + [ J TT ] T + J T 0

After sorting the unknown variables h , T by finite nodes into a single vectorψ , Equation (6.34)
will read

∂ψ
[cc ] + kk ψ + cc 0 + kk 0 = [ J ]ψ + J 0 (6.35)
∂t

The right-hand side (6.35) is non-zero only for non-zero prescribed boundary conditions and
hence it has character of “load” vector in a static analysis.

In (6.31) we used Green theorem. It states:

∫ u div(v )dV = v∫ u n v dS − ∫ ⎡⎣∇u ⎤⎦ v dV

T
s
V S V

(6.36)

∫ u div(v )dV = v∫ u n v dS − ∫ ⎡⎣∇u ⎤⎦ v dV

T
s
V S V

where

⎡ ∂u ∂u ∂u ⎤
⎡⎣∇u ⎤⎦ = ⎢
⎣ ∂x ∂y ∂z ⎥⎦

⎡ ∂u1 ∂u1 ∂u1 ⎤

⎢ ∂x ∂y ∂z ⎥
⎢ ⎥
⎢ ∂u2 ∂u2 ∂u2 ⎥
⎡⎣∇u ⎤⎦ = ⎢ ∂x ∂y ∂z ⎥
⎢ ⎥
⎢ ... ... ... ⎥
⎢ ⎥
⎢ ∂un ∂un ∂un ⎥
⎢⎣ ∂x ∂y ∂z ⎥⎦ (6.37)
208

6.2 Numerical solution of the transport problem – temporal

discretisation

The heat and moisture transfer governing equations (6.35) can be written in the form:

∂ t +Δt
t +Δt
K t +Δtψ +t +Δt C
∂t
( ψ ) =t +Δt J (6.38)

where t +Δt K , t +Δt C = are unsymmetrical problem matrices defined in the previous section,
t +Δt
J =vector of concentrated nodal fluxes (both moisture and heat) and t +Δtψ is vector of
unknown variables. All of these apply for time t + Δt . Equation (6.38) is solved iteratively. i.e.
the vector t +Δtψ is searched for in the incremental form:

t +Δt
ψ =t +Δt (i ) ψ =t +Δt (i −1) ψ + t +Δt (i ) Δψ (6.39)

where index (i )
indicates number of iteration and t +Δt ( i )
Δψ is increment of the unknowns for time
t + Δt and iteration ( i ) :

t +Δt ( i )  -1 t +Δt (i ) J
Δψ =t +Δt (i −1) K (6.40)

 -1 and t +Δt (i ) J is derived from

The matrix and vector t +Δt (i −1) K t +Δt ( i −1)
K -1 , t +Δt (i −1)C-1 and Δt based
on temporal integration method being used:

CCStructureTransport module currently supports θ Crank Nicholson (Wood. 1990) and Adams-
Bashforth (Diersch and Perrochet 1998) integration scheme. The former scheme is linear, i.e.
it’s a first-order integration procedure. The latter scheme is a second-order integration procedure.
It is supposed to be more accurate, however, it is also more CPU and RAM expensive and it is
more difficult to predict its real behaviour. Hence, the θ Crank Nicholson scheme is typically
preferred. It has been more tested and verified in the CCStructureTransport module and thereby
it is more recommended.
 209

6.2.1 θ -parameter Crank Nicholson scheme

This scheme comprises a number of well established integration procedures. It depends, what
value of the parameter θ is used. The set of equations (6.38) is solved for time t + Δt θ ,
whereby the vector of unknown variables is calculated as a linear combination of the
corresponding vectors at time t and t + Δt . Hence

t +Δt
ψ =t ψ (1 − θ ) + t +Δt ψθ (6.41)

Depending on a particular value of the parameter θ we get the well known Euler implicit
integration (for θ =1), trapezoidal Crank Nicholson scheme (for θ =0.5), Galerkin integration
method (for θ =2/3) or even Euler explicit scheme (for θ =0), which is only conditionally stable.

Solution predictor:
∂ tψ
t +Δt
ψ =t ψ + Δt (6.42)
∂t
Solution corrector:
∂ t +Δtψ 1
= ( t +Δtψ −t ψ ) (6.43)
∂t Δt

 Ji .
Using the above after some mathematical manipulation we derive final expressions for K,
These read:

 = ⎛ Kθ + 1 C ⎞
K ⎜ ⎟
⎝ Δt ⎠

1
Ji = J − K (θ t +Δtψ + (1 − θ ) tψ ) − C ( t +Δtψ − tψ ) (6.44)
Δt

( )
−1

Δψ = K Ji

6.2.2 Adams-Bashforth integration scheme

Solution predictor:

Δt ⎡⎛ ⎞ ∂ tψ Δt ∂ ψ prev ⎤
t
Δt
t +Δt
ψ = ψ + ⎢⎜ 2 +
t
⎟⎟ − ⎥ (6.45)
2 ⎢⎣⎜⎝ Δt prev ⎠ ∂t Δt prev ∂t ⎥⎦
210

where
index prev indicates that the entity comes from time preceding time t Note that we assume that all
entities from time t are already known and we solve for their values at time t + Δt .

Solution corrector:
∂ t +Δtψ 2 ∂ tψ
= ( t +Δtψ −t ψ ) − (6.46)
∂t Δt ∂t

∂ tψ Δt prev ⎛ t +Δtψ −t ψ ⎞ Δt ⎛ tψ −t ψ prev ⎞

= ⎜ ⎟+ ⎜ ⎟⎟ (6.47)
∂t Δt + Δt prev ⎝ Δt ⎠ Δt + Δt prev ⎝⎜ Δt prev ⎠

 Ji :
Similar to (6.44) we have here for K,

 = Δt ( KΔt ( Δt + Δt ) + C ( 2Δt + Δt ) )
K n −1 n n n −1 n n −1

Ji = −K t +Δtψ (( Δt ) Δt + Δt ( Δt ) ) +
n
2
n −1 n n −1
2

+ C ( − ψ ( 2Δt Δt + ( Δt ) ) + ψ ( 2Δt
t +Δt
n −1 n n −1
2 t
n −1 )
Δtn + ( Δtn −1 ) + ( Δtn ) − t −Δtψ ( Δtn )
2 2 2
) (6.48)
+ J ( ( Δt ) Δt + Δt ( Δt ) )
2 2
n n −1 n n −1

( )
−1

Δψ = K Ji

6.2.3 Reduction of oscillations and convergence improvement

The transport governing equations are prone to suffer from oscillations. As reported in (Jendele
2001) this can be improved by introducing a sort of Line Search method damping η . The basic
idea is that Equation (6.39) gets replaced by

t +Δt
ψ =t +Δt ( i ) ψ =t +Δt ( i −1) ψ + t +Δt (i ) η Δψ (6.49)

where η is a new damping factor. The factor is typically set to something in range < 0.3...1 >
depending on current convergence behaviour of the problem.
 211

6.3 Material constitutive model

The previous section referred to a material constitutive model, i.e. it was assumed that we know
how to compute material diffusivity matrix Dh , (see(6.4)), and material capacity w = w(h) ,
(see(6.1). Calculation of these entities is described here.

Currently, ATENA has only two constitutive model available for the transport analysis. The first
one, i.e. CCModelBaXi94 is characterized as follows and the second one, i.e.
CCTransportMaterial is briefly described later in this section.

CCModelBaXi94
For heat transport a simple constant linear model is implemented. For moisture transport a
nonlinear model based on the model (Xi, Bazant et al. 1993; Xi, Bazant et al. 1994) has been
developed.
It can be used for temperatures in range T =< 5 ...75 0C > and moisture H =< 0 ...1 > . It is
important to note that the model was originally written only for mortar hence, it is inaccurate for
concrete with an aggregate having higher permeability (i.e. diffusivity) and/or absorption. The
model has the following main parameters

• Type of cement
w
• Water-cement ratio wc =
c

As already pointed out, the model does not account for aggregate, i.e. it predict moisture move
only in pores filled by water-cement paste.
w
The main entity of the model is water content w = w(h, t , T , ) . It is defined as follows:
c

Gw
w= (6.50)
Gw,0 + G c

where

⎡ kg ⎤
Gw is water content in mortar at time t , ⎢ 3 ⎥,
⎣ m of morter ⎦
⎡ kg ⎤
Gw,0 is water content at time zero, ⎢ 3 ⎥,
⎣ m of morter ⎦
212

⎡ kg ⎤
Gc is amount of cement at time zero, ⎢ 3 ⎥.
⎣ m of morter ⎦

Mortar here stands for mixture of water and cement. If concrete material is to be considered, then
w can be calculated by

Vconcrete
Gw iw
Vmortar G
w=  (6.51)
Vconcrete V i w,0 + G
G ic
Gw,0 + G c concrete
Vmortar Vmortar

Vconcrete i
where is ratio of total volume to (only) volume of mortar (i.e. water and cement) and G
Vmortar
are corresponding amounts of water and cements in concrete, (i.e. not only in
⎡ kg ⎤
mortar!) ⎢ 3 ⎥.
⎣ m of concrete ⎦

∂w
The model itself already accounts for moisture used by hydration process. i.e. ≠ 0 . As a
∂t
result, wh according to (6.19) need not be implemented.
On the other hand, if moisture losses due to hydration are to be computed by the model based on
∂w
(6.19), it is important to fix = 0 and to modify wh , so that it predicts “relative” moisture
∂t
content w used throughout whole derivation CCStructuresTransport. The original function for
wh was written for absolute weight of water and hence, for “relative” content of water Equations
(6.8) must be rewritten into

i c ⎛ te ⎞
3
0.21 G ⎜ ⎟ 1 1

⎝ τ + t ⎠ = i
G c ⎛ t ⎞ 3
G ⎛ t ⎞ 3
wh = e e
0.21 ⎜ e ⎟ = c
0.21 ⎜ e ⎟ (6.52)
i w,0 + G ic i w,0 + G
ic
G G ⎝ τ e + te ⎠ Gw,0 + Gc ⎝ τ e + te ⎠

and the constant c from (6.8) becomes equal to

Gc ic
G
c= = (6.53)
Gw,0 + Gc Gi w,0 + G
ic
 213

More detailed description of the model is beyond scope of this document and the reader is
referred to in (Xi, Bazant et al. 1993; Xi, Bazant et al. 1994).

