ATENA - Theory Nonlinear RC
ATENA - Theory Nonlinear RC
ATENA - Theory Nonlinear RC
cz Web: http://www.cervenka.cz
Theory
Trademarks: ATENA is registered trademark of Vladimir Cervenka. Other names may be trademarks of their respective owners. Copyright 2000-2011 Cervenka Consulting Ltd.
CONTENTS
1
1.1 1.2 1.3
9
9 10 12 12 12 14 14 14 14 15 17 17 19 20 21 22
1.3.1 Cauchy Stress Tensor 1.3.2 2nd Piola-Kirchhoff Stress Tensor 1.4 Strain Tensors
1.4.1 Engineering Strain 1.4.2 Green-Lagrange Strain 1.5 1.6 1.7 1.8 1.9 Constitutive tensor The Principle of Virtual Displacements The Work Done by the External Forces Problem Discretisation Using Finite Element Method Stress and strain smoothing
1.9.1 Extrapolation of stress and strain to element nodes 1.10 1.11 Simple, complex supports and master-slave boundary conditions. References
2
2.1
CONSTITUTIVE MODELS
Constitutive Model SBETA (CCSbetaMaterial)
24
24 24 27 33 34 34 36
2.1.1 Basic Assumptions 2.1.2 Stress-Strain Relations for Concrete 2.1.3 Localization Limiters 2.1.4 Fracture Process, Crack Width 2.1.5 Biaxial Stress Failure Criterion of Concrete 2.1.6 Two Models of Smeared Cracks
4 2.1.7 Shear Stress and Stiffness in Cracked Concrete 2.1.8 Compressive Strength of Cracked Concrete 2.1.9 Tension Stiffening in Cracked Concrete 2.1.10 Summary of Stresses in SBETA Constitutive Model 2.1.11 Material Stiffness Matrices 2.1.12 Analysis of Stresses 2.1.13 Parameters of Constitutive Model 2.2 FracturePlastic Constitutive Model (CC3DCementitious, CC3DNonLinCementitious, CC3DNonLinCementitious2, CC3DNonLinCementitious2User, CC3DNonLinCementitious2Variable, CC3DNonLinCementitious2SHCC, CC3DNonLinCementitious3) 2.2.1 Introduction 2.2.2 Material Model Formulation 2.2.3 Rankine-Fracturing Model for Concrete Cracking 2.2.4 Plasticity Model for Concrete Crushing 2.2.5 Combination of Plasticity and Fracture model 2.2.6 Variants of the fracture plastic model 2.2.7 Tension stiffening 2.2.8 Crack spacing 2.2.9 Fixed or Rotated Cracks 2.2.10 Fatigue 2.2.11 Strain Hardening Cementitious Composite (SHCC, HPFRCC) material 2.2.12 Confinement-sensitive constitutive model 2.3 2.4 2.5 2.6 2.7 Von Mises Plasticity Model Drucker-Prager Plasticity Model User Material Model Interface Material Model Reinforcement Stress-Strain Laws 38 38 39 39 40 41 42
44 44 45 45 47 51 54 57 57 58 58 61 65 68 71 72 72 76
5 2.7.1 Introduction 2.7.2 Bilinear Law 2.7.3 Multi-line Law 2.7.4 No Compression Reinforcement 2.7.5 Cycling Reinforcement Model 2.8 Reinforcement bond models 76 76 77 78 78 79 80 81 83 84 85 89
2.8.1 CEB-FIP 1990 Model Code 2.8.2 Bond Model by Bigaj 2.8.3 Memory Bond Material 2.9 Microplane material model (CCMicroplane4)
3
3.1 3.2 3.3 3.4 3.5 3.6 3.7
FINITE ELEMENTS
Introduction Truss 2D and 3D Element Plane Quadrilateral Elements Plane Triangular Elements 3D Solid Elements Spring Element Quadrilateral Element Q10
93
93 95 99 106 108 121 123 123 126 127 128 130 134 136
3.7.1 Element Stiffness Matrix 3.7.2 Evaluation of Stresses and Resisting Forces 3.8 3.9 3.10 3.11 3.12 External Cable Reinforcement Bars with Prescribed Bond Interface Element Truss Axi-Symmetric Elements. Ahmad Shell Element
6 3.12.1 Coordinate systems. 3.12.2 Geometry approximation 3.12.3 Displacement field approximation. 3.12.4 Strain and stresses definition. 3.12.5 Serendipity, Lagrangian and Heterosis variant of degenerated shell element. 3.12.6 Smeared Reinforcement 3.12.7 Transformation of the original DOFs to displacements at the top and bottom of the element nodal coordinate system 3.12.8 Shell Ahmad elements implemented in ATENA 3.13 Curvilinear nonlinear 3D beam element. 138 143 144 145 146 152 152 156 157 157 161 162 168 170 173
3.13.1 Geometry and displacements and rotations fields. 3.13.2 Strain and stress definition 3.13.3 Matrices used in the beam element formulation 3.13.4 The element integration 3.14 3.15 Global and local coordinate systems for element load References
4
4.1
174
174 175 176 176 180 182 183 186 186 187
4.1.1 Direct Solver 4.1.2 Direct Sparse Solver 4.1.3 Iterative Solver 4.2 4.3 4.4 Full Newton-Raphson Method Modified Newton-Raphson Method Arc-Length Method
4.4.1 Normal Update Method 4.4.2 Consistently Linearized Method 4.4.3 Explicit Orthogonal Method
7 4.4.4 The Crisfield Method. 4.4.5 Arc Length Step 4.5 4.6 4.7 4.8 Line Search Method Parameter Band width optimization References 188 189 189 190 192 195
5
5.1
197
197 197 199 201 202 204 211
5.1.1 Basic theoretical assumptions 5.2 5.3 5.4 5.5 5.6 Approximation of compliance functions (t , t ') by Dirichlet series. Step by step method Integration and retardation times Creep and shrinkage constitutive model References
6
6.1 6.2
TRANSPORT ANALYSIS
Numerical solution of the transport problem spatial discretisation Numerical solution of the transport problem temporal discretisation
213
217 225 226 227 228 229 247 247 248 249 251
6.2.1 -parameter Crank Nicholson scheme 6.2.2 Adams-Bashforth integration scheme 6.2.3 Reduction of oscillations and convergence improvement 6.3 6.4 Material constitutive model Fire element boundary load
6.4.1 Hydrocarbon fires 6.4.2 Fire exposed boundary 6.4.3 Implementation of fire exposed boundary in ATENA 6.5 References
DYNAMIC ANALYSIS
253
8
8.1
259
259 260 261 262 263 265 266
8.1.1 Rayleigh-Ritz method 8.1.2 Jacobi method 8.1.3 Inverse iteration method 8.1.4 Algorithm of Inverse subspace iteration 8.1.5 Sturm sequence property check 8.2 References
9
9.1
267
267 268 271 275 275 276 277 278 279
9.1.1 Single CBC 9.1.2 Multiple CBCs 9.2 Application of Complex Boundary Conditions
9.2.1 Refinement of a finite element mesh 9.2.2 Mesh generation using sub-regions 9.2.3 Discrete reinforcement embedded in solid elements 9.2.4 Curvilinear nonlinear beam and shell elements 9.3 Reference
INDEX
280
10 However the relative deformations (strains) are still small. The complete form of the geometric 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.
11 the same 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 + t by replacing index t by t + t .
M0
Configuration 0
M1
Configuration t
M2
Configuration t+t
[ X1, X1,
t+t
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 +t 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 +0t ij denotes element i , j of stress tensor at time t + t 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 +t 0 i, j
u =
X j
t +t
ui
(1.2)
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:
12 The principle of virtual displacements, The principle of virtual forces, 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.
Sij =
0 t 0 t X i , m mn t X j ,n
(1.3)
where
0 t
is the ratio of density of the material at time 0 and t , t 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 PiolaKirchhoff stress tensor, i.e.:
t
mn =
t 0
t t t 0 X m ,i 0 S mn 0 X n , j
(1.4)
The elements 0 X i ,m are usually collected in the so-called Deformation gradient matrix: t
13
t 0
X = ( 0 t X T )
(1.5)
where:
, 0 , 0 0 = 0 X1 X 2 X 3
T
t
X T = t X1, t X 2 , t X 3
0
The ratio
= t 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 Piola0 Kirchhoff and Cauchy stress tensors because 0 X = E , i.e. unity matrix and the density
0
ratio
= 1.,
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. Theyre 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 u u = t m + t n 2 Xn Xm
(1.7)
1 t ( 0 ui , j + 0t u j ,i + 0t uk ,i 0t uk , j ) 2
(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 ( 0 ij ) =
t X m t X n ( t emn ) 0 X i 0 X j
(1.9)
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.
(1.10)
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 t
Cmnpq =
t t t t t 0 xm ,i 0 xn , j 0Cijrs 0 x p , r 0 xq , s
(1.11)
15
t 0
Cijrs =
0
t
0 0 0 0 t t xi , m t x j , n t Cmnpq t xr , p t xs , q
(1.12)
Using constitutive tensor (1.11) and Almansi strains tt , we can write for Cauchy stresses (with respect to coordinates at time t ):
t
ij =tt Cijrs tt rs
(1.13)
(1.14)
1 t ( t ui , j +tt u j,i tt uk ,i tt uk , j ) 2
(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 .
(
V
t + dt 0
Sij
(
(
t + dt 0
t ij ) dV = 0+ dt R
(1.16)
(
V
t + dt t
Sij
t + dt t
ij ) dV = tt + dt R
(1.17)
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 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 + t (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
16 beginning of time increment, i.e. the beginning of load step. The following table compares the above-mentioned formulations:
Table 1.6-1 Comparison of different Lagrangian formulation. Transform each load Transfor m stress increment and strain IP state Material IP state Material for output variable properti variable properties s es s Transform each iteration No Yes No No Yes No No Yes Yes No Yes Yes Yes No No Calculate
t +t ( i 1) t i, j
t
Lagrangian formulation
u for
eij
Yes No Yes
Governing equations:
t
t +t t
(1.18)
nd
( Siji ) =
t +t
t ( i ) t +t ( i ) t t +t i ,m t mn t +t
) x (ji,n
(1.19)
( t +tt iji ) =
1 t +t ( i ) ( t ui, j + 2
t +t ( i ) t i, j
u +
t +t ( i ) t +t ( i ) t k ,i t k, j
(1.20)
(1.21)
t +t ( i ) t ij
t +t ( i 1) t ij
( + t iji )
(1.22)
t ( ( ( iji ) = t eiji ) + tiji )
(1.23)
(1.24)
17
Using constitutive equations in form:
t +t ( i ) t ij
( = t Cijrs t rsi )
(1.25)
(i ) where t Cijrs is tangent material tensors and noting that ( t + tt ij( i ) ) = ( t ij( i ) ) , an incremental form of (1.18) can be derived:
Cijrs
(e
(i ) t ij
t +t t
(1.26)
Cijrs
(e
(i ) t ij
(1.27)
t +t ( i 1) t ij
( ( tiji ) ) dV = t
t +t
t +t t
( i 1) ij
( e
(i ) t ij
) dV
t
(1.28)
( Note that the term ( t eiji ) ) = ( t eij ) is constant, i.e. independent of t ui( i ) , hence it is on RHS
of (1.28).
R=
t +t
fbi ( t u i ) dV +
t
t +t
fsi ( t u i ) dS + t
t
2 t +tt ui( i 1) dV t 2
(1.29)
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).
18
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.
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 +t t ( (i ( i ) = ( t BL 0 + t +tt BLi11) + t +tt BNL1) ) t +tU ( i )
(1.31)
U ( i ) is vector of displacements, st t +t ( i 1) t +t ( i 1) 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 +t t
t +t
S (i ) =
t +t t
C (i )
t +t t
(i )
(1.32)
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
t 2
t +t
(i U ( i ) + ( t K L + t +tt K NL 1) ) t +tU ( i ) = t +t R t +t F ( i 1)
(1.33)
where
t
t +t t
t +t
(i )
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
T K L U ( i ) = t BL t C t BL dV U ( i ) t V (i K NL 1) U ( i ) = t V F ( i 1) =
t
t +t t
t +t t
(i BNL 1)T
t +t t
( Siji 1)
t +t t
(i BNL 1) dV U ( i ) t V
t
t +t t
( ( Siji 1) ( tiji ) ) dV t
t +t
t +t t
( Siji 1) dV
t
t +t t
( ( Siji 1) ( t eiji ) ) dV
(1.34)
t +t
R=
H
A t +t
T t +t t
f A dV dA +
H
V
T t +t t
f B dV t +t R
t 2
U ( i ) = H T t H dV 2 t t V
t +t
U (i )
t +t ui( i ) t +t ui( i ) t dV 2 t 2 t t V
important to note 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.
20
The whole technique is described briefly. All details and derivations can be found e.g. (ZIENKIEWICZ, TAYLOR 1989) and ERVENKA et. al. 1993.
The extrapolation is done as follows (for each component of structural stress and strain ). Let us define a vector of stresses xx at element nodes i such as xx = { xx ,1 , xx ,2 ,.... xx ,n } ,
T
where the 2nd index indicates element node number. Let us also define a vector
Pxx ,i = hi xx d e
e
(1.35)
The nodal value xx (with values of xx at nodes i =1..n ) is then calculated as follows:
xx = [ M ] Pxx
inv
(1.36)
where:
M ij = hi h j d e
e
(1.37)
In the above xx is an extrapolated field of stress of xx calculated by FEM. It is typically discontinuous. n is number of element nods, e 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 ij d e
e k =1, n
(1.38)
k =1, n
M ij = hi ij d
e
(1.39)
where 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 i = { xx , yy ,..... xz }i in a global node i that is participated by all elements k with incidence at the global node i .
21
=
i k k ek
ek
(1.40)
where is vector of stresses i = { xx , yy ,..... xz }i at a node i , ek is volume of element k that has incidence of global node i . It should be noted that in ATENA the same extrapolation techniques is used for other integration point quantities as well such as: fracturing strains, plastic strains and others.
(1.41)
(1.42)
ATENA software supports that any constrained degree of freedom can be a linear combination of other degrees of freedom plus some constant term:
i i k uD = uD,0 + k uN k
(1.43)
i where uD,0 is the constant term and k are coefficients of the linear combination. Of course,
the equation (1.43) can include also the term constant term. The free degree of freedom are then solved by
u
l l
l D
uN = ( K NN )
and the dependent RD are solved by
(R
K ND RD )
(1.44)
RD = K DN uN + K DD uD
(1.45)
22
The ATENA simple support boundary conditions mean that the boundary conditions use i only constant terms are uD,0 , (i.e. 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 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 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, Inc., Englewood Cliffs, New Jersey 07632, ISBN 0-13-317305-4. Prentice-Hall,
ERVENKA, 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
24
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 = { x , y , xy } , e = { x , y , 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 x , y , xy , Fig. 2-1, and the strain components x , y , xy , Fig. 2-2, where 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 sc , 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.
25
The reinforcement stress vector ss is the sum of stresses of all the smeared reinforcement components:
s s = s si
i =1 n
(2.3)
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 si by the transformation:
, s si = T pi si
(2.4)
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 . The angle 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 ) = T s ( x )
(2.5)
26
cos( ) 2 sin( ) 2 2 cos( ) sin( ) 2 2 2 cos( ) sin( ) sin( ) cos( ) T = cos( ) sin( ) cos( ) sin( ) cos( )2 sin( ) 2
(2.6)
s ( u ) = { u , v , uv } , s ( x ) = { x , y , xy }
T
(2.7)
cos( ) 2 sin( ) 2 cos( ) sin( ) 2 2 cos( ) sin( ) T = sin( ) cos( ) 2 cos( ) sin( ) 2 cos( ) sin( ) cos( ) 2 sin( ) 2
(2.8)
e(u ) = { u , v , uv } , e( x ) = { x , y , xy } .
T T
The angles of principal axes of the stresses and strains, Fig. 2-1, Fig. 2-2, are found from the equations:
tan(2 ) = 2 xy
x y
, tan(2 ) =
xy x y
(2.9)
where is the angle of the first principal stress axis and is the angle of the first principal strain axis. In case of isotropic material (un-cracked 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.
27
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.
The nonlinear behavior of concrete in the biaxial stress state is described by means of the soef called effective stress c , and the equivalent uniaxial strain eq . The effective stress is in most cases a principal stress. The equivalent uniaxial strain is introduced in order to eliminate the Poissons effect in the plane stress state. (2.10) Eci The equivalent uniaxial strain can be considered as the strain, that would be produced by the stress 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 ci . The details can be found in CHEN (1982). The complete equivalent uniaxial stress-strain diagram for concrete is shown in Fig. 2-4.
eq =
ci
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.
28
Unloading is a linear function to the origin. An example of the unloading point U is shown in ef Fig. 2-4. Thus, the relation between stress c and strain eq is not unique and depends on a load 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 fcef and in tension ftef 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
c (2.11) eq It is used in the constitutive equation to calculate stresses for the given strain state, Sect. 2.1.12.
Ec =
s
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, f t 'ef is the effective tensile strength derived from the biaxial failure function, Section 2.1.5.2.
c ef = Ec eq , 0 c ft ' ef
2.1.2.3 Tension after Cracking
(2.12)
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.
29
(1) Exponential Crack Opening Law
Fig. 2-5 Exponential crack opening law. This function of crack opening was derived experimentally by HORDIJK (1991).
f t ' ef
(2.13)
where w is the crack opening, wc is the crack opening at the complete release of stress, 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
c ef
ft ' ef
2G ft ' ( wc w) , wc = ' f wc ft
(2.14)
30
(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. Parameters:
c1 =
f1 ft
' ef
, c2 =
f2 ft
' ef
, wc =
2G f
f1 + f 2
31
Parameters: c1 =
f1 ft
' ef
, c2 =
f2 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.
c ef = f c ' ef
E kx x 2 ,x = ,k = o c Ec 1 + (k 2) x
(2.15)
Fig. 2-10 Compressive stress-strain diagram. Meaning of the symbols in the above formula in: cef - concrete compressive stress, f c'ef - concrete effective compressive strength (See Section 2.1.5.1) x - normalized strain, - strain, c - strain at the peak stress fcef , k - shape parameter, Eo - initial elastic modulus, f ' ef Ec - secant elastic modulus at the peak stress, Ec = c .
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.
32
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 d 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:
d = c +
wd L'd
(2.16)
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.
33
Lc Lt
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 = Lt , Ld = Ld
' '
= 1 + ( max 1)
45
0; 45
(2.17)
34
An angle is the minimal angle ( min (1 , 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 =1.0 for the direction parallel with element sides, and = max , for the direction inclined at 45o. The recommended (and default) value of max =1.5.
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.)
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
35
1 + 3.65a ' f c , a = c1 2 (1 + a ) c2
(2.19)
where c1 , c 2 are the principal stresses in concrete and fc 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 c1 = 0 , c 2 = f c' into the tension-compression region with the linearly decreasing strength:
f c ' ef = f c' rec , rec = (1 + 5.3278
c1
f c'
),
(2.20)
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 ft. In the tension-compression state, the tensile strength is reduced by the relation:
f t ' ef = f t ' 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.
ret = 1 0.95 ret =
c2
f c'
(2.22) (2.23)
A + ( A 1) B , B = Kx + A, x = c '2 AB fc
The relation in Eq.(2.22) is the linear decrease of the tensile strength and (2.23) is the hyperbolic decrease.
36
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 r
a b 0.5 0.5
point x
0.4 0.2
parameters A
0.75 1.0625
K
1.125 6.0208
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.
37
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 1 and 2 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 c1 and c2 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
Gt =
c1 c 2 2(1 2 )
(2.24)
38
1000 u ln c1 G = rg Gc , rg = c3 c2
(2.25)
Gc =
Ec 2(1 + )
(2.26)
where Ec is the initial elastic modulus and is the Poisson's ratio. The strain 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 users 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 uv = G is limited by the tensile strength ft. The secant and tangent shear moduli of cracked concrete are equal.
(2.27)
For the zero normal strain, , 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.
40
(2.28)
In the above E is the concrete elastic modulus derived from the equivalent uniaxial law. The Poisson's ratio 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 21 0 E E2 1 1 1 1 12 0 2 (2.29) 2 = E1 E2 1 0 0 G First we eliminate the orthotropic Poissons ratios for the cracked concrete, because they are commonly not known. For this we use the symmetry relation 12 E2 = 21E1 . Therefore, in (2.29) there are only three independent elastic constants E1 , E2 , 21 . Assuming that 21 = is the Poisson's ratio of the uncracked concrete and using the symmetry relation, we obtain
12 =
L
E1 E2
(2.30)
The stiffness matrix Dc is found as the inverse of the flexibility matrix in (2.30): 0 D c = H 1 0 , 0 0 G E = 1 ,H = E1 (1 2 ) E2
L
(2.31)
41 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 modules are E1 zero, the matrix Dc is set equal to the null matrix. The matrix Dc is transformed into the global coordinate system using the transformation matrix
T from (2.8).
