A Multi-Scale Framework for Predicting Additive-Manufacture Fatigue Life

Thesis Abstract

Additive manufacturing (AM) introduces complex microstructural heterogeneities and residual stress fields that strongly influence fatigue performance, challenging traditional life-prediction models that assume uniform material behavior and conventional loading conditions. This study addresses the need for a credible multi-scale framework capable of integrating process-structure-property couplings across scales to predict fatigue life of AM components with improved accuracy and reliability. The aim is to develop a predictive framework that links AM process parameters to microstructural evolution, damage mechanisms, and macroscopic fatigue life. Specific objectives include (i) characterizing the relationships between laser powder bed fusion (LPBF) process parameters (laser power, scan strategy, hatch spacing) and resulting microstructural features (grain size, texture, porosity, and defect distribution) using in-situ and ex-situ characterization; (ii) developing a mesoscale damage model that captures defect interaction, vein formation, and crack initiation under representative cyclic loading; (iii) formulating a homogenized macroscale fatigue-life model that embeds the mesoscale insights through scale-bridging parameters; (iv) validating the framework against a multi-parameter experimental database comprising 3D-printed Ti-6Al-4V specimens tested under axial-tension–compression and bending fatigue; and (v) performing a sensitivity and uncertainty analysis to quantify the influence of process-induced variability on fatigue life. The research adopts a sequential exploratory mixed-methods approach under a philosophy of critical realism. The population comprises LPBF-manufactured metallic specimens, with a target sample size of 180 specimens for fatigue testing, distributed across three build orientations and four process condition sets. Data collection employs laser-confocal microscopy, electron backscatter diffraction (EBSD), X-ray computed tomography (XCT), and digital image correlation (DIC) to capture microstructure, defect statistics, and strain fields. Mechanical testing utilizes high-cycle fatigue tests (R = 0.1) up to 10 million cycles, with subset tests for ratcheting and overload scenarios. At the mesoscale, integrated finite element (FE) modeling with cohesive zone elements and phase-field fracture formulations will be used to simulate defect-driven crack initiation and propagation, calibrated with EBSD- and XCT-derived microstructural descriptors. At the macroscale, a probabilistic fatigue framework based on a stochastic continuum damage model will be developed, incorporating defect statistics as random fields and utilizing Weibull and lognormal distributions to describe life variability. The analysis employs regression-based calibration, Bayesian updating for parameter uncertainty, and ANOVA to assess the effects of process variables. Model validation uses a hold-out dataset of 40 specimens, with performance metrics including root-mean-square error (RMSE) in life prediction, coefficient of determination (R2), and hazard-consistency indices. The study anticipates key findings (i) quantified links between LPBF defect populations, microstructural anisotropy, and initiation life, (ii) a robust mesoscale damage law that captures defect coalescence and localized thinning that governs crack nucleation, (iii) a validated multi-scale bridging strategy that reduces macro life prediction error by 25–40% compared with traditional S-N based methods, and (iv) a quantified uncertainty propagation framework that informs design safety factors under AM variability. The contribution to knowledge includes (a) a novel multi-scale framework that harmonizes process physics, microstructure, and macroscopic fatigue response for AM metals, (b) an integrated modeling workflow enabling rapid scenario evaluation for process optimization, and (c) an empirical database establishing cross-scale relationships for Ti-6Al-4V produced by LPBF. The study concludes that incorporating defect statistics and mesoscale damage mechanics into a probabilistic macro-fatigue model significantly enhances prediction fidelity and informs design guidelines for AM parts under cyclic loading. Recommendations include adopting in-situ monitoring strategies to reduce process-induced uncertainty, extending the framework to other alloys and AM processes, and integrating the model into design-for-additive-manufacturing (DfAM) software tools for industry deployment.