Search Fracture Mechanics Website

FEA Meshing Guidelines

 home > LEFM > fea meshing
Prev Up Next

Introduction

This page and the next are a bit of a detour from the theoretical development presented thus far. Instead of classical theory, these pages address the very practical tasks of (i) generating finite element meshes for linear elastic fracture mechanics (LEFM) analyses, and (ii) determining stress intensity factors, \(K\), from displacement data. This page will focus on meshing guidelines at crack tips.


Meshing Guidelines for LEFM Analyses

Proper LEFM FE Mesh
The sketch here shows the ideal finite element mesh at a crack tip for performing LEFM analyses. The key feature of the mesh is the placement of the midside nodes of the inner ring of elements at quarter-point locations. The reason for this is related to the element shape functions and will be explained in the next section. But first, here is a more complete list of meshing guidelines.

  1. As stated already, the inner ring of elements are quadratic and their midside nodes are at 25% of the edge length. This is the one guideline here that borders on being an absolute requirement.

  2. Note that all element edges are straight in the figure. It may be tempting to move the midside nodes so the outer edges of the elements form a perfect circle. But this would be wrong. The reason is buried deep in quadratic element shape function theory.

  3. The midside nodes on the outer ring of elements are at the midpoints of the edges, not at the quarter points. This should in fact be true of all elements that are not part of the inner ring.

  4. Finally, an anti-guideline. While the sketch here shows eight elements forming the inner ring around the crack tip, this number is not cast in stone. Six elements are also popular. Any number greater than five should be acceptable.


Quadratic Element Shape Functions

Crack Opening Example
The reason for the quarter point node locations is... finite element shape functions. Unfortunately, the subject of shape functions is far too expansive to thoroughly review here. Nevertheless, it is absolutely necessary to be familiar with them in order to follow the discussion below. This pdf file is a pretty good reference for shape functions.

For LEFM analyses, elements at the crack tip should be quadratic with quarter-point nodes, regardless of whether they are 2-D or 3-D. Elements with triangular cross-sections, 2-D triangles and 3-D wedges, are by far the most popular.

An example element is highlighted in the lower chart to the right. It includes the element's 6 nodes and local \(s\) and \(t\) coordinates.

The mapping equations for position, \({\bf x}\), and displacement, \({\bf u}\), are

\[ {\bf x} = \sum_{i=1}^6 {\bf x}_i \, \phi_i \qquad \qquad \text{and} \qquad \qquad {\bf u} = \sum_{i=1}^6 {\bf u}_i \, \phi_i \]
Proper LEFM FE Mesh
and the 6 shape functions are

\[ \begin{eqnarray} \phi_1 & = & (1 - s - t) (1 - 2s - 2t) \\ \\ \phi_2 & = & s ( 2s - 1 ) \\ \\ \phi_3 & = & t ( 2t - 1 ) \\ \\ \phi_4 & = & 4 s (1 - s - t) \\ \\ \phi_5 & = & 4 s t \\ \\ \phi_6 & = & 4 t (1 - s - t) \end{eqnarray} \]
where both \(s\) and \(t\) vary between 0 and 1.

The crack edge corresponds to nodes 1, 4, and 2, and \(t = 0\). So setting \(t = 0\) and focusing on the three key nodes gives

\[ \begin{eqnarray} \phi_1 & = & (1 - s) (1 - 2s) \\ \\ \phi_2 & = & s ( 2s - 1 ) \\ \\ \phi_4 & = & 4 s (1 - s) \end{eqnarray} \]
Rename the three nodes to make the following discussions easier to follow. Rename "1" to "0" because it is at the crack tip. Rename "4" to "mid" because it is the midside node. And rename "2" to "L" because it is at the full length of the finite element. The shape functions for the position vector, \({\bf x}\), and the displacement vector, \({\bf u}\), can now be written as

\[ \begin{eqnarray} {\bf x} = {\bf x}_0 \, (1 - s) (1 - 2s) + {\bf x}_\text{mid} \, 4 s (1 - s) + {\bf x}_L \, s (2s - 1) \\ \\ {\bf u} = {\bf u}_0 \, (1 - s) (1 - 2s) + {\bf u}_\text{mid} \, 4 s (1 - s) + {\bf u}_L \, s (2s - 1) \end{eqnarray} \]
The components of the position vector are \({\bf x} = (x, y, z)\) and we only care about the \(x\) position. Also, the components of the displacement vector are \({\bf u} = (u_x, u_y, u_z)\) and we only care about the vertical displacement, \(u_y\). This leads to