CCTransportMaterial
CCTransport material is a simple constitutive law that allow users to enter laboratorily measured
moisture and heat characteristics. Refering to Equations (6.1) and (6.5) heat and moisture flow
governing equations can be written in the following general form:

Heat :
∂h ∂T ∂w ∂
CTh + CTT + CTw + CTt = − ( KTh grad (h) + KTT grad (T ) + KTw grad ( w) + KTgrav )
∂t ∂t ∂t ∂x
(6.54)
Moisture :
∂h ∂T ∂w ∂
Cwh + CwT + Cww + Cwt = − ( Dwh grad (h) + DwT grad (T ) + Dww grad ( w) + Dwgrav )
∂t ∂t ∂t ∂x

The parameters CTh , CTT … K wgrav are calculated as:

CTh = CTh0 f ChTh (h) fCTTh (T ) fCtTh (t )

CTT = CTT
0
f ChTT (h) f CTTT (T ) f CtTT (t )
CTw = CTw
0
fChTw (h) f CTTw (T ) f CtTw (t )
CTt = CTt0 fChTt (h) fCTTt (T ) fCtTt (t )
Cwh = Cwh
0
f Chwh (h) f CTwh (T ) fCtwh (t )
CwT = CwT
0
fChwT (h) f CTwT (T ) fCtwT (t )
Cww = Cww
0
fChww (h) fCTww (T ) fCtww (t )
Cwt = Cwt0 fChwt (h) fCTwt (T ) fCtwt (T )

KTh = KTh0 f KhTh (h) f KTTh (T ) f Kt Th (t )

KTT = KTT
0
f KhTT (h) f KTTT (T ) f Kt TT (t )
KTw = KTw
0
f KhTw (h) f KTTw (T ) f Kt Tw (t )
KTgrav = KTgrav
0
f KhTgrav (h) f KTTgrav (T ) f Kt Tgrav (t )
Dwh = Dwh
0
f Dhwh (h) f DTwh (T ) f Dt wh (t )
DwT = DwT
0
f DhwT (h) f DTwT (T ) f Dt wT (t ) (6.55)
Dww = Dww
0
f Dhww (h) f DTww (T ) f Dt ww (t )
Dwgrav = Dwgrav
0
f Dhwgrav (h) f DTwgrav (T ) f Dt wgrav (t )
214

0
and the constant parameters CTh 0
thru Dwgrav and functions fChTh (h) thru f DTwgrav (T ) are input
parameters, (to be possibly obtained from some experiments). The functions are defined as
multilinear functions and only their ids are input into CCTransportMaterial model definition.
Note that gravity terms in RHS of (6.54) have a little physical justification in heat and moisture
diffusion gathered transports, nevertheless, they are included to allow using this material law for
solution of other kinds of transport problems.

6.4 Fire element boundary load

When undertaking heat transfer calculations it is important that relevant thermal properties of
materials and heat transfer coefficients at the boundaries are defined for the entire temperature
interval of the load.

6.4.1 Hydrocarbon fires

Hydrocarbon fires are those sustained by hydrocarbon-based products, such as chemicals, gas
and petroleum. Depending on the heat load different HC-curves can be derived in accordance
with Equation (6.56). The magnitude of the maximum temperature of the radiation source ( T1 ) is
crucial for the time temperature development. The nominal HC-curve is represented by the heat
load 200 kW/m2 and reaches maximum temperature of 1100 °C. The curve representing 345
kW/m2 is called the "modified" or "increased" HC-curve for tunnel applications. It reaches at
maximum 1300 °C.

T (t ) = T1 (1 − 0.325e −0.167 t − 0.204e −−1.417 t − 0.417e −15.833t ) (6.56)

where:
T (t ) = temperature of radiation source as function of time [°C],
T1 = maximum temperature of radiation source [°C] according to (6.56)
t = time [minutes]

Time development of temperature of radiation source is depicted in the figure below. For time
t → 0 Equation (6.56) yields T (0) = 0 and hence, it is necessary to supplement (6.56) by
requirement T (t ) ≥ Tambiant ,ini , where Tambiant ,ini is initial ambient temperature prior the fire broke
up, (typically something about 20 °C).
 215

Fig. 6-3 Temperature of radiation source

6.4.2 Fire exposed boundary

The nature of the structural ambient conditions is essential for the determination of the
temperature fields. Depending on the geometry, view factors and ambient conditions, various
types of boundary conditions may be considered.

Fire exposed boundary

The heat is transferred from the fire gas to the exposed structure through radiation and
convection. At high temperatures the radiation dominates. The radiation is expressed by the
resulting emissivity factor, which takes into account emissivity of the fire source, ε , and
absorptivity of the heated surface, α . The convection is calculated from the temperature
difference between the structure and ambient gas, depending on the gas velocity. Emissivity and
convection factors used for exposed surfaces are shown below

ε r = 0.56, [−]
⎡ W ⎤ (6.57)
hc = 50, ⎢⎣ m 2 K ⎥⎦

The convection and emissivity heat flux on a boundary exposed to fire is calculated as follows:

qn = hc (Tg − Tb ) + ε rσ (Tg4 − Tb4 ) (6.58)

where
σ = Stefan-Boltzmann constant [5.67x10-8 W/m2 K4],
Tg = absolute temperature of radiation source [K],
216

Tb =boundary temperature of the structure,

ε r = resulting emissivity factor of the radiation source and the heated surface [-],
qn = heat flow at the fire exposed boundary [W/m2],
hc = convection heat transfer coefficient [W/m2K].

Adiabatic boundary
Adiabatic boundary surface refers to a boundary surface, where no heat can pass in (and/or out)
the structure. Structural symmetry lines and areas are good example of this boundary conditions.

6.4.3 Implementation of fire exposed boundary in ATENA

The described fire boundary load conditions are ATENA modelled by

CCFireElementBoundaryLoad load. It is essentially an element boundary load that applies the
heat flow qn at the element boundary, i.e at a surface exposed to fire. Unlike other loads in
ATENA (that are of incremental nature and constant within one load step) this load is considered
variable and has kind of a total load.
Four type of heat source definitions are implemented:
• Nominal HV fire – Temperature of the heat source is calculated by (6.56) and T1 (unless
it is manually input as temp_g_ref) is set to 1100 [°C].
• Modified HC fire – This definition is much the same as the above with the only
difference that default value for T1 is 1300 [°C].
• Generic fire, (also refered to as User curve fire) - Temperature of the heat source is
assumed constant and is set value of temp_g_ref . If temp_g_ref is not inputed, then 1100
[°C] is used.
In any case, the generated (or directly inputed) curve for T (t ) can be additionally modified in
time by a user supplied function time_id. The function takes one parameter, which is time of the
fire and it specifies a coffecient, by which the originally generater (or inputed) boundary
conditions should be multiplied. Of course, load variation in space can be modified by coeff_x,
coeff_y coefficients etc. in the same way as for any other generated element load, (for more
details see Atena Input file manual).
 217

6.5 References

ACI_COMMITTEE_209 (1978). Prediction of Creep, Shrinkage and Temperature Effects in