L D c = TT D c T
(2.32)
The angle 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
(2.33)
The angle 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:
D = Dc + D si
i =1 n
(2.34)
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 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.
s c = Dc e
(2.35)
42 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.
s
ft = 0.24 f
'
2 ' 3 cu
= 0.2
wd = 0.0005mm 1 exponential, based on GF c = 0.8
st = 0.
variable (Sect.2.1.7) linear
GF = 0.000025 f t ' ef [MN/m]
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 fcu (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. 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.
43 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.
44
2.2 FracturePlastic 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 Mentrey-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.
45
e ij = ij + ijp + ijf
(2.36)
(2.37)
where the increments of plastic strain ijp and fracturing strain ijf must be evaluated based on the used material models.
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, iit identifies the trial stress and f ti 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. ijt = ijn 1 + Eijkl 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 = iin f ti = iit E iikl klf f ti = 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 ij = = ik ij
f
(2.41)
After substitution into (2.40) a formula for the increment of the fracturing multiplier is recovered.
= t kk f tk E kkkk = t kk f t( wkmax ) E kkkk
and wkmax = Lt ( kkf + )
(2.42)
This equation must be solved by iterations since for softening materials the value of current max tensile strength f t( wk ) is a function of the crack opening w , and is based on Hordijks formula (defined in SBETA model).
The crack opening w is computed from the total value of fracturing strain kkf in direction k , plus the current increment of fracturing strain , and this sum is multiplied by the
46 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.
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:
(2.43)
Equation (2.43) is violated for softening materials only when snap back is observed in the stressstrain 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 f (0) f t(0) E should be used, where t denotes the initial slope elements smaller then L < kkkk w w of the crack softening curve.
(2.44)
47
cr The fourth order crack tensor Eijkl represents the cracking stiffness in the local material directions.
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.
(2.45)
(2.46)
(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).
ij
0.18 f c , i j 24 w 0.31 + ag + 16
(2.48)
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 exceed 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 T (i.e. global to local strain transformation matrix) can be used to transform the local secant stiffness matrix to the global coordinate system.
E s = T Es T
T
(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.
2.2.3.1 Unloading Direction
Crack closure stiffness is controlled by the unloading factor (material parameter) 0 fU 1. The value of 0 corresponds to unloading to origin (default value for backward compatibility), fU =1 means unloading direction parallel to the initial elastic stiffness.
48
(n)
(2.51)
The plastic corrector ijp is computed directly from the yield function by return mapping algorithm.
t t F p ( ij ijp ) = F p ( ij lij ) = 0
(2.52)
The crucial aspect is the definition of the return direction lij , which can be defined as lij = E ijkl
t t G p ( ij ) G p ( kl ) then ijp = kl ij
(2.53)
where G ( ij ) is the plastic potential function, whose derivative is evaluated at the predictor stress
t state ij to determine the return direction.
The failure surface of MENETREY, WILLAM is used in the current version of the material model.
2 c = 0 F = 15 ' + m . r ( , e) + ' fc 3 f c' 6 fc p 3P
(2.54)
where f '2 f '2 e m=3 c ' ' t , r ( , e) = fc ft e + 1 4(1 e 2 ) cos2 + (2e 1) 2 2(1 e ) cos + (2e 1) 4(1 e ) cos + 5e 4e
2 2 2 2
1 2
In the above equations ( , , ) are Heigh-Vestergaard coordinates, f c and f t is compressive strength and tensile strength respectively. Parameter e 0.5,1.0 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.
p eq = min( ijp )
(2.55)
For Mentrey-Willam surface the hardening/softening is controlled by the parameter c 0,1 , which evolves during the yielding/crushing process by the following relationship:
p f c( eq ) c= f c 2
(2.56)
p In the above two formulas the expression f c( eq ) 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:
49
= f co + ( f c f co )
p c eq 1 c
(2.57)
f' c
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 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 Mentry-Willam surface. Return direction is given by the following plastic potential G p ( ij ) = 1 3 I1 + 2 J 2 (2.58)
where determines the return direction. If < 0 material is being compacted during crushing, if = 0 material volume is preserved, and if > 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.
50
Fig. 2-23. Schematic description of the iterative process (2.73). For clarity shown in two dimensions.
51
( n 1)
ij , ( n 1) ijp , ( n ) ij )
t ij = ( n1) ij + Eijkl ( n ) kl
t f Ap = F p ( ij , ( n 1) ijp ) , A = 0
(2.59) (2.60)
If failure criterion is violated i.e. Evaluate return direction: Return mapping: Evaluate failure criterion: Secant iterations (i ) as long as
f Ap > 0
mij =
t F p ( ij B E mij ,
t G p ( ij )
ij
( n 1)
ijp ) = 0 B
( n 1)
t f Bp = F p ( ij B E mij ,
ijp + B mij )
A B > e
New plastic multiplier increment: New return direction: Evaluate failure criterion:
(i )
= A f Ap
mij =
(2.65) (2.66)
(i )
t G p ( ij E
mij )
ij
(i )
t f p = F p ( ij E
mij ,
( n 1)
ijp +
mij )
(2.67)
f Bp < 0 f Bp 0
End of secant iteration loop
f Bp = f p , B = f Bp = f p , B =
(2.68) (2.69)
f Ap = f Bp , A = B ,
(n)
t ij = ij B E (i ) mij
(2.70)
(2.71) (2.72)
Each inequality depends on the output from the other one, therefore the following iterative scheme is developed.
52 Algorithm 2:
f cor p p Step 1: F p ( ( n 1) ij + Eijkl ( kl (i 1) kl + b ( i 1) kl ( i ) kl )) 0 solve for (i ) kl
f p f Step 2: F f ( ( n 1) ij + Eijkl ( kl (i ) kl (i ) kl )) 0 solve for (i ) kl cor Step 3: (i ) ij = (i ) ijf ( i ) ijf
(2.73)
Iterative correction of the strain norm between two subsequent iterations can be expressed as
cor cor (i ) ij = (1 b) f p ( i ) ij
(2.74)
where =
f
, =
p
(i ) ijp (i ) ijp
cor ij
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 f and p , such that the convergence of the iterative scheme can be assured. The parameters f and 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) f 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 f or 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 0 , (2.58). This is a sufficient but not necessary condition to ensure that f and p < 1 . It can be shown that the values of f and 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 f and p . There are three possible scenarios, which must be handled, for the appropriate calculation of b :
f p
, where is related to the requested convergence rate. For linear rate it can be < 1 , then the convergence would be too slow. In this case b can be estimated
< f p
as b = 1
1
f p
b = 1
53 This approach guarantees convergence as long as the parameters p , 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 and finite element size. It is advantageous to further stabilize the algorithm by smoothing the parameter b during the iterative process: b = ( ( i )b + (i 1)b) / 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: kkf ft p eq f c (2.77)
The compressive strength reduction (b) is based on the following formula based proposed by Collins:
c = rc f c
rc = 1 , rclim rc 1.0 0.8 + 170 1
(2.78)
54 Where 1 is the tensile strain in the crack. In ATENA the largest maximal fracturing strain is used for 1 and the compressive strength reduction is limited by rclim . If rclim is not specified then no compression reduction is considered.
f c'0 f c' 0
(2.79) (2.80)
f c'0 0,
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).
55
t/ft
loc
f1 ~
Fig. 2-24. An example of a user defined tensile behavior for CC3DNonLinCementitious2User material.
c/fc
c
G/Gc
loc
~ eq p <0
Fig. 2-25. An example of a user defined compressive behavior for CC3DNonLinCementitious2User material.
1.0 (1 ) Lt/Lch
loc
sh
sh
loc
~f
Fig. 2-26. An example of a user defined shear retention factor for shear stiffness degradation after cracking.
56 In the user defined material mode II and III crack stiffness are evaluated with the help of the shear retention factor rg as:
cr Eijij =
rg G 1 rg
(2.81)
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 Lc represents a size for which the tensile and compression ch 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. 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:
f if 1f < loc
1f = 1f
else
f f 1f = loc + (1f loc ) Lt
Ltch
(2.82)
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 loc , L and Lch should be used. It should be noted that the strain 1f is the strain that is calculated from the strain tensor at the finite element integration points, while the strain 1f 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 Shear strength of cracked concrete
c = rc (1f ) f c
(2.83) (2.84)
ij f sh (1f ) ft
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 1f after the localization transformation (2.82). The shear strength law is specified as a value relative to ft . The compressive strength reduction is specified as a function relative to f c .
57
t/ft 1.0
1.0
3/fc
Fig. 2-27. An example of a user defined tensile strength degradation law due to lateral compressive stress.
ft c ts ft
58 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.
2.2.10 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 (2.2.10.1). It has an additional parameter, fatigue , and additional data attributes for base , N , and fatigue , used in the damage calculation as described in section 2.2.10.2. For details and validation against tests conducted by KESSLER-KRAMER (2002) see ERVENKA, PRYL (2007).
2.2.10.1 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).
rm N = C
(2.85)
where 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 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 semilogarithmic diagram of the form:
max
f
= 1 (1 R ) log N
(2.86)
min , min is minimum max stress and is a material constant. The equation (2.86) holds for both compressive and tensile stresses, however, the value of is not neccessarily the same for tensile and compressive behavior of a material. The value should be determined from experiments. For example, =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.
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).
N
i =1
ni
i
=1
(2.87)
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.
60
2.2.10.2 Fatigue Damage Calculation In the implemented model, fatigue damage consists of a contribution based on cycling stress (2.2.10.2.1), and an additional contribution from crack opening and closing in each cycle (2.2.10.2.2). The former is dominant before cracking occurs, the latter in already cracked regions. 2.2.10.2.1 Stress Based Contribution The number of cycles to failure N is determined from a simple stress based model, so called SN or Whler curve as described in the previous section 2.2.10.1.
upper 1 f fatigue (1 R )
upper
f
base . upper
Then, the damage due to fatigue after n cycles is calculated as an increase of the maximum
fracturing strain ij f
fatigue =
w fail
n w fail and the failing displacement for the given stress N ElemSize = invert _ soft _ law( upper ) (see Fig. 2-30).
w fatigue
, where w fatigue =
61 In ATENA 4.0, a single value of 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.
2.2.10.2.2 Crack Opening Based Contribution The damage due to cracks that open and close during the cycling is determined as w fatigueCOD = fatigueCOD , where w fatigueCOD = n fatigue c fatigueCODload COD , and COD denotes the ElemSize difference between the maximum and minimum crack opening during a cycle. The resulting fatigueCOD is added to fatigue before the fatigue damage is introduced into the material. 2.2.10.3 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.
62
Fig. 2-31: Representative volume element with cracks. 2.2.11.1 Basic assumptions a) multiple cracking regime (hardening)
A set of parallel planar multiple cracks forms when maximum principal stress max = fc (first crack strength). Crack planes are perpendicular to the direction of max (-axis). The direction of a crack set is fixed. Secondary crack set may form in direction perpendicular to primary set if the maximum normal stress in the corresponding direction (-axis) exceeds fc. 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 , mc , , ij (notation: lower indices components of tensor or vector, upper indices ij multiple or localized crack mc, lc and association with primary or secondary crack direction , ) 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 mcmb (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 wc or wc). 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:
ij s ij mc , ij mc , ij lc, ij lc , ij
(2.88)
63
2.2.11.2 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.
mcmb
cracking strain mc
Fig. 2-32: Stress vs. cracking strain relations in crack-normal direction. 2.2.11.3 Crack sliding model The model for crack sliding phenomena is implemented by means of a variable shear retention factor . The shear retention factor is defined as a ratio of the material post-cracking shear stiffness Gc to its elastic shear stiffness G,
Gc . (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 , then the total engineering shear strain has only one non-zero component, which is obtained as:
2 2 2 2 2 , which can be rewritten with use of the shear bridging model (Kabele, 2000) as:
1 G 1
mc ,
mc ,
mc ,
lc ,
lc ,
(2.90)
1
mc ,
1 wc L Vf k Gf 2
1 wc L
1 Gc
(2.91)
M ( ) =
(2.92)
64
Vf k Gf 1 L() = 2 2 0 4 k G f 1 + 3Ef d f
, for 0
(2.93) L ( ) = 0 , for > 0 Here Vf is the fiber volume fraction, Gf is the fiber shear modulus, Ef is the fiber Youngs modulus, df is the fiber diameter, and k is the fiber cross-section shape correction factor. The quantity and indicates the crack opening in direction and respectively. The parameter 0 represents the limiting value of the crack opening displacement, when no tensile stress can be transferred across the crack, i.e. the point when the stress-displacement diagram in Fig. 2-32 drops to zero. These parameters are to be supplied by the user with the exception of the parameter 0 , which is automatically extracted from the provided stress-strain law for tension. The shear retention factor is then expressed as
1 1 1 1 1 1+ G + + + mc , mc , M ( ) M ( ) wc L( ) wc L( )
(2.94)
0 and Note that for an element containing only multiple cracks (before localization) mc , mc , 0 and 1/M and 1/L 1/L terms approach zero. For an uncracked element, approach zero, giving =1.
65
where p v,t
p p (2.96) k() = k( ) = k o + (1 k o ) 1 v,t p v v,t is the plastic volumetric strain at uniaxial concrete strength (onset of softening) and ko
p v
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: 1 2 c() = c( p ) = n1 1 v 1+ n2 1 where :
2
(2.97)
n1 = p / p v v,t
(2.98)
66
n 2 = ( p + t) / p (2.99) v,t v,t 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.
k() / c()
1.0
c k
0.8
0.6
0.4
0.2
ko
0.0
v,t
= pv
Fig. 2-33: Evolution of hardening (k) and softening (c) functions with respect to the plastic volumetric strain. 2.2.12.2 Plastic potential function The present plasticity model incorporates a non-associated flow rule using a polynomial plastic potential function (g), with Lode angle () dependency and adjustable order (n) :
1 g = A + a (2.100) + C + (B C)(1 cos3) 2 k c fc k c fc k c fc 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 () 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.
67
Fig. 2-34: Direction () of the incremental (a) and total (b) plastic strain vectors. 2.2.12.3 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 (Pa) c (MPa) ft (MPa) t e fco (MPa) 20 30 40 50 60 70
24377 0.2 1.917 1.043 0.5281 -4.32 4.9210-4 1.3310-3 7.342177 -8.032485
27530 0.2 2.446 1.227 0.5232 -9.16 6.5410-4 2.0010-3 5.436344 -6.563421
30011 0.2 2.906 1.376 0.5198 -15.62 8.0010-4 2.6710-3 4.371435 -5.73549
32089 0.2 3.323 1.505 0.5172 -23.63 9.3510-4 3.3310-3 3.971437 -5.430334
33893 0.2 3.707 1.619 0.5151 -33.14 1.0610-3 4.0010-3 3.674375 -5.202794
35497 0.2 4.066 1.722 0.5133 -44.11 1.1810-3 4.6710-3 3.43856 -5.021407
p v,t
t A B
68
C n Gf (MN/m) fc (Pa) c (MPa) ft (MPa) t e fco (MPa)
-3.726514 3 4.8710-5
80
-3.25626 3 6.4710-5
90
-3.055953 3 7.9210-5
100
-2.903173 3 9.2610-5
110
-2.797059 3 1.0510-4
120
-2.719067 3 1.1710-4
36948 0.2 4.405 1.816 0.5117 -56.50 1.3010-3 5.3310-3 3.245006 -4.871993 -2.659098 3 1.2910-4
38277 0.2 4.728 1.904 0.5104 -70.30 1.4110-3 6.0010-3 3.082129 -4.745867 -2.611426 3 1.4010-4
39506 0.2 5.036 1.986 0.5092 -85.48 1.5210-3 6.6710-3 2.942391 -4.637358 -2.572571 3 1.5010-4
40652 0.2 5.333 2.063 0.5081 -102.01 1.6210-3 7.3310-3 2.820644 -4.542587 -2.540158 3 1.6110-4
41727 0.2 5.618 2.136 0.5071 -114.00 1.7310-3 8.0010-3 2.713227 -4.458782 -2.512681 3 1.7110-4
p v,t
t A B C n Gf (MN/m)
(2.101)
where
k (
p eq
)=
J2
1 3
p y ( eq ) is the maximal shear stress and y is the uniaxial yield stress. This
y (
p eq
) =
+ H ,
p eq
=
p eq i =1
N inc
: p )
(2.102)
69
p y is the yield stress, H the hardening modulus and eq 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 ( ij ) = F P ( 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 p ( X ) : ( X ) k ( eq ) (r 1)k0 = 0
(2.104)
(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 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 Fig. 2-35.
70
Fig. 2-35: Effect of material parameter choice on cyclic response for E=210 GPa and y = 200 MPa.
71
(2.106)
Where 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. = 00 , the formulas are:
2sin , 3 ( 3 sin )
k=
6 c cos 3 ( 3 sin )
(2.107)
This corresponds to a outer cone to the Mohr-Coulomb surface. The inner cone, which passes through the tensile meridian where = 600 has the constants given by the following expressions:
2sin , 3 ( 3 + sin )
k=
6 c cos 3 ( 3 + sin )
(2.108)
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.
p eq = min( 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.
k = k0
p f c( eq )
f c
(2.110)
p In the above two formulas the expression f c( eq ) indicates the hardening/softening law, which is
based on the uniaxial compressive test. The law is shown in Fig. 2-36.
72 where determines the return direction. If < 0 material is being compacted during crushing, if = 0 material volume is preserved, and if > 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.
1 Ktt 2 = 0 0
0 Ktt 0
0 v1 0 v2 K nn u
(2.112)
For two-dimensional problems second row and column are omitted. The initial failure surface corresponds to Mohr-Coulomb condition (2.113) with ellipsoid in tension regime. After stresses violate this condition, this surface collapses to a residual surface which corresponds to dry friction.
c , 0
(2.113) c , c = ft 2 , 0 < ft c 2 ft
( c ) , = = 0 1 0 2 ( ft c )
2
c2 1 2 ( ft c )
= 0 , > ft
73 In tension the failure criterion is replaced by an ellipsoid, which intersect the normal stress axis at the value of ft with the vertical tangent and the shear axis is intersected at the value of c (i.e. cohesion) with the tangent equivalent to . The parameters for the interface model cannot be defined arbitrarily; there is certain dependence of the some parameters on the others. When defining the interface parameters, the following rules should be observed:
ft < c
(2.114)
c > 0,
It is recommended that parameters c, f t , are always greater than zero. In cases when no cohesion or no tensile strength is required, some very small values should be prescribed.
Trial stress
1 Final stress Residual surface
Initial surface
ft
Fig. 2-37: Failure surface for interface elements.
(2.115)
In general three-dimensional case in Fig. 2-37 and equation (2.113) is calculated as:
= 12 + 22
74
c Ktt 1
(a)
Ktt
min
ft
Knn 1
(b)
Knn
min
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 tt = (2.116) 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 K nn =
75 meters. It is suitable due to numerical reasons if stiffness is about 10 times of the stiffness of adjacent finite elements. There are two additional stiffness values that need to be specified in the ATENA input. They are min min denoted in Fig. 2-38 as K nn and K tt . 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).
s1 = ft , 4 v1 = 0.75 GF ft
(2.117)
/ft
I GF
c/c 0
II GF
s1 u1
s1c
u u eq f
v 1c
eq uv f
The evolution law depends on the equivalent nonlinear interface relative displacement
f f ueq = u 2 + v 2 1 + v 2 2 in 3D and ueq = u 2 + v 2 in 2D f f f f f
(2.118)
Where u f and v fi are the inelastic components of the relative interface displacement on the basis of their decomposition into elastic and nonlinear, i.e. fracturing part.
u = ue + u f vi = vi + v fi
(2.119)
This approach ensures that the degradation in shear affects also tensile strength and vice versa. For instance, when the interface is damaged in shear, the tensile strength is reduced as well. The typical behavior of the interface model with the softening evolution laws is shown in Fig. 2-38 by the dotted lines. The default behavior when no softening law is given is brittle with immediate drop to zero in tension and to the residual dry friction in shear. The behavior is shown in Fig. 2-38 by the solid black line. When user softening laws are defined for the interface material, it is recommended that the softening law for cohesion is always more ductile then the one for tensile strength, i.e. the cohesion should be higher than the tensile strength at any time during the softening process.
76
Fig. 2-40: Example of a cyclic response of the model in shear under constant normal pre-stress.
77
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 L represents limited ductility of steel.
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 as shown in Fig. 2-43.
78
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).
(2.120)
'
where s is the steel stress between the cracks (the steel stress in smeared reinforcement), scr the discrete reinforcement the steel stress is always s .
'
is the steel stress in a crack. If no tension stiffening is specified ts =0 and scr = s . In case of
= ( 0 r ) * + r
where
(2.121)
c1 c2 +
* = b * +
(1 + )
*R
(1 b ) *
1
R
* =
r , 0 r
R = R0
(2.122)
79
where R0 , c1 and c2 are experimentally determined parameters. The Fig. 2-44 shows the meaning of strain values r , 0 , and stress values r and 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)
* = f R ( eq ) , eq =
Nincr . i =1
i eq
(2.123)
Fig. 2-44: Cycling reinforcement model based on Menegotto and Pinto (1973).
