Parametrically upscaled model-based predictive platform for fatigue with location-specific microstructural linkages

Fatigue in metallic materials is a cost-intensive engineering challenge due to unpredictable occurrences when undergoing cyclic loading. Particularly vulnerable is dwell fatigue, where the hold time significantly reduces life. The unpredictability is attributed to microstructure-dependent crack nucleation (a significant portion of life) in polycrystalline microstructures with evolving mechanisms. In titanium alloys like Ti-6Al-4V, microstructural heterogeneity, including micro-textured regions, anisotropic crystallographic properties, and strain-rate dependence play key roles in fatigue crack evolution. This paper develops a spatiotemporal multiscale computational platform for predicting the probabilistic manifestations of component-scale dwell fatigue crack nucleation, with linkages to the location-specific underlying microstructure. It integrates physics-based modeling, machine learning, temporal acceleration, and probabilistic analysis to introduce parametrically upscaled constitutive and crack nucleation models (PUCM-PUCNM) for efficient prediction at macro and micro scales. Nucleation studies with the experimentally-validated computational platform show its promise in multiscale fatigue predictions, exploring the competing effects of geometry and microstructure with loading. It also demonstrates the effectiveness of specimen test data-calibrated PUCM-PUCNMs in component fatigue predictions, demonstrating its promise for fatigue-resistant structure-material design.