Micromechanische Modelleringsstrategieën voor Hooggetrouwe Digitale Tweelingen van Polykristallijne Materialen

Reliable and efficient computational models are critical for achieving high-fidelity digital twins of engineering components. Since material behavior at the component scale is governed by mechanisms operating at the microscale and below, the challenge of modeling this multiscale phenomenon lies in bridging the wide span of time and length scales. While modeling techniques at the extremes of the time and length scales---empirical macroscopic continuum plasticity and atomistic simulations---are effective at their respective scales, their applicability diminishes either in capturing lower-scale fundamental mechanisms or in scaling up to engineering applications. The mesoscale offers a natural bridge, but accurate representation of the material at this scale requires exhaustive parameterization to describe both the material state and its deformation behavior. With rapid advances in experimental characterization now providing unprecedented amounts of high-throughput data, mesoscale simulations can be further enhanced to appropriately exploit these advancements. The central aim of this thesis is therefore to assess and improve mesoscale modeling strategies in order to ultimately achieve high-fidelity digital twins of polycrystals. To this end, five interconnected research questions are posed and addressed. First, a rigorous assessment of the predictive capabilities of contemporary mesoscale crystal plasticity models is performed against large-scale experimental data obtained from in situ high-energy diffraction microscopy during a uniaxial tensile test. The mesoscopic grain-averaged elastic responses at multiple stages of the macroscopic loading were juxtaposed with two conceptually distinct crystal plasticity models---one phenomenological and the other physics-based---each calibrated to reproduce the experimental macroscopic stress--strain behavior. The assessment revealed that while both elastoplastic models accurately predict the elastic-dominated regime, their predictive quality diminishes once plastic mechanisms become active, thereby highlighting limitations in the constitutive descriptions. In pursuit of addressing the above question, the incorporation of \emph{additional} experimental data into model initialization led to substantial improvements in predictive capabilities. This was accomplished through a novel iterative strategy developed to initialize grains with experimentally measured elastic strains while simultaneously preserving compatibility and equilibrium conditions. The approach is shown to yield markedly improved mesoscale predictions of elastic strains at multiple stages of macroscopic loading, with the greatest impact in the elasticity-dominated loading regime, thereby demonstrating the importance of fully leveraging available experimental information when constructing digital twins. The third question highlights a major shortcoming of phenomenological models, namely their reduced reliability outside the calibrated data range when compared to physics-based approaches, despite their well-known efficiency and simplicity. A potential improvement was identified in the parametrization of the hardening moduli matrix governing slip--slip interactions, which has traditionally relied on simplifications dating back several decades. This was revisited by introducing a physics-informed parametrization that demonstrated the ability to reproduce both deformation behavior and material state more consistently beyond the calibration regime, including load path changes. With this refinement, phenomenological models retain their advantages while achieving closer alignment with physics-based predictions across a wider range of loading conditions. The fourth question extended the investigation to materials with higher anisotropy, specifically the hexagonal close-packed phases of Ti-6Al-4V, where both phenomenological and physics-based models face challenges in reliable parameter calibration. This difficulty is more pronounced in the former, as the reduced physical grounding of the parameters often makes them function as fitting variables, complicating their unique identification. To address this, a multistage optimization framework was developed, leveraging extensive in situ high-energy X-ray diffraction data collected during tensile testing. By incorporating both macro- and mesoscale responses into the objective function, the calibrated parameters not only aligned with literature values but also demonstrated consistent predictive performance in subsequent validation studies. Beyond calibration, this study provided new insights into slip activity, grain reorientation, and texture evolution during deformation, emphasizing the critical role of well-identified parameters for accurate modeling of hexagonal close-packed systems and, ultimately, for achieving reliable digital twins. Finally, the calibrated parameters were applied in a structure--property linkage study connecting microstructural features to residual stress states in a hot-rolled Ti-6Al-4V alloy induced after an abrasive wheel cutting operation. Localized out-of-plane residual stress measurements, obtained via focused ion beam--digital image correlation ring-core technique near the cut surface, revealed unexpectedly high tensile and compressive magnitudes, with subtle correlations to layered macro-textured regions. To test for causality and identify the governing conditions, crystal plasticity simulations were performed on representative microstructures under simplified loading. The study confirmed that the orientation of the micro-textured regions---governing both elastic and plastic responses and shaped by neighborhood interactions---combinatorially influences the residual stress state. The study confirmed qualitative and, to a considerable extent, quantitative agreement with experiment, and additionally also demonstrated how reliable parameters can enable digital twin frameworks to meaningfully link microstructure to macroscopic performance. Overall, it is shown in this thesis that the five areas of investigation not only address their respective questions with confidence but also, taken together, advance mesoscale modeling strategies toward achieving high-fidelity digital twins of polycrystals. At the same time, they show that the potential for improvement at this scale remains vast, and the present work only begins to scratch the surface of opportunities for further refinement and innovation.