80
s = max s 1
0 s s1
(2.124)
b =
max
s1 < s s 2
(2.125)
b =
max
( max f
ss ) , s s
2 3 2
f
s 2 < s s3
(2.126)
b =
s3 < s
(2.127)
81
Table 2.8-1: Parameters for defining the mean bond strenth-slip relationship for ribbed bars.
2 Value
Unconfined concrete* Bond conditions Good All other cases 0.6 mm 0.6 mm 2.5 mm 0.4
1.0 f
C
Confined concrete** Bond conditions Good All other cases 1.0 mm 3.0 mm clear rib spacing 0.4
2.5 f
C
S1 S2 S3
max
2.0
f
C
1.25
f
C
0.15 max
* Failure by splitting of the concrete **Failure by shearing of the concrete between the ribs
0.40 max
Table 2.8-2: Parameters for defining the bond strength-slip relationship for smooth bars.
Values
s1 = s 2 = s3
0.01 mm
0.1 mm 0.5
0.5
0.1 f
C
max = f
0.05
f
C
0.3
f
C
0.15
f
C
82
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.
83
Table 2.8-3: Parameters for defining the bond strength-slip relationship for smooth bars.
Point 1
Point 2 0.020 3.000 0.030 2.000 0.040 1.000 0.012 2.500 0.020 1.900 0.025 1.100
Point 3 0.044 0.700 0.047 0.700 0.047 0.700 0.030 0.900 0.030 0.900 0.030 0.900
Point 4 0.480 0.000 0.480 0.000 0.480 0.000 0.340 0.000 0.340 0.000 0.340 0.000
s/D
' b / 0.8 f cu
0.000 0.500 0.000 0.500 0.000 0.500 0.000 0.600 0.000 0.600 0.000 0.600
f c' < 60
Good
s/D
' b / 0.8 f cu
Bad
s/D
' b / 0.8 f cu
Excelent
s/D
' b / 0.88 f cu
f c' > 60
Good
s/D
' b / 0.88 f cu
Bad
s/D
' b / 0.88 f cu
max the maximum bond stress (max. value from the bond strength bond slip function).
In the figure, s is the current slip value, smax the maximum of the absolute slip value ever reached (damage variable), = f ( s ) is the bond strength function.
84
The response for a slip change si = si 1 + si is defined separately for 2 cases: (1) Loading range = f (s) (2) Unloading range
s 0 = 1 s 0 = 1
s smax
85
1 (2.128) ( mi n j + m j ni ) ij , L = 1 ( li n j + l j ni ) ij 2 2 where mi and li are chosen orthogonal vectors lying in the microplane and defining the shear
N = ni n j ij , M =
strain components. The constitutive relations for the microplane strains and stresses can be generally stated as:
N (t ) = Ft=0 [ N ( ), L ( ), M ( ) ] M (t ) = Gt =0 [ N ( ), L ( ), M ( )] L (t ) = Gt =0 [ N ( ), L ( ), M ( )]
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 . After the integration, the following expression for the macroscopic stress tensor is recovered (BAZANT 1984):
Nm 3 ( ) ij = sij d 61 w sij , where sij = N ni n j + 2M ( mi n j + m j ni ) + 2L ( li n j + l j ni ) 2 =
(2.129)
(2.130)
where the integral is approximated by an optimal Gaussian integration formula for a spherical surface; numbers label the points of the integration formula and w are the corresponding optimal weights.
86
b u u u b u u u ij , 1 ij , 2 ij , 3 ij and ij , 1 ij , 2 ij , 3 ij
where superscript b denotes the quantities in the localization band and the symbol superscripts u and m defines the quantities in the elastic spring in the direction m .
x u with
ij ,3ij
u
ij ,ij
b 3
1 u1
ij ,ij
L
3
h
1
2 u2
ij ,ij
h
1
87
Finite element
Localization element
Elastic springs
u
2
1
ij
ijb
2
iju
Localization band
h
2
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:
b b ij = F ( ij ) m u u ij = Dijlk m kl for m = 1...3
(2.131) (2.132)
The first formula (2.131) represents the evaluation of the non-linear material model, which in our case is the microplane model for concrete. The second equation (2.132) is a set of three elastic constitutive formulations for the three linear zones (springs) that are involved in the arrangement at Fig. 2-48. 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.
88
1 b1 1 u 1 11 h + 11 L 1h L 1 b u 22 = 2 22 2h + 2 22 2 L 2h L
11 =
33 =
1 b 3 u 33 h + 3 33 3 L 3h L 1 1 b 1 b u u 12 = 1 12 1h + 112 1L 1h + 2 12 2h + 212 L 2 L
3
L 2h
(2.133)
1 1 b 1 b u u 23 = 2 23 2h + 2 23 ( 2 L 2h ) + 3 23 3h + 3 23 ( 3 L 3h ) 2 L L 1 1 b 1 b u u 13 = 1 13 1h + 113 ( 1L 1h ) + 3 13 3h + 313 ( 3 L 3h ) 2 L L
(2.134)
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:
b u ij m e j = m ij m e j
for m = 1...3
(2.135)
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
(2.136)
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:
b u ij m e j = n ij m e j
(2.137)
The equation (2.137) is equivalent to a static constraint on the remaining stress and strain components of the elastic springs. Formulas (2.135) and (2.137) 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:
b u ij = m ij
for m = 1...3
(2.138)
b This means that the macroscopic stress must be equal to ij , 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 b microplane stress ij . 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: Initialization:
b u ij , ij , ij , ij b u ij = ij = ij
(2.139) (2.140)
L jh + j L ih ( Cijkl rkli 1) i j 2L L
u = ij
( i 1)
u ij
(i )
u + d ij
(i )
2iL jL 2 i L j L i L jh j L ih u ij ij i i L jh + j L ih L j h + j L ih
b u rij(i ) = ij ij
(i ) (i )
where Cijlk is the compliance tensor. The above iterative process is controlled by the following convergence criteria;
u d ij (i )
ij
<e ,
rij( i )
b ij
<e ,
u rij( i ) T d ij ( i ) b ij ij
<e
(2.145)
b The macroscopic stress is then equal to the stress in the localization band ij . 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. BAANT, 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. BAANT, 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 BAANT, 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 DInformation No 188, Fatigue of concrete structures, State of the art report.
90 CERVENKA, V., GERSTLE, K. (1972) - Inelastic Analysis of Reinforced Concrete Panels: (1) Theory, (2) Experimental Verification and application, Publications IABSE, Zrich, 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 Zrich, Switzerland, ISSN 10168664, 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. ERVENKA, J., BAANT 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. ERVENKA, 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
91 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 TwoDimensional 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 stressstrain 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 Ermdungsbeanspruchung, Schriftenreihe des Instituts fr 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 fr Gerissene Stahbetonscheiben, Nr.6 Forschungsbericht, Massivbau, Gesamthochschule Kassel. KOLMAR, W. (1986) - Beschreibung der Kraftuebertragung ber Risse in nichtlinearen FiniteElement-Berechnungen von Stahlbetontragwerken", Dissertation, T.H. Darmstadt, p. 94. KUPFER, H., HILSDORF, H.K., RSCH, 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.
92 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, IIComputational 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 21612185. 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.
93
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 3 7 4 8 r 1 9 5 6 h2 h1 2 s
t 3 7 4 8 r 1 t 3 7 4 8 r 1 9 5 h9 6 2 s 9 5 6 h8 2 s r 4 8 3 7
t h6 2 5 1 s
6 9
The available elements can be divided into three groups: plane elements for 2D, 3D and axisymmetric 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 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
94 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 6 3 1 10 7 4 2 11 8 5 9 6 3 1 Standard solution 4 2 10 7 8 5 11
95
y 1
s 3 2 r CCIsoTruss<xx> CCIsoTruxx<xxx> x
Node i
Function hi
1 (1 r ) 2 1 (1 + r ) 2
1 h3 2 1 h3 2
(1 r 2 )
Table 3.2-2 Sample points for Gauss integration of 1 node CCIsoTruss<xx> element.
Integration point
Coordinate r
Weight
96 1 0. 2.
Table 3.2-3 Sample points for Gauss integration of 2 and 3 nodes CCIsotruss<xxx> elements.
Integrati on point
Coordinate r
Weight
1 2
0.577350269189626 -0.577350269189626
1. 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 1 x1 = t x1 h1 + t x12 h2 + t x13 h3
1 2 3 x2 = t x2 h1 + t x2 h2 + t x2 h3
(3.1)
1 2 3 x3 = t x3 h1 + t x3 h2 + t x3 h3 t +t
X ( i 1) :
t +t
(3.2)
t +t
X (i )
t +t
(3.3)
t +t
97
(i Increment of Green Lagrange strain t 11) = to configuration at time t ) is calculated: t +t ( i ) t +t t 11
(i )
with
(i ) t 11
t +t ( i ) 2 t +t ( i 1) 2 l l r r 1 = 2 t 2 l r
2 2 2
(3.4)
(3.5)
Substituting (3.5), (3.3) into (3.4) after some math manipulation it can be derived:
h1 h1 t 1 h1 h2 r r x1 + r r h2 h1 t x1 + h2 h2 r r 1 r r h h h h 1 3 1 t x1 + 3 2 r r r r h1 h1 t 1 h1 h2 x2 + r r r r 1 h2 h1 t 1 h2 h2 t +t x2 + t BL 0 = 2 r r t l r r r h3 h1 t 1 h3 h2 x + r r 2 r r h h h h 1 1 1 t x3 + 1 2 r r r r h2 h1 t 1 h2 h2 x3 + r r r r h3 h1 t 1 h3 h2 r r x3 + r r
t
h1 h3 t 3 x1 r r h2 h3 t 3 t 2 x1 + x1 r r h h t 2 x1 + 3 3 t x13 r r h h 3 t 2 x2 + 1 3 t x2 r r h2 h3 t 3 t 2 x2 + x2 r r h3 h3 t 3 t 2 x2 + x2 r r h h 3 t 2 x 3 + 1 3 t x3 r r h2 h3 t 3 t 2 x3 + x3 r r h3 h3 t 3 t 2 x3 + x3 r r x12 +
(3.6)
98 h1 h1 t +t 1( i 1) h1 h2 u1 + r r r r h2 h1 t +t 1( i 1) h2 h2 u1 + r r r r h h h3 h2 1( 3 1 t +t u1 i 1) + r r r r h1 h1 t +t 1( i 1) h1 h2 u2 + r r r r 1 h2 h1 t +t 1( i 1) h2 h2 = u2 + 2 r r t l r r r h3 h1 t +t 1( i 1) h3 h2 u2 + r r r r h h h h 1( 1 1 t +t u3 i 1) + 1 2 r r r r h2 h1 t +t 1( i 1) h2 h2 u3 + r r r r h3 h1 t +t 1( i 1) h3 h2 u3 + r r r r h1 h3 t +t 3( i 1) u1 r r h2 h3 t +t 3( i 1) t +t 2( i 1) u1 + u1 r r h h t +t 2( i 1) u1 + 3 3 t +t u13( i 1) r r h1 h3 t +t 3( i 1) t +t 2( i 1) u2 + u2 r r h2 h3 t +t 3( i 1) t +t 2( i 1) u2 + u2 r r h h t +t 2( i 1) 3( u2 + 3 3 t +t u2 i 1) r r h1 h3 t +t 3( i 1) t +t 2( i 1) u3 + u3 r r h h t +t 2( i 1) 3( u3 + 2 3 t +t u3 i 1) r r h3 h3 t +t 3( i 1) t +t 2( i 1) + u3 u3 r r
t +t
u12( i 1) +
t +t t
( BLi11)
(3.7)
and
t +t t
(n BNL1)
h1 r 1 = t 0 l r 0
0 h1 r 0
0 0 h1 r
h2 r 0 0
0 h2 r 0
0 0 h2 r
h3 r 0 0
0 h3 r 0
0 0 h3 r
(3.8)
( i 1)
(i t +tt S111) = 0 0
t +t ( i 1) t 11
0 S 0
0 0 , t +t ( i 1) t S11
t +t t
(i S ( i 1) = [ t +tt S111) ]
(3.9)
t +t t
(i X 1,1) , which
t +t t
(i X 1,1)
(3.10)
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) ):
99
(i 11)
t +t ( i ) 2 t +t ( i 1) 2 1( l ) ( l ) = t 2 l 2
(3.11)
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).
y 6
2 5
100
Table 3.3-1: Interpolation functions of CCIsoQuad<...> elements.
Node i
Function hi
i=5 1
1 (1 + r )(1 + s ) 4 1 (1 r )(1 + s ) 4 1 (1 r )(1 s ) 4 1 (1 + r )(1 s ) 4 1 (1 r 2 )(1 + s ) 2 1 (1 s 2 )(1 r ) 2 1 (1 r 2 )(1 s ) 2 1 (1 + r )(1 s 2 ) 2 1 (1 r 2 )(1 s 2 ) 2 1 h5 2 1 h5 2
I=6
i=7
i=8
1 h8 2
i=9
1 h9 4 1 h9 4
1 h6 2 1 h6 2 1 h7 2 1 h7 2 1 h8 2
1 h9 4 1 h9 4 1 h9 2 1 h9 2 1 h9 2 1 h9 2
101
Table 3.3-2: Sample points for Gauss integration of 4 nodes CCIsoQuad<...> element.
Integrati on point
Coordinate r
Coordinate s
Weight
1 2 3 4
1. 1. 1. 1.
Table 3.3-3: Sample points for Gauss integration 5 to 9 nodes CCIsoQuad<...> elements.
Integrati on point
Coordinate r
Coordinate s
Weight
1 2 3 4 5 6 7 8
Equations (3.12) through (3.21) present CCIsoQuad<...> axisymmetric element formulation. 2D element formulation is simply obtained by removing terms associated with circumferential (i (i strains and stresses t +tt 33) , t +tt S33) . Incremental strains:
(i ) t 11
1 2 1 2
(( u
(i ) 2 t 1,1
) +( u ) )
t (i ) 2 2,1
(( u
(i ) 2 t 1,2
) +( u ) )
t (i ) 2 2,2
(i ) t 12
1 (i (i ( t u1,2) + t u2,1) ) + 2 1 t +t ( i 1) ( i ) t +t ( i 1) ( i ) t +t ( i 1) ( i ) t +t ( i 1) ( i ) ( t u1,1 t u1,2 + t u2,1 t u2,2 + t u1,2 t u1,1 + t u2,2 t u2,1 ) + 2 1 (i (i (i (i ( t u1,1) t u1,2) + t u2,1) t u2,2) ) 2
t +t
(i ) 33
u(i ) = t1 + x1
( x)
t 1
u1( i ) u1( i )
2
1 u (i ) + t1 2 x1
(3.12)
Displacement derivatives:
t
(i ) i, j
t +t ( i 1) t i, j
(i ) =
t +t t
( BLi 1) U ( i )
(3.14)
(3.15)
103 Linear strain-displacement matrix constant part: t h1,1 0 t +t BL 0 = t h1,2 t h1 t x1 where hi t x j (3.17) 0 t h1,2 t h1,1 0 h2,1 0 t h2,2 h2 t x1
t
0 0 t hn ,2 h ... t n x1
0 t hn ,2 t hn ,1 0
(3.16)
hi , j =
x1 = hk t x1k
k =1
l l
h h
t +t t
( BLi11)
h + 0
l l
h h
l l
h h
t +t ( i 1) 2 33 t 1
h x
0 h
... ...
t +t ( i 1) 11 t n ,1 t +t ( i 1) 11 t n ,2 t +t ( i 1) t +t ( i 1) 11 t n ,2 11 t n ,1
l l
h h
t +t ( i 1) 21 t n ,1 t +t ( i 1) 21 t n ,2 t +t ( i 1) t +t ( i 1) 21 t n ,2 21 t n ,1
l l
h h
(3.18)
t +t ( i 1) n 33 t 1
h x
104 where
t +t ( i 1) 11
= t hk ,1 t +tt u1k ( i 1)
k =1 n
t +t ( i 1) 12
= t hk ,2 t +tt u1k ( i 1)
k =1 n
t +t ( i 1) 21
k = t hk ,1 t +tt u2 ( i 1) k =1
(3.19)
t +t ( i 1) 22
k = t hk ,2 t +tt u2 ( i 1) k =1
t +t ( i 1) 33
1 n hk t x1 k =1
t +t k ( i 1) t 1
t +t t
(i BNL 1)
0 0 t h2,1 t h2,2 0
hn ,1 t hn ,2 0 0 h ... t n x1
t
0 0 t hn ,1 t hn ,2 0
0 0 t +t ( i 1) t S33 0 0
(3.20)
S S
0 0
t +t ( i 1) t 11 t +t ( i 1) t 21
0 0
t +t ( i 1) t 12 t +t ( i 1) t 22
S (i 1)
0 0 0
t +t t (i S 221)
S S 0
S S 0
(3.21)
t (i S ( i ) = t +tt S111) t +t t (i S 211) t +t t (i S331)
In case of the simplified 3D analysis, i.e. elements CCIsoQuad2_5<...>, the equations are further extended as follows:
105 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:
(3.22)
(3.23)
where:
xlocal ,i , xglobal ,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. 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.
106
x3 3
xl o c a l , 3 4 X O X 2 xl o c a l , 2
1 x1 xl o c a l , 1
x2
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 +t t
(i BNL 1)
t h1,1 h t 1,2 0 = 0 0 0
0 0
t 1,1 t 1,2
0 0 0 0
t 1,1
h2,1 0 0 t h2,2
t
0 0
h3,1 0 0 t h3,2
t
0 0
hN ,1 t hN ,2
t t t
0 0 hN ,1 hN ,2 0 0
h 0
h 0 t h1,2
0 0 0 0 t hN ,1 t hN ,2
(3.24)
107 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 6
1 4 2 r
Node i
Function hi
i=4 1
i=5
i=6
1 h6 2
1 r s
1 h4 2 1 h4 2 1 h5 2 1 h5 2
s
4 r (1 r s )
1 h6 2
4 5 6
4rs
4 s (1 r s )
108
Table 3.4-2: Sample point for Gauss integration of 3 nodes CCIsoTriangle<...> elements.
Integration point
Coordinate r
Coordinate s
Weight
1/3
1/3
1/2
Table 3.4-3: Sample points for Gauss integration of 3 to 6 nodes CCIsoTriangle<...> elements.
Integration point
Coordinate r
Coordinate s
Weight
1 2 3
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.
109
t z
10 8 1 7 5 2 6 r 4 9 3 CCIsoTetra<xxxx> ............. CCIsoTetra<xxxxxxxxxx>
t
Fig. 3-7 Geometry of CCIsoTetra<...> elements.
110
t z
10 8 1 7 5 2 6 r 4 9 3 CCIsoTetra<xxxx> ............. CCIsoTetra<xxxxxxxxxx>
t
Fig. 3-9 Geometry of CCIsoWedge<...> elements. Table 3.5-1 Interpolation functions of CCIsoTetra<...> elements.
Node i
Function hi
i=5
i=6
i=7
i=8
i=9
i = 10
1 r s t
1 h5 2 1 h5 2 1 h6 2 1 h6 2
1 h7 2 1 h8 2 1 h7 2 1 h8 2 1 h9 2 1 h9 2
1 h10 2
s
t
4 r (1 r s t ) 4 rs (1 t ) 4 s (1 r s t ) 4 rt (1 s )
1 h10 2
5 6 7 8
111 9 10
4 s t (1- r ) 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
1/4
1/4
1/4
1/6
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 2 3 4
Function hi
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 + r )(1 + s)(1 + t ) h 9 8 2
1 h12 2
1 h17 2
112 2
1 (1 r )(1 + s )(1 + t ) 8 1 1 h9 h10 2 2 1 h18 2
1 (1 r )(1 s )(1 + t ) 8
1 1 h10 h11 2 2
1 h19 2
1 (1 + r )(1 s )(1 + t ) 8
1 1 h11 h12 2 2
1 h20 2
1 (1 + r )(1 + s )(1 t ) 8
1 h13 2
1 1 h16 h17 2 2
1 (1 r )(1 + s )(1 t ) 8
1 1 h13 h14 2 2
1 h18 2
1 (1 r )(1 s )(1 t ) 8
1 1 h14 h15 2 2
1 h19 2
1 (1 + r )(1 s )(1 t ) 8
1 1 h15 h16 2 2
1 h20 2
1 (1 r 2 )(1 + s )(1 + t ) 4
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
113 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
114
Table 3.5-5 Sample points for Gauss integration of 8 nodes CCIsoBrick<...> element.
Integrati on point
Coordinate r
Coordinate s
Coordinate t
Weight
1 2 3
0.5773502691896 0.577350269189626 26 0.5773502691896 26 0.577350269189626 0.577350269189626 0.5773502691896 26 0.5773502691896 0.577350269189626 26 0.5773502691896 0.577350269189626 26 0.5773502691896 26 0.577350269189626 0.577350269189626 0.5773502691896 26 0.5773502691896 0.577350269189626 26
1. 1. 1.
1.
1.
1.
1.
1.
Table 3.5-6 Sample points for Gauss integration of 9 to 20 nodes CCIsoBrick<...> element.
