Beruflich Dokumente
Kultur Dokumente
CHAPTER 25
MODELING RECOMMENDATIONS
25.1 Choosing element types, defining boundary conditions and
assigning DOF constrains
The finite element model can be a 2-dimensional or a 3-dimensional one,
according to the problems significant dimensions. It is always
recommended to create a 2D model when the main hypotheses allow such a
simplification. In either 2D or 3D space, the model can be made of point
elements, line elements and/or solid elements, according to the structures
make up. The integration of different kinds of finite elements in the same
model should be made carefully, in order to maintain the appropriate
compatibility among their degrees of freedom.
The element type should be chosen according to the required refinement of
the finite element model. For example, the model of a roof beam made of
assembled steel components may be represented roughly by pin-jointed
truss elements, or in detail, using shell elements for meshing each steel
component and its assembling features. In the first case only the overall
results are available (joint displacements, axial strains and stresses, axial
forces), while in second case a finer distribution of results is achieved,
emphasizing those regions where each component or assembling detail may
be improved.
The most convenient method for model generation is also important when
choosing the element types.
Frame and truss structures are usually modeled using various types of line
(or so called one-dimensional) elements, such as trusses, beams or pipes, in
both 2D and 3D space. Two examples of modeling one-dimensional
member structures are given in figures 2.1 and 25.1. Solid elements in the
2D space are used for thin planar structures working in plane stress state or
for analyzing constant cross sections of long structures which fulfill the
plane strain hypothesis. The 2D model of an embankment cross section is
represented in figure 25.2. The embankment is considered long enough to
provide plain strain conditions for the transversal cross section. The 2D
263
Fig. 25.1 3D frame structure model made of various beam and truss elements
264
265
limits, placed far enough from the dam-foundation interaction region, where
chosen on purpose parallel with the global Cartesian coordinate system, in
order to simplify the constrain procedure.
A special care should be taken when joining elements with different number
of DOF per node, because inconsistencies will be expected at their interface.
In case of inconsistency, some of the corresponding nodal forces or bending
moments will be not transferred from one element to the other. To be
consistent, the joining elements must have the same DOF (for example
displacement DOF and rotational DOF) and, furthermore, these DOF must
be continuous across the element boundaries.
Suppose the 2D structure made of a diaphragm wall and some planar frames
shown in figure 25.4. Modeling the diaphragm wall by using 2D linear solid
elements with 2 DOF per node (displacements along x and y axes) and the
frames by using 2D elastic beam elements with 3 DOF per node
(displacements along x and y axes and rotation around the normal axis to the
xy plane), the bending moment at common nodes will be not transferred
from the beam elements to the diaphragm wall. Consequently, the joints will
act like hinges between the beam elements and the solid elements.
A
Connection detail A
uy u
x
2D solid with
2DOF/node
z
ux u
y
2D beam with
3DOF/node
266
In order to embed the beam into the diaphragm wall, two possibilities are
available. The first one is represented in figure 25.5, where the beam is
extended over the diaphragm in order to connect both element types in at
least 2 common nodes (3 nodes recommended). Although the embedment
bending stiffness is not identical with the real one, acceptable overall results
are expected (including the bending moment and shear force at the left hand
side of the beam).
2 DOF/node: ux, uy
3 DOF/node: ux, uy, rotz
i
Fig. 25.5 Embedment of 2D elastic beam element into a diaphragm wall modeled
with 2D linear solid elements
267
m ux,m
z,i
i
n
ux,n
Fig. 25.6 The procedure of assigning constrain equations
13
14
15
269
Mx
My
z
y
etr
mm e 1
y
S lan
p
My
L2
Sy
m
p l a me t
ne ry
2
Mx
L1
Constrained displacements
Constrained rotations
z
y
270
To create the finite element model, only a quarter of the floor plate should
be taken into account, with appropriate boundary conditions. There are two
symmetry planes, along the x and y axes. Points lying in the symmetry plane
x = 0 have zero displacements on x direction and zero rotation around the y
axis, while points lying in the y = 0 plane have zero displacements on y
direction and zero rotation around the x axis. The middle of the plate (in this
case corresponding to the coordinate system origin) has only vertical
displacement (both x and y displacements are zero) and no rotation at all. In
order to reproduce the correct behavior of the entire plate using the reduced
model with 3D shell elements (6 DOFs per node) the following boundary
conditions should be prescribed along the symmetry planes:
-
For the other boundaries (x = L1/2 and y = L2/2) all DOFs are prescribed
zero.
