GSA Notes On 2D Element
GSA Notes On 2D Element
GSA Notes On 2D Element
2D Element Analysis
1 Use of 2D elements
When to use them
slabs (not rafts)
shear walls
cores
stress analysis
n c 3 c1 c 4 c 2
where c is the coordinates on a point on the element, i.e. the coordinates of the node, cn, plus any
offset, o, at that topology position.
The exception to this rule is when the x axis of the axis set is close to the element normal in which
case the y or z axis of the axis set is use for projection.
For Cylindrical and Spherical axes the z axis of the axis set is projected on to the element to become
the local y axis.
If an orientation angle is defined these axes are rotated by the orientation angle in a positive direction
about the element z axis.
f K u
where the load vector f and the stiffness K are known and u are the displacements to be calculated.
Once the displacements are known the forces in individual elements can be calculated from
fe K e ue
Depending on the element type it will have different nodal degrees of freedom
Plane stress elements – 2 degrees of freedom per node (x,y)
Fabric elements – 2 degrees of freedom per node (x,y) (like plane stress elements but with no
thickness and properties per unit width)
Plate elements – 3 degrees of freedom per node (z,xx,yy) – because of the ‘missing’ degrees of
freedom these elements should be used with care.
Shell elements (bilinear) – 5 degrees of freedom per node (x,y,z,xx,yy)
Shell elements (Allman-Cook) – 6 degrees of freedom per node (x,y,z,xx,yy,zz) – linear
elements only
2D elements are relatively simple elements and the mesh must be designed so that the elements are
sufficient to model the actual stiffness of the structure. The size and shape of the finite elements will be
determined by:
The type of element being used
Most 2D elements are variants on isoparametric elements. In an isoparametric element the element
displacements are interpolated in the same way as the geometry, eg a plane stress element. In a
superparametric or degenerate isoparametric element the interpolation of geometry is of a higher
order than the interpolation of the displacements, eg a plate element. In a subparametric element the
interpolation of the geometry is of a lower order than the interpolation of the displacements eg an 8
node straight sided quad element, where different interpolation functions are used for the geometry and
the displacements. The term isoparametric is often used to describe any of these types.
For a plane stress problem there is a matrix C relating stress and strain (for a linear material). The
displacements in a local coordinate system, the strains and stresses are
xx xx
u
u ε yy σ yy
v
xy xy
1 0
E
C 1 0
1 2 1
0 0
2
Notes: There is an out of plane strain in the zz direction which can be ignored in the element
formulation.
u v u v
xx yy xy
x y y x
The simplest elements to consider are 4 node and 8 node quadrilateral elements, where the 4 node
quad is a simplification of an 8 node element. A typical 8 node element can have curved edges.
In addition to the ‘local’ coordinate system based there is also a ‘natural’ coordinate system which is
used to map the curvilinear element on to a square.
4 7 3
8 6 r r
1 5 2
x
n n
x hi xi y hi y i
i 1 i 1
h1 14 (1 r )(1 s) 12 [h8 h5 ]
h2 14 (1 r )(1 s) 12 [h5 h6 ]
h3 14 (1 r )(1 s) 12 [h6 h7 ]
h4 14 (1 r )(1 s) 12 [h7 h8 ]
h5 14 (1 r 2 )(1 s)
h6 14 (1 s 2 )(1 r )
h7 14 (1 r 2 )(1 s)
h8 14 (1 s 2 )(1 r )
and the h1 – h4 without the […] terms are the interpolation functions for quad 4 elements. These
interpolation function are chosen so that at node i the interpolation function hi has a value of 1 and all
the other interpolation functions have a value of 0.