Integrati on point
Coordinate r
Coordinate s
Coordinate t
Weight
115 1 2 3 4 5 6 7 0.7745966692414 83 0.7745966692414 83 0.7745966692414 83 0.7745966692414 83 0.7745966692414 83 0.7745966692414 83 0.7745966692414 83 0.7745966692414 83 0. 0. 0. 0. 0. 0. 0. 0.7745966692414 0.774596669241483 83 0.7745966692414 83 0. 0.1714677641 0.2743484225 0.1714677641 0.2743484225 0.4389574760 0.2743484225 0.1714677641
0.2743484225
10 11 12 13 14 15 16
17
0.
0.4389574760
19
0.1714677641
20
0.2743484225
21
0.1714677641
22
0.2743484225
23
0.
0.
0.4389574760
24
0.
0.774596669241483
0.2743484225
25
0.1714677641
26
0.2743484225
27
0.7745966692414 0.774596669241483 83
0.1714677641
117
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 = 4 s (1 r s ) 1+ t 2 1 t hv2 = 2 hv3 = (1 t 2 ) hv1 =
Node I
Function hi
1 2 3 4 5 6 7 8 9 10 11 12
hh1 hv1
1 h7 2 1 h7 2 1 h8 2 1 h8 2
1 h9 2
1 h13 2
hh2 hv1
1 h14 2
1 h9 2 1 h10 2 1 h10 2 1 h11 2 1 h11 2 1 h12 2 1 h12 2 1 h13 2 1 h14 2 1 h15 2 1 h15 2
hh3 hv1
hh1 hv2
hh2 hv2
hh3 hv2
hh4 hv1 hh5 hv1 hh6 hv1 hh4 hv2 hh5 hv2 hh6 hv2
118 13 14 15
hh1 hv3 hh2 hv3 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 2 3 4 5 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 2 3 4 5 6 7 8
(3.25)
Displacement derivatives: u
(i ) i, j
t +t ( i 1) t i, j
(i (i ) = t 11)
(i 22)
(3.27)
U ( i ) =t +t U ( i ) t +t U (i 1) =
1( u1 i )
u1(i ) 2
1( u3 i )
u12(i )
2( u2 i )
2( u3 i ) ... u1n (i )
n u2 ( i )
n u3 (i )
(3.28)
0
t 1,2
0 0
t 1,3
0 0
... ...
hn ,1 0 0 hn ,2 0 hn ,3
t
0 hn ,2 0 hn ,1 t hn ,3
t
0 h t h1,3
t 1,1
h2,3 ...
t
0 t h1,2
t 1,1
h2,1 ...
0 0 hn ,3 t 0 t hn ,2 t hn ,1
(3.29)
120
hi , j =
hi t x j
(3.30)
t +t t
( BLi11)
l l l
h h h
l l l
h + h + h +
l l l
h h h
l l l
h h h
l l l
h + h + h +
l l l
h h h
l l l
h h h
l l l
h + h + h +
l l l
h h h
(3.31)
... ... ... ... ... ...
t +t ( i 1) 31 t n ,1 t +t ( i 1) 31 t n ,2 t +t ( i 1) 33 t n ,3 t +t ( i 1) t +t ( i 1) 31 t n ,2 32 t n ,1 t +t ( i 1) t +t ( i 1) 32 t n ,3 33 t n ,2 t +t ( i 1) t +t ( i 1) 31 t n ,3 33 t n ,1
l l l
h h h
l l
h + h + h +
l l l
h h h
where
t +t ( i 1) ij
= t hk , j
k =1
t +t k ( i 1) t i
(3.32)
0 0 0 0 0 0
t 1,1 t 1,2
t t t
h2,1 ... h2,2 ... h2,3 ... 0 ... 0 0 0 0 0 ... ... ... ... ...
t t t
hn ,1 hn ,2 hn ,3 0 0 0 0 0 0
0 0 0 t hn ,1
t
t +t t
(i BNL 1)
h t h1,3 0 0 0
hn ,2 t hn ,3 0 0 0
h t h1,3
0 0 0 0 0 0 t hn ,1 t hn ,2 t hn ,3
(3.33)
t +t ( i 1) t 12 t +t ( i 1) t 22 t +t ( i 1) t 32
S S
S ( i 1)
S 0 0
t +t ( i 1) t 13 t +t ( i 1) t 23 t +t ( i 1) t 33
S S
0 0
t +t ( i 1) t 11 t +t ( i 1) t 21 t +t ( i 1) t 31
0 0
t +t ( i 1) t 12 t +t ( i 1) t 22 t +t ( i 1) t 32
0 0
t +t ( i 1) t 13 t +t ( i 1) t 23 t +t ( i 1) t 33
0 0 0 0 0
t +t ( i 1) t 11 t +t ( i 1) t 21 t +t ( i 1) t 31
0 0 0 0 0
t +t ( i 1) t 12 t +t ( i 1) t 22 t +t ( i 1) t 32
S 0 0
0 S S
0 S S
0 S S
0 0 0 0
t +t t
(i S221)
0 0 0 0
t +t t
(i S331)
S 0 0 0
S 0 0 0
S 0 0 0
0 S S S
0 S S S
0 0 0 0 t +t ( i 1) t S13 t +t ( i 1) t S 23 t +t ( i 1) t S33 0 0
(i S ( i ) = t +tt S111)
t +t ( i 1) t 12
t +t t
(i S231)
t +t ( i 1) t 13
(3.34)
Ri =
where
ui kA n direction
(3.35)
k is spring material stiffness parameter set by &MATERIAL SPRING command, (parameter k has character of multi-linear Young modulus),
122
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
CCPlaneSpring CCSpring
area A y
123
1 =
A1 A A , 2 = 2 , 3 = 3 A A A
(3.36)
1 + 2 + 3 = 1
Using the quadratic interpolation function, the displacement components u( i ), v( i ) is written in the terms of triangular coordinates i and nodal displacement vectors : (3.37) u ( i ) = F ( i )T u, v ( i ) = F ( i )T v The displacement vectors u, v contain six components of the nodal displacements and the vector F ( 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)
4 3 1}
T
F ( i ) = { 1 (2 1 1) 2 (2 2 1) 3 (2 3 1) 4 1 2
4 2 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)
x =
u ( x, y ) x
, y =
v ( x, y ) y
, =
u ( x, y ) y
v ( x, y ) x
(3.41)
which is written in terms of the natural coordinates i and the nodal displacements vectors u, v:
( i ) = F T
The stiffness matrix:
K = F DF dV
T V
u v
(3.42)
(3.43)
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 = Bd
e x U O u e y = O V g V U v
(3.44)
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 = { x1 x 2
x 3 } , e y = { y1 y 2 y 3 } , g = { x1 x 2 x 3 }
T T
(3.45) (3.46)
u = { u1 u2 u3 u4 } , v = { v1 v2 v3 v4 }
T
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
125 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
4b2 4b1 .
4a2 4a1 . (3.47)
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 = uu K vu
Kuv K vv
(3.48)
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 + H T ) + d33 C K vv = St d 22 C + d 23 (H + H T ) + d 33 A
K uv = St d12 H + d13 A + d 23 C + d 33 H T
(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 Where the area integration matrix Q is: (3.50)
2 1 1 1 Q= 1 2 1 12 1 1 2
(3.51)
The element stiffness matrix of the 5-node quadrilateral, Fig. 3-13(b), is composed of the two 4node 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:
126
K K10 = ee K ie
K ei K ii
(3.52)
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:
K = K ee K ei K ii K ie
(3.53)
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.
(3.54)
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.
127
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 1, 2, They are introduced as variables to be determined by the analysis.
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: F2 = F1 exp( i ) Q (3.55)
128 In this expression i is the angular change of the cable direction at the deviator i and is the friction coefficient. In the above equation the constant part of the friction Q = k R i , 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 Ri 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 1, 2 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 + 2 1 )
(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 i . 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.
129
u
i
u lo l
i+1
i+1
i =
(ui +i + i +1 ui i ) E li
(3.57)
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 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 .
c = c ( )
The equilibrium condition for reinforcement bar with prescribed bond yields:
p c x A
(3.58)
(3.59)
where 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):
A( i i 1 ) li + li 1 p c 2
(3.60)
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:
130
right =
i +1li +1 + i li
li +1 + li (3.61)
left =
i li + i 1li 1
li + li 1
c =
( right left ) A 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.
131
Linear geometry v
2 4 3 CCIsoGap<xxxx> 1
2D
r,u(r)
2
Nonlinear geometry v 5
1 r, u(r)
4 6 3 CCIsoGap<xxxxx_x>
w
2 5 3 6 1 4
3D
r,u(r,s)
2 8 5 11 7
10 9 3 12 6
w
1 4
r,u(r,s)
s,v(r,s)
CCIsoGap<xxxxxx>
s,v(r,s)
CCIsoGap<xxxxxxxxxxxx> 13 1 5 4 8 2 9 10 3 7 11 15 5
w
2
1 12 4 8 16
r,u(r,s) 6
r,u(r,s) 6
14
3 7 CCIsoGap<xxxxxxxx>
s,v(r,s)
s,v(r,s)
CCIsoGap<xxxxxxxxxxxxxxxx>
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 F = = tt F 0
0 u = Du K nn v
(3.62)
132 in which u , v 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. F , F 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 v, u and are defined as follows:
1 h1 = (1 + r ), 2 1 h2 = (1 r ) 2
h u = 1 0
0 h1
h2 0
0 h2
h2 0
0 h2
h1 0
(3.63)
The rest of the element derivation is the same as in case of any other elements, i.e. the stiffness matrix K = T DBdV , vector of internal forces Q = T 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 F >Rti (Rti is the interface tensile strength force) and the corresponding constitutive law is (stress free interface):
F 0 = F 0
op op The stiffness coefficients are set to small, but nonzero values K tt , K nn .
(3.64)
The interface element is considered closed if F Rti. The stiffness coefficients are set to large
cl cl values K tt , K nn . 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
133 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 = 2 x
(3.65)
where x is the distance from the axis of symmetry. There are two special options for processing the gap elements: Initial gap opening It is possible to "open" gap at a particular load step, typically the first step of the analysis, i.e. we can introduce to the gaps something like initial element strains in case of ordinary finite elements. This is achieved by LOAD INITIAL GAP ... INIT_STEP_ID step_id command. Upon that, during calculation of the (gap) element at the step step_id an artificial opening of the interface is introduced. Its value is the distance between upper and lower element surfaces/lines (with reference to undeformed structural shape). The GAP element load is typically used as follows: we have a structure with a base and upper blocks. The upper block falls down towards the base block that is typically fixed. The structure is solved by introducing a layer of gap elements between the base and upper blocks and applying the GAP element load (for these gaps elements) in the 1st step. As a results, in the first steps the gaps will open to the distance between the blocks. It involves some tensional forces, but as the interface material usually sustains only compression forces, they can be neglected. In next steps the upper block gradually is falling down to the base block until it hits it. At this moment interface gaps get fully closed, they change their regime form tension to compression and the upper block gets fully supported by the base block. Moving gaps 1 Suppose we have a structure has a base block and an upper block sitting on the base block. The base block is fixed, the upper block is dragged on the upper surface of the base block. The blocks are not mutually interconnected, only some friction and cohesion forces exist between them. Such problems can be modelled by the RESET_DISPLS n flag for the CC2DInterface / CC3DInterface. If this flag is input, then the upper and bottom surface/lines for all corresponding elements are realigned at the end of each step as shown for 2D elements in the following picture. The 3D gaps element are realigned in the same way. Of course, the boundary surface/lines projection of the gap interface (and thus its "moving" can be used in more complex situation, but the essence of the described technique remains the same. The layer of interface elements are typically connected to the bottom/ upper block of structure by MASTER SLAVE NODAL LISTS boundary conditions, where we must not forget to use the flag PROCESS_FLAG USE_CURRENT_COORDS. It will assure that after realigning the interface gets properly connected to the rest of the (deformed) structure.
134
Fig. 3-20 Moving gap 2D element Note that the option of the gap's initial opening and the reset displacements flag can be combined. Both these special processing options are possible, because the ATENA software uses incremental approach to solve the structure. Thus changing shape of the gap (at the end of the steps) will not harm governing equilibrium equations.
y 2
1 CCCircumferentialTruss x
1 CCCircumferentialT russ2 x
135 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 centrepoint 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.). The truss element center has at reference time t and t + t ( i 1) co-ordinates t X = [t x1 , t x1 ] and
t +t ( i 1)
X = [t +t t l = 2 t x1 and
( i 1)
( i 1)
(i )
with
(i ) t 11
t +t ( i ) 2 t +t ( i 1) 2 1( l ) ( l ) = 2 2 (tl)
t
(3.66)
1( u1 i ) is co-ordinate increment
t +t ( i )
1 1( 1( = 2 t ( t x1 + t u1 i 1) + t u1 i ) ) . Note that
( t +t x1 t +t
(i)
( i 1)
(i ) t 11
4 2
(( x + u
t
1 1
1( i 1) 1
1( 1 1( + t u1 i ) ) ( t x1 + t u1 i 1) ) 2
(x)
t
1 2 1 1( i ) t 1 t 1 1 2
)=
(3.67)
1( i ) t 1 t 1 1
u u u + x ( x1 )
1( i 1) 1( i ) 1 t 1 t 1 2
1 u + 2 x
Separating
t +t t
BL 0 =
1 1 x1
t 1 2 1
(3.68) (3.69)
t +t t
( i 1) L1
(x)
1
t
1( u1 i 1)
and
t +t t
(n BNL1) =
(x)
1 2 1
(3.70)
S ( i 1) =
t +t t
(i S ( i 1) = [ t +tt S111) ]
(3.71)
136 The formulation is completed by relationship for element deformation gradient yields:
t +t t
t +t t
(i X 1,1) , which
(i ) 1,1
= 1+ e
(i ) t 11
1 1( x1 + t u1 i ) )
1 x1
(3.72)
(i ) t 11
t +t ( i ) t
l l
2 t 1 t 1( i ) 2 2 t 1 2 4 ( x1 + u1 ) 4 ( x1 ) = = 2 1 4 2 ( t x1 )
2 1 1( 1 x1 + t u1 i ) ) t x1 t 1 x1
(3.73)
137 The essential point in the elements 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 elements 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 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. 3-25. 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. 3-25. 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 elements vectors and matrices are derived based on the internal geometry as depicted also in Fig. 3-25. 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. 3-22 (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. 3-22 (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. 3-23 is not recommended, as the master-slave relations induced by the fixed thickness of the shell may cause numerical problems.
Shell1
Brick
Shell2
(a)
Shell3
(b)
138
139
t
s X3
L t
V3k
t t
X2
G X3
node k
t
V2k
X 1L
t
V1k
t t
X1
X 2G
The vectors V1k , V2k , V3k are defined as follows: Firstly, two auxiliary vectors V1 , V3 are calculated. Vector V3 at a point is defined as a line joining bottom and top coordinates at the node k (prior any deformations, i.e. at reference configuration). The second vector V1 is normal to V3 and is parallel to plane of global X 1G and
t t
0 0
G X 3 . Hence:
(3.74)
t V1 = V3 2 , 0, 0
t t
(3.75)
After that, the coordinate system V1k , V2k , V3k itself is defined. The vector V3k is constructed in 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 t
V2k = V3k V1
t t
(3.76) (3.77)
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 , .
140 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. Sometimes, it is advantageous to modify the nodal coordinate system so that V3k remains unchanged but V1k and V2k are rotated (along V3k ) to a certain direction. Note however, that mutual orthogonality of V1k , V2k , V3k must not be damaged.
t t t t t t t t t
t
(3.78)
X 2L = X 3L V1k X 1L = X 2L X 3L
t t t t
(3.79)
As the nodal coordinate system V1k , V2k , V3k can rotate along V3k , the local coordinate system would X 1L , X 2L , X 3L rotate simultaneously along X 3L . This definition allows for user defined shell local coordinate system that is common for all shell elements, irrespective of their incidences. Note that unlike V3k the vector X 3L is always normal to the element mid-plane surface.
t t t t t t
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. 3-24. 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.
141
11 4 20 15 8 r 16 12
t3 1 9 19 17
10 2 7 14 6 13 5
Input geometry
18
11 4 12 15 8 r 16
t3 117 9
10 2 7 18 5 14
External geometry
s 6 13
Internal geometry
3 4 r 2 5 1 9 8 6 7 s
142
Fig. 3-26 Degenerate shell Ahmad element coordinate systems and degree of freedoms (DOFs)
143
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 k k function, r,s,t are isoparametric coordinates (see Fig. 3-25), t x2 ,top and t x2 ,bot are vector of k k t x3 ,top t x3 ,bot top and bottom coordinates of point k, see Fig. 3-27.
node k
Using the above the equation (3.80) can be rewritten in the following form: t k ,mid t V3k 1 x1 x1 N t t t k ,mid t t k x2 = x = hk x2 + 2 V3 2 [thick ]k k =1 k t x3 t x3 ,mid t k V3 3
t
(3.81)
where [thick ]k is element thickness in node k (i.e. distance between top and bottom points) and
144
t x1top t x1bot t x1mid 1 t top t bot t mid x2 = 2 x2 + x2 mid t xtop t xbot t x3 k 3 k 3 k are coordinates of mid surface.
(3.82)
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 elements DOFs , see (10). Every point thus has five degree of freedom,
mid mid t u1mid ,t u2 ,t u3 ,t ,t . Displacement vector is calculated by: T
t V2k 1 t k V 22 t k V2 3
V1k 1 t k t k V1 2 t k t k V1 3
t T
(3.83)
mid mid The original displacement vector at point k has the form t u1mid ,t u2 ,t u3 ,t ,t . 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.
145
k u k ,mid = [ t u1k ,mid , t u2 , mid , t u3k , mid ] t
V3k
node k
t t
V2k
V1k
The general definition for Green-Lagrange strain tensor has the form (see eq. (1.8)):
t 0 ij
1 t ( 0 ui, j + 0t u j ,i + 0t uk ,i 0t uk , j ) 2
(3.84)
Using the above equation and applying the Von-Karman assumption, Eqn. (3.84) can be written as:
t u1 2 t t x1 1 u3 t t u2 2 x1 t 0 11 t 2 1 t u3 t x2 0 22 t u t u t 2 0t12 = t 1 + t 2 + 2 x2 = 0t L + 0t NL t x2 x1 t u t u 3 3 2 0 13 t u1 t u3 t x + t x t 2 1 0 23 t x + t x 2 3 1 0 t t u2 + u3 0 t x t x 2 3
(3.85)
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.
146
and 0t 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).
t 0 L
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):
t t 0 S k = 0 S11
t 0
S22
t 0 12
t 0 13
t 0
S23
(3.86)
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.
147 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 ) 2 The actual values in center point can be calculated by:
a9 = ai hi (r = 0, s = 0) + a9
i =1
8
(3.87)
(3.88)
where hi are values of interpolation function at point (0,0), ai are corresponding node values, a9 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. 3-29.
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.
148
Fig. 3-29 Node notation for element variants of the Ahmad shell 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.
149
Problem of membrane and shear locking for linear analysis are summarized in Fig. 3-30. 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)
Elements 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. 3-31.
150
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. ):
(3.89)
where fi are element nodal values of function f and hi' are interpolation functions corresponding to two by two interpolation and a node i.
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. 3-32. 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. 3-33. 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.
