Effects of algorithmic simulation parameters on the prediction of extreme value fatigue indicator parameters in duplex Ti-6Al-4V

Fatigue Indicator Parameters (FIPs) serve as surrogate measures of the driving force to form and grow transgranular fatigue cracks in metals and alloys. FIPs are volume averaged over sub-grain regions using a crystal plasticity finite element method (CPFEM) model of polycrystalline microstructures. Questions naturally arise regarding the ability to computationally simulate a sufficient volume of microstructure samples to objectively build a converged estimate of the distribution of maximum FIPs above some threshold, referred to as an Extreme Value Distribution (EVD). A reliable EVD can be estimated by considering FIPs computed from samples comprised of a sufficient number of grains. A recently developed statistical technique is exploited in this work to predict the maximum FIPs in a large volume of duplex Ti-6Al-4V. Effects of several crucial algorithmic parameters in the CPFEM simulations are explored, including the sub-grain FIP averaging volume, the statistical volume element (SVE) size of each microstructure instantiation, number of SVEs necessary to project extreme value FIPs for larger volumes, and finite element mesh density. The SVE size defines the number of grains in a single simulation and therefore controls the number of nearest and second nearest neighbor grain interactions that dominate FIP EVDs. This interplays closely with the number of SVEs needed to establish reliable estimates of FIP EVDs for larger volumes of microstructure. We demonstrate that the number of grains sampled is a critical consideration in these types of CPFEM simulations aimed at EVDs. The statistical technique for estimation of EVD FIPs is relatively insensitive to both the SVE size and mesh density of a single simulation, provided a sufficient number of individual grains are sampled to capture dominant effects of microstructure heterogeneity. Furthermore, averaging FIPs over different sub-grain volumes results in consistent predictions of the maximum FIPs.