Although, the bounding box cohesive element selection method is capable of large computational savings, the method has some drawbacks. This method is not self starting, instead it requires some cohesive elements to be present in regions where cohesive failure is expected. For fairly predictable problems, it can provide good results, but when the failure regions are not known a priori, the method is not optimal. An improved insertion method uses the average stresses of the neighbor volumetric elements to determine if the interface between these elements should be made cohesive. In effect, this allows us to begin a cohesive element free solution and only insert the elements as the stresses build to some predefined levels  defined by Equation 2.34.
As a test of the stress based insertion method, we use an Langle problem presented in Figure 4.20. The top and right boundaries are fixed in place and a vertical velocity of is placed in shear along the left boundary. The bulk material of the domain is PMMA with a Young's Modulus , Poisson's Ratio, , and density . The cohesive elements have a maximum stress , initial strength parameter and a normal and tangential critical separations of . The domain is discretized into nodes, edges and volumetric elements. Taking into account the instability of the cohesive elements to be inserted, the critical time step is reduced to .
We run four different simulations for a duration of time steps or , with the stress insertion selection occurring every time steps. The first represents the reference solution where cohesive elements are inserted everywhere in the domain at the start. The other three use the stress based insertion method for stress level of , and , respectively.
In order to verify the accuracy of the various solutions we observe the crack tip distance versus time, presented in Figure 4.22. From this figure, we can see that the crack tip distance, and indirectly the speed, are very close to the reference solution. Furthermore, from Figures 4.23 through 4.26, we can see that the crack profiles at the end of the simulation are very similar. In addition, to the crack profiles, we can see that the cohesive elements tend to concentrate in the high stress regions with the fewest elements present for the stress insertion level. Even though we achieve the greatest savings for the larger stress insertion levels, these results contain greater instabilities in the solution. This is visually apparent by the greater number of failing cohesive elements on the fringes of the domain  represented by dashed lines. For lower stress levels, as well as for the reference solution, the failing cohesive elements tend to be limited to only the immediate vicinity of the crack.
The timing results for the reference and stress insertion cases are presented in Table 4.5. The largest savings, of , occurs for the stress insertion of , which uses the fewest number of cohesive elements to obtain the solution, as seen in Figure 4.21.