152
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
153 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 V1 1 t u1k , N ,top t V k t k , N ,top 2 1 u2 t k k t u3 , N ,top V3 1 t k , N ,bot = u1 tV k t k , N ,bot u 11 2 t k , N ,bot t u3 V2k 1 t k V3 1
t
1 thick 2 0 0 1 thick 2 0 0
V2k 2 V3k 2
V2k 3 V3k 3
(3.90)
The next step in the elements derivation is to write transformation of the left-hand side vector of (3.90) from nodal to global coordinate system. It reads:
tV k 11 t k ,top u1 t k t k ,top V1 2 u2 t k k t u3 ,top V1 3 t k ,bot = u1 0 t u k ,bot t 2 ,bot 0 k u3 0
t
V2k 1
V3k 1
0 0 0
t
0 0 0
t
V2k 2
V3k 2 V3k 3 0 0 0
t
V2k 3 0 0 0
k 1 1
k 2 1
V1k 2 V1k 3
V2k 2
V2k 3
0 t k , N ,top u t u1k , N ,top 0 t 1k , N ,top t k , N ,top u u2 t 2 , N ,top k 0 uk t u3 , N ,top 3 = T2 t k , N ,bot t k t k , N ,bot u1 u1 V3 1 t u k , N ,bot t u k , N ,bot t k 2 t 2 , N ,bot V3 2 t k , N ,bot k u3 u3 t k V3 3
(3.91)
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
154 t u1k ,top t u1k ,mid t u1k ,mid t k ,top u2 t k ,mid t k ,mid k u2 u2 t u3 ,top k k t k ,bot = T2 T1 t u3 ,mid = T t u3 ,mid u1 t k t k t u k ,bot 2 tk tk t k ,bot u3 where T
2 2 2 t k t t V1 1 + V2k 1 + V3k 1 t k t k t k t k t k t k V1 1 V1 2 + V2 1 V2 2 + V3 1 V3 3 t k t k t k t k t k t k V1 1 V1 3 + V2 1 V2 3 + V3 1 V3 3 2 2 2 t k t t V1 1 + V2k 1 + V3k 1 t k t k t k t k t k t k V1 1 V1 2 + V2 1 V2 2 + V3 1 V3 3 t k t k t k t k t k t k V1 1 V1 3 + V2 1 V2 3 + V3 1 V3 3
(3.92)
( ) ( ) ( )
( V ) +( V ) +( V )
t k 1 2 2 t k 2 2 t 2 t k 3 2 t t t t t t t t t t t
V1k 2 V1k 3 + V2k 2 V2k 3 + V3k 2 V3k 3 V1k 1 V1k 2 + V2k 1 V2k 2 + V3k 1 V3k 3
t
( V ) +( V ) +( V )
t k 1 3 2 t k 2 3 t 2 t k 3 3 t t t t t t t t t t
( ) ( ) ( )
( V ) +( V ) +( V )
t k 1 2 2 t k 2 2 t 2 t k 3 2 t t t t t
( V ) +( V ) +( V )
t k 1 3 2 t k 2 3 2 t k 3 3
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 u1k , mid t k ,top t k ,mid u2 u2 t u k ,top k t u3 ,mid = T' t 3 ,bot k t k u1 t u k ,bot tk t 2 ,bot k u3 T13 T11 T12 2 2 2 T23 T21 T22 2 2 2 T33 T31 T32 2 2 2
t
(3.93)
V1k 2
t
V1k 3
t
V1k 1 thick k
t
V1k 2 thick k
t
V2k 1
V2k 2
thick k
thick k
155 As noted earlier, the original set of DOFs at a node comprises 5 DOFs, whilst the new one has
k k k k six DOFs. Consequently, one DOF from t u1k ,top ,t u2 ,top ,t u3 ,top ,t u1k ,bot ,t u2 ,bot ,t u3 ,bot T
must be
fixed.
The presented
k k work prefers to constrain t u3 ,bot but u1k ,bot or u2 ,bot are also good
k candidates, if t u3 ,bot can not be fixed due some numerical problems, usually due to a special
k k u3 ,top + (T16 T13 ) u1k ,mid + (T26 T23 ) u2 ,mid + .... (T56 T53 ) k = t t t k u3 ,top thick k V1k 3 k + thick k V2k 3 k t t t t
(3.94)
k k k k k t u1k ,top ,t u2 ,top ,t u3 ,top ,t u1k ,bot ,t u2 ,bot ,t u3 ,bot , which is then used to constrain t u3 ,bot as a linear
k k k combination of t u1k ,top ,t u2 ,top ,t u3 ,top ,t u1k ,bot ,t u2 ,bot :
k top k top k bot k u3 ,bot = c1top t u1k ,top + c2 t u2 ,top + c3 t u3 ,top + c1bot t u1k ,bot + c2 t u2 ,bot
(3.95)
where:
top 1
(
t
V1k 3
t
) (
2 2
+ V2k 3
t
) 1
2 t 2 2
top 2
top c3
) ( ) 1 1 ( V ) ( V ) yy = ( V ) + ( V ) 1
t
k 1 3
k 2 3
2 z
k 1 3 t
k 2 3 t
c1bot =
( V ) + ( V ) 1
t k 1 3 t 2 t k 2 3 2 t
bot 2
V1k 3
) (
2
+ V2k 3
) 1
2
(3.96)
156
k The DOFs u1k ,bot or u2 ,bot can be eliminated in the same way. During the execution of the
A special attention needs to be paid to the 9th mid-plane node of Hetherosis element, when we k k have to additionally constrain t u1k ,mid , t u2 , mid , t u3 ,mid . Thus, of the 6 DOFs we need to constrain 4 of them. For example, suppose we want to keep free three more equations. These are: t u1k ,top t k ,top u t k , mid ' ' ' ' ' ' u1 T11 T12 T13 T14 T15 T16 t 2 ,top 0 k u3 t k ,mid ' ' ' ' ' ' (3.97) u2 = T21 T22 T23 T24 T25 T26 t u k ,bot = 0 ' ' ' ' ' ' 1 t k , mid 0 u3 T31 T32 T33 T34 T35 T36 t u k ,bot t 2 ,bot k u3 t k ,top t k ,bot t k ,bot t k ,bot Equations (3.94) and (3.97) are then solved for u1 , u1 , u2 , u3 as a linear combination
k k of t u2 ,top and t u3 ,top .
t k u2 ,top and t k u3 ,top and we need to
k k k fix t u2 ,top ,t u1k ,bot ,t u2 ,bot ,t u3 ,bot . Equation (3.94) from the previous paragraph needs to be added by
' ' ' t u1k ,top T11 T14 T15 k ,bot ' ' ' u1 = T21 T24 T25 k ' ' ' u2 ,bot T31 T34 T35 t k ,bot top 1 c1bot c1 u3
k k ' ' T12t u2 ,top + T13t u3 ,top ' t k ,top ' t k ,top T22 u2 + T23 u3 k k ' ' T32t u2 ,top + T33t u3 ,top top t k ,top top t k ,top c2 u2 + c3 u3
(3.98)
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.
157
Table 3.12-1 Ahmad shell elements.
Element name
Type of approximation
Comment
CCAhmadElement33L9
Lagrange
CCAhmadElement32L9
Lagrange
3 3 3
2 3 2 Good compromise between locking and spurious energy modes Fast, but spurious modes
CCAhmadElement22S8
Serendipity
158 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.
159
t t
b2
a2
Geom etry
Vt
t
b3
a3 6 13 5 16 15 8
a1
t
Vs s 18 19 17 20
14 7
Bric k nodes
b1
6 5 7 8
Beam 3D nodes
2 10
10 9 1 12 4 11 9 12
3 11
14
Beam 1D nodes
2 1 3 4 15
w
13
z
2
2 s
y x y v
160
(3.99)
t s 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. hi = hi (r ) is i-th nodal interpolation function i described in Section 3.2.
T
t X i ,t Yi ,t Z i
T
are
t s 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 t vectors Vt , Vs depicted in Fig. 3-34, 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 s x = hi t + dt X i + t ai t + dtVi tx + t bi t + dtVi sx 2 2 t s t s t + dt y = hi t + dt Yi + t ai t + dtVi y + t bi t + dtVi y 2 2 t s t + dt z = hi t + dt Z i + t ai t + dtVi tz + t bi t + dtVi sz 2 2
t + dt
(3.100)
u = t + dt x t x v = t + dt y t y w=
t + dt
(3.101)
z z
t
(3.102)
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 = Vi y = Vi tz = Vi sx = Vi y = Vi sz =
s t
( (
(
t + dt
Vi tz iy Vi t x iz
t
t + dt t + dt
Vi y iz
t
)
)
t + dt t + dt
Vi tz ix Vi t x iy
s
Vi y ix Vi sz iy Vi sx iz
s
t + dt
) )
)
(3.103)
( (
(
t + dt
t + dt t + dt
Vi y iz
t + dt t + dt
Vi sz ix Vi sx iy
Vi y ix
t + dt
The parameters ix , iy , iz are rotations around the global axis, with respect to beginning of the current load step. Note that (3.103) is valid only approximately.
162
df dx dx dr df = dx dy ds df dx dz dt
dy dr dy ds dy dt
dz dr dz ds dz dt
df df dr dr df = J 1 df ds ds df df dt dt
(3.104)
t z 1 J 33 = = hi t ai tVi tz t 2
(3.105)
163
The matrix
t + dt t
BNL . :
u x u y u z v x v = y v z w x w y w z
t + dt t
BNL U
(3.106)
The detailed expressions for calculating t + dt BNL are given in (3.109) and (3.110). The equations t are important because they present the way, how spatial derivatives of all the displacements are calculated. The entries in t + dt BNL are thus used to setup also the matrix t + dt BL 0 and t + dt BL1 . t t t These matrices are computed as follows:
t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t
t + dt t
BNL (1,i ) BNL (5,i ) BNL (9,i ) BNL (4,i ) + t + dt BNL (2,i ) t BNL (6,i ) + t + dt BNL (8,i ) t BNL (7,i ) + t + dt BNL (3,i ) t
t + dt t t + dt t t + dt t t + dt t t + dt t
(3.107)
164
t + dt t t + dt t
BL1(1,i ) = BL1(2,i )
t + dt u t + dt t + dt v t + dt t + dt w t + dt BNL (1,i ) + BNL (4,i ) + t t t BNL (7,i ) x x x t + dt u t + dt t + dt v t + dt t + dt w t + dt = BNL (2,i ) + BNL (5,i ) + t t t BNL (8,i ) y y y
t + dt t t + dt t
BL1(3,i ) = BL1(4,i )
t + dt u t + dt t + dt v t + dt t + dt w t + dt BNL (3,i ) + BNL (6,i ) + t t BNL (9 i ) t z z z t + dt u t + dt t + dt u t + dt t + dt v t + dt BNL (2,i ) + BNL (1,i ) + = t t t BNL (5,i ) + x y x t + dt v y
t + dt t
BNL (4,i ) +
t + dt t
BL1(5,i )
t + dt u = y t + dt v z
t + dt t
t + dt u BNL (3,i ) + z
t + dt v BNL (2,i ) + y
t + dt t
BNL (6,i ) +
t + dt t
t + dt t
BL1(6,i )
t + dt u t + dt t + dt u t + dt t + dt v t + dt = t BNL (1,i ) + t BNL (3,i ) + t BNL (4,i ) + z x z t + dt v t + dt t + dt w t + dt t + dt w t + dt BNL (6,i ) + BNL (7,i ) + t t t BNL (9,i ) x z x
(3.108)
165
t + dt t t + dt t t + dt t t + dt t t + dt t 1 BNL (1,1) = J1,1
hi r
hi t t t + dt tz s t t + dt sz 1 1 t t + dt sz 1 1 hi t t + dt tz ai Vi ai Vi + bi Vi + J1,2 hi bi Vi + J1,3 r 2 r 2 2 2 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 + dt 1 hi ai Vi t BNL (1,6) = J1,1 ai Vi bi Vi J1,2 hi bi Vi + J1,3 r 2 r 2 2 2 1 hi t + dt t BNL (2,1) = J 2,1 r t + dt BNL (2,2) = 0 t
t + dt t t + dt t
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 ai Vi ai Vi + bi Vi + J 2,2 hi bi Vi + J 2,3 r 2 r 2 2 2 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 + dt 1 hi ai Vi t BNL (2,6) = J 2,1 ai Vi bi Vi J 2,2 hi bi Vi + J 2,3 r 2 r 2 2 2 1 hi t + dt t BNL (3,1) = J 3,1 r t + dt t BNL (3,2) = 0
t + dt t t + dt t t + dt t t + dt t
t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t
BNL (3,6)
hi t t t + dt tz s t t + dt sz 1 1 t t + dt sz 1 1 hi t t + dt tz ai Vi ai Vi + bi Vi + J 3,2 hi bi Vi + J 3,3 r 2 r 2 2 2 t s 1 1 h 1 1 s t t s 1 h = J 3,1 i t ai t + dtVi y t bi t + dtVi y J 3,2 hi t bi t + dtVi y + J 3,3 i t ai t + dtVi y r 2 r 2 2 2 hi r hi r hi r hi r hi t t t + dt tz s t t + dt sx 1 1 t t + dt sz 1 1 hi t t + dt tz ai Vi ai Vi bi Vi J 2,2 hi bi Vi J 2,3 r 2 r 2 2 2 t t t + dt tx s t t + dt sx 1 1 t t + dt sx 1 1 hi t t + dt t x ai Vi ai Vi + bi Vi + J1,2 hi bi Vi + J1,3 r 2 2 2 2 t t t + dt tz s t t + dt sx 1 1 t t + dt sz 1 1 hi t t + dt tz ai Vi ai Vi bi Vi J1,2 hi bi Vi J1,3 r 2 2 2 2 t t t + dt tx s t t + dt sx 1 1 t t + dt sx 1 1 hi t t + dt tx ai Vi ai Vi + bi Vi + J1,2 hi bi Vi + J1,3 r 2 2 2 2
BNL (4,1) = 0
1 BNL (4,2) = J1,1
BNL (4,3) = 0
1 BNL (4,4) = J1,1
BNL (4,5) = 0
1 BNL (4,6) = J1,1
BNL (5,1) = 0
1 BNL (5,2) = J 2,1
BNL (5,3) = 0
1 BNL (5,4) = J 2,1
(3.109)
166
t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t
BNL (6,1) = 0
1 BNL (6,2) = J 3,1
BNL (6,3) = 0
1 BNL (6,4) = J 3,1
BNL (6,5) = 0
1 BNL (6,6) = J 3,1
t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t
hi r 1 1 h 1 1 s s t t s 1 h t = J1,1 i t ai t + dtVi y + t bi t + dtVi y + J1,2 hi t bi t + dtVi y + J1,3 i t ai t + dtVi y r 2 r 2 2 2 t s 1 1 h 1 1 1 h = J1,1 i t ai t + dtVi tx t bi t + dtVi sx J1,2 hi t bi t + dtVi sx J1,3 i t ai t + dtVi tx r 2 r 2 2 2
t + dt t t + dt t t + dt t t + dt t t + dt t t + dt t
hi r 1 h = J 2,1 i r
t + dt t t + dt t
hi r s 1 1 1 1 h t t s s 1 h t = J 3,1 i t ai t + dtVi y + t bi t + dtVi y + J 3,2 hi t bi t + dtVi y + J 3,3 i t ai t + dtVi y r 2 r 2 2 2 1 1 h t s 1 1 1 h = J 3,1 i t ai t + dtVi tx t bi t + dtVi sx J 3,2 hi t bi t + dtVi sx J 3,3 i t ai t + dtVi tx r 2 r 2 2 2
(3.110)
BNL (9,6) = 0
167
t +t t
t +tt xx t +t t Sij =
t +t t xz t +t t yz t +t t zz t +t t
xx
t +t t xy t +t t yy
t +t t xz t +t t yz t +t t zz t +t t
sym.
xx
t +t t xy t +t t yy
(3.111) t +t t xz t +t t yz t +t t zz
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 T , t + dtT transformation matrices for strain and stress transformation from global to isoparametric coordinate system, so that: t +tt xx t +t t +t t yy t rr t +t t +t t + dt t zz rs = T t +t t t xy t +tt rt t +t t +tt yz t xz (3.112) t +tt xx t +t t +t t yy t rr t +t t +t t + dt t zz t rs = T t +t t xy t +tt rt t +t t +tt yz t xz Then the transformation matrices are calculated by:
t + dt V rx 2 t + dt T = t + dt V rx t + dtV sx t + dt V rx t + dtV tx
(
t + dt t + dt
t + dt
ry
(
sy ty t + dt t + dt
t + dt
V rz )
t + dt t + dt
2t + dt V rx
t + dt
t + dt
ry ry t + dt ry t + dt
2t + dt V V sx V tx
t + dt
ry t + dt
V rz
t + dt t + dt
V V
ry t + dt
V V
V rz V rz
V sz V sz
V rx V rx
t + dt
V y + t + dt V V y + t + dt V
t
V V
ry t + dt
V sz + t + dt V rz V tz + t + dt V rz
V V
sy ty
ry t + dt
t + dt
t + dt
t + dt
ry t + dt
t + dt rx t + dt s z V V + t + dt V rz t + dtV sx t + dt rx t + dt t z V V + t + dt V rz t + dtV tx 2t + dt V rx
t + dt
V rz
(3.113)
168
t + dt V rx 2 t + dt T = 2t + dt V rx t + dtV sx 2t + dt V rx t + dtV tx
(
2
t + dt
ry
(
sy ty
t + dt
V rz )
t + dt
t + dt t + dt
V rx
s
t + dt
ry ry t + dt ry t + dt
t + dt
ry t + dt
V rz
t + dt
2t + dt V
t + dt
ry t + dt
V V
2t + dt V rz 2
t + dt
V sz
sz
V rx V
t + dt
V y + t + dt V V +
ty t + dt
V sx
tx
t + dt
V V
ry t + dt
V sz +t + dt V rz V +
tz t + dt
sy ty
ry t + dt
rz t + dt
t + dt
rx t + dt
t + dt
ry t + dt
rz t + dt
t + dt rx t + dt sz V V + t + dt V rz t + dtV sx rz t + dt t x t + dt rx t + dt t z t + dt V V + V V
t + dt
V rx
t + dt
V rz
(3.114)
s t 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 of unity length from Fig. 3-34. The remaining vector is calculated as a vector product of the previous two vectors:
where vectors
t + dt
t + dt
V r = t + dt V rx
t + dt
ry
t + dt
V rz =t + dt V s t + dtV t
(3.115)
T1 =t + dt TT (3.116) T =
1 t + dt
t + dt
Coordinate r
Weight
1 2 1
2 3
1 2 3
169 4 1 2 5 3 4 5 1 2 6 3 4 5 6 0.861136311594053 0.906179845938664 0.538469310105683 0. -0.538469310105683 -0.906179845938664 0.932469514203152 0.661209386466265 0.238619186083197 -0.238619186083197 -0.661209386466265 -0.932469514203152 0.347854845137454 0.236926885056189 0.478628670499366 0.568888888888889 0.478628670499366 0.236926885056189 0.171324492379170 0.360761573048139 0.467913934572691 0.467913934572691 0.360761573048139 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.
ds1 ds2
dsns 2
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).
170 Nice feature of the ATENAs 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.
2D
YG YL N1 XL XG N2 YG
3D
ZG N1 YL YG
XL
N2
171
XG
Fig. 3-36 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 elements edge or an elements 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 matters is the order in which they appear in the element incidences. When processing a boundary load, ATENA loops thru all elements 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
Truss Triangle
Type
Edge Surface Edge
Surface/node incidences
(n1 , n2 ) (n1 , n2 , n3 ) (n1 , n2 ); (n2 , n3 ); (n3 , n1 ) (n1 , n2 , n3 , n4 ) (n1 , n2 ); (n2 , n3 ); (n3 , n1 ); (n4 , n1 ) (n1 , n2 , n3 , n4 ); (n5 , n6 , n7 , n8 ); (n1 , n4 , n8 , n5 ); (n2 , n3 , n7 , n6 ); (n1 , n2 , n6 , n5 ); (n4 , n3 , n7 , n8 ); (n1 , n2 ); (n2 , n3 ); (n3 , n4 ); (n4 , n1 ); (n5 , n6 ); (n6 , n7 ); (n7 , n8 ); (n8 , n5 ); (n1 , n5 ); (n2 , n6 ); (n3 , n7 ); (n4 , n8 )
Quad
Surface Edge
Hexahedron, (brick)
Surface Edge
Tetrahedron
Surface Edge
(n1 , n2 , n3 ); (n1 , n2 , n4 ); (n1 , n3 , n4 ); (n2 , n3 , n4 ) (n1 , n2 ); (n2 , n3 ); (n3 , n1 ); (n4 , n1 ); (n4 , n2 ); (n4 , n3 ) (n1 , n2 , n3 ); (n4 , n5 , n6 ); (n1 , n2 , n5 , n4 ); (n6 , n5 , n2 , n3 ); (n4 , n6 , n3 , n1 )
Wedge
Surface
172 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 elements surfaces or edges are to be loaded, use more boundary load definitions. Violation of this rule causes an error report and skipping of the offending boundary load.
3D edge load planar element ZG n1
XL
ZL
n2
YL n3 YG
XG
n5
n6 YG XG
n5
n6 YG
Fig. 3-37 Examples of positioning local coordinate system used by surface and element load for 2D and 3D elements
173
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.
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, Englewood Cliffs, New Jersey 07632, ISBN 0-13-317305-4. Prentice-Hall, Inc.,
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 TwoDimensional 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)
174
(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.
a15 a11 a12 a13 a 21 a22 a23 a24 a25 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
matrixs profile:
d = [ a11 u = [ a12
a22 a13
a33 a23
a25 a52
a35 a53
a45 a54
a46 a64
a56 a65
a67 ] a76 ]
(4.3)
p = [ 0 1 3 5 9 11 12]
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
175
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). 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
a15 a11 0 a13 0 a a23 a24 0 22 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 c =[ 1 3 5 r =[ 1 2 4 3 4 3 1 2 7 3 ... 7 6 21
] ]
23 (4.5)
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) .