Concrete Structures. Detroit, 2nd draft, ACI.
BATHE, K. J. (1982). Finite Element Procedures in Engineering Analysis. Englewood Cliffs,
New Jersey 07632, Prentice Hall, Inc.
BAZANT, Z. and T. SPENCER (1973). "Dirichlet Series Creep Function for Aging Concrete."
ASCE Journal of Engineering and Mechanical Division: 367-387.
BAZANT, Z. P. (1986). Mathematical Modelling of Moisture Diffusion and Pore Pressure,
Chapter 10. Concrete at High Temperature. Z. P. Bazant: 198-237.
BAZANT, Z. P. (1988). Mathematical Modeling of Creep and Shrinkage of Concrete. New
York, John Wiley & Sons.
BAZANT, Z. P. and S. BAWEJA, Eds. (1999). Creep and Shrinkage Predicition Model for
Analysis and design of Concrete Structures: Model B3. Creep and Shrinkage of Concrete,
ACI Special Publicatino.
BAZANT, Z. P. and J. K. KIM (1991a). "Improved Prediction Model for Time-Dependent
Deformation of Concrete: Part 1- Shrinkage." Materials and Structures 24: 327-345.
BAZANT, Z. P. and J. K. KIM (1991b). "Improved Prediction Model for Time-Dependent
Deformation of Concrete: Part 2- Basic Creep." Materials and Structures 24: 409-421.
BAZANT, Z. P. and J. K. KIM (1991c). "Improved Prediction Model for Time-Dependent
Deformation of Concrete: Part 3- Creep at Drying." Material and Structures 25: 21-28.
BAZANT, Z. P. and J. K. KIM (1991d). "Improved Prediction Model for Time-Dependent
Deformation of Concrete: Part 4- Temperature Effects." Material and Structures 25: 84-
94.
BAZANT, Z. P. and L. PANULA (1978a). "Practical Prediction of Time-dependent
Deformations of Concrete; Part 1: Shrinkage." Material and Structures 11 (65): 301-316.
BAZANT, Z. P. and L. PANULA (1978b). "Practical Prediction of Time-dependent
Deformations of Concrete; Part 3: Drying Creep." Material and Structures 11 (65): 415-
423.
BAZANT, Z. P. and L. PANULA (1978c). "Practical Prediction of Time-dependent
Deformations of Concrete; Part 4: Temperature Effect on Basic Creep." Material and
Structures 11 (66): 424-434.
BAZANT, Z. P. and L. PANULA (1978d). "Practical Prediction of Time-dependent
Deformations of Time-dependent Deformation of Concrete; Part 2: Basic Creep."
Material and Structures 11 (65): 317-328.
BAZANT, Z. P. and L. PANULA (1978e). Simplified Prediction of Concrete Creep and
Shrinkage from Strength and Mix. Struct. Engng. Report No. 78-10/6405. Evanston,
Illinois, Northwestern University, Dep. of Civ. Engng.
BAZANT, Z. P. and W. THONGUTHAI (1978). Pore Pressure and Drying of Concrete at High
Temperature. Proceedings of the ASCE.
218

BAZANT, Z. P. and F. H. WHITTMAN (1982). Creep and shrinkage in Concrete Structures.

New York, John Wiley & Sons.
BAZANT, Z. P. and Y. XI (1995). "Continuous Retardation Spectrum for Solidification Theory
of Concrete Creep." Journal of Engineering Mechanics 121(2): 281-287.
BETON, C. E.-I. D. (1984). CEB Design Manual on Structural Effects on Time Dependent
Behaviour of Concrete. Saint Saphorin, Switzerland, Georgi Publishing Company.
CELIA, M. A. and P. BINNING (1992). "A Mass Conservative Numerical Solution for Two-
Phase Flow in Porous Media with Application to Unsaturated Flow." Water Resour. Res
28(10): 2819-2828.
CELIA, M. A., T. BOULOUTAS, et al. (1990). "A General Mass-Conservative Numerical
Solution for the Unsaturated Flow Equations." Water Resour. Res 27(7): 1438-1496.
DIERSCH, H. J. G. and P. PERROCHET (1998). On the primary variable switching technique
for simulating unsaturated-saturated flows, http://www.wasy.de/eng/prodinfo/flow/
swpool/swpool.htm#fef_manuals.
HAGEMAN, L. and D. YOUNG (1981). Applied Iterative Methods. New York, Academic
Press.
HUGHES, J. R. (1983). Analysis of Transient Algorithms with Particular Reference to Stability
Behaviour. Computational Methods for Transient Analysis, Elsevier Science Publishers
B.V.
JENDELE, L. (1987). The Orthogonalization of Multiple Eigenvectors in Subspace Iteration
Method. IKM - XI. Internationaler Kongress ueber Anwendungen der Mathematik in der
Ingenieurwissenschaften, Weimar.
JENDELE, L. (2001). ATENA Pollutant Transport Module - Theory. Prague, Edited PIT, ISBN
80-902722-4-X.
JENDELE, L. and D. V. PHILLIPS (1992). "Finite Element Software for Creep and Shrinkage
in Concrete." Computer and Structures 45 (1): 113-126.
REKTORYS, K. (1995). Přehled užité matematiky. Prague, Prometheus.
SEAGER, M. K. and A. GREENBAUM (1988). A SLAP for the Masses, Lawrence Livermore
National Laboratory.
WOOD., W. L. (1990). Practical-Time Stepping Schemes. Oxford, Clarenton Press.
XI, Y., Z. P. BAZANT, et al. (1993). "Moisture Diffusion in Cementitious Materials, Adsorbtion
Isotherms." Advn. Cem. Bas. Mat. 1: 248-257.
XI, Y., Z. P. BAZANT, et al. (1994). "Moisture Diffusion in Cementitious Materials, Moisture
Capacity and Diffusivity." Advn. Cem. Bas. Mat. 1: 258-266.
ZIENKIEWICZ, O. C. and R. L. TAYLOR (1989). The Finite Element Method, Volume 1:
Basic Formulation and Linear Problem. London, McGraw-Hill, 4th edition.
 219

7 DYNAMIC ANALYSIS

Currently ATENA software support two methods to carry out dynamic analysis: Newmark’s β
and Hughes α method (Hughes 1983). If the α parameter in Hughes- α method equals zero,
the Hughes- α degrades to Newmark’s β . Therefere, only Hughes α is described here.
The method is based on Taylor series expansion of the displacements and velocities at time t:

t +Δt ∂u t ∂ 2u t Δt 2 ∂ 3u t Δt 3
u =u + t
Δt + 2 + 3 (7.1)
∂t ∂t 2! ∂t 3!
∂ 2u t ∂ 3u t +Δt Δt 2
u t +Δt = u t + 2 Δt + (7.2)
∂t ∂t 3 2!

The above two equations represent the approximate displacement and velocity ( u t +Δt and u t +Δt ) by a
truncated Taylor series. Looking at the remainder term (last term of the above displacement),

∂ 2u t +Δt ∂ 2u t
t +Δt − 2
∂u
3
Δt 3
∂t Δt
3
 ∂t
2
R1 =
∂t 3 3! Δt 3!
Δt 3
 ( u t +Δt − u t ) (7.3)
3!
 β ( u − u ) Δt 2
t +Δt t

Similar expression can be written for velocities:

∂ 2u t +Δt ∂ 2u t
t +Δt − 2
∂u
3
Δt 2
∂ t 2
∂t Δt 3
R2 =  (7.4)
∂t 3 2! Δt 2!
 γ ( u − u ) Δt
t +Δt t

The above parameters β , γ are so caller Newmark β , γ parameters. Their value is essential for
1 1
convergence of the this time marching scheme. It can be shown that γ = , β = corresponds
2 6
1 1
to a linear acceleration within the time step, γ = , β = yields constant acceleration. In any
2 4
1 γ
case, the scheme is unconditionally stable, if γ ≥ , β ≥ and it is only conditionally stable for
2 2
1 γ
γ ≥ , β < provided that the stability limit is fulfilled:
2 2
220

1 ⎞ ⎡γ 1 ⎞ ⎤2
2
⎛ 2⎛
ξ ⎜γ − ⎟ + ⎢ − β + ξ ⎜γ − ⎟ ⎥
⎝ 2 ⎠ ⎣⎢ 2 ⎝ 2 ⎠ ⎦⎥
ωΔtcrit = (7.5)
⎛γ ⎞
⎜ −β⎟
⎝2 ⎠
where ξ is the damping parameter.
The above defines the condition for time increment Δt for a linear conditionally stable case:

Δt 1
≤
Tn 2
π 2 γ− (7.6)
3
≤ 0.551

More details on the methods’ convergency can be found in (Hughes 1983) and (Wood. 1990).
The final expression for structural displacements and velocities is obtained by substituting (7.3)
and (7.4) into Equation (7.1) and (7.2):

⎡⎛ 1 ⎞ ⎤
u t +Δt = u t + Δt u t + Δt 2 ⎢⎜ − β ⎟ u t + β u t +Δt ⎥ (7.7)
⎣⎝ 2 ⎠ ⎦
u t +Δt = u t + Δt ⎡⎣(1 − γ ) u t + γ u t +Δt ⎤⎦ (7.8)

The expressions above are used in governing equations:

Mu t +Δt + (1 + α ) Cu t +Δt − α Cu t + (1 + α ) Ku t +Δt − α Ku t = (1 + α ) R t +Δt − α R t (7.9)

where
M, C, K is mass, damping and stiffness matrix respectively,
R is vector of external forces, i.e. concentrated loads,
α is Hughes damping pameter.

