Fatigue modeling of U720 – A multi-scale approach in understanding grain boundary effects on crack initiation

Excessive scatter is observed in the fatigue response of a nickel-based superalloy, Udimet 720. This scatter can be partly attributed to the variability in the microstructure of this material. Hence, there is great interest in linking the microstructure to fatigue properties using a multi-scale approach that focuses on integrating the results of atomic simulations to the continuum level in the form of a micro-mechanical model. Thus, to capture the physics at the grain boundary (GB) interface, it is necessary to investigate this problem at a smaller scale. Molecular Dynamics (MD) simulations are used to obtain the energy barriers for slip nucleation and transmission across various GB characters, which are used to service our fatigue model. In this study, we construct a model for prediction of fatigue crack initiation based on the material’s microstructure. Our approach is to model the energy of a persistent slip band (PSB) structure and use its stability with respect to dislocation motion as our failure criterion for crack initiation. The components that contribute to the energy of the PSB are identified, namely, the stress field resulting from the applied external forces, dislocation pile-ups, and work-hardening of the material is calculated at the continuum scale. Further, energies for dislocations creating slip in the matrix/precipitates, interacting with the GBs, and nucleating/agglomerating within the PSB are computed via MD. The predicted fatigue life is driven by the microstructure such as grain orientations, widely distributed grain sizes, precipitates, PSB-GB interactions, as well as the effect of neighboring grains. The results predict that cracks initiate near twin boundaries from PSBs spanning a single large grain with a favorable orientation or multiple grains connected by low-angle GBs. Additionally, by varying the neighboring grains, we can account for scatter in the fatigue life. The uniqueness of our approach is that it avoids the large number of parameters prevalent in previous fatigue models and provides deterministic results. Excellent agreement is shown between the model predictions and experimental data