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 = L1b x = ( DU ) v
1
(4.7)
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)
176
177 important for the case of preconditioned iterative solvers. This is where a preconditioning matrix is created. 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'
Comparing A and A' , it can be written
(4.9)
(4.10)
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
memory Description
D I I I I I
For large symmetric well-posed problems For large symmetric problems, recommended For large nonsymmetric wellposed problems For large nonsymmetric problems, recommended
DCGN
ssd2s
scgn
S,NS
LUCN
ssilus
scgn
S,NS
4*(13+4*n+nl+nl)+8*(1 +8*n+nl+nu)
178 DBCG LUBC DCGS LUCS DOMN LUOM I I I I I I ssds ssilus ssds ssilus ssds ssilus sbcg sbcg scgs scgs somn somn S,NS S,NS S,NS S,NS S,NS S,NS 4*(11)+8*(1+8*n) 4*(13+4*n+nl+nu)+8*(1 +8*n+nu+nl) 4*(11)+8*(1+8*n) 4*(13+4*n+nl+nu)+8*(1 +8*n+nu+nl) 4*(11)+8*(1+4*n+nsave +3*n*(nsave+1)) 4*(13+4*n+nu+nl)+8*(1 +nl+nu+4*n+nsave+3*n *(nsave+1)) 4*(31)+8*(2+n+n*(nsav e+6)+nsave*(nsave+3)) 4*(33+4*n+nl+nu)+8*(2 +n+nu+nl+n*(nsave+6)+ nsave*(nsave+3))
DGMR LUGM
I I
ssds ssilus
sgmres sgmres
S,NS S,NS
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.
Description Preconditioned Iterative Refinement sparse Ax = b solver. Routine to solve a general linear system Ax = b using iterative refinement with a matrix splitting. Preconditioned Conjugate Gradient iterative Ax=b solver. Routine to solve a symmetric positive definite linear system Ax = b using the Preconditioned Conjugate Gradient method. 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. Solve a Non-Symmetric system using Preconditioned BiConjugate Gradient. Preconditioned BiConjugate Gradient Sparse Ax=b solver. Routine to solve a Non-Symmetric linear system Ax = b using the Preconditioned BiConjugate Gradient method. Preconditioned Orthomin Sparse Iterative Ax=b Solver. Routine to solve a general linear system Ax = b using the Preconditioned Orthomin method.
scg
scgn
sbcg scgs
somn
179 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.
Description Diagonal Scaling Preconditioner SLAP Set Up. Routine to compute the inverse of the diagonal of a matrix stored in the SLAP Column format. 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. 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. 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.
ssics
ssd2s
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:
180 r 1 = b A x1 z 1 = M 1 r 1 r i zi = i 1 i 1 r z
i
p i = z i + i p i 1 r i zi = i p Ap i
i
(4.11)
) ( )
x i +1 = x i + i p i r i +1 = r i 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.
(4.12)
where: q is the vector of total applied joint loads, f ( p ) is the vector of internal joint forces, p 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)
181 and non-linearity in the stiffness matrix K ( p ) K ( p + p ) 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: (4.14)
K ( pi 1 ) pi = 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 + pi
(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):
piT pi piT pi rel . disp
rel . force
(4.17)
rel .energy
abs . force
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 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:
182
Loading
Loading increment
Deformation
(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 NewtonRaphson method and later, i.e. near extreme points, switch to the modified procedure to avoid divergence.
Loading q
Loading increment
p0
p1
p2
p3
p4
Deformation
183
(4.19)
K ( pi 1 ) pi = q f ( pi 1 ) = q f i 1
pi = pi 1 + pi = pi 1 + i 1 i 1 pi = pi 1 + i 1 i 1
(4.20)
(4.21)
(4.22)
i = i 1 + i 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.
184 The scalar is an additional variable introduced by the Line-search method, which will be discussed later. The scalar is used to accelerate solutions in cases of well-behaved loaddeformation relationships or to damp possible oscillations, if some convergence problems arose, e.g. near bifurcation and extreme points.
q 0 q 0 q 1 q 2 q 3 q 1 g0 Loading R1 T 0 q start p 00 p0
0
R2
Load increment
0
1
p 2 p 3
11
p1
2 2 Deformation
p2
(4.24)
(4.25)
Substituting (4.21) through (4.25) into (4.20), the deformation increment i 1 can be calculated from:
K i 1 = RHS i 1 = i 1q gi 1
(4.26)
Hence:
i 1 = i 1 + i 1 T
where
(4.27)
i 1 = K 1 gi 1
(4.28)
T = K 1 q
It remains only to set the additional constraint for i 1 and i 1 and the whole algorithm is defined. Thus compared to the Newton-Raphson methods in which we solve n dimensional nonlinear 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 i 1 and i 1 .
185 If we set i 1 = 1 , then we deal with an (n + 1) dimensional problem that correspond to pure Arclength 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 i 1 , T are of order (n + 1). Their (n + 1)-th coordinate corresponds to the loading dimension 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 = pi 1 + ( i 1 start )
(4.29) (4.30)
ni 1 = i 1 + i 1 where:
start is a (n+1) dimensional vector similar to i 1 , however its (n + 1)-th coordinate equal
to start .
t1
n1
n2 t3
n3
t2
equations (4.20) through (4.32) lead to the final expression for the unknown i 1 (noting that
piT1 i 1 = piT1 i 1 = 0 ):
186
i 1 =
Ri 1 piT1 i 1
piT1 T + 2 (i 1 start )
(4.33)
To obtain i 1 by (4.33) the residual Ri 1 must be defined. In fact, it also define type of Arclength constrain being used. The types supported in ATENA are described below.
t1
n1
n2 t3
n3
t2
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.
ni 1 cos( ) = ti 1 ( ti 1 s )
(4.34)
The step length s and angle are depicted in Fig. 4.3-4. The norm of the vector calculated using (4.29):
ti 1
2
ti 1 is
(4.35)
= piT1pi 1 + 2 ( i 1 start ) 2
187
ti-1 current s required s ti s = step length
i-1
Substituting (4.34) and (4.35) in (4.33) we obtain the final expression for i 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 p diagrams with sudden breaks and discontinuities, e.g. for materially nonlinear problems.
ni 1 cos( ) = ti 1
tl 1 t 'l
ri 1
(4.36)
Based on the similar triangles (see Fig. 4.4-), the following can be derived:
rl 1 t 'l 1 s
=
(4.37)
Ri 1 =
s 2 ( ti ' s ) ti '
(4.38) (4.39)
2
ti ' = ti 1 + ni'1
ti '
' = ti 1 + 2 i2 1 + 2 i 1
(4.40)
' The vector ti 1 is calculated using (4.35). By substituting the above equations into (4.33) final
expression for i 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 i 1 in order to satisfy the constraint ti 1 = ti = s .
188
||ri-1||
||ti-1||
ti-1 ti
ni-1 s
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 i 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.
(4.41)
where:
T a1 = 2 T T + 2
T a2 = 2 2 ( i 1 start ) + 2 T i 1 2
(4.42)
a3 = 2 ( i 1 start ) 2 + 2 i T1 i 1 s 2
Equation (4.41) has generally two roots i 1 and hence we must decide which of them to use. There exist several strategies but ATENA chooses that root i 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).
189
si = si = si =
where
4
ni 1 si 1 n ni 1 si 1 n n si 1 ni 1
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.
( p ) = ( p0 ) + g ( p )T dp = minimum
po
(4.46)
Hence:
190
( p ) d
= 0+
p p p T = g ( p )T =0 p g ( p) p 0
(4.47)
Interpolating linearly out-of-balance forces between points p0 and p0 + ' g ( p0 + ' ) g ( p0 ) g ( p0 + ' ) g ( p0 ) g ( p0 + ) = g ( p0 ) + || p0 + p0 ||= g ( p0 ) + || p0 + ' p0 || ' (4.48) and using :
p = p0 + p
The final expression for ' can be derived:
(4.49)
= '
g ( p0 )T g ( p0 )T g ( p0 + ' )T
(4.50)
Thus, the Line search method can be summarized: Use any method to calculate displacement increment , (see Fig. 4-3 and (4.28)). The parameter ' can be set from the last load increment or simply to unity. Calculate out-of-balance forces for both g ( p0 ) and g ( p0 + ' ) . Use (4.50) to calculate new value for . As all the above equations are nonlinear, the parameter must be solved by iterations until
g ( p0 + g ( p0 )
Practical experience suggests that the value of parameter should be kept in interval < 0.1 5>.
4.6 Parameter
The parameter scales the deformation space p to the loading dimension . If = 0 , the solution for i 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 > 0 , the solution is carried out in the same way on ellipsoidal or spherical space. The higher value of , the higher "weight factor" for changes in loading space compared to displacement increments. ATENA currently supports the following formulae for setting and optimization of (for current step j ). They are reviewed below.
191 The first strategy requires the load to displacement increment ratio (4.51) is constant throughout all steps, (e.g. input value req )
=
Then, at the end step j-1 we can calculate
j 1
( p)
= req
(4.51)
j 1
j 1
( j 1 p )
(4.52)
This value (due to nonlinearities) will not match req . Therefore, for step j we will modify j as follows:
= req = req
j 1
j 1
(4.53)
j 1
j 1
req
j 1
j 1
( j 1 p ) req = req j 1 j 1 j 1 ( j 1 p )
The above optimisation process is initialized in the first step by assuming that , where T is displacement corresponding to master Arc-length 0 = 1, 0 = 1, ( j 1 p ) = T load increment defined earlier in this chapter. Hence
= 0 =
req
j 1
(4.54)
The parameters j in all subsequent steps are calculated using (4.53). If ratio of displacements changes ( j p ) to load changes ( j ) in the last load step increase, then the equation (4.54) (4.55) increases in the current step, thereby puts higher weight factor on loads compared to displacements. Hence, the equation (4.54) tends to keep constant importance of loading space irrespective of displacements. Note that the equation (4.54) corresponds to BETA_FORCES_DISPLS_RATIO_CONSTANT. The second supported strategy is different. In ATENA it is referred to as BETA_RATIO_CONSTANT method and it tries to keep constant coefficients, whilst managing the coefficients . Thus, it works in opposite way as compared to the first strategy described above. From (4.52) we can write for steps (j-1) and j
192
j 1
=
j
j 1
( j 1 p) j 1
j =
( j p) j
Now requiring
j 1
= j we have
j
( j p) j
j
j 1
( j 1 p) j 1 (4.56)
( j 1 p)
j 1
j 1 ( j p) j
and if we assume
j 1 ( j 1 p )
j ( j p)
, then
j 1
( j 1 p)
j
= j 1
j 1 ( j p) j
(4.57)
( j 1 p )
If
j 1 ( j p) j
in subsequent steps changes, the procedure is trying to compensate for that by re-
adjusting the coefficients . In other words, this strategy is trying to keep relative importance of load vs. displacement spaces).
( p)
constant, (i.e.
193 By default, i.e. without any optimization, the structural degree of freedom are allocated sequentially starting from the node 1 up to the last node n, i.e. 4. Hence, the jth degree of freedom at the node i has number ndof (i 1) + j , where ndof is number of dofs per node.
If the structural nodes are numbered as indicated, then the beam 1,2 and 3 has nodal incidences 1-3, 3-4 and 4-2, respectively and the final stiffness matrix K has the pattern from the leftbottom part of Fig. 4-5. Note that the matrix K must store also the entries depicted as circles without filling. Although they are initially zero, they may turn non-zero during the matrix decomposition needed to solve the problem, i.e. we must store the matrix with 69 entries and maximum half-band width 9. On the other hand, if nodal degrees of freedom are numbered as shown in the right-bottom part of Fig. 4-5, then the matrix K must store only 51 entries and has maximum half bandwidth only 6. The two examples document, how important efficient numbering of the degrees of freedom of the structure is. If the structure (to be solved) is simple, then a suitable dofs' numbering can be done manually by appropriate numbering of the structural nodes. However, in the more complex cases (and in particular if a model of the structure is generated automatically), an optimal dofs mapping must be calculated. There are number of algorithms that deliver more or less efficient dofs mapping. Probably the best established algorithm of that kind is Cuthill-McKee algorithm (Cuthill, McKee 1969). This is not due to its superior property, but due it has been developed as first. The algorithm produces an ordered n-tuple R of vertices which is the new order of the structural vertices. It numbers the vertices according to a particular breadth-first traversal, where neighboring vertices are visited in order from lowest to highest vertex order. The reverse CuthillMcKee algorithm (RCM) is the alternative of the Cuthill-McKee algorithm, in which the vertices are visited in reverse order, i.e. form the highest to the lowest vertex. ATENA implements Gibbs and Sloan dofs optimization algorithms:
194
The Sloan algorithm (Sloan, Randolf (1983) In an effort to obtain an optimum elimination order, the algorithm first renumbers the nodes, and then uses this result to resequence the elements. This intermediate step is necessary because of the nature of the frontal solution procedure, which assembles variables on an element-byelement basis but eliminates them node by node. To renumber the nodes, a modified version of the King algorithm is used. In order to minimize the number of nodal numbering schemes that need to be considered, the starting nodes are selected automatically by using some concepts from graph theory. Once the optimum numbering sequence has been ascertained, the elements are then reordered in an ascending sequence of their lowest-numbered nodes. This ensures that the new elimination order is preserved as closely as possible. For meshes that are composed of a single type of high-order element, it is only necessary to consider the vertex nodes in the renumbering process. This follows from the fact that mesh numberings which are optimal for low-order elements are also optimal for high-order elements. Significant economies in the reordering strategy may thus be achieved.
195
The Gibbs et. al. algorithm (Gibbs et. al. 1976) This algorithm typically produces bandwidth and profile, which are comparable to those of the commonly-used reverse CuthillMcKee algorithm, yet it requires significantly less computation time. Nevertheless, it delivers dofs mapping that is usually slightly less efficient than that by the Sloan algorithm and therefore, it is less preferred option the optimization. Note that the above algorithms optimize dofs numbering by reordering the structural nodes. They do not account for possible different number of dofs within a particular node. Note also that in order to minimize cost of the dofs remapping, the optimization is carried out before assembling the structural global matrices and vectors. Thus, they are assembled directly into their final, optimized location. Iterative solvers use data storage scheme (4.3). As the storage scheme stores only non-zero elements, the solution is less sensitive to a bad dofs mapping. For huge analyses it is nevertheless suggested to carry out a dofs mapping optimization, as it typically yields individual elements entries stored closer to each other with positive effect on solution convergence and RAM data management. A detailed description of the above algorithms is above scope of the publication. For more information the reader is suggested to study the given references.
4.8 References
BATHE, K.J.(1982), Finite Element Procedures In Engineering Analysis, Englewood Cliffs, New Jersey 07632, ISBN 0-13-317305-4. Prentice-Hall, Inc.,
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). Pehled uit 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. CUTHILL, E. and J. MCKEE (1969). Reducing the Bandwidth of Sparse Symmetric Matrices. Proc. 24th Nat. Conf. ACM.
196 SLOAN, S. W. and M. F. RANDOLF (1983). "Automatic Element Reordening for Finite Element Analysis with Frontal Solution Schemes." Int. Journal for Numerical Methods in Eng. 19: 1153-1181. Gibbs, N. E., W. G. Poole, et al. (1976). "An Algorithm for Reducing the Bandwidth and Prole of a Sparse Matrix." SIAM Journal of Numerical Analysis 13(2).
197
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.
198 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.
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 ) = (t , t ')
t'
d + 0 (t )
(5.1)
(t ) =stress at time t ,
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 + t depends only on current material state at time t and stress increment at time t + t , i.e. d . =
(t ) = (t , ) B( ( ))
t'
B( ( )) ( ) d + 0 (t ) +
(5.2)
where:
(t ) = is stress vector at time t , (note the bar atop of a symbol indicates vector),
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 (t , t ') , whereas the short-term behaviour is comprised in the matrix B( ( )) . 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. Substituting t = t '+ t , t 0 into (5.2) and applying load increment (t ') = (t ') (i.e. loading from zero level) at time t ' , it can be derived (5.3) Comparison of (5.3) with similar equations for constitutive relations for short-term loading conditions, i.e. t '+ t t ' , yields instantaneous secant material rigidity matrix:
D(t ') = ( B ( (t ')) (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
(5.5)
200
t t ' n 1 1 1 e '(t , t ') = + E (t ') =1 E (t ')
(5.6)
where :
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 '(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.
201
(tr ) = r = r 1 + r
(tr ) = r = r 1 + r
1 E r 1/ 2 = 1 Er 1/ 2 + (1 ,r )
=1
n
1 E ,r 1/ 2
(5.10)
t r r = 1 e
n
tr
(5.11)
t i * r = 1 e r 1 + r0 =1
(5.12) (5.13)
(tr ) = r = r 1 + r
= e
*
r
t r
* +
r 1
E r 1/ 2 r
r
Er 1 / 2
(5.14)
tr ,
202
1 ( E (tr ) + E (tr 1 ) ) = 1 ( Er + Er1 ) = constant average value of Dirichlet 2 2 coefficient E at tr ,
E r 1/ 2 =
B r 1/ 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 r 1/ 2 and initial strain increments
r to account for creep and shrinkage. After each step these have to be updated. It involves
mainly update of r and r . With these values a new E r 1/ 2 is calculated and the next temporal analysis step is carried out.
203 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 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 1 = 0.1 t ' . As for the upper limit for , it is required:
t 2
(5.16)
The above limits are applicable for the case, when the coefficients E (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 E (t ') of Dirichlet series in (5.6). In this case, Inverse Laplace transformation (Bazant and Xi 1995) is used instead. This method requires 1 0 , typically 1E-3 and
n 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 E (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
tr = 10
r 1 l
( t1 t ')
(5.18)
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.
where l 2 is number of time increments per unit of log(t t ') and t1
204
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. 1 and 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.
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, 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,
205 6. CCModelBP1_DATA (Bazant and Panula 1978; Bazant and Panula 1978; Bazant and Panula 1978; Bazant and Panula 1978), relatively efficient and complex model; now it is superceeded by CCModelBP_KX or CCModelB3, 7. CCModelBP2_DATA (Bazant and Panula 1978), simplified version of the above model, 8. CCModelBP_KX (Bazant and Kim 1991; Bazant and Kim 1991; Bazant and Kim 1991; Bazant and Kim 1991), 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 (t , t ') and 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
Description Type of concrete according to ACI. Type 1 is Portland cement etc. Types 1,3 accepted for static analysis, types 1-4 accepted for transport analysis. Cross section thickness defined as ration of sections surface to volume Material cylindrical strength in compression at time 28 days Short-term material Young modulus at 28 days, i.e. inverse compliance at 28.01 days loaded at 28 days Ambient relative humidity. Accepted range (0.4..1). Total aggregate/cement weight ratio. Water/cement weight ratio.
Units
Default 1
0.0767m 35.1MPa
F ( f cyl 28 )
206 Ratio as Total aggregate/find sand weight ratio. as = sa 1 . Fine/total aggregate weight ratio. sa = as 1 Coarse gravel/fine aggregate weight ratio. 2.8
Ratio sa Ratio g s
0.4 1.3
Fine aggregate/cement weight ratio. Cross section shape factor. It should be 1, 1.15, 1.25, 1.3, 1.55 for slab, cylinder, square prism, sphere, cube, respectively. Result of material slump test. Material volumetric air content. Weight of cement per volume of concrete length % mass/ length3
1.8 1.25
Concr. density
Material density used to evaluate strength and Young modulus at 28 days.. Curing conditions. It can be either in water (i.e. WATER) or air under normal temperature (i.e. WATER) or steamed curing (i.e. STEAM). Time at beginning of drying, i.e. end of curing.
mass/ length3
Curing type
days
Load time t
207 Tot.water loss w Total water loss (up to zero humidity and infinite time). It is measured in an oven in a laboratory and it is used to enhance prediction of shrinkage infinite t sh 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 laboratory. It is used to enhance drying creep and shrinkage prediction. See also description of total water loss w. Measured shrinkage at time t. It is used to enhance drying creep and shrinkage prediction. See also description of total water loss w. Measured material compliance at time t. It is 1 used to improve overall creep and shrinkage /stress prediction of the model. History of humidity in a material point. 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 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). History of temperature in a material point. See also description of h(t ) Celsia kg N/A kg N/A
Improvem.
Shrink. 0 (t )
N/A
Compl. (t , t ')
N/A
Humidity h(t )
N/A
Hist.
Temperat. T (t )
208 Compl. (t , t ') Measured compliance at time t loaded at time t. This and the next two parameters 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 data are only used to calculate the parameters of Dirichlet series approximation. Measured shrinkage at time t . See the parameter above. Measured shrinkage at time t . See the parameter above 1/ stress
Direct
N/A
Table 5.5-2: Input parameters needed by individual creep and shrinkage prediction models
Model name Model No. Concrete. Type Thickness S / V Strength f cyl 28 Young m. E28 Ambient humid. h Ratio ac Ratio wc Ratio as Ratio sa Ratio g s Ratio sc
B3 3 x x x x x x x
B3impr 4 x x x x x x x
BPKX 8 x x x x x x x
CEB 2
ACI 1 x
CSN 5 x x x x
BP1 6 x x x
BP2 7 x x x
Gen eral 9
x x x x
x x
x x x
x x x
x x x
x x x
x x x
209 Shape factor Slump Air content Cement mass Concr. density Curing type End of curing Current time t
I/D
x x x
x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x
Water loss w(t) Shrink. 0 (t ) Compl. (t , t ') Humidity h(t ) Temperat. T (t ) Compl. (t , t ') Shrink. 0 (t ) Strength f cyl (t )
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.