As the elements are isoparametric the displacements in the elements are related to the nodal
displacements:
n n
u hi u i v hi vi
i 1 i 1
To evaluate the stiffness matrix it is necessary to establish the strains in terms of the displacements –
the strain-displacements transformation matrix. The element strains are obtained in terms of
derivatives of the element displacements with respect to the local coordinate system(x,y). The element
displacements are in the natural coordinate system (r,s) so it is necessary to relate derivatives with
respect to x,y to derivatives with respect to r,s. This is achieved through the Jacobian J
where
x y
r
J r
x y
s s
Since x and y are known in terms of the interpolation functions the Jacobian can be easily found and
the derivatives with respect to x and y are then
J 1
x r
This requires that the inverse of the Jacobian exist , which is satisfied if there is a one to one
correspondence between natural and local coordinates. This will be the case provided the element is
not grossly distorted and that it does not fold back on itself.
Constructing the partial derivatives then allows the strains and nodal displacements to be related
through the strain-displacement transformation matrix B:
ε Buˆ
The element stiffness corresponding to the local element degrees of freedom is then
K B
T
CBdV
V
The elements of B are functions of the natural coordinate system r,s. Therefore the integration extends
over the natural coordinates volume, so the volume differential has to be written in terms of the natural
coordinates
dV det Jdrds
The volume integral is not normally amenable to an explicit integration so a numerical technique is
normally used. The integral can be written
K Fdrds
V
where
F B T CB det J
and the integral is performed in the natural coordinate system of the element. This is convenient as the
limits of the integration are then ±1. The stiffness can then be calculated
K ij Fij
i, j
where Fij is the matrix F evaluated at the Gaussian integration points and ij are the Gaussian
weights.
In a similar way the mass matrix and the load vectors can be established. Writing the displacement in
terms of the interpolation functions and the nodal displacements
u Huˆ
gives
M H HdV
T
V
RB H f dV T
B
V
H f dS
T
RS S S
S
RI B τ dV T
I
V
where the subscripts B, S and I refer to the body forces, surface forces and initial stresses respectively.
In the past the out-of-plane performance of the linear elements limited their usefulness, so the obvious
choice was to use the parabolic elements. The problem with the conventional (Mindlin) linear elements
was than the elements were susceptible to locking (so they would be too stiff). One solution to this was
to use reduced integration which reduced the over-stiffness problem, but as there were non-stiff modes
of deformation in the element it was susceptible to hourglassing problems.
With hourglassing there is a loss of accuracy in the results but it is usually very obvious.
From GSA 8.5 on there is also the option of switching from Mindlin elements to MITC (mixed
interpolation of tensorial components) elements. This provides appropriate stiffness for the elements,
removing the hourglassing problem.
In general it is simpler to understand what is happening with linear elements, especially where other
features such as tied interfaces are introduced into the model
Checks are made before assembling the structure matrix so that only the required degrees of freedom
are set up. A pass is made through the elements
For elements in global planes only degrees of freedom relevant to the element type are set up
For elements in non-global planes automatic constraints check stiffness in each direction and
set up local constraints if required
The success with which the structure matrix represents the actual structure depends on the individual
elements and the relationship between adjacent elements.
Error Occurs in
An analysis is only as good as the data supplied. Thus the results will depend on the input data –
material properties, loading and representation of the real structure in the finite element model. Given
these limitations there are a number of factors related more directly to the method used to solve for the
displacements which need to be considered.
The shape of the elements is significant in the accuracy of the solution. This is discussed in more detail
below.
Certain assumptions are made about the material of which the element is composed. If an isotropic-
elastic material model is used it will give, at best, approximate results if the material is in fact
orthotropic. If the material may yield or crack, because of local high stresses a non-linear material may
be required. Or if the displacements of the structure are large a geometrically non-linear analysis may
be required.
From these it is obvious that there is no single factor which will lead to errors and it is difficult to
quantify the error resulting from these factors. However there are some general principles which can
be applied.
We make a distinction between flat shell and curved shell elements. With flat shell elements the
formulation assumes that the element is flat so problems may arise with warping of the element. In
general where a flat shell element is used to model a floor slab this will not be a problem unless it is
used to model a spiral ramp or similar feature. Curved shell elements will also have limitations on the
out-of-plane curvature. In particular the element cannot fold back on itself (or the Jacobian becomes
singular), but other factors will limit the degree of out-of-plane curvature the element will allow. In the
case of shell elements the user should consult the user manual or contact the supplier.