\[ \begin{eqnarray} x & = & x_0 \, (1 - s) (1 - 2s) & + & x_\text{mid} \, 4 s (1 - s) & + & x_L \, s (2s - 1) \\ \\ u_y & = & u_{y,0} \, (1 - s) (1 - 2s) & + & u_{y,\text{mid}} \, 4 s (1 - s) & + & v_{y,L} \, s (2s - 1) \end{eqnarray} \]
Now set the following variables. For the \(x\) position, set \(x_0 = 0\), \(x_L = L\), and finally set \(x_\text{mid}\) equal to a fraction, \(f\), of the element length, \(L\), or \(x_\text{mid} = f*L\) where \((0 \le f \le 1)\). So the position function can finally be written as

\[ \begin{eqnarray} x & = & f \, L \, 4 s (1 - s) + L \, s (2s - 1) \\ \\ & = & \Big[ f \, 4 s (1 - s) + s (2s - 1) \Big] \, L \end{eqnarray} \]
Also set \(u_{y,0} = 0\) and \(u_{y,L} = \delta_L\) where \(\delta_L\) is the vertical displacement of Node 2 at \(x_L\). These are the easy displacements. The remaining displacement at \(x_\text{mid}\) is the tricky one. Its vertical displacement is \(u_{y,\text{mid}} = \sqrt{f \,} \, \delta_L\) because the crack profile follows a square root function. All this gives

\[ \begin{eqnarray} u_y & = & \sqrt{f\,} \, \delta_L \, 4 s (1 - s) + \delta_L \, s (2s - 1) \\ \\ & = & \Big[ \sqrt{f\,} \, 4 s (1 - s) + s (2s - 1) \Big] \, \delta_L \end{eqnarray} \]
So in summary, we have

\[ \begin{eqnarray} x & = & \Big[ f \, 4 s (1 - s) + s (2s - 1) \Big] \, L \\ \\ u_y & = & \Big[ \sqrt{f\,} \, 4 s (1 - s) + s (2s - 1) \Big] \, \delta_L \end{eqnarray} \]
Now we can finally start choosing different values for \(f\) and analyzing the results. We'll start by showing how and why \(f = 1/4\) is the proper value for LEFM analyses. Insert \(f = 1/4\) and \(\sqrt{f\,} = 1/2\) into the equations. (They lead to remarkable simplifications.)

\[ \begin{eqnarray} x & = & \Big[ f \, 4 s (1 - s) + s (2s - 1) \Big] \, L \\ \\ & = & \Big[ (1/4) \, 4 s (1 - s) + s (2s - 1) \Big] \, L \\ \\ & = & s^2 \, L \end{eqnarray} \]

\[ \begin{eqnarray} u_y & = & \Big[ \sqrt{f\,} \, 4 s (1 - s) + s (2s - 1) \Big] \, \delta_L \\ \\ & = & \Big[ \sqrt{1/4\,} \, 4 s (1 - s) + s (2s - 1) \Big] \, \delta_L \\ \\ & = & s \, \delta_L \end{eqnarray} \]
In summary, \(x = s^2 L\) and \(u_y = s \, \delta_L\). The final step is to combine the two equations by eliminating \(s\). Do this by solving the \(x\) equation for \(s\) and substituting the expression into the other for \(u_y\). Solving the \(x\) equation for \(s\) gives

\[ s = \sqrt{ x \over L} \]
and inserting this into the \(u_y\) equation results in

\[ u_y = \sqrt{ x } \, \left( \delta_L \over \sqrt{L} \right) \]
This shows the vertical displacement in a quadratic element is indeed dependent on \(\sqrt{x}\), just as LEFM theory requires. It cannot be overstated how critical this is. The quadratic element with the quarter-point node properly captures the square-root dependence, and just as important, the infinite slope of \(d u_y/dx\) at \(x = 0\) that arises from the square-root function. This infinite slope at \(x = 0\) is known as the singularity.

Any other meshing strategy, either linear elements or quadratic elements with midside nodes at any location other than the quarter-point, will not capture this singularity regardless of how fine the mesh is because the slope at \(x = 0\) will not be infinite. It may be very large, but even very large is still finite, not infinite. In fact, one could say that very large is still infinitely far from infinite.

Displacements of Quadratic Elements

We return again to the example of the 60mm crack in the infinite plate on displacements.html, to see how well quadratic elements can reproduce the elliptical crack displacement profile with the singularity. Recall the displacement profile under plane stress conditions was given by

\[ u_y = 0.061 \, \text{mm} \; \sqrt{1 - \left( x \over a \right)^2} \]
where \(a = 30\,\text{mm}\) and \(-30\,\text{mm} \le x \le +30\,\text{mm}\).