210 Cement type I and Type IA 2 Description General purpose cements suitable for all uses where the special properties of other types are not required. Type II cements contain no more than 8% tricalcium aluminate (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 are ground finer to produce higher early strengths. Used in massive concrete structures where the rate and amount of heat generated from hydration must be minimized. It develops strength slower than other cement types. Contains no more than 5% C3A for high sulfate resistance. Portland blast furnace slag cement Portland-pozzolan cement. General use High early strength Moderate sulfate resistance High sulfate resistance Moderate heat of hydration low heat of hydration
2 3
Air-entraining cements
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.
4
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.
211
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. 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.
212 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). Pehled uit matematiky. Prague, Prometheus. SEAGER, M. K. AND A. GREENBAUM (1988). A SLAP for the Masses, Lawrence Livermore National Laboratory.
213
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 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. A 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 ) t t
(6.1)
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].
J w = Dwwe where
(6.2)
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
w = w(h)
(6.3)
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.
Q T = ( CT (T Tref ) ) = CT = div ( J T ) t t t
(6.5)
215
Q ( CT (T Tref ) + Qh ) = CT T + th = div( JT ) t t
(6.6)
JT = KT grad (T )
and KT stands for heat conductivity, e.g. [J/(day.m.K)].
(6.7)
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:
t 3 wn = wh 0.21 c e e + te where
(6.8)
te = h T dt
(6.9)
where dt is time increment after the mould has been removed and coefficients T , h are calculated by
h =
1 1 + (3.5 3.5h) 4 U 1 1 h R T0 T
(6.10)
T = exp
(6.11)
216
Uh the symbol U h stands for activation energy of hydration and R is gas R Uh constant. According to (Bazant 1986) = 2700 0 K . T , T0 are real and reference concrete R temperature expressed in 0 K . The reference temperature is given by
In the fraction
T0 = 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 .
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:
217
The solution is based on Equations (6.1) thru(6.7). Note that the unknown variables are
(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, RHS h , RHST . Notice that RHS expressions do not include the divergence operator!
218
LHS h =
( w + wh ) t T Qh + t t
(6.14)
LHST = CT
(6.15)
RHSh = J w = J h
RHST = J T
(6.16)
(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
wh = wh (te ) te = h T t wh wh te wh = = h T t te t te
(6.19)
For the next derivation, let us write Equations(6.14), (6.15) in a general form:
219
LHS h = chh h w T + chw + chT + ch 0 t t t
(6.20)
LHST = cTh h w T + cTw + cTT + cT 0 t t t
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 [ ].
The parameter cTT is in ATENA an input material parameter, ch 0 is computed from(6.19), i.e.
ch 0 = wh w w . Their values depend on a 0; h T . The solution also includes expressions te h T
constitutive model being used in the solution. For more information please refer to Section Material constitutive model.
220 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).
h = NT h; w = N T w; T = NT T ;
h = N h w = N w T = N T
T T
(6.24)
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.
N1 x T N N = 1 y N1 z
N 2 x N 2 y N 2 z
Using (6.24) Equations (6.20) and (6.21) can be written in the form
LHSh = chh N T
LHST = cTh N T
and
221
(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 N ( LHS
div( RHSh ) dV = 0
div( RHST ) dV = 0
(6.27)
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( RHSh ) and
div ( RHST ) are scalars !
In the next derivation the two parts of (6.27) are dealt with separately.
hh
NN T dV
h w + chw NN T dV + ... t V t
h0
NdV =
(6.28)
[cchh ]
h w + [ cchw ] + ...cc h 0 t t
222 N ( LHS ) dV = N c
T V
Th
NT
[ ccTh ]
h w + [ ccTw ] + ...ccT 0 t t
... cc h 0 = ch 0 NdV
V
(6.30)
[ccTh ] = cTh NN
V
The second part of (6.27) are calculated using Green theorem (6.36):
N ( div( RHS ) ) dV = N ( n
h S
T s
RHS h dS + N RHSh dV =
V T T T
= N nsT
S
([k ] N h + [k ] N w + [k ] N T + k ) dS + + N ( [ k ] N h + [ k ] N w + [ k ] N T + k )dV
hh hw hT h0 T T T hh hw hT h0 V
(6.31)
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 ]
223
[ kkhh ] = N [ khh ] N
V
dV
[ kkhw ] = N [ khw ] N
V
dV
kk h 0 = N kh 0 dV
V
(6.32)
...
[ kkTT ] = N [ kTT ] N
V
dV
kk T 0 = N kT 0 dV
V
and also
[ J hh ] = N
S
nsT [ khh ] N dS
T
[ J hw ] = N
S
nsT [ khw ] N dS
T
...
[ JTT ] = N
S
nsT [ kTT ] N dS
T
J h0 =
N n
S
T s
kh 0 dS
JT 0 =
N n
S
T s
kT 0 dS
(6.33)
Using (6.28) to (6.33) the original governing equations (6.27) can be written as follows:
224
[cchh ]
h w T + [ 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 + [ 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
[ccTh ]
After sorting the unknown variables h , T by finite nodes into a single vector , Equation (6.34) will read
[cc ]
+ kk + cc 0 + kk 0 = [ J ] + J 0 t
(6.35)
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.
u div(v )dV = u n
S
T s
v dS u v dV
V
(6.36)
V
u div(v )dV = u n
S
T s
v dS u v dV
V
where
225
u u = x u1 x u2 u = x ... un x u y u1 y u2 y ... un y u z u1 z u2 z ... un z
(6.37)
t +t
K t +t + t +t C
t +t ( ) =t +t J t
(6.38)
where t +t K ,
t +t
t +t
is vector of
unknown variables. All of these apply for time t + t . Equation (6.38) is solved iteratively. i.e. the vector
t +t
=t +t (i ) =t +t (i 1) +t +t (i )
t +t ( i )
(6.39)
where index
(i )
t + t and iteration ( i ) :
t +t ( i )
=t +t (i 1) K -1 t +t (i ) J
(6.40)
t +t ( i 1)
K -1 and
t +t ( i )
J is derived from
t +t ( i 1)
226
CCStructureTransport module currently supports Crank Nicholson (Wood. 1990) and AdamsBashforth (Diersch and Perrochet 1998) integration scheme. The former scheme is linear, i.e. its 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 Crank Nicholson scheme is typically preferred. It has been more tested and verified in the CCStructureTransport module and thereby it is more recommended.
t +t
=t (1 ) +t +t
(6.41)
Depending on a particular value of the parameter we get the well known Euler implicit integration (for =1), trapezoidal Crank Nicholson scheme (for =0.5), Galerkin integration method (for =2/3) or even Euler explicit scheme (for =0), which is only conditionally stable.
Solution predictor:
t +t
t = + t t
t
(6.42)
Solution corrector:
t +t 1 = ( t +t t ) t t (6.43)
Using the above after some mathematical manipulation we derive final expressions for K, J . These read:
227 1 K = K + C t J = J K ( t +t + (1 ) t ) C = K 1 t +t ( t ) t (6.44)
( )
=t +
t t 2 + 2 t prev
t t t prev t t prev t
(6.45)
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 + t .
Solution corrector:
t +t 2 t = ( t +t t ) t t t
t prev t +t t t t = + t t + t prev t t + t prev t t prev t prev
(6.46)
(6.47)
228
K = tn 1 ( Ktn ( tn + tn 1 ) + C ( 2tn + tn 1 ) ) J = K t +t
(( t ) t + t ( t ) ) + + C ( ( 2t t + ( t ) ) + ( 2t + J ( ( t ) t + t ( t ) )
2 2 n n 1 n n 1 2 t +t n 1 n n 1 2 t 2 n n 1 n n 1
n 1
tn + ( tn 1 ) + ( tn ) t t ( tn )
2 2
) (6.48)
= K
( )
t +t
=t +t (i ) =t +t (i 1) +t +t (i )
(6.49)
where 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.
229
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
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
w=
Gw Gw,0 + G c
(6.50)
where
230
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 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
Gw w=
Vconcrete Vmortar
Gw G w,0 + G c
(6.51)
where
Vconcrete 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
The model itself already accounts for moisture used by hydration process. i.e. result, wh according to (6.19) need not be implemented.
w 0 . As a t
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
(6.52)
c=
Gc Gc = Gw,0 + Gc G w,0 + G c
(6.53)
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 : Q h T w = CTh + CTT + CTw + CTt = ( KTh grad (h) + KTT grad (T ) + KTw grad ( w) + KTgrav ) t t t t x Moisture : w h T w = Cwh + CwT + Cww + Cwt = ( Dwh grad (h) + DwT grad (T ) + Dww grad ( w) + Dwgrav ) t t t t x (6.54) The parameters CTh , CTT K wgrav are calculated as:
232
0 CTh = CTh fChTh (h) f CTTh (T ) fCtTh (t ) 0 CTT = CTT fChTT (h) f CTTT (T ) fCtTT (t ) 0 CTw = CTw f ChTw (h) fCTTw (T ) fCtTw (t ) 0 CTt = CTt f ChTt (h) f CTTt (T ) f CtTt (t ) 0 Cwh = Cwh fChwh (h) fCTwh (T ) fCtwh (t ) 0 CwT = CwT f ChwT (h) fCTwT (T ) fCtwT (t ) 0 Cww = Cww fChww (h) fCTww (T ) fCtww (t ) 0 Cwt = Cwt fChwt (h) f CTwt (T ) f Ctwt (T )
T t 0 KTh = KTh f KhTh (h) f KTh (T ) f KTh (t ) T t 0 KTT = KTT f KhTT (h) f KTT (T ) f KTT (t ) T t 0 KTw = KTw f KhTw (h) f KTw (T ) f KTw (t ) T t 0 KTgrav = KTgrav f KhTgrav (h) f KTgrav (T ) f KTgrav (t ) 0 h T t Dwh = Dwh f Dwh (h) f Dwh (T ) f Dwh (t ) h T t 0 DwT = DwT f DwT (h) f DwT (T ) f DwT (t ) h T t 0 Dww = Dww f Dww (h) f Dww (T ) f Dww (t ) h T t 0 Dwgrav = Dwgrav f Dwgrav (h) f Dwgrav (T ) f Dwgrav (t )
(6.55)
0 T 0 and the constant parameters CTh thru Dwgrav and functions fChTh (h) thru f Dwgrav (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.
CCTransportMaterialLevel7 material
CCTransport materialLevel7 is an extension of the above CCMaterialTransport material in the way it automatically computes moisture and temperature capacity and conductivity/difussivity incl. "sink" terms regarding hydration, (i.e. rate of hydration heat and mosture consumption during connrete hydration). In terms of the above nomenclature this upper material level calculates CTT , KTT , CTt , Cwh , Dwh , Cwt . As already mentined, the presented material adds on its
bottom level, i.e. CCMaterialTransport. All parameters and characterisctics from the bottom level, (ie. those from CCMaterialTransport) can still be input and used. They typically serve for a refinement/addition of parameters generated by the upper material level. The result from the bottom an upper level are simply added to form final characteristics of the material model
233
CCTransportMaterialLevel7. Note that default values of CTT , KTT , CTt , Cwh , Dwh , Cwt in the bottom level are by default set to zero.
The most important part of the presented model is computation of concrete hydration maturity factor. It is accompanied by calculation of generated hydration heat and consumed hydration moisture. The analysis is based on the affinity hydration model, which provides a framework for accommodating all stages of cement hydration. Consider hydrating cement under isothermal temperature 25oC a relative humidity h = 1 . At this temperature, the rate of hydration maturity factor , 0...1 can be expressed by chemical affinity A25 = A25 ( ) :
= A25 t
(6.56)
where A stands for the chemical affinity, [ s 1 ], The expression already include coefficient E exp a . Hence A25 is not normalized and refers to temperature 25oC. For different RT
temperature it is replaced by A , see (6.60). R is gas constant 8314.41
J , T is temperature, kmol K
[K] and Ea is 40 kJ/mol. It is worthy to note the incorporation of the maturity method into (6.56) . A characteristic time might be introduced to express an affinity A (Bernard, Ulm et al. 2003). The affinity property can be obtained experimentally or analytically. Using experimental approach, heat flow q (t ) that corresponds to the hydration heat Qh = Qh (t ) is meassured in an isothermal calorimetry. Alternatively, the hydration material parameters are computed by an analytical micro-scale model that accounts for the majority of underlying chemical reactions as well as topology of cement grains (with the consequence to hydration kinetics). The solution stems from (Smilauer and Bittnar 2006) and it employs discrete hydration model CEMHYD3D (Bentz 2005) allowing to account for particle size distribution of cement, chemical composition of cement, temperature and moisture history in concrete etc.
234
Qh Qh , pot
1 Qh , pot Qh = A25 t t
(6.57) (6.58)
Qh under Qh , pot
where Qh, pot is potential hydration heat, [J/kg]. Hence the normalized heat flow isothermal 25oC equals to chemical affinity A25 .
Cervera et al. (Cervera, Oliver et al. 1999) developed an analytical form of the affinity which was refined in (Gawin, Pesavento et al. 2006). A slightly modified formulation is proposed here
B A25 = B1 2 + ( ) exp
Ea exp RT
(6.59)
where B1 ,[ s 1 ], B2 , [-] are coefficients to be calibrated, is the ultimate hydration degree, [-], and represents microdiffusion of free water through formed hydrates, [-]. (6.59)express isothermal hydration at 25C. When hydration proceeds under varying temperature, maturity principle expressed via Arrhenius equation scales the a nity to arbitrary temperature T E 1 1 AT = A25 exp a R 273.15 + 25 T (6.60) The parameters in
For example, simulating isothermal hydration at 35oC means scaling A25 with a factor of 1.651 at a given time. This means that hydrating concrete for 10 hours at 35oC 35C releases the same amount of heat as concrete hydrating for 16.51 hours under 25C. Note that setting Ea = 0 ignores the e ect of temperature and proceeds the hydration under 25C. Gawin et al. (Gawin, Pesavento et al. 2006), among others, added the effect of relative humidity. The extension of (6.58) leads to
1 Qh , pot Qh = AT h t t
(6.61)
235
h =
1 1 + ( a ah )
4
(6.62)
where h = h (h) accounts for the reduction of capillary moisture. h is relative humidity r, (Bazant and Najjar 1972). a is material parameter, typically a = 7.5 . Depending on curing conditions is calculated as follows: Sealed curing:
=
Saturated curing:
w/c , 1 0.42
(6.63)
=
w / c is water-cement ratio.
w/c , 1 0.36
(6.64)
Substituting (6.59) and (6.62) into (6.61) yields final equation to predict development of hydration heat. As it is difficult to express function analytically (from (6.59), (6.61)), the above equations are integrated numerically. Note that Qh is calculated in the same unit as is entered the parameter Qh, pot . If the governing equations are written for unit volume and Qh, pot is given per cement unit weight, then Qh must be multiplied by fraction of cement mass mcement and total volume of concrete Vtot .
Heat capacity
The model assumes the following components of concrete: aggregate, filler, water and cement. Total mass of concrete in one cubic meter results from individual masses of components: mconcr = maggregate + m filler + m paste m paste = mcement + mwater (6.65)
where mconcr is mass of concrete per a unit volume. Similarly for mass of aggregate maggregate , mass of filler m filler , mass of water mwater and mass of cement mcement . Corresponding volumes are Vaggregate = maggregate / aggregate , V filler = m filler / filler etc. i stands for mass density of the phase i. Having total volume Vconcr = Vaggregate + V filler + Vwater + Vcement , we can calculate phase fractions f aggregate = Vaggregate / Vconcr and similarly for the remaining phases.
236
Heat capacity and its evolution of cement paste (cement+water) was studied in (Bentz 2007) at 230C for w/c between 0.3 and 0.5. The capacity of fresh cement paste yields
(6.66)
where Cconcrete is concrete capacity (per unit volume) and akin for aggregate, filler and cement paste. The last term, i.e. C paste depends also on degree of hydration and is calculated by
(6.67)
The heat capacity of structural concrete spans the range between 0.8 and 1.17 Jg-1K-1. A former Czech standard CSN 731208 declares 840 and 870 Jkg-1K-1 for dry and saturated mature concrete, respectively. Caggregate is approximately 840 Jkg-1K-1 for basalt and limestone, 790 Jkg-1K-1 for granite, 800 Jkg-1K-1 for sand. Ccement is about 750 Jkg-1K-1 and Cwater is 4180 Jkg-1K-1 .
Heat conductivity
The thermal conductivity of cement paste was found to remain in the range 0.9-1.05 Wm-1K-1 for arbitrary degree of hydration, for both sealed and saturated curing conditions, and for w/c from 0:3 to 0.4 (Bentz 2007). Water in the capillaries has the thermal conductivity 0.604 Wm-1K-1 (Bentz 2007). The thermal conductivity of hardened concrete varies between 0.85 and 3.5 Wm1 -1 K (Neville 1997) p.375, depending strongly on an aggregate type. Thermal conductivity also depends on the saturation state of concrete. For example, a structural concrete made from normal-weight aggregate with a unit mass of 2240 kg/m3 yields = 1.696 Wm-1K-1 for protected and 1.904 for weather exposed conditions (Neville 1997) , p. 376.
237
Figure 6-1. Thermal conductivity of concrete according to the Czech code CSN 731209.
Figure 6-1 summarizes thermal conductivities for ordinary concrete depending on concrete unit mass and saturation conditions, according to (Neville 1997) and a former Czech standard CSN 731208. The latter considers 1.5 for a dry concrete and 1.7 Wm-1K-1 for a water-saturated concrete. Faria et al. (Faria, Azenha et al. 2006) applied the evolution of concrete conductivity with regards to
= 0 (1.0 0.248 )
where is the conductivity of fully hardened concrete, i.e. in infinite time.
The model implemented in Atena, i.e. CCTransportMaterialLevel3 stems from homogenization theories. Consider conductivity of cement paste paste and aggregates aggregate such that
paste aggregate . Corresponding volume fractions are f paste , f aggregate . Hashin-Shtrikman lower concrete,low and upper bounds concrete,upper are (Bentz 2007)
concrete,low, = paste +
3 f aggregate paste ( aggregate paste ) 3 paste + f paste ( aggregate paste ) 3 f paste aggregate ( paste aggregate )
(6.68)
concrete,upper , = aggregate +
concrete =
concrete.low, + concrete,upper ,
2
(1.33 0.33 )
(6.69)
aggregates varies from 0 to 1. Important thermal conductivities: limestone 1.26 - 1.33, sandstone 1.7, granite 1.7 - 4.0 Wm-1K-1 . The above equations for homogenization are written for phases paste-aggregates. In atena the homogenization is carried out as follows: 1. homogenize phases cement - water -> phase paste. 2. homogenize phases paste - filler -> phase paste with filler 3. homogenize phases paste with filler - air -> phase paste with filler and air 4. homogenize phase paste with filler and air - aggregates -> concrete Note that filler and aggregate is in this case treated as one component and the same applies for water and cement, (being the component paste). Volume averaging technique is used to calculate corresponding properties of paste and mixed aggregate.
239 It is assumed that 1 kg of cement (in concrete) approximately consumes during full hydration process about Qw, pot of water. Typically Qw, pot =0.42 kg of water per 1 kg of cement. Thus, e.g. concrete mixture with 300kg cement per 1m3 of concrete needs 300*0.42=126kg o water per 1m3 of concrete. Assuming linear dependence of water hydration consumption wh on concrete hydration level , ( = 0 for fresh concrete and = 1 for fully hydrated concrete) the water sink term due to hydration is
Ch , t =
wh wh = t t
(6.70) (6.71)
wh = Qw, pot c ,[kg] where c stands for weight of cement in 1m3 of concrete.
Moisture capacity
w = wf
( b 1) h
bh
(6.72)
where w f ,[kgm-3 ] is the free water saturation and b is dimensionless approximation factor, which must always be greater than one. It can be determined from the equilibrium water content w80 at relative humidity h = 0.8 by substituting the corresponding numerical values in equation (6.72):
b=
h( w f w80 ) w f h w80
(6.73)
w w f ( b 1) b = 2 h (b h)
(6.74)
240 The above expression is applicable for analyses using reference unit volume. If reference unit weight of the structure is preferred, then we employ moisture capacity C = C / , [ kg/kg ] , where
Moisture diffusion
The present model accounts for diffusivity mechanism of moisture transport. It is valid for dense concrete, which has not mutually connected pores and moisture conduction thru pores (being
kg driven by water pressure) can be neglected. Hence, moisture flux qh , 2 is calculated by the m s kg equation qh = Dh h , where total moisture diffusivity Dh , is calculated as sum of water ms
Dhw and watervapour Dhwv diffusivity: Dh = Dhw + Dhwv
(6.75)
w Dhw = Dw
w h
(6.76)
w Dw =
3.8 A21000
w 1 wf
(w )
f
(6.77)
241
wv Dp =
(6.78)
where is the water vapor diffusion resistance factor and is the vapor diffusion coefficient
kg in air m s Pa
1.81
(6.79)
Atmospheric pressure pa = 101325Pa , gas constant R = 8314.41Jkmol-1K -1 and molar mass of water is M w = 18.01528kgkmol-1 As in the presented model relative humidity h is the primary variable used to analyze moisture wv transport, Dp must be transformed to Dhwv . This is done by:
wv Dhwv = D p
p wv ( psat h ) wv = Dp = D p psat h h
(6.80)
Any expression to calculate pressure of saturated water vapour can be used. The presented model uses
psat = 611e
aT T0 +T
, [ Pa ]
(6.81)
In the above T is temperature [ o C ] and the remaining T 0 : T0 = 234.18o C , a = 17.08; T < 0 : T0 = 272.44o C , a = 22.44
parameters
are
Fitted parameters for cement paste hydration need to be considered for each concrete separately. Due to high cement variability, it is impossible to assign one particular cement to one concrete grade. The user needs rst to select the cement parameters from the following table:
242
The above table is based on fitting predicted results from CEMHYD3D analysis by (6.59), see Table 6-4 and Figure 6-3. The simulations were carried out on CEMHYD3Ds microstructures 50 50 50 m and with the activation energy 38.3 kJ/mol. Saturated curing conditions were assumed, since sealed conditions will be obtained from coupling with moisture transport. Table 6-4 species input data for selected Portland cements.