Given these constraints, it is still possible to have problems if the assumed displacements in the
elements do not accurately reflect the real displacements in the structure. In most cases this is where
the displacement over a single element varies so much that the assumptions of the element
interpolation functions are no longer valid. In these cases it is necessary either to use higher order
elements, eg move from 4 node quads to 8 node quads, or to refine the mesh.
It is particularly important to use a mesh that is appropriate to the analysis and this will mean that in
areas of high stress gradients a fine mesh of nearly square elements should be used.
The other option is to store the basic geometry in GSA and generate the mesh as required. Provided the
mesh generation is a straightforward (and reliable) process, this allows for a minimum amount of data
to define the structure, with what is required for analysis being added as required.
In GSA the mesh generation is based on regions. Regions exist in the design layer and are defined by
geometric entities:
Areas – which can be voids
Lines
Grid nodes
Each of the geometric entities holds data that control the mesh to be created – mesh density, spacing,
element size, etc. Once the region is defined the mesh can then be generated automatically by GSA.
Because the mesh generation is automatic changes in geometry can be accommodated without a large
amount of manual meshing.
There are a number of sculpting options in GSA to allow mesh refinement. These can be useful to
make some local mesh refinements but should, in general, be avoided as these adjustment will be lost
if re-meshing is required.
From these guide-lines it is obvious that there is no such thing as an ideal mesh since the type of mesh
needed depends on the analysis to be carried out.
Irrespective of how the mesh is generated there are some rules which can usefully be applied:
Use fill & edge check to ensure that elements are all facing the same way up and properly
connected
Use quads in preference to triangles
For linear analysis use parabolic elements in preference to linear elements
Avoid linear elements attached to parabolic elements
Hmax = dist of 3rd corner node from plane of element / longest side
Notes: Mid-side node locations not checked - should be approximately 1/2 way along edge
The simplest type of loading is the point load. This is a load which is applied directly to a node. The
implicit assumption with a point load is that the stress at the point of application of the load is infinite.
The finite element code will not calculate infinite stresses as the stresses are calculated from the strain
in the whole element. This means that the stresses found in a structure due to a point load at the centre
will be good enough except in the area adjacent to the load.
When details of the stresses around the load point are required, either the mesh should be refined or
some form of distributed load should be applied. A uniform face pressure applied to a quad 8 element
leads to a set of nodal forces which are not intuitive:
Body forces are those which apply to the body as a whole. In most analyses these will be restricted to
gravity loads. The other body loads which may be encountered are constant accelerations (much like
gravity) and electromagnetic loads. These loads are calculated using the interpolation functions:
RB V
H T f B dV
and result in a set of forces at the nodes, which are equivalent to the body load.
The face and edge pressures are similar. A face pressure is applied over the face of an element, for
example the load on a floor slab would be a face load, while a distributed load on beam modelled with
plane stress elements would be an edge load. In both cases the definition is
H
T
RS S f S dS
S
where the surface is determined by the type of loading. As with the body force it is better to allow the
program to calculate the equivalent nodes.
Another problem is where some of the degrees of freedom of an element are missing. Take for
example a square slab modelled with plate/shell elements. If this is to be loaded with a torsional in-
plane load this could either be loaded with torsion about the z axis or with forces in the x and y
directions. The type of loading which can be used will depend on the type of elements that are
available, as many plate elements do not have a local z rotational degree of freedom. This may also
affect the choice of finite element mesh.
If you have doubts about your results then consult someone else and / or build a more detailed model
and compare the results of the different models.
The displacements in a local coordinate system, the strains and stresses are
xx xx
u
u yy yy
v
xy xy
It was previously noted that we can establish a material matrix which relates stress and strain and for a
plane-stress elastic-isotropic material this is
1 0
E
C 1 0
1 2 1
0 0
2
u v u v
xx yy xx
x y y x
So if we can calculate the strains it is then a straightforward matter to calculate the stresses.
It was noted before that we can evaluate the partial derivatives of the displacements and thus construct
the strain-displacement transformation matrix. The element stresses can then be calculated from
σ CBu
ˆ
The strain displacement matrix has previously been established in the calculation of the element
stiffness, as has the material matrix so given the nodal displacements we can calculate the stresses.