In nonlinear mechanics the terms Ku t +Δt and Ku t are calculated directly, i.e. we calculate
directly vectors of forces F t +Δt , F t . The above equation then changes to

Mu t +Δt + (1 + α ) Cu t +Δt − α Cu t = (1 + α ) ( R t +Δt − F t +Δt ) − α ( R t − F t ) (7.10)

 221

The parameter α specify the amount of an artificial damping that is to be applied in the
governing equation. As alredy pointed out, if α = 0 , then the time marching scheme reduces to
1
the Newmark method. The value α = − provides maximum damping that results in a damping
3
ratio about 6%, when the time increment is 40% of the period of oscillation of the mode being
studied and smaller, if the oscillation period increases. A recommended value is about
α = −0.05 .

For the actual execution it is advantages to rewrite (7.7) and (7.8) as follows

u t +Δt = u t +Δt + Δt 2 β u t +Δt (7.11)

u t +Δt = u Δt + γ u t +Δt (7.12)

where
⎡⎛ 1 ⎞ ⎤
u t +Δt = u t + Δt u t + Δt 2 ⎢⎜ − β ⎟ u t ⎥ (7.13)
⎣⎝ 2 ⎠ ⎦

u t +Δt = u t + Δt ⎡⎣(1 − γ ) u t ⎤⎦ (7.14)

The expressions for u t +Δt and u t +Δt can be calculated at the beginning of each time increment, as
their values depend only on values at time t. Equation (7.10) is then solved iteratively for
displacement increment Δukt +Δt = ukt +Δt − ukt +Δ
−1 . The displacement Δ uk
t t +Δt
will involve velocity and
acceleration increments

Δukt +Δt = uk − uk −1

t +Δt t +Δt

− u t +Δt uk −1 − u t +Δt
t +Δt t +Δt
uk
= −
Δt 2 β Δt 2 β
t +Δt
Δu
= 2k
Δt β
(7.15)
Δu kt +Δt = u k − u k −1
t +Δt t +Δt

= u Δt + γ uk − u Δt − γ uk −1

t +Δt t +Δt

= γΔuk
t +Δt

t +Δt
γΔuk
=
Δt β
222

This (after some manipulations) yields the final governing equations :

K teff+Δt Δukt +Δt = Refft +Δt (7.16)

where
1 (1 + α ) γ C +
K teff+Δt = M+ (1 + α ) K (7.17)
Δt β
2
Δt β

Refft +Δ,kt = (1 + α ) ( R t +Δt − F t +Δt ) − α ( R t − F t ) − ( Mu t +Δt + (1 + α ) Cu t +Δt − α Cu t ) (7.18)

The vector Refft +Δ,kt is typically splitted in two parts R1,t +Δ t t +Δt
eff , k and R2, eff , k :

eff , k = −α ( R − F ) + α Cu
R1,t +Δ t t t t
(7.19)
eff , k = (1 + α ) ( R
R2,t +Δ − F t +Δt ) − ( Mu t +Δt + (1 + α ) Cu t +Δt )
t t +Δt

The first one is computed only once at the beginning of the time steps and in each iteration it is
necessary to re-caulculate only R2,t +Δ t
eff , k . This concept also allows to forget values R t , F t , u t
during the iterarting procedure, whilst we have to “rember” only vectors u t +Δt and u (and of
course, also R1,t +Δ t
eff , k ) .

7.1 Structural damping

As for damping matrix C , Atena supports the well known proportional damping:
C = αd M + βd K (7.20)

where α d , β d are user defined damping coefficients. These coefficients can be directly set as
user input data or they can be generated based on knowledge of modal damping parameters ξ .
The parameters ξ are defined by:
φiT Cφi = φiT (α d M + β d K )φi = 2ωiξi (7.21)

where:
φi is i-th structural eigenvector,
ωi is i-th structural eigenmode,
ξi is modal damping parameter associated with ωi and φi .
 223

Using the fundamental properties of eigenmodes φiT M φ = 1, φiT K φ = ωi we can rewrite (7.21):

α d + β d ωi = 2ωiξi (7.22)

Equations (7.20) introduces 2 parameters for damping and thus, if only 2 values of ξi are to be
used, they are directly substutited in (7.21), (resp. (7.22)) and solved for from this set of
equations.
However, in practice structural damping is more complicated and some sort of compromise must
be done. In this case structural damping properties are typically measured for more eigenmodes
and optimal values of coefficients α d , β d are calculated by least square method, i.e. we are
seeking minimum of the expression (α d + β d ωi − 2ωiξi ) . It yields the following set of equations
2

wi2 wi2ξi
αd ∑ +β d ∑ wi2 = 2∑
i ωi2 i i ωi
(7.23)
α d ∑ wi2 +β d ∑ wi2ωi2 = 2∑ wi2ωiξi
i i i

which is used to calculate the required damping parameters α d , β d .

There exists other assumptions to account for structural damping, however their use is typically
significantly more complex and more costly in terms of both RAM and CPU.
224

8 EIGENVALUES AND EIGENVECTORS ANALYSIS

This section describes methods used by ATENA software to calculate structural eigenvalues and
eigenvectors. Good knowledge of eigenmodes of a structure is in many cases essential for
understanding its behaviour and selection of a method for its further analysis. It applies to statics
and particularly to dynamic analyses, in which case it helps choosing a proper time increment in
subsequent loading steps. It also help in avoiding or reducing oscillation of the structure.

8.1 Inverse subspace iteration

Currently ATENA uses Inverse subspace iteration method to compute the eigenvalues and
eigenvectors. The method is in details described in (Bathe 1982) and hence, only its main
features are presented here. The current implementation can be used only of symmetric matrices.
The same applies about Jacobi and Rayleigh-Ritz method that are mentioned later in this section.
It consinst of three methods, each of them is capable of solving eigenvalue problem on its own.
However, if they are used simultaneously, they yield a very efficient scheme for solving
eigenvalues and eigenvectors of large sparse structural systems. The significant advantage of
this approach is that it is possible to search for a selected number of the lowest eigenmodes only.
The lowest eigenmodes are typically the most important for behavior of the structure because
they represent the highest energy that the structure can absorb. On the other hand, the highest
eigenmode are of low importance, can be neglected and thereby save a lot of CPU time and other
computational resources.
The Inverse subspace iteration consists of

• Inverse iteration method

• Rayleigh-Ritz method
• Jacobi method

It solves general eigenvalues and eigenvector problem of the following form:

Ku = ω 2 Mu (8.1)
where
K, M is stiffness and mass matrix of structure,
u is vector of eigenvector’s nodal displacements,
ω is circular eigenfrequency

We are looking for a non-trivial solution, so that we solve for ω 2 that comes from

det(K − ω 2 M ) = 0 (8.2)
 225

8.1.1 Rayleigh-Ritz method

This method is used to transform the original eigenproblem of dimension n into an associated
eigenproblem of dimension m<<n. The solution is search for in a space Vm << Vn . Let vectors
ψ k constitute linearly independent bases in Vn . An eigenvector ui is computed as a linear
combination ci of the base vectors ψ k , i.e.

ui = Ψci (8.3)
where
Ψ is matrix of base vectors ψ k , k = 1..m ,
ci is vector of coefficients of the linear combination.

Rayleigh quotient is defined as

uiT Kui
ρ (ui ) = (8.4)
uiT Mui

It can be proved that ρ (ui ) converges from upper side to the corresponding circular frequency
ωi2 . The condition of minimum of ρ (ui ) yields:

∂ρ (ui )
= 0, k = 1..m (8.5)
∂ci , k

where ci ,k is k component of the vector ci

If we introduce
A = ΨT KΨ, B = ΨT MΨ (8.6)

the condition (8.5), after substituting (8.6), results in

Aci = ωi2 Bci (8.7)

which is an equation of eigenproblem of matrices A,B. This problem has dimension m , which is
significantly smaller than the original dimension n.

8.1.2 Jacobi method

226

Jacobi method is used for solution of full symmetric eigensystems of lower dimension. In the
frame of Inverse subspace iteration method it is used to solve (8.7). (Note however, that that the
eigenproblem (8.7) can be used by any other method).
The Jacobi method is based on the property that if we have a matrix A, a orthogonal matrix C
and a diagonal matrix D, whereby

CT AC = D (8.8)

then the matrices A and D have identical eigenvalues and they are diagonal elements of the
matrix D. The transformation matrix C is calculated in iterative manner

C = S1 S 2 .......S k , k = 1..∞ (8.9)

where the individual S k has the following form

⎡1 0 0 0⎤
⎢ 1 ⎥
⎢ ⎥
⎢ 1 ⎥
⎢ ⎥
S k = ⎢0 cos(α ) 0 − sin(α ) 0⎥ (8.10)
⎢ 0 1 0 ⎥
⎢ ⎥
⎢0 sin(α ) cos(α ) 0⎥
⎢0 0 0 1 ⎥⎦
⎣

The entries cos(α ), ± sin(α ) are put in i,j rows and columns and they are constructed in the way
that they will zeroize aij after the transformation. The other diagonal elements are equal to 1 and
the remaining off-diagonal elements are 0.

