XFEM is often proposed to be a mesh-independent method, i.e.
the mesh does not have to follow
the crack geometry. But the modeling method showed a higher mesh dependency than expected, where the mesh size at the crack tip showed to be the aspect that affected the accuracy the most. XFEM enables for a relatively simple/flexible crack modeling in terms of mesh construction, but the accuracy is varying with the flexibility. A conformed mesh around the crack tip, similar to what is used in conventional crack modeling, gives very good results within 5 percent error. However, this kind of mesh refinement is hard to accomplish for complex structures. An unstructured mesh is more flexible and easier to model but is shown to give lower accuracy (within 10 percent error).
Caution should be taken when the crack breaks the specimen surface. The contour integral is evaluated poorly at the surface which results in erroneous SIFs there. It is advised to disregard this value and extrapolate the SIF from the closest evaluation points. Submodeling has proven to work well with XFEM and is recommended to handle larger structures. Using both stress and displacement based submodels is advised to find the limits where the real solution lies within. Mixed tetrahedral-hexahedral FE-meshes with tie constraints have also shown good results. It simplifies the mesh modeling, thus increasing flexibility while the accuracy is maintained. The approach is recommended for more complex geometries, where the crack region is meshed using hexahedral elements while tetrahedral elements are used outside the crack region. This is only proven for a straight crack case, whereas a more complex loading situation can be disturbed by the constraint between the meshes to a greater extent. Elements with reduced integration scheme give for pure tension loads a reduction of the calculation time by 10-20 percent. The loss in accuracy is less than 1 percent and the reduced integration scheme is therefore concluded to be applicable for XFEM. It should be noted that this is not proven for complex load situations.
The requirements on the SIF solution in the thesis are thought to be too high for practical use. Larger mesh sizes are therefore suggested to be investigated further, especially for conformed meshes. The results for notched specimens differ vastly and therefore further studies of cracks with stress raisers are suggested. In this context, the usage of reduced integration could be examined as a potential problem. Other benchmark cases with stress raisers could be investigated, but also the cases analyzed in this thesis could be further studied to strengthen or disapprove these results. Furthermore, the models could be tested using conventional crack modeling techniques using conformed meshes.