In case of axial-symmetric 3D objects (see chapter 13) the two dimensional
solution is available when taking into account some basic rules:
- if the axis of symmetry coincide with the global Cartesian y axis, negative
x coordinates are not permitted;
- when the global Cartesian y direction represents the axial direction and the
Cartesian x direction represents the radial direction, the global Cartesian z
axis corresponds to the circumferential direction;
- when the 3D axial-symmetric object has an axial hole, a distance equal to
the minimal hole radius should be provided from the y axis to the 2D axialsymmetric model.
25.3 The detailing level
A first important rule when creating the finite element model is to discard
all irrelevant details of the real object in order to maintain the model as
simple as possible. Small details that are unimportant to overall behavior
should not be included in the finite element model, because they create a
compulsory, local density of nodes. Small details which disrupt the overall
271
Thus, nonzero terms of the global matrix are usually grouped (confined)
along the main diagonal, within a band. The width of this band depends
on the nodal (or degrees of freedom) numbering*.
The main diagonal term i expresses the relationship between the
displacement along the degree of freedom i and the load applied on its own
direction, while the other nonzero terms corresponding to line (or column) i
express the relationships between the same displacement and the loads
applied on other degrees of freedom directions, belonging to elements
connected in the same node (or vice versa).
The band shape of the matrix is due to the fact that only a limited number of
DOFs can be connected through elements converging in the same node.
However, not only the possible number of connections are defining the band
width, but also the node (or DOF) numbering. Although the node numbering
can be arbitrary chosen (it does not affect the overall dimension of the
matrix nor the number of nonzero terms), different numbering schemes may
lead to different widths of the matrix band MB. Usually the lowest width is
obtained by numbering the nodes along the smallest dimension of the model
(that one to which a lower number of nodes is assigned).
274
the lower integration order of a mesh with linear elements. Besides, for
quadratic or cubic elements, the load scheme due to elements own weight or
to distributed pressures is moving away from the engineering intuition,
which is also not convenient. In case of square linear elements, the nodal
forces are equal distributed, while for square quadratic elements, the load
distribution is consistent with the element shape function but is far from an
intuitive repartition. Higher order elements exhibit also some abnormal
behavior in the proximity of concentrated loads.
The users choice regarding the element type to be utilized should also take
into account the dominant stress or deformation trend. The linear element
with incompatible modes (or extra displacement shape) is appropriate for
solids subjected to pure bending, in order to reproduce this phenomenon
without false shear effects. However, the same element has a worse
behavior then the compatible linear element when used for modeling solids
subjected to only tension or compression.
The above examples denote the fact that no general or common rule exists
for choosing the element type. It is up to the user, knowing the capabilities
of the available elements, to find the most suitable ones for the specific
analysis.
25.5.3 Elements shape
The problem concerns only the 2D and 3D solid elements, whose geometry
is defined by the nodes position. Element shape distortion yields the
increasing of numerical errors. Roughly, the numerical errors are
proportional with the ratio (Lmax/Lmin)2m-1, where Lmax, Lmin are the maximal
and minimal dimensions of the element and m the order of the equation with
partial derivatives governing the phenomenon.
Obviously, elements with low distortion are recommended:
-
275
Generally, the mesh is coarser at the domain limit (where the boundary
conditions are applied) while in the region of interest (close to the structure)
the minimal element size is used.
The elements size should change gradually according to the expected results
distribution. For example, in regions with steep stress changes, smaller
elements should be used. By contrary, in those regions where minor changes
in the stress distribution are expected or stresses are even constant, larger
elements are appropriate. In order to apply this rule, the models behavior
and a qualitative distribution of results should be anticipated, based on
experience and engineering judgment.
277