In principle we could now calculate the stress at any point in the element by calculating the strain-
displacement transformation matrix. In practice this does not lead to very good results.
The stresses are based on the strains, which in turn are based on the displacement gradients in the
element. Thus the strains in an element which has a parabolic displacement field are linear, and in one
with a linear displacement field are constant. In practice we normally want to get as good an
approximation to the stress as possible. So in practice we calculate the stresses at selected points and
extrapolate to the nodal values.
It has been found in practice that the stresses are most accurately calculated at the Gaussian integration
points (for which we have already established the strain displacement relationships). The extrapolation
to the nodal values can be done in several ways such as
Least squares fit using an assumed shape function
The second of these methods is used by GSS. The calculation of the stress in the element can be
summarized as follows:
Calculate the stress at the Gauss points
Extrapolate the stresses to get the nodal values
This procedure while seeming a cumbersome way of calculating the stress in fact leads to better stress
values at the nodes than by direct calculation at the nodes.
The stress is calculated on an element by element basis (unlike the displacement) so it is possible, and
in fact very likely, that the stresses at a node in one element will not match up with the stress at the
same node in an adjacent element. In some cases, where there is a discontinuity in the material
properties or the thickness of the elements, it is correct that there is a local stress discontinuity,
however in other cases this stress difference will be a measure of the error in the solution.
When contour plots of the stresses are required, or when a stress value is required for further
calculation it is better to have a single value. Most post-processors will give the option to calculate an
averaged value at the node. However it should be remembered that the stresses which look smooth and
continuous on the plots are in fact calculated a series of discontinuous stress patches.
It is useful to be able to carry out some quick checks on the stresses to get an idea of the accuracy of
the analysis. There is no simple answer as to how good an analysis is, but a number qualitative rules
can be established
Check that the stress gradients are low within the elements, particularly if the stresses in the
element are high
Check that the stress discontinuities between elements are small – look at the stress errors
Check the results against some simple hand calculations, to establish that the results are in line
with your expectations.
Provided you can satisfy yourself with these checks then your analysis should be OK. If not then look
at how you can refine the mesh to give better results.
From the basic stress results it is possible to examine different stress measures those available in GSA
are:
Average av 13 ( x y z )
Principal stress
Flow of stress in metals, concrete
Directions of stress relative to reinforcement
Average stress
Checking if material is in overall compression/tension
When dealing with elements with no thickness, such as fabrics, or composite materials, such as
reinforced concrete, it is more useful to work with stress resultants than stresses. (The stress values for
concrete are based on the properties of an equivalent isotropic material.) The checks used for stress
results noted above can also be applied to force results. For fabric elements the force resultants are
calculated directly but for elements with thickness they are calculated from the stress results at through
the thickness of the element
xx xy xz
yx yy yz
zx zy zz
and the element thickness, t. There are two ways in which these are defined. The preferred method,
used by Timoshenko1 is as follows.
This means that a moment Mx is based on the stress in the x direction. With the Timoshenko
convention if a slab is in compression on the top face in both the x and y directions the moments are
both negative. Consequently starting from the assumption that tensile stress is positive, we have the
following relationships for the forces:
t t t
N x t xx dz N y t yy dz N xy t yx dz
2 2 2
2 2 2
1
Timoshenko & Woinowsky-Krieger, Theory of Plates and Shells, McGraw-Hill
Q x 2t xz dz Q y 2t yz dz
2 2
and moments:
t t t
M x 2t xx z dz M y 2t yy z dz M xy 2t yx z dz
2 2 2
Following from this a plate which has a positive in-plane stress in x/y will have a positive force
resultant Nx/Ny and a positive bending stress in x/y (i.e. positive stress at the top surface relative to the
bottom surface) will have a positive moment Mx/My.
Q x xz t Q y yz t
and
t2 t2 t2
M x xxb M y yy
b
M xy xyb
6 6 6
where the superscript p and b refer to in-plane and bending stress terms.