In the case of general eigenproblem the whole procedure of constructing S k is very similar. The
matrices S k now adopt the shape

⎡1 0 0 0⎤
⎢ 1 ⎥
⎢ ⎥
⎢ 1 ⎥
⎢ ⎥
Sk = ⎢0 1 0 a 0⎥ (8.11)
⎢ 0 1 0 ⎥
⎢ ⎥
⎢0 b 1 0⎥
⎢0 0 0 1 ⎥⎦
⎣
 227

Notice that the matrix S k is not orthogonal anymore. The two variables a,b are calculated to
zeroize off-diagonal elements i,j of the both matrices K and M. Eigenmodes of the problem are
then calculated as

aii'
ωi2 = (8.12)
bii'

where aii' , bii' are diagonal elements of transformed (and diagonalized) matrices A, B.
Eigenvectors of the problem are columns of the transformation matrix C.

8.1.3 Inverse iteration method

Inverse iteration method is carried out as follows: Starting with an initial transformation of
eigenvector ui ,1 we calculate vector of corresponding inertia forces, (step 1)

fi ,1 = M ui ,1 (8.13)

Knowing fi ,1 , we can compute a new approximation of ui , (step 2)

ui ,2 = K −1 fi ,1 (8.14)

and repeat the step 1. Hence, for iteration k we have

fi ,k = M ui , k
(8.15)
ui ,k +1 = K −1 fi ,k

and the iterating is stop, when ui ,k +1 ≈ ui ,k . The above described algorithm tends to converge to
the lowest eigenmodes. If any of these are to be skipped, the initial eigenvector ui ,1 must be
orthogonal to the corresponding eigenvectors. In practice, the vector ui ,k must orthogonalized
with respect to the skipped eigenvectors even during the iterating procedure, as the initial
orthogonality may get (due to some round-off errors) lost.

8.1.4 Algorithm of Inverse subspace iteration

Having briefly described the above three methods we can now proceed to the actual solution
algorithm of Inverse subspace iteration method itself:
228

Step1- Inverse iteration method:

KU = MU
k +1 k

Step 2 - Raylegh quotient method:

A =U  T KU

k +1 k +1 k +1
 T MU
B k +1 = U 
k +1 k +1
(8.16)

Step 3 − Jacobi method:

A k +1Ck +1 = B k +1Ck +1Δ 2

Step 4 - Correct the eigenvectors:

UT = U T C
k +1 k +1 k +1

In the above
m is number of projection eigenmodes, (reasonably higher than the number of required
eigenmodes),
U k is matrix of columnwise arranged m einvectors after k- th iteration,
A k +1 , B k +1 are transformed stiffness and mass matrices of the problem, (having dimension
m<<n),
Ck +1 is matrix of eigenvectors of A k +1 , B k +1 , see (8.9)

Δ 2 is matrix with eigenmodes (on its diagonal). Notice that eigenmodes for transformed and the
original eigenmode problem are the same.

The steps 1 thru 4 are repeated until the difference between the two subsequent operations is
negligible.

The solution algorithm (8.16) is in ATENA a bit modified in order to reduce CPU time and
RAM resources and is described below:
 229

Step1- Inverse iteration method:

ˆ = MU
U k +1 k

 =U
KU ˆ
k +1 k +1

Step 2 - Raylegh quotient method:

A =U  T KU
 =U T U ˆ
k +1 k +1 k +1 k +1 k +1

ˆ = MU
U 
k +1 k +1

B =U  T MU  T U
=U ˆ
k +1 k +1 k +1 k +1 k +1

Step 3 − Jacobi method:

A k +1Ck +1 = B k +1Ck +1Δ 2

Step 4 - Correct the eigenvectors:

UT = U T C (8.17)
k +1 k +1 k +1

The advantage of this procedure over the one defined in (8.16) is that now you don’t need to
store the original and factorised form of the matrix K. Only the factorised form is needed during
the iterations.

A special issue in this method is how to setup the initial vectors U1 . This is what we do in
ATENA. The first vector contains the diagonal elements of M. The next vectors are constructed
in the way that they have zeros everywhere except one entry. This entry correspond to maximum
m ii
and is set to 1.
kii
The procedure as it is , (because of Inverse iteration method), cannot solve for zero eigenmodes.
This may be a problem, especially if we want to analyze structural rigid body motions or
spurious energy modes. If this is the case, shift matrix K by an arbitrary value λs , i.e. solve the
associated eigenproblem

(K − λs M )us = ωs2 Mus (8.18)

The original eigenvalues and eigenvectors are then calculated by

u = us
(8.19)
ω 2 = ωs2 − λs
Another problem of Inverse subspace iteration is to compute multiple eigenvectors.
Unfortenatly, it is not that rare case and it happens e.g., if the structure has an axis of symmetry.
Occurrence of multiple eigenmodes in the structure may yield non-orthogonal eigenvectors and
thus some eigenmodes can be missed. There are some techniques for resolve this problem
(Jendele 1987), however, they have not been implemented in ATENA yet. Good news is that in
230

reality no eigenmodes are usually quite identical due to some round-off errors. The case of
multiple structural eigenmodes thus typically causes only some worsening of accuracy and no
eigenmode gets missed.
Nevertheless, if we want to be sure that no eigenmode was missed, we can assess it by Sturm
sequence property test.

8.1.5 Sturm sequence property check

This property says (Bathe 1982) that if we have an eigenproblem (8.1), perform a shift λs and
factorise that matrix, (i.e. D is diagonal matrix, L is lower triangular matrix ),
K − λs M = L D LT (8.20)
then number of negative diagonal elements in D equal to the number of eigenvalues smaller than
the shift λ . This way we can simply test, whether we missed an eigenvalue with the calculated
set of m eigenmodes or not

There are other methods that can be used to compute eigenvalues and eigenvectors of large
sparse eigensystem. Particularly popular is e.g. Lanczosh method (Bathe 1982). There exist also
several enhancements for the present Inverse subspace iteration method. For instance using
shifting technique may siginificantly improve convergency of the method, (especially if some
eigenvalues are close each other).
These improved technique may be implemented in the future. In any case, the current ATENA
implementation of eigenmodes analysis proves to solve the eigenmodes problem in most case
quite successfully.

8.2 References

BATHE, K. J. (1982). Finite Element Procedures in Engineering Analysis. Englewood Cliffs,

New Jersey 07632, Prentice Hall, Inc.
JENDELE, L. (1987). The Orthogonalization of Multiple Eigenvectors in Subspace Iteration
Method. IKM - XI. Internationaler Kongress ueber Anwendungen der Mathematik in der
Ingenieurwissenschaften, Weimar.
WOOD., W. L. (1990). Practical-Time Stepping Schemes. Oxford, Clarenton Press.
 231

9 GENERAL FORM OF DIRICHLET BOUNDARY CONDITIONS

A unique feature of ATENA software is the way, in which it implements Dirichlet boundary
conditions. It supports to constraint any degree of freedom (DOF) by a linear of any number of
other structural DOFs. The proposed method of applying and processing the boundary conditions
is computationally efficient and memory economical, because all constraint degrees of freedoms
(DOFs) are eliminated already during assembly of structural global stiffness matrix and load
vectors. The adopted concept has wide range of use and several its possibilities are discussed. At
the end of the Section a few samples are given.

9.1 Theory behind the implementation

A crucial part of a typical finite element analysis, (whether linear or nonlinear) is solution of a
set of linear algebraic equations in the following form
n

∑K
j =1
ij u j = ri , i = 1..n (9.1)

where K ij is an element i, j of a predictor matrix K, (i.e. usually structural stiffness matrix), ri

is an external force, (or unbalanced force) applied into i-th structural degree of freedom (DOF)
and finally ui is displacement (or displacement increment) at the same DOF. Such a set of
equations is always accompanied by many boundary conditions (BCs). They can be one of the
following:
Von-Neumann boundary conditions, (also called right-hand side (RHS) BCs). Number and type
of these BCs has no impact on dimension n of the problem (9.1). They are accumulated in the
vector r . This vector is assembled on the per-node basis for concentrated nodal forces and/or
per-element basis for nodal forces being equivalent to element loads.

The second type of boundary conditions are Dirichlet boundary conditions, (also called left-hand
side (LHS) BCs). ATENA implementation of this type of BCs is now described. A simple form
of such BCs reads:

ul = 0, l ∈< 1, n >
(9.2)
ul = ul 0 , l ∈< 1, n >

These kinds of BCs typically represent structural supports with no displacements, (the first
equation) or with prescribed displacements ul 0 , (the second equation). Although most LHS BCs
are of the above form, (and only a few finite element packages offer anything better), they are
cases, when a more general LHS BC is required. Therefore, ATENA software provides solution
for implementing a form of Dirichlet BCs, where each degree of structural freedom can be a
linear combination of any other degrees of freedom. Mathematically, this is expressed by:
232

ul = ul 0 + ∑
k ∈<1, n >
α lk uk , l ∈< 1, n > (9.3)

There are many cases, in which the above form of Dirichlet conditions proves helpful. Some
examples are discussed later in the Chapter. The important point about implementing Equations
(9.3) is that they are utilised already during assembling of the problem (9.1). It means that, if we
have m of these BCs, then final dimension of the matrix K becomes only (n − m) . This fact
significantly reduces requirements towards computer storage.