Figure 6-3 Fit of selected cements to the affinity model, w/c = 0.4
The majority of concretes is produced from blended cements (CEM II - CEM V), hence it is necessary to scale down Q pot by approximately 30 %. This is a common Portland clinker substitution in the majority of blended cements in Europe.
243 There are other default parameters, which are not specied here: QW POT= 0.42, TH INIT = 0, ALPHA INIT = 0, TEMPERATURE INCR MAX =0.1, H80 = 0.8, TEMP0 = 234.18, A WV = 17.08, TEMP0 ICE = 272.44 ,A WV ICE = 22.44 The parameter A 7.5 expresses hydration slow-down with regards to relative humidity. The hydration practically stops at 0.8. Parameters in Figure 6-1 are computed for saturated state. When = 1, the hydration proceeds as there is saturated water environment around concrete. Under standard circumstamces, hydration consumes water, which decreases relative humidity in the calculation. Three parameters are related to moisture transport and are given for ordinary structural concrete:
W80 expresses total mass of free water at =80%. Standard value is 50 kg/m 3 for structural concrete. A W is water absorption coe cient, whose value spans the range 0.25-0.846 kgm 2 h 0.5 ]. MI WV is the water vapor di usion resistance factor, spanning 210-260 [-] for structural concrete.
Parameters specifying specic heat capacity for concrete components are summarized in Table 6-2. Values are obtained from http://www.engineeringtoolbox.com/density-solids-d_1265.html, http://www.engineeringtoolbox.com/specific-heat-solids-d_154.html Parameters specifying specic heat conductivity for concrete components are summarized in Table 6-3. Sources from http://www-odp.tamu.edu/publications/192_SR/109/109_5.htm Concrete strength classes strongly depend on the amount of cement in concrete. Table 6-5 species approximate compositions for major concrete classes used in EN 206-1. The calculation assumes 5 % of entrained air in the concrete, cement density 3220 kg/m 3 and aggregate density 2800 kg/m 3 .
Table 6-2 Parameters specifying density and specic heat capacity for concrete components
244
Table 6-3 Parameters specifying specic heat conductivity for concrete components
Ready-mix concrete is assumed, which requires rather higher w/c due to workability and pumping issues. The parameters CEMENT DENSITY, WATER DENSITY, AGGREGATE DENSITY, FILLER DENSITY are provided in Table 6-2 in the units [kg/m 3 ].
245
246
Table 6-5 Approximate composition for major concrete classes used in EN206-1
247
where:
T (t ) =
temperature maximum
of
radiation of
source
as source
function [C]
T1 =
temperature
radiation
according
t = time [minutes]
Time development of temperature of radiation source is depicted in the figure below. For time t 0 Equation (6.82) yields T (0) = 0 and hence, it is necessary to supplement (6.82) by requirement T (t ) Tambiant ,ini , where Tambiant ,ini is initial ambient temperature prior the fire broke up, (typically something about 20 C).
248
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, , and absorptivity of the heated surface, . 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
r = 0.56, []
hc = 50,
W m2 K
(6.83)
The convection and emissivity heat flux on a boundary exposed to fire is calculated as follows:
249
(6.84)
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.
Nominal HV fire Temperature of the heat source is calculated by (6.82) 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).
250
251
6.5 References
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. and W. THONGUTHAI (1978). Pore Pressure and Drying of Concrete at High Temperature. Proceedings of the ASCE. CELIA, M. A. and P. BINNING (1992). "A Mass Conservative Numerical Solution for TwoPhase 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. 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. (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). Pehled uit 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.
252
253
7 DYNAMIC ANALYSIS
Currently ATENA software support two methods to carry out dynamic analysis: Newmarks and Hughes method (Hughes 1983). If the parameter in Hughes- method equals zero, the Hughes- degrades to Newmarks . Therefere, only Hughes is described here. The method is based on Taylor series expansion of the displacements and velocities at time t:
t +t
u t 2u t t 2 3u t t 3 =u + t + 2 + 3 t t 2! t 3!
t
(7.1) (7.2)
u t +t = u t +
2u t 3u t +t t 2 t + t 2 t 3 2!
The above two equations represent the approximate displacement and velocity ( u t +t and u t +t ) by a truncated Taylor series. Looking at the remainder term (last term of the above displacement),
R1 =
u t 3
3
t +t
t 3!
2u t +t 2u t 2 3 t 2 t t 3! t 3 ( u t +t u t ) t 3! t +t t ( u u ) t 2
(7.3)
R2 =
u t 3
3
t +t
t 2!
2u t +t 2u t 2 3 t 2 t t t 2! t +t t ( u u ) t
(7.4)
The above parameters , are so caller Newmark , parameters. Their value is essential for
1 1 convergence of the this time marching scheme. It can be shown that = , = corresponds 2 6 1 1 to a linear acceleration within the time step, = , = yields constant acceleration. In any 2 4
254
1 case, the scheme is unconditionally stable, if , and it is only conditionally stable for 2 2 1 , < provided that the stability limit is fulfilled: 2 2
tcrit =
where is the damping parameter.
2 1 1 2 + + 2 2 2 2 2
(7.5)
The above defines the condition for time increment t for a linear conditionally stable case:
t Tn
2 3
(7.6)
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 +t = u t + t u t + t 2 u t + u t +t 2
(7.7)
(7.8)
Mu t +t + (1 + ) Cu t +t Cu t + (1 + ) Ku t +t Ku t = (1 + ) R t +t R t
(7.9)
where
255
In nonlinear mechanics the terms Ku t +t and Ku t are calculated directly, i.e. we calculate directly vectors of forces F t +t , F t . The above equation then changes to
Mu t +t + (1 + ) Cu t +t Cu t = (1 + ) ( R t +t F t +t ) ( R t F t )
(7.10)
The parameter specify the amount of an artificial damping that is to be applied in the governing equation. As alredy pointed out, if = 0 , then the time marching scheme reduces to 1 the Newmark method. The value = 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 = 0.05 .
For the actual execution it is advantages to rewrite (7.7) and (7.8) as follows
u t +t = u t +t + t 2 u t +t u t +t = u t + u t +t where 1 u t +t = u t + t u t + t 2 u t 2 u t +t = u t + t (1 ) u t
(7.11) (7.12)
(7.13)
(7.14)
The expressions for u t +t and u t +t 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 ukt +t = ukt +t ukt +t . The displacement ukt +t will involve velocity and 1 acceleration increments
256
ukt +t = uk =
t +t
uk 1
t +t
uk
t +t
u t +t uk 1 u t +t t 2 t 2
t +t
t +t
u = 2k t
(7.15)
ukt +t = uk
t +t
uk 1
t +t
t +t
= u t + uk = uk
t +t
u t uk 1
t +t
uk = t
t +t
(7.16)
where
+ K teff t =
1 t
2
M+
(1 + ) C +
t
(1 + ) K
(7.17)
t + Reff ,kt = (1 + ) ( R t +t F t +t ) ( R t F t ) ( Mu t +t + (1 + ) Cu t +t Cu t )
(7.18)
t + t t The vector Reff ,kt is typically splitted in two parts R1,+t, k and R2,+t,k : eff eff
(7.19)
The first one is computed only once at the beginning of the time steps and in each iteration it is t necessary to re-caulculate only R2,+t,k . This concept also allows to forget values R t , F t , u t eff during the iterarting procedure, whilst we have to rember only vectors u t +t and u (and of t course, also R1,+t, k ) . eff
257
iT Ci = iT ( d M + d K )i = 2ii
where:
(7.21)
i is i-th structural eigenvector, i is i-th structural eigenmode, i is modal damping parameter associated with i and i .
d + d i = 2ii
(7.22)
Equations (7.20) introduces 2 parameters for damping and thus, if only 2 values of i 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 d , d are calculated by least square method, i.e. we are seeking minimum of the expression ( d + d i 2ii ) . It yields the following set of equations
2
d
i
i2
wi2
+ d w = 2
2 i i i
wi2i
(7.23)
258 which is used to calculate the required damping parameters d , 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.
259
Ku = 2Mu
where
(8.1)
is circular eigenfrequency
We are looking for a non-trivial solution, so that we solve for 2 that comes from
260
det(K 2 M ) = 0
(8.2)
ui = ci
(8.3)
where
is matrix of base vectors k , k = 1..m ,
(ui ) =
(8.4)
It can be proved that (ui ) converges from upper side to the corresponding circular frequency
(ui ) = 0, k = 1..m ci ,k
where ci ,k is k component of the vector ci
(8.5)
If we introduce
A = T K , B = T M
(8.6)
261
Aci = i2 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.
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)
0 0 0 1
(8.10)
The entries cos( ), sin( ) 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.
262
In the case of general eigenproblem the whole procedure of constructing S k is very similar. The matrices S k now adopt the shape
0 0 0 1 1 1 Sk = 0 1 0 a 0 0 1 0 b 1 0 0 0 0 0 1
(8.11)
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
i2 =
(8.12)
' ' where aii , bii are diagonal elements of transformed (and diagonalized) matrices A, B.
(8.13)
(8.14)
263
fi , k = M ui ,k ui ,k +1 = K 1 f i ,k
(8.15)
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.
(8.16)
In the above m is number of projection eigenmodes, (reasonably higher than the number of required eigenmodes),
264
A k +1 , B k +1 are transformed stiffness and mass matrices of the problem, (having dimension
m<<n),
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:
(8.17)
The advantage of this procedure over the one defined in (8.16) is that now you dont 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
265 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 s , i.e. solve the associated eigenproblem
(K s M )us = s2 Mus
(8.18)
The original eigenvalues and eigenvectors are then calculated by (8.19) 2 = s2 s 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 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.
u = us
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
266
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.
267
K
j =1
ij
u j = ri , i = 1..n
(9.1)
where Kij 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 lefthand side (LHS) BCs). ATENA implementation of this type of BCs is now described. A simple form of such BCs reads:
(9.2)
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
268
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:
ul = ul 0 +
k<1, n >
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).
(9.4)
j =1, j l
Kij u j + Kil ul =
j =1, j l
(9.5)
(K
j =1
ij
+ K il lk kj ) u j = ri Kil 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 . kj stands for k. j Kronecker delta tensor. The trouble is,
269
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 lk multiple of the row l , i.e.
lk ( K lj + K ll lk kj ) u j = lk ( rl K ll ul 0 )
j =1
(9.7)
(K
j =1
kj
+ K kl lk kj ) u j = rk K kl ul 0
(9.8)
(K
n j =1 n
kj
+ K kl lk kj + lk ( K lj + K ll lk kj ) u j = + lk K lj + ( K kl lk + lk lk K ll ) kj ) u j =
(K
j =1
kj
(9.9)
rk K kl ul 0 + lk ( rl lk K ll ul 0 )
Hence, the final form of the governing set of equations will read:
(K
j =1
ij
2 + K il lk kj + ik lk K lj + ik kj lk K ll ) u j =
(9.10)
ri K il ul 0 + ik lk ( rl K ll ul 0 )
K
j =1
ij
u j = ri , i = 1..n
(9.11)
where
270
K= K11 ... K i1 ... K k 1 + lk K l1 ... K j1 ... K n1 ... ... ... ... ... ... ... ... K1i ... K ii ... ... K ji ... K ni ... ... ... ... ... ... ... ... K1k + K1l lk ... K ik + K il lk ...
2 lk
... K ki + lk K li
... K kj + lk K lj
(9.12)
(9.13)
Providing the original matrix K is symmetric, i.e. K ij = K ji , then the matrix K is now also symmetric, i.e. K ij = K ji .
The displacement ul constrained by Equation (9.4) has a constant part ul 0 and a variable part
lk u k , in which ul depends only on a single u k . 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 + lk uk
k
(9.14)
In this case, the displacement ul is calculated as a constant part ul 0 plus a linear combination
lk of displacements u k . 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
271
K + ( K
j =1 ij il k ,k l
k ,k l ri Kil ul 0 + ( ik lk ( rl Kll ul 0 ) )
lk
2 kj + ik lk Klj + ik kj lk Kll ) u j =
(9.15)
(9.16)
The problem is, however, that displacements u k 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
272
ul = ul 0 + A u k
(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 . 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 lk = 0 and variable fixed DOFs, i.e. that are those dependent on other (master) DOFs and having 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 nonzero 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.
273 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.
a11 0 0 A=0 0 0 0
0 a23 a33 0 0 0 0
0 0 0 a44 0 0 0
0 0 0 0 a55 0 0
(9.21)
It is stored as follows ( A.data stores the actual data, A.rowbase stores base indices for nonzero entries in rows, A.rowlength contains dimension of non-zero data chunks; all arranged by rows):
274 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 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
(9.22)
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 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
275 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.
276
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).
277
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.
278
i m l n p j k bar 1 bar 2
For example, if we want to connect the node n to the embedding solid element, i.e. to nodes 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
i =1 4
are element interpolation function and Ui are nodal displacements for the underlying solid element, respectively. For displacement at the node n we can write u (rn , sn ) = hi (rn , sn ) U i .
i =1 4
( rn , sn ) are coordinates of the node n. Comparing this formula with (9.3), it is obvious that
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
279 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 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 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 Reference
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
280
INDEX
A
Adam-Bashfoth integration ............................ 227 convergence44, 52, 53, 89, 181, 182, 184, 188, 189, 203, 228, 253 convergency............................................ 254, 266 cracking ................................................ 26, 34, 62 Cracking ........................................................... 28 Crank-Nicholson integration .......................... 226 creep ............................................................... 197 basics.......................................................... 197 constitutive models .................................... 204 Dirichlet series ........................................... 199 parameters needed by models .................... 208 retardation times......................................... 202 solution parameters .................................... 205 Step by Step Method .................................. 201 creep model CCModelB3 ............................................... 204 CCModelB3Improved................................ 204 CCModelBP_KX ....................................... 205 CCModelBP1_DATA ................................ 205 CCModelBP2_DATA ................................ 205 CCModelCEB_FIP78 ................................ 204 CCModelCSN731202 ................................ 204 CCModelGeneral ....................................... 205 creep model CCModelACI78 ......................................... 204
B
boundary condition........................................... 21 complex ....................................................... 22 simple........................................................... 22
C
Clapeyron divergent theorem ........................... 11 constitutive model ............................................ 24 after peak behaviour .................................... 32 Bigaj............................................................. 81 CC3DCementitious...................................... 44 CC3DNonLinCementitious ......................... 44 CC3DNonLinCementitious2 ....................... 44 CC3DNonLinCementitious2Fatigue ........... 58 CC3DNonLinCementitious2User................ 44 CC3DNonLinCementitious2Variable.......... 44 CEB-FIP 1990 ............................................. 79 compressive failure...................................... 34 compressive stress ....................................... 38 crack opening law ........................................ 29 crack spacing ............................................... 58 definition.................24, 93, 174, 197, 213, 267 Drucker Prager............................................. 71 equivalent uniaxial law ................................ 27 fracture process............................................ 34 Hookes law................................................. 24 HORDIJK law ............................................. 29 interface model ............................................ 72 localization limiters ..................................... 33 Microplane................................................... 84 peak stress.................................................... 31 Rankine fracturing ....................................... 45 reinforcement............................................... 76 reinforcement bond...................................... 79 SBETA model.............................................. 26 SBETA model parameters ........................... 42 shear and stiffness in crack .......................... 38 size effect..................................................... 33 smeared cracks............................................. 36 smeared cracks-fixed ................................... 36 smeared cracks-rotated ................................ 37 tensile failure ............................................... 35 tension stiffening ................................... 39, 57 transformation.............................................. 25 variants......................................................... 54 Von Mises.................................................... 68 constitutive tensor ............................................ 14 constutive model plasticity and crushing ................................. 47
D
damping .......................... 228, 254, 255, 257, 258 Dirichlet conditions ........................................ 267 discretisation spatial ......................................................... 217 temporal ..................................................... 225 dynamic .......................................................... 253
E
eigenvalues ..................................... 259, 261, 265 eigenvectors ............................ 259, 263, 264, 265
F
fatigue ............................................................... 58 fiber reinforced concrete............................. 30, 61 finite element Ahmad element .......................................... 136 axisymmetric element ................................ 134 brick quadrilateral element......................... 108 extwernal cable element............................. 127 hexahedral element .................................... 108 interface element ........................................ 130 nonlinear 3D beam element ....................... 157 plane quadrilateral element .......................... 99 Q10 element ............................................... 123 Q10Sbeta element ...................................... 123 reinforcement bar with prescribed bond..... 128
281
shell element .............................................. 136 spring element............................................ 121 triangular element ...................................... 106 truss 2D/3D.................................................. 95 Fire analysis ................................................... 247 flux heat............................................................. 214 moisture ..................................................... 213 formulation....................................................... 10 Euler............................................................. 10 Lagrange ...................................10, 16, 67, 157 Updated Lagrange........................................ 10
J
Jacobi method ................................................. 261
L
Lagrangian element ........................ 136, 148, 157
M
moisture .......................................................... 213 Multipoint constraint ...................................... 267
G
governing equations ......................................... 16 Green theorem ................................................ 224
N
Newmark ................................................ 253, 255 nonlinearity......................................................... 9 types ............................................................... 9 nonlinearity classification................................... 9
H
heat ................................................................. 214 Heterosis element ............................136, 148, 157 hierarchical formulation ................................... 93 high performance fiber reinforced concrete ..... 61 HPFRCC .......................................................... 61 Hughe alpha metod......................................... 253 hydration ................................................ 215, 230 Hydrocarbon fires........................................... 247
O
oscilations ....................................................... 228
P
Palmgren-Miner hypothesis.............................. 59 principle of virtual displacements............... 11, 15 principle of virtual forces.................................. 11 problem............................................................. 10 configuration ................................................ 11 FEM discretisation ....................................... 17 formulation................................................... 10 general.......................................................... 10
I
integration points CCBeamNL element.................................. 168 CCIsoBrick element................................... 114 CCIsoQuad element................................... 101 CCIsoTetra element ................................... 111 CCIsoTrianle element................................ 108 CCIsoWedge element ................................ 118 Q10/Q10Sbeta element .............................. 125 shell/Ahmad element ................................. 147 truss 2D/3D element .................................... 95 integration shell element-summart ................. 156 interpolation function ....................................... 93 CCBeamNL element.................................. 160 CCIsoBrick element................................... 111 CCIsoQuad element................................... 100 CCIsoTetra element ................................... 110 CCIsoTriangle element .............................. 107 CCIsoWedge element ................................ 117 problem discretisation.................................. 18 Q10/Q10Sbetaelement............................... 123 shell/Ahmad element ................................. 147 truss 2D/3D element .................................... 95 introduction ........................................................ 9 Inverse Iteration method................................. 262 Inverse Subspace Iteration method................. 259 isoparametric formulation ................................ 93
R
Rayleigh Ritz method ..................................... 260
S
Serendipity element ........................ 136, 147, 157 shape function................................................... 93 SHCC................................................................ 61 S-N curve.................................................... 58, 60 solver .............................................................. 174 Arc length................................................... 183 Arc Length step.......................................... 189 Consistently Linearized Method ................ 186 Crisfield Method ........................................ 188 direct sparse................................................ 176 Explicit Orthogonal Method ...................... 187 Cholesky .................................................... 175 iterative ...................................................... 176 iterative-types............................................. 177 linear .......................................................... 174 Linear search .............................................. 189 Modified Newton Raphson ........................ 182
282
Newton Raphson........................................ 180 nonlinear .................................................... 180 Normal Update Method ............................. 186 scaling parameter.................................... 190 strain................................................................. 14 Almansi........................................................ 15 engineering .................................................. 14 Green Lagrange ........................................... 14 Strain Hardening Cementitious Composite ...... 61 Strain Hardening Cementitious Composites .... 61 stress................................................................. 12 2nd Piola-Kirchhoff ..................................... 12 Cauchy ......................................................... 12 stress/strain smoothing ..................................... 19 lumped ......................................................... 20 variational .................................................... 20 Sturm sequence check .................................... 265
T
time equivalent ............................................... 215 Transport analysis........................................... 213
W
Whler curve .................................................... 60
X
Xi-Bazant model............................................. 229