Harnessing Digital Microstructure for Simulation‐Guided Optimization of Permanent Magnets

Optimization of functional materials with properties governed by complex microstructures remains a central challenge in materials science. Accurate modeling of microstructure‐property relationships is often limited by idealized geometries and insufficient integration between experimental characterization and simulation. Here we present an experimental‐to‐computational workflow that combines 3D focused ion beam‐scanning electron microscopy tomography, deep learning‐based segmentation, and physics‐based postprocessing to enable high‐fidelity, structure‐aware modeling. Although this approach can be applied on a broad range of materials, we demonstrate it on a heavy‐rare‐earth‐free Nd‐Fe‐B magnet. We reconstruct a 13 × 29 × 33 µm 3 volume containing ~800 Nd 2 Fe 14 B grains and use patch‐wise micromagnetic simulations to identify microstructural features that control magnetization reversal. As a result, we estimated an upper bound on coercivity and analyzed local switching fields governed by a combination of microstructural factors, revealing the critical role of large nonmagnetic inclusions in strongly textured, exchange‐coupled regions. This methodology extends to multiphysics simulations, enabling data‐driven design of energy‐conversion and other functional materials, while also providing high‐quality data for generative AI and reduced‐order modeling.