In the following we shall call such boundary conditions as “Complex Boundary Conditions”, or
CBCs, (see also ATENA Input file manual, where the same name is used).

9.1.1 Single CBC

The procedure of implementing Dirichlet BCs of the form (9.3) is now presented. Let us start
with just one BC equation (9.4). It says that ul equals to a constant prescribed displacement ul 0
plus α lk multiple of a displacement uk .

ul = ul 0 + α lk uk (9.4)
Substituting (9.4) into the Equation (9.1) yields

n n

∑
j =1, j ≠ l
K ij u j + K il ul = ∑
j =1, j ≠ l
K ij u j + K il (ul 0 + α lk uk ) = ri , i = 1..n (9.5)

which after some manipulation can be simplified into the form

∑(K
j =1
ij + K ilα lk δ kj ) u j = ri − K il ul 0 , i = 1..n (9.6)

The above set of equations could be already used to solve for the unknown displacements (or
displacement increments) u j . δ kj stands for k . j Kronecker delta tensor. The trouble is, however,
that even though the matrix K might be symmetric, the set of equations (9.6) is not symmetric
anymore. Thus, to preserve the symmetry, add an α lk multiple of the row l , i.e.

⎛ n
⎞
α lk ⎜ ∑ ( K lj + K llα lk δ kj ) u j ⎟ = α lk ( rl − K ll ul 0 ) (9.7)
⎝ j =1 ⎠
to the row k, i.e.
 233

∑(K
j =1
kj + K klα lk δ kj ) u j = rk − K kl ul 0 (9.8)

This results in the row k getting the form:

∑(K )
n

kj + K klα lk δ kj + α lk ( K lj + K llα lk δ kj ) u j =
j =1
n

∑(K
j =1
kj + α lk K lj + ( K klα lk + α lkα lk K ll ) δ kj ) u j = (9.9)

rk − K kl ul 0 + α lk ( rl − α lk K ll ul 0 )

Hence, the final form of the governing set of equations will read:

∑(K
j =1
ij + K ilα lk δ kj + δ ikα lk K lj + δ ik δ kjα lk2 K ll ) u j =
(9.10)
ri − K il ul 0 + δ ikα lk ( rl − K ll ul 0 )

The above equations can be written as

n

∑ K
j =1
ij u j = ri , i = 1..n (9.11)

where
K =

⎡ K11 ... K1i ... K1k + K1lα lk ... K1 j ... K1n ⎤

⎢ ... ... ... ... ... ... ... ... ... ⎥
⎢ ⎥
⎢ K i1 ... K ii ... K ik + K ilα lk ... K ij ... K in ⎥
⎢ ⎥
⎢ ... ... ... ... ... ... ... ... ... ⎥
⎢ K k1 + α lk K l1 ... K ki + α lk K li ... K kk + 2 K klα lk + δ kk δ kkα lk2 K ll ... K kj + α lk K lj ... K kn + α lk Kln ⎥
⎢ ⎥
⎢ ... ... ... ... ... ... ... ... ... ⎥
⎢ ⎥ (9.12)
K j1 ... K ji ... K jk + K jlα lk ... K jj ... K jn
⎢ ⎥
⎢ ... ... ... ... ... ... ... ... ... ⎥
⎢ K n1 ... K ni ... K nk + K nlα lk ... K nj ... K nn ⎥
⎣ ⎦
234

r =
⎡ r1 − K1l ul 0 ⎤
⎢ ... ⎥
⎢ ⎥
⎢ ri − K il ul 0 ⎥
⎢ ⎥
⎢ ... ⎥ (9.13)
⎢ rk − K kl ul 0 + α lk ( rl − K ll ul 0 ) ⎥
⎢ ⎥
⎢ ... ⎥
⎢ rj − K jl ul 0 ⎥
⎢ ⎥
⎢ ... ⎥
⎢ rn − K nl ul 0 ⎥
⎣ ⎦

 is now also
Providing the original matrix K is symmetric, i.e. K ij = K ji , then the matrix K
symmetric, i.e. K = K .
ij ji

The displacement ul constrained by Equation (9.4) has a constant part ul 0 and a variable part
α lk uk , in which ul depends only on a single uk . A more general form of this BC would be, if
ul depends on more displacements. It corresponds to the following form of the boundary
condition:

ul = ul 0 + ∑ α lk uk (9.14)
k

In this case, the displacement ul is calculated as a constant part ul 0 plus a linear combination
α lk of displacements uk . k can be any displacement, i.e. k ∈< 1..n > . Replacing BC defined by
Equation (9.4) by the above Equation (9.14), the equation will change to the form

n
⎛ ⎞
∑⎜ K ij + ∑ (K α il lk δ kj + δ ikα lk K lj + δ ik δ kjα lk2 K ll ) ⎟ u j =
j =1⎝ k ,k ≠l ⎠ (9.15)
ri − K il ul 0 + ∑ (δ ikα lk ( rl − K ll ul 0 ) )
k ,k ≠l

9.1.2 Multiple CBCs

The previous paragraph derived all the necessary relations for implementing a single boundary
condition. Now we will proceed to the case of multiple boundary conditions. Each particular BC
is again written in the form (9.14).
 235

ul = ul 0 + ∑ α lk uk , l ∈< 1, n >, l = {l1 , l2 ,...lr } (9.16)

k

The problem is, however, that displacements uk in (9.16) need not be free but fixed by another
BC, k can run also through l, (resulting in a recursive formulation), more BCs can be specified
for the same ul , a particular BC can be specified more times and in more forms etc. For
example, we may have a set boundary equations that contains BCs

u1 = u2 , u2 = u1 (9.17)

or it can contain
u1 = u2 , u2 = u1 , u1 = 0.003 (9.18)

Both of these are correct. Unfortunately, the set can also contain

u1 = u2 , u2 = −0.003, u2 = u1 , u1 = 0.003 (9.19)

which is definitely wrong. Therefore, before any use of such set of BCs it is necessary to detect
and fix all redundant and contradictory multiple BCs present in it. It is easily done in case of a
simple set of BCs like the one above, but in real analyses with thousands of BCs in the form
(9.16), (some of them quite complex, i.e. k runs through many DOFs) the only way to proceed is
to treat (9.16) as a set of equations to be solved prior their use in (9.13). Redundant BCs are
ignored and contradictory BCs are fulfilled after their summation. Let us suppose that all
structural constraints are specified in the set of equation (9.16). This can be written in matrix
form

ul = ul 0 + Auk (9.20)

The above relationship represents a system of algebraic linear equations. The system is typically
non-symmetric, sparse and has different number of rows (i.e. number of BCs) and columns, (i.e.
number of master and slave DOFs). Moreover, it is often ill-conditioned, with a number of
equations being linear combinations of the others, e.g. see the example in (9.17). At the
beginning it is often not known, which DOF is dependent, (i.e. slave) and which is independent,
(i.e. master), (e.g. see also (9.17)).

Based on the above properties the following procedure has been developed to solve the problem
(9.20):

1. Allocate "columns" for all slave and master DOFs, i.e. loop through all BCs in (9.16) and
allocate DOFs for both slave (i.e. LHS) and master (i.e. RHS) displacements ui .
236

2. Allocate storage for the matrix A and vectors ul , ul 0 in (9.20). The matrix has lr number
of rows (see (9.16)) and lc number of columns. lc is dimension of the DOFs map created
in the point add. 1.
3. Assemble the matrix A and the vectors ul , ul 0 .
4. Detect constant BCs, i.e. ul = ul 0 and swap rows of A so that the rows corresponding to
constant BCs are pushed to the bottom.
5. Detect constant fixed DOFs, i.e. those with α lk = 0 and variable fixed DOFs, i.e. that are
those dependent on other (master) DOFs and having α lk ≠ 0 .
6. Swap columns of A , so that the former DOFs are pushed to the right and the latter DOFs
to the left. The operations described at the point 5 and 6 are needed to assure order, in
which the constrained DOFs are eliminated. This is important for later calculation of the
structural reactions.
7. Using Gauss method triangulize the set of BC equations. The triangulization is carried
out in the standard way with the following differences.

a. Before eliminating entries of A located in column below akk , check for a non-
zero entry in the row k. If all its entries are zero, then ignore this row and proceed
to the next one. (It is the case of multiple BCs having the same content).
b. Check, whether the row k specifies BC for constant or variable DOF, (see
explanation in the point 5 above). In the former case push the row k to the bottom
and proceed to the next row.
c. Swap columns < k ...lc > so that abs (akk ) becomes maximum.
d. If akk = 0 , swap lines < k ...lr > to find a non-zero entry in akk . Thereafter, swap
columns < k ...lc > to find maximum akk .
e. Eliminate entries below akk as usually.

As it was already mentioned, the matrix A is typically very sparse. Hence, a special storage
schemes is used that stores only non-zero entries of A. The data are stored by rows. Each row
has a number of data series, i.e. sequences or chunks of consecutive non-zero data (within the
row). The data are in a three-dimensional container. For each such chunk of data we also need to
store its first position and length. This is done in two two-dimensional containers.

