Shape Function
Shape Function
Shape Function
Related terms:
SHAPE FUNCTIONS
M.V.K. Chari, S.J. Salon, in Numerical Methods in Electromagnetism, 2000
The shape functions for the triangular prism are obtained as the product of the
shape functions of the triangular faces (cross section) and the shape function for
the line element corresponding to the z dimension. This procedure is applicable
to both Lagrange and serendipity types of elements. The main advantage of using
triangular prisms either by themselves or in combination with hexahedral elements
for discretizing the geometry is that they fit irregular boundaries better and more
easily than hexahedral elements alone. It must be noted from Figure 5.28 that the
sides along the z dimension are rectangles.
For the preceding right triangular first-order prism, the shape functions are obtained
as
(5.186)
where i is the shape function of the triangular prism at node i, i is the shape
function of the triangular face at node i and i is the half length of the prism along
the z dimension.
Shape functions for high-order elements can also be obtained by following the same
procedure. For purposes of illustration, the shape functions for linear and quadratic
prism elements from reference [49] are presented next.
Linear element (6 nodes):
(5.187)
(5.188)
(5.189)
(5.190)
7.4.4 Remarks
The shape functions used to interpolate the coordinates in Eq. (7.70) are the same
as those used for interpolation of the displacements. Such an element is called an
isoparametric element. However, the shape functions for coordinate and displacement
interpolations do not necessarily have to be the same. Using different shape func-
tions for coordinate and displacement interpolations will lead to the development
of what is known as subparametric or superparametric elements. These elements have
been studied in academic research, but are less often used in practical applications.
Details of such elements will not be covered in this book.
(12.128)
where Ni (i = 1, 2, 3) are the three shape functions defined by Eq. (7.22) and
(7.23)(7.22)(7.23), and i (i = 1, 2, 3) are the nodal values of temperature at the 3-nodes
of the triangular element shown in Figure 12.13.
Note that the strain matrix B is constant for triangular elements, and can be
evaluated similar to Eq. (7.38). Using Eq. (12.127), can be easily evaluated as the
integrand and is a constant matrix if the material constants Dx and Dy do not change
within the element.
(12.129)
(12.130)
(12.131)
The above integral can be carried out using the factorial formula given in Eq. (7.43),
since the shape functions are equal to the area coordinates—just as we did for the
mass matrix, Eq. (7.44), in Chapter 7. For example,
(12.132)
Using the area coordinates and the factorial formula in Eq. (7.43), the matrix is found
as
(12.133)
The element force vector defined in Eq. (12.122) also involves the integration of
shape functions and can also be obtained using the factorial formula in Eq. (7.43):
(12.134)
It is assumed that Q is a constant within the element. Note that the heating rate,
QA, is equally shared by the three nodes of the triangular element.
(7.18a)
(7.18b)
(7.18c)
or by recognizing that the constraint can produce the third dependent coordinate
from any two independent coordinates. The name “area coordinates” arises because
their value relates to the area of the shaded subtriangles in Fig. 7.32(b) to the area
of the whole triangle
(7.19a)
(7.19b)
(7.19c)
(7.20b)
To obtain the element mass matrix me and the element stiffness matrix ke, we have
to use the energy functions given by Eqs. (8.8) and (8.9) and Hamilton’s principle.
Substituting Eq. (8.15) into the kinetic energy function, Eq. (8.9) gives
(8.18)
(8.19)
The above integration can be carried out analytically, but it will be omitted here. In-
terested readers can refer to Petyt (1990) for details. In practice, we often perform the
integration numerically using the Gauss integration scheme, discussed in Chapter
7.
To obtain the stiffness matrix ke, we substitute Eq. (8.15) into Eq. (8.6), from which
we obtain
(8.20)
The first term in the above equation represents the strain energy associated with the
in-plane stress and strain. The strain matrix BI has the form of
(8.21)
where
(8.22)
Using the expression for the shape functions in Eq. (8.14), we obtain
(8.23)
In deriving Eq. (8.23), the relationship given in Eq. (7.47) has been used.
