Influence of p-method finite element parameters on predictions of crack front geometry

•K is calculated more accurately at free surfaces using smaller extraction radii.•Corner crack results are dependent on the number of elements along the crack front.•Predicted crack front geometry depends on the shape of the K profile along the crack front.•High p-levels do not necessarily compensate for a coarse mesh.•The energy norm should not be used as the sole indicator of solution convergence.

The effect of various p-method finite element model parameters on the prediction of planar crack front geometry in three dimensional structures is evaluated. An automatic crack growth method was developed using the commercial software StressCheck coupled with Microsoft Excel. The geometry of the evolving crack front was predicted using iterations of three dimensional finite element models to determine the stress intensity values at discrete points along the crack front and incrementing the crack length locally using the Paris growth law. Modeling parameters such as the global mesh size, p-level, crack tip mesh size, and the extraction radius used for determining the stress intensity factors were all considered. The effect of these parameters on the predicted shape of a growing elliptical crack front was evaluated through a series of simulations of a corner crack emanating from a centered hole in a plate. A single model of a through crack with a straight crack front was also used to compare the effects of the model parameters on a different geometric configuration. This study has shown that the corner crack geometry and the growth series are more sensitive to the number of elements along the crack front and the extraction radius than the through thickness crack. In addition, convergence of the predicted crack front geometry does not guarantee convergence of the distribution of stress intensity factors along the crack front. It was also determined that for the models considered, using a smaller overall mesh size is a more effective way of decreasing discretization error than increasing the p-level.