Vous êtes sur la page 1sur 5

Not long ago, I wrote a blog titled “How Do I Know If My Mesh is Good Enough?”.

This
post dealt with the issue of mesh density and how it affects solution accuracy in a finite
element analysis. It also mentioned that element type and shape also affect accuracy.
This article focuses on element shapes: specifically, what’s meant by poor element
shapes and how can they affect solution accuracy.
Many finite element analysts are too preocuppied with other pre-processing
considerations to consider element shapes unless their meshing tool checks shape
quality automatically. This can be a mistake, however, as poor element shapes will often
cause convergence problems in nonlinear analyses and can produce inaccurate results,
especially when they’re located regions of a model where critical results are essential.
A number of element shape metrics can be used to evaluate shape quality. These
include Jacobian ratio, aspect ratio, corner angle, mid-side node position, parallel
deviation, warping factor, and composite quality measures. In all cases, they measure
the ability of an element to map data between element (numerical) space and real,
physical space, which is a critical step in FEA. Some of these metrics are shown in
Figure 1, above. The ideal element shape is one that has near 90 degree angles for
quadrilaterals and hex elements, and near 60 degree corners for triangles and tetradral
elements, while also having low elongation (aspect ratio). The element surfaces should
also be relatively flat, as measured by the warping factor.
The ANSYS software is a good example of a finite element code that includes automatic
element shape checking. By default, it checks both Jacobian ratio and a composite
quality measure. These two measures together are quite comprehensive because they
account for the placement of mid-side nodes, skewness, warping, curvature, aspect
ratio, and ultimately the element’s ability to map data. Figure 2 shows the Jacobian
ratios for different 2D element shapes. As you can see, large curvatures due to midside
nodes deviating from a straight edge and large corner angles result in large the
Jacobian ratios. The element quality number used by ANSYS as a composite measure
of shape quality is based on the ratio of the volume of an element to a sum of its edge
length. In 2D, a value of 1 indicates a perfect cube, square, or triangle while zero
indicates an element that has zero or negative area. A simple example is shown in
Figure 3 for a single 2D quad element.
Figure 2: Jacobian Ratios for Different 2D Element Shapes

Figure 3: Element Quality for a Poorly-Shaped Quadrilateral

All of this element shape information is great, but how much do poorly-shaped elements
really affect critical results? To illustrate this, we can look at the example for the same
2D plane stress model used in the earlier study of mesh density. In this model (Figure
4), a 2D bracket was constrained at its top end and subjected to a shear load at the
edge on the lower right. The maximum Von Mises stress converged to a value of
14,754 psi with a very high mesh density and high quality element shapes in the critical
region, as shown in Figure 4. Figure 5 shows the mesh used in the current study. This
mesh is coarser than the converged mesh used in the earlier study but fine enough to
predict stresses within 1.5% of the conveged stress with well-shaped elements.
Meshes with a range of element quality were evaluated. The best and worst-shaped of
these evaluated meshes in the critical region are shown in the upper right of Figure 5. A
comparison of the maximum nodal averaged and unaveraged stress results are listed in
Table 1 for these different meshes. Notice that the maximum stress solution deviates
from the converged value by 0.75% for a Jacobian ratio of 3.75, 2.4% for a ratio of 29.4,
and 8.1% for a ratio of 34. This last ratio corresponds to the mesh with highly skewed
elements shown in the upper right of Figure 5. Table 1 also shows that the difference
between the unaveraged stress and the converged stress is considerably greater for the
mesh with significantly skewed elements.

While this doesn’t represent a comprehensive test of the effects of poor element shapes
on results accuracy, it provides a solid example of its effects. It‘s important to note,
however, that the influence of element shape on results is not easy to predict, especially
because so many other factors affect results accuracy. These factors include the mesh
density, element type selection, response type (linear or nonlinear), and the accuracy of
inputs such as loads, constraints, and material properties. So the overall error can be
considerably different from the error caused by element shapes alone.

Finite element codes such as ANSYS will automatically check the Jacobian ratio and
composite element quality, and will warn users if the values exceed specified limits.
However, it’s good practice to check these error measures yourself, especially the
Jacobian ratio, which provides a direct indication of the ability of a mesh to map data
from element space to real space. As illustrated by the bracket example, poorly-shaped
elements in regions of your model where results accuracy is critical can cause
significant inaccuracies which can ultimately influence important design decisions. We
are very interested to know of any experience you have with poorly-shaped elements
affecting your results. Have they caused real problems, how did you find them, and
what did you do to fix them?
Figure 4: Highly-Quality Refined Mesh with Converged Stress Results
Figure 5: 2D Bracket Model Geometry and Mesh
Table 1: Maximum Stresses For Meshes With Good to Poorly-Shaped Elements

Vous aimerez peut-être aussi