We can imagine meshing two elements on the left hand tip at \(x = -30\,\text{mm}\) with \(L = 10\,\text{mm}\) elements. Nodes 1, 2, and 3 make up the edge of element 1 and nodes 3, 4, and 5 make up the edge of element 2. The \(x\) coordinates of the nodes, and their vertical displacements computed from the above elliptical equation, are

   Node ID     x (mm)      uy (mm)     Node Description
1 -30 0.0000   Crack Tip
2 -27.5 0.0244   Quarter-Point
3 -20 0.0455   Vertex Node
4 -15 0.0528   Midside Node
5 -10 0.0575   Vertex Node


Note - the five node numbers here are independent from the six nodes identified in the above sketch of the single quadratic triangle element.

Recall from above that the general shape function expressions for position, \(x\), and vertical displacement, \(u_y\), along the crack face are given by the following equations.

\[ \begin{eqnarray} x & = & x_0 \, (1 - s) (1 - 2s) & + & x_\text{mid} \, 4 s (1 - s) & + & x_L \, s (2s - 1) \\ \\ u_y & = & u_{y,0} \, (1 - s) (1 - 2s) & + & u_{y,\text{mid}} \, 4 s (1 - s) & + & u_{y,L} \, s (2s - 1) \end{eqnarray} \]
Also recall that \(s\) varies from 0 to 1 in these equations.

In this example, the expressions for element 1, containing nodes 1, 2, and 3, are

\[ \begin{eqnarray} x_\text{ Elem 1} & = & x_1 \, (1 - s) (1 - 2s) & + & x_2 \, 4 s (1 - s) & + & x_3 \, s (2s - 1) \\ \\ u_{y,\text{ Elem 1}} & = & u_{y,1} \, (1 - s) (1 - 2s) & + & u_{y,2} \, 4 s (1 - s) & + & u_{y,3} \, s (2s - 1) \end{eqnarray} \]
and element 2 contains nodes 3, 4, and 5.

\[ \begin{eqnarray} x_\text{ Elem 2} & = & x_3 \, (1 - s) (1 - 2s) & + & x_4 \, 4 s (1 - s) & + & x_5 \, s (2s - 1) \\ \\ u_{y,\text{ Elem 2}} & = & u_{y,3} \, (1 - s) (1 - 2s) & + & u_{y,4} \, 4 s (1 - s) & + & u_{y,5} \, s (2s - 1) \end{eqnarray} \]
Substitution of numerical values gives the following for element 1

\[ \begin{eqnarray} x_\text{ Elem 1} & = & (-30) \, (1 - s) (1 - 2s) & + & (-27.5) \, 4 s (1 - s) & + & (-20) \, s (2s - 1) \\ \\ u_{y,\text{ Elem 1}} & = & (0.0) \, (1 - s) (1 - 2s) & + & (0.0244) \, 4 s (1 - s) & + & (0.0455) \, s (2s - 1) \end{eqnarray} \]
and for element 2...

\[ \begin{eqnarray} x_\text{ Elem 2} & = & (-20) \, (1 - s) (1 - 2s) & + & (-15) \, 4 s (1 - s) & + & (-10) \, s (2s - 1) \\ \\ u_{y,\text{ Elem 2}} & = & (0.0455) \, (1 - s) (1 - 2s) & + & (0.0528) \, 4 s (1 - s) & + & (0.0575) \, s (2s-1) \end{eqnarray} \]
Keep in mind that \(s\) varies from 0 to 1 in all these equations. They are plotted by selecting many values of \(s\) between 0 and 1 and evaluating the equations for each selected value.

Plotting these equations on the ellipse, which is the exact solution, produces



The shape functions clearly do an excellent job of capturing the singularity and square-root dependency at the crack tip, as well as the overall elliptical behavior.

Finally, there are a couple of important points to highlight. First, note that node #2 is indeed a quarter-point node. Once again, this is essential in order to obtain the infinite slope of \(du_y/dx\) at the crack tip. Second, note that node #4 is a traditional midpoint node in element #2 because it does not contain any singularities. In fact, it would be incorrect to make node #4 a quarter-point node because doing so would force a singularity into element #2 when none should be present.

Example withOUT Quarter-Point Node

Let's repeat the above example, but with the quarter-point node, Node #2, moved to the midpoint. All the other nodes remain unchanged. Keep in mind the accuracy of this mesh will be vastly inferior to the above example. The chart at the end of this example will show why.

Recall the displacement profile under plane stress conditions is given by

\[ u_y = 0.061 \, \text{mm} \; \sqrt{1 - \left( x \over a \right)^2} \]
where \(a = 30\,\text{mm}\) and \(-30\,\text{mm} \le x \le +30\,\text{mm}\).