As an example, suppose that we have the following matrix A:

 237

⎡ a11 0 0 0 0 0 0⎤
⎢0 a22 a23 0 0 a26 a27 ⎥⎥
⎢
⎢0 0 a33 0 0 0 0⎥
⎢ ⎥
A=⎢ 0 a42 0 a44 0 0 0⎥ (9.21)
⎢0 0 0 0 a55 a56 a57 ⎥
⎢ ⎥
⎢0 a62 0 0 0 a66 0⎥
⎢0 0 0 0 0 0 a77 ⎥⎦
⎣

It is stored as follows ( A.data stores the actual data, A.rowbase stores base indices for non-zero
entries in rows, A.rowlength contains dimension of non-zero data chunks; all arranged by rows):
A.data (1)(1)(1) = a11
A.data (2)(1)(1) = a22 , A.data(2)(1)(2) = a23 , A.data (2)(2)(1) = a26 , A.data (2)(2)(2) = a27
A.data (3)(1)(1) = a33 ,
A.data (4)(1)(1) = a42 , A.data(4)(2)(1) = a44
.......
A.rowbase(1)(1) = 1
A.rowbase(2)(1) = 2, A.rowbase(2)(2) = 6
(9.22)
A.rowbase(3)(1) = 3
A.rowbase(4)(1) = 2, A.rowbase(4)(2) = 4
.....
A.rowlength(1)(1) = 1
A.rowlength(2)(1) = 2, A.rowlength(2)(2) = 2
A.rowlength(3)(1) = 1
A.rowlength(4)(1) = 1, A.rowlength(4)(2) = 1

A number of optimisation techniques are used to speed up the process of triangularization of the
matrix A. These are summarized below:
The data are stored by rows and the elimination is also carried out by rows. (Row-based storage
is also more convenient during assembling the A from (9.16)). All the operations needed for the
elimination are carried out only for nonzero data. Their horizontal position is stored in
A.rowbase and A.rowlength , hence it is no problem to skip all zero entries. A typical total
number of columns lc , see (9.16), is of order from thousands to hundred thousands DOFs. On
the other hand a.rowlength is on average only of order of tens. This is where the CPU savings
comes from.
By the way, the same mapping of non-zero entries is also used for columns. This is achieved by
additional arrays A.columnbase and A.columnlength that are also included in the storage
scheme A. (Their construction is similar to A.rowbase and A.rowlength ; instead by rows they
are arranged by columns). These two additional arrays make possible to skip all zero entries
during column-base operations. The resulting significant increase of triangularization speed pays
238

off for a small amount of an extra RAM that is needed to store A.columnbase and
A.columnlength .
The adopted procedure of triangularization many times swaps lines and/or columns of A. In view
of the adopted storage scheme it can be quite expensive procedure. To alleviate this problem, the
storage scheme includes four additional arrays, namely A.rowindex , A.rowinverseindex ,
A.columnindex and A.rowinverseindex . At the beginning A.rowindex(i ) = i and similarly
A.rowinverseindex(i ) = i, i = 1...lc . When a row r1 should be swapped with a row r2 , the data in
A.data remains unchanged and we swap only corresponding row indices in A.rowindex , (and
accordingly also entries in the array for inverse mapping A.rowinverseindex ). The same
strategy is used for swapping the columns. As a result any swapping operation does not require
any moving of actual data, (except of swapping corresponding indices for mapping the rows and
columns) and thus it is extremely fast. On the other hand, in order to access aij we must use ai ' j ' ,
where i ' = rowindex(i ) and j ' = columnindex( j ) . The incurred CPU overhead is well
acceptable, because the matrix A is very sparse.

9.2 Application of Complex Boundary Conditions

This section presents several examples, where the developed Dirichlet boundary conditions are
advantageously used. In each case the corresponding finite element model exploits the general
form of BC defined by Equation (9.16).

9.2.1 Refinement of a finite element mesh

Suppose we need to refine a mesh as shown in Fig. 9-1. The mesh should refine from 5 elements
per row to 10 elements per row. The figure depicts three possible techniques to achieve the goal.
In the case A the fine and coarse parts of the mesh (consisting of quadrilateral elements) are
connected by a row of triangular elements. This way of mesh refinement is used the most often.
However, mixing quadrilateral and triangular elements is not always the best solution.
In the case B the refinement is achieved by using hierarchical finite elements, see (Bathe 1982).
The coarse mesh near the interface employs five nodes hierarchical elements. This refinement is
superior to the others, however, it requires a special finite elements and special mesh generator;
both of these rarely available in a typical finite element package.
In the case C the fine and coarse parts of the mesh are generated independently. After generation
of all nodes and elements the interface nodes are connected by complex boundary conditions.
For example, we can use ui = um , uk = un , u j = 0.5um + 0.5un . The main advantage of this
approach is that it is simple for both finite element pre/postprocessor and finite element
modeller, (namely its finite element library). Hence it is preferable!
 239

Fig. 9-1 Mesh refinement

Note that all the above techniques are supported in ATENA finite element package, the last one
requiring implementation of CBCs in the form (9.14).

9.2.2 Mesh generation using sub-regions

This example demonstrates another advantage of using the proposed CBCs: It is possible to
generate meshes within sub-regions without requirement of nodes coincidence on their
interfaces. Because mesh structure on the sub-regions’ surfaces is not prescribed, this approach
provides more flexibility to mesh generation. This feature is heavily used by Atena3D
preprocessor.

Compatible meshes on the contact between the blocks

Fig. 9-2 Mesh generation from simple blocks
240

Incompatible meshes on the contact between the blocks using CBCs

Fig. 9-2 (cont) Mesh generation from simple blocks

In the above example two blocks are connected to form a structure, where the top (smaller) block
is placed atop of the bottom (larger) block. Position of the top block is arbitrary with respect to
the bottom block. Unless the concept of CBCs is used, the meshes on interface of the two blocks
must be compatible, (see top of Fig. 9-2). On the other hand, the proposed CBCs allow using of
incompatible meshes, (see the bottom of Fig. 9-2). In this case the mesh in each block is
generated independently, which is significantly simpler. After they are done, the proposed CBCs
are applied to connect the interface nodes. (Typically the surface with the finer mesh is fixed to
the surface with the coarse mesh). The latter approach also demonstrates possibility of a mesh
refinement while still using well-structured and transparent meshes. This is particularly useful in
the case of complex numerical models.

9.2.3 Discrete reinforcement embedded in solid elements

In this example, the described boundary conditions are used to simplify modelling of reinforced
concrete beam, see Fig. 9-3. The procedure to create the model is as follows. Firstly, the mesh
for solids, i.e. concrete elements is generated. It poses no problem, as it is a regular mesh
consisting of 48 quadrilateral elements. At this point, no attention needs to be paid to the
geometry of reinforcing bars present in the beam. Thereafter, the reinforcing bars are inserted
and their meshes are generated based on the existing mesh of solid elements. First step is to find
all nodes, where the bar changes direction. These nodes are called principal nodes; see e.g. node
n in Fig. 9-3. Then, intersection of all straight parts of the bar with underlying solid elements are
detected, e.g. the nodes m,p. Thus, all end nodes of embedded bar elements are defined. The last
step is to link displacements of the nodes of the bar to the underlying solid elements.

i j
m n
l p k

bar 1
bar 2

Fig. 9-3 Discrete bar reinforcement

 241

For example, if we want to connect the node n to the embedding solid element, i.e. to nodes
4
i,j,k,l, see Fig. 9-3, we use the standard interpolation u (r , s ) = ∑ hi (r , s ) U i , where hi ( r , s), U i are
i =1
element interpolation function and Ui are nodal displacements for the underlying solid element,
4
respectively. For displacement at the node n we can write u (rn , sn ) = ∑ hi (rn , sn ) U i . ( rn , sn ) are
i =1
coordinates of the node n. Comparing this formula with (9.3), it is obvious that
α ni = hi (rn , sn ), un 0 = 0 . Consequently, the bar DOFs are always kinematically dependent on the
DOFs of underlying solid elements.
This technique can be also applied, when bond elements are inserted between solid and
embedded bar elements. This is treated in a separate paper by authors in ref . (Jendele, 2003).
Currently, ATENA software can generate discrete reinforcement to all 2D and 3D linear and
nonlinear elements, (triangles, quads, tetrahedral elements, wedges, bricks…). The user only
draws position of the principal nodes of reinforcement bars and the rest is done automatically.

9.2.4 Curvilinear nonlinear beam and shell elements

In the following text, another possible use of the present boundary conditions is presented. A
curvilinear nonlinear beam from Chapter 3.13 is discussed. A particular feature we would like to
point out here is that although it has originally only three displacements and three rotations in the
nodes 13,14,15, see Fig. 3-33, its implementation in ATENA has also 3 displacements in the
nodes 1 to 12. However, these DOFs are linked to the original DOFs in the nodes 13 to 15 by the
proposed CBCs. This concept has several advantages.