The second term in Eq. (8.20) relates to the strain energy associated with the
off-plane shear stress and strain. The strain matrix BO has the form
(8.24)
where
(8.25)
The integration in the stiffness matrix ke, Eq. (8.20) can be evaluated analytically
as well. Practically however, the Gauss integration scheme is used to evaluate the
integrations numerically. Note that when the thickness of the plate is reduced, the
element becomes over-stiff, a phenomenon that relates to so-called “shear locking.”
The simplest and most practical means to alleviate this problem is to use 2 × 2 Gauss
points for the integration of the first term, and use only one Gauss point for the
second term in Eq. (8.20).
As for the force vector, we substitute the interpolation of the generalized displace-
ments, given in Eq. (8.15), into the usual equation, as in Eq. (3.81), assuming that
there is a distributed transverse force, fz, acting on the surface of the plate:
(8.26)
If the load is uniformly distributed in the element, fz is constant, and the above
equation becomes
(8.27)
Equation (8.27) implies that the distributed force is divided evenly into four concen-
trated forces of one quarter of the total force.
(3.69)
where the subscript e stands for the element. Note that the volume integration over
the global domain has been changed to that over the elements. This can be done
because we assume that the assumed displacement field satisfies the compatibility
condition (see Section 3.3) on all the edges between the elements. Otherwise,
techniques discussed in Chapter 11 are needed. In Eq. (3.69), B gives the strain
in the element using the element nodal displacements, and is often called the
strain–displacement matrix or simply strain matrix. It is defined by
(3.70)
where L is the differential operator that is defined for different problems in Chapter
2. This implies that the strain matrix is determined once the shape functions are
obtained. For 3D solids, L is given by Eq. (2.7). By denoting
(3.71)
which is called the element stiffness matrix, Eq. (3.69) can be rewritten as
(3.72)
(3.73)
which shows that the transpose of matrix ke is itself. In deriving the above equation,
c = cT has been employed. Making use of the symmetry of the stiffness matrix, only
half of the terms in the matrix need to be evaluated and stored.
By substituting Eq. (3.6) into Eq. (3.3), the kinetic energy can be expressed as
(3.74)
By denoting
(3.75)
which is called the mass matrix of the element, Eq. (3.74) can be rewritten as
(3.76)
Finally, to obtain the work done by the external forces, Eq. (3.6) is substituted into
Eq. (3.5):
(3.77)
where the surface integration is performed only for elements on the force boundary
of the problem domain. By denoting
(3.78)
and
(3.79)
(3.80)
Fb and Fs are the nodal forces acting on the nodes of the elements, which are
equivalent to the body forces and surface forces applied on the element in terms
of the work done on a virtual displacement. These two nodal force vectors can then
be added up to form the total nodal force vector fe:
(3.81)
Substituting Eqs. (3.72), (3.76), and (3.80) into Lagrangian functional L (Eq. (3.2)), we
have
(3.82)
(3.83)
Note that the variation and integration operators are interchangeable, hence we
obtain
(3.84)
To explicitly illustrate the process of deriving Eq. (3.84) from Eq. (3.83), we use a
two-degree of freedom system as an example. Here, we show the procedure for
deriving the second term in Eq. (3.84):
In Eq. (3.84), the variation and differentiation with time are also interchangeable, i.e.,
(3.85)
Hence, by substituting Eq. (3.85) into Eq. (3.84), and integrating the first term by
parts, we obtain
(3.86)
Note that in deriving Eq. (3.86) as above, the condition de = 0 at t1 and t2 have been
used, which leads to the vanishing of the first term on the right-hand side. This is
because the initial condition at t1 and final condition at t2 have to be satisfied for any
de (admissible conditions (c) required by Hamilton’s principle), and no variation at t1
and t2 is allowed. Substituting Eq. (3.86) into Eq. (3.84) leads to
(3.87)
To have the integration in Eq. (3.87) as zero for an arbitrary integrand, the integrand
itself has to vanish, i.e,
(3.88)
Due to the arbitrary nature of the variation of the displacements, the only insurance
for Eq. (3.88) to be satisfied is
(3.89)
Equation (3.89) is the FEM equation for an element, while ke and me are the stiffness
and mass matrices for the element, and fe is the element force vector of the total
external forces acting on the nodes of the element. All these element matrices and
vectors can be obtained simply by integration for the given shape functions of
displacements.