The new values for Node #2 are highlighted in the table below. All other values remain unchanged.

   Node ID     x (mm)      uy (mm)     Node Description
1 -30 0.0000   Crack Tip
2 -25 0.0337   Midside Node
3 -20 0.0455   Vertex Node
4 -15 0.0528   Midside Node
5 -10 0.0575   Vertex Node


Substitution of numerical values gives the following for element 1.

\[ \begin{eqnarray} x_\text{ Elem 1} & = & (-30) \, (1 - s) (1 - 2s) & + & (-25) \, 4 s (1 - s) & + & (-20) \, s (2s - 1) \\ \\ u_{y,\text{ Elem 1}} & = & (0.0) \, (1 - s) (1 - 2s) & + & (0.0337) \, 4 s (1 - s) & + & (0.0455) \, s (2s - 1) \end{eqnarray} \]
Element 2 remains unchanged.

\[ \begin{eqnarray} x_\text{ Elem 2} & = & (-20) \, (1 - s) (1 - 2s) & + & (-15) \, 4 s (1 - s) & + & (-10) \, s (2s - 1) \\ \\ u_{y,\text{ Elem 2}} & = & (0.0455) \, (1 - s) (1 - 2s) & + & (0.0528) \, 4 s (1 - s) & + & (0.0575) \, s (2s-1) \end{eqnarray} \]
Plotting these equations on the ellipse produces



The above chart shows that the shape function for element 1 is substantially different from the exact, elliptical solution. This has a significant impact on finite element simulations. The shape function error will cause the computed displacements of all nodes in the above example to differ from the exact elliptical solution. The exact amount of error will depend on particulars of the finite element mesh. Keep in mind that this nodal displacement error is not reflected in the above chart.

Limits on Midpoint Nodes

We have seen how the quarter-point location (25%) is the ideal placement of midside nodes for LEFM analyses. And we've seen how placing the midside node at any point other than the quarter-point makes it impossible to accurately capture the singularity at the crack tip.

This time, we address the very big problem that arises when midside nodes are placed at locations less than 25% along the edge. For example, let's place the midside node at 10%. The results of this move are presented below.

Recall the displacement profile under plane stress conditions is given by

\[ u_y = 0.061 \, \text{mm} \; \sqrt{1 - \left( x \over a \right)^2} \]
where \(a = 30\,\text{mm}\) and \(-30\,\text{mm} \le x \le +30\,\text{mm}\).

The new values for Node #2 are highlighted in the table below. All other values remain unchanged.

   Node ID     x (mm)      uy (mm)     Node Description
1 -30 0.0000   Crack Tip
2 -29 0.0156   10% Node
3 -20 0.0455   Vertex Node
4 -15 0.0528   Midside Node
5 -10 0.0575   Vertex Node


Substitution of numerical values gives the following for element 1.

\[ \begin{eqnarray} x_\text{ Elem 1} & = & (-30) \, (1 - s) (1 - 2s) & + & (-29) \, 4 s (1 - s) & + & (-20) \, s (2s - 1) \\ \\ u_{y,\text{ Elem 1}} & = & (0.0) \, (1 - s) (1 - 2s) & + & (0.0156) \, 4 s (1 - s) & + & (0.0455) \, s (2s - 1) \end{eqnarray} \]
Element 2 remains unchanged.

\[ \begin{eqnarray} x_\text{ Elem 2} & = & (-20) \, (1 - s) (1 - 2s) & + & (-15) \, 4 s (1 - s) & + & (-10) \, s (2s - 1) \\ \\ u_{y,\text{ Elem 2}} & = & (0.0455) \, (1 - s) (1 - 2s) & + & (0.0528) \, 4 s (1 - s) & + & (0.0575) \, s (2s-1) \end{eqnarray} \]
Plotting these equations on the ellipse produces



Once again, the shape function for element 1 is substantially different from the exact, elliptical solution. So far, this statement is no different than the previous example in which the midside node was placed at 50%. But it is the details of how this shape function is different that are critical.

The key is that the shape function folds over on itself. It does so because, for example, it crosses the \(x=-30\,\text{mm}\) vertical grid line twice. It does so first when \(s=0\), and then again when \(s=0.375\). This amounts to space folding over on itself, something that absolutely cannot happen in reality.

This is why some FEA programs will check for this occurrence and refuse to run if the midside node is less than 25% by any amount, no matter how small. For example, 24.999% would still be considered unacceptable.

Actually, an FEA program could still run. It would not divide by zero, and it would not crash. I think it should run and simply print a "strongly worded" warning message to the output file. Oh well.