• The beam finite element has native 3D geometry and its pre- and postprocessing
visualisation is more realistic than using its original 1D geometry.
• It is simple to connect such beam elements to any adjacent 3D finite elements, e.g. brick
elements.
• Mesh generation is easily done by any 3D solid element generator that can pull off
nonlinear hexahedral elements. It suffices to generate only the nodes 1 to 12 (with 3
displacement DOFs) and the three original beam nodes (each beam node has 3
displacement and 3 rotation DOFs) are generated automatically. The preprocessor need
not to support rotational DOFs.
• Postprocessing of this element and an ordinary nonlinear hexahedral element is the same.
Consequently, this element does not need any extra support for visualisation of the
results. It makes its implementation and use simple.
Derivation of all α ij coefficients and ui 0 constants for all nodes 1 to 12 is beyond the scope of
this document. Nevertheless, a similar procedure is used, as it was in the previous example.

ATENA package covers also Ahmad element for curved shell structures, see Chapter 3.12. The
usual 2D shape of shell element is in the same manner replaced by geometry of a 3D nonlinear
hexahedral element. Originally, the shell element has 3 displacements and 2 rotations at each
node located at middle thickness of the shell. These 5 DOFs are in by use of CBCs replaced by 3
displacements at the top and 2 displacements at the bottom at the respective nodes from the
hexahedron, (i.e. brick) geometry. Advantages of this approach are the same as those in the case
242

of the curvilinear beam above: simpler pre/post processing, simpler connection to the adjacent
3D elements, no need to support rotational DOFs during pre/post processing, no need for extra
support for geometry of the shell element.

9.3 References
BATHE, K.J. (1982), Finite Element Procedures In Engineering Analysis, Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey 07632, ISBN 0-13-317305-4.
JENDELE, L, CERVENKA, J, CERVENKA V., " Bond in Finite Element Modelling of
Reinforced Concrete", Proceedings Euro-C 2003, Computatinal Modelling of Concrete
Structures, Swets & Zeitlinger, Lisse, The Netherlands, ISBN 90 809 536 3, 793-8036
 243

INDEX
convergence43, 51, 52, 85, 172, 173, 175, 180,
A 181, 189, 210, 219
convergency .............................................. 220, 230
Adam-Bashfoth integration............................... 210 cracking................................................... 25, 33, 60
Cracking.............................................................. 27
B Crank-Nicholson integration............................. 209
creep.................................................................. 184
boundary condition.............................................. 21 basics ............................................................ 184
complex .......................................................... 21 constitutive models....................................... 190
simple ............................................................. 21 Dirichlet series.............................................. 186
parameters needed by models....................... 193
C retardation times ........................................... 188
solution parameters....................................... 191
Clapeyron divergent theorem .............................. 11 Step by Step Method .................................... 187
constitutive model ............................................... 23 creep model
after peak behaviour ....................................... 31 CCModelB3 ................................................. 191
Bigaj ............................................................... 79 CCModelB3Improved .................................. 191
CC3DCementitious......................................... 43 CCModelBP_KX ......................................... 191
CC3DNonLinCementitious ............................ 43 CCModelBP1_DATA .................................. 191
CC3DNonLinCementitious2 .......................... 43 CCModelBP2_DATA .................................. 191
CC3DNonLinCementitious2Fatigue .............. 57 CCModelCEB_FIP78................................... 190
CC3DNonLinCementitious2User................... 43 CCModelCSN731202 .................................. 191
CC3DNonLinCementitious2Variable............. 43 CCModelGeneral ......................................... 191
CEB-FIP 1990 ................................................ 77 creep model
compressive failure......................................... 33 CCModelACI78 ........................................... 190
compressive stress .......................................... 37
crack opening law........................................... 28
crack spacing .................................................. 57
D
definition................... 23, 89, 166, 184, 197, 231 damping .............................210, 220, 221, 222, 223
Drucker Prager................................................ 69 Dirichlet conditions........................................... 231
equivalent uniaxial law................................... 26 discretisation
fracture process............................................... 33 spatial ........................................................... 201
Hooke’s law.................................................... 23 temporal........................................................ 208
HORDIJK law ................................................ 28 dynamic............................................................. 219
interface model ............................................... 70
localization limiters ........................................ 32
Microplane...................................................... 81
E
peak stress....................................................... 30 eigenvalues.................................224, 226, 229, 230
Rankine fracturing .......................................... 44 eigenvectors .......................224, 227, 228, 229, 230
reinforcement.................................................. 74
reinforcement bond......................................... 77 F
SBETA model ................................................ 25
SBETA model parameters .............................. 41 fatigue ................................................................. 57
shear and stiffness in crack ............................. 37 fiber reinforced concrete ............................... 29, 60
size effect........................................................ 32 finite element
smeared cracks................................................ 35 Ahmad element ............................................ 128
smeared cracks-fixed ...................................... 35 axisymmetric element................................... 126
smeared cracks-rotated ................................... 36 brick quadrilateral element........................... 103
tensile failure .................................................. 34 extwernal cable element ............................... 120
tension stiffening ...................................... 38, 56 hexahedral element....................................... 103
transformation................................................. 24 interface element .......................................... 123
variants ........................................................... 53 nonlinear 3D beam element.......................... 149
Von Mises....................................................... 66 plane quadrilateral element ............................ 95
constitutive tensor ............................................... 14 Q10 element ................................................. 116
constutive model Q10Sbeta element ........................................ 116
plasticity and crushing .................................... 46 reinforcement bar with prescribed bond....... 121
244

shell element................................................. 128 J

spring element .............................................. 114
triangular element ......................................... 102 Jacobi method ................................................... 226
truss 2D/3D..................................................... 91
Fire analysis ...................................................... 214 L
flux
heat ............................................................... 198 Lagrangian element........................... 128, 140, 149
moisture ........................................................ 197
formulation.......................................................... 10 M
Euler ............................................................... 10
Lagrange ..................................... 10, 16, 65, 149 moisture ............................................................ 197
Updated Lagrange........................................... 10 Multipoint constraint......................................... 231

G N
governing equations ............................................ 16 Newmark................................................... 219, 221
Green theorem................................................... 207 nonlinearity ........................................................... 9
types ................................................................. 9
nonlinearity classification ..................................... 9
H
heat.................................................................... 198 O
Heterosis element .............................. 128, 140, 149
hierarchical formulation ...................................... 89 oscilations ......................................................... 210
high performance fiber reinforced concrete ........ 60
HPFRCC ............................................................. 60 P
Hughe alpha metod ........................................... 219
hydration ................................................... 198, 212 Palmgren-Miner hypothesis ................................ 58
Hydrocarbon fires ............................................. 214 principle of virtual displacements ................. 11, 15
principle of virtual forces.................................... 11
problem ............................................................... 10
I configuration .................................................. 11
integration points FEM discretisation ......................................... 17
CCBeamNL element..................................... 160 formulation ..................................................... 10
CCIsoBrick element ..................................... 108 general ............................................................ 10
CCIsoQuad element........................................ 97
CCIsoTetra element...................................... 105 R
CCIsoTrianle element................................... 103
CCIsoWedge element ................................... 111 Rayleigh Ritz method ....................................... 225
Q10/Q10Sbeta element................................. 118
shell/Ahmad element .................................... 139 S
truss 2D/3D element ....................................... 91
integration shell element-summart .................... 148 Serendipity element .......................... 128, 139, 149
interpolation function .......................................... 89 shape function ..................................................... 89
CCBeamNL element..................................... 152 SHCC .................................................................. 60
CCIsoBrick element ..................................... 106 S-N curve ...................................................... 57, 58
CCIsoQuad element........................................ 96 solver................................................................. 166
CCIsoTetra element...................................... 105 Arc length..................................................... 174
CCIsoTriangle element................................. 102 Arc Length step ............................................ 181
CCIsoWedge element ................................... 110 Consistently Linearized Method................... 178
problem discretisation..................................... 17 Crisfield Method .......................................... 180
Q10/Q10Sbetaelement.................................. 116 direct sparse.................................................. 168
shell/Ahmad element .................................... 139 Explicit Orthogonal Method......................... 179
truss 2D/3D element ....................................... 91 Cholesky....................................................... 167
introduction ........................................................... 9 iterative......................................................... 168
Inverse Iteration method ................................... 227 iterative-types ............................................... 169
Inverse Subspace Iteration method ................... 224 linear............................................................. 166
isoparametric formulation ................................... 89 Linear search ................................................ 181
Modified Newton Raphson .......................... 173
Newton Raphson .......................................... 172
 245

nonlinear ....................................................... 172 Sturm sequence check....................................... 230

Normal Update Method ................................ 177
β scaling parameter....................................... 182 T
strain.................................................................... 14
Almansi........................................................... 15 time equivalent.................................................. 199
engineering ..................................................... 14 Transport analysis ............................................. 197
Green Lagrange .............................................. 14
Strain Hardening Cementitious Composite......... 60 W
Strain Hardening Cementitious Composites ....... 60
stress.................................................................... 12 Wöhler curve....................................................... 58
2nd Piola-Kirchhoff........................................ 12
Cauchy............................................................ 12 X
stress/strain smoothing ........................................ 19
lumped ............................................................ 20 Xi-Bazant model ............................................... 211
variational ....................................................... 19