L1(2L1-1); L2 (2L2-1); L3(2L3-1); 4L1L2; 4L2L3; 4L3L1,and those concerning the pressure
for nodes 1,2,3 are L1,L2, L3. (Fig. 6.1.2).
The derivatives of the interpolation functions with respect to the global coordinates
are of the form
(6.1.1)
(6.1.2)
(6.1.3)
where W1 and S1 denote the weights and integration points of the quadrature rule.
The integration must be carried out over the elemental volume of the axisymmetric
geometry rdrd dz. Since the solution is independent of the coordinate, the inte-
gration with respect to yields a multiplicative constant 2π.
(5.17)
(5.18)
Note that Eqs. (2.48) and (5.1) have been used for the derivation of the above
expression for B. Since the expressions for Ni have been derived in Eq. (5.14), we
have
(5.19)
where
(5.20)
> Read full chapter
(10.15)
where Ci are arbitrary constants and r is the radial distance from the origin or pole,
which is usually the “central” point of the problem with the infinite domain. Consider
the 1D mapping of the line OPQ, which coincides with the x axis, as shown in Figure
10.5. Like the finite element formulation for isoparametric elements, the coordinates
are interpolated from the nodal coordinates and thus let us propose that
(10.16)
From Eq. (10.16), it is observed that = 0 corresponds to x = xQ, = 1 corresponds
to x = , and = −1 corresponds to . As mentioned, r is the distance measured from
the origin or pole and we assume that the pole is at O. Therefore,
(10.17)
(10.18)
(10.19)
Then, substituting Eq. (10.18) into (10.19) would give us a series of decaying terms
in the form of Eq. (10.15) with the linear shape function in corresponding to 1/r
terms, quadratic to 1/r2, and so on.
(10.20)
Then we complete the interpolation and map the whole domain of ( ) by adding a
standard interpolation in the ( ) directions. Thus, for element PP1QQ1RR1 in Figure
10.5, we can write
(10.21)
with
(10.22)
and map the points as shown. In a similar manner, quadratic interpolations could
be used as well. These infinite elements can be joined to a standard finite element
mesh as shown in Figure 10.6.
Figure 10.6. Infinite elements attached to the boundary of a standard finite element
mesh.
Consider the triangular element of Figure 6.12, with area coordinates L1, L2, and L3.
The area coordinate L1 has magnitude 1 at node 1 and is zero at node 2. Similarly,
L2 is 1 at node 2 and zero at node 1. This is illustrated in Figure 6.13. If edge 1
(connecting vertex 1 and 2) is of length ℓ12, then the gradient of L1 along the edge
is simply . By a similar argument, the gradient of L2 along the edge is . We also
note that the sum of L1 and L2 at any point along the edge is equal to 1. Thus if we
construct a vector function of the form
(6.276)
we find that this function has a constant component along edge 1 equal to . Also,
because the area coordinates are zero on the opposite edges, the function w12 has
zero component along edges 2 and 3. Let us examine further the properties of w12.
Taking the divergence, we see
(6.277)
(6.278)
Here we have also used the fact that the curl of the gradient of any scalar function
is zero.
We can now take for a shape function of edge 1, ℓ12 times w12. This is to scale the
result so that the function has a value of 1 along its associated edge. In this way the
three shape functions for the triangle are
(6.279)