Machine Learning-Driven Design and Optimization of Neodymium-Substituted R-Type Hexagonal Ferrites for Enhanced Electrical Polarization Properties
https://doi.org/10.5281/zenodo.17527188
Keywords:
Machine Learning, Neodymium Substitution, Hexagonal Ferrites, Polarization Properties, R-Type Structure, Materials Optimization, MultiferroicsAbstract
Multiferroic materials of the R-type hexagonal ferrites have simultaneous magnetic and electrical properties, which is why they are technologically important. Most recently, a promising trend in the control of the structural symmetry and polarization processes within these systems is the replacement of rare-earths by neodymium (Nd3 +). The present-day empirical studies offer the design and optimization of Nd-substituted Ba2Co2Fe16O27-type R-phase ferrites which are formed in a sol-gel auto-combustion process. The experimental data (n=420 samples) comprising of compositional ratios, sintering temperatures, descriptors of grain morphology, and dielectric parameters was trained on an ensemble model with a supervisor applying Random Forest Regression and Gradient-Boosting Algorithms. It was predicted and minimized polarization magnitude (P) and dielectric loss (tan d) with R2 = 0.963 on invisible information which is much better than the conventional regression techniques. At moderate levels of neodymium replacement (x 0.10-0.15), X-ray diffraction (XRD), scanning electron microscopy (SEM), and ferroelectric hysteresis loop analysis validation provided a gain in lattice distortion to allow spontaneous polarization without magnetic order decay.
Machine-learning optimization further revealed a nonlinear coupling between macrostrain and domain wall motion, establishing data-driven processing windows for high-performance multiferroic ceramics.
Results show that artificial intelligence (AI) can successfully fill in structure-property gaps in complex oxides and minimize the experimental trial-and-error process and allow rational design of state-of-the-art ferrites. The research will offer practical findings to researchers who aim at developing sustainable and smart materials engineering since it can be concluded that algorithmic discovery is supplementary but not substitutive of physical experimentation.
