Machine learning guided processing, microstructure and coercivity mapping in M type strontium hexaferrite.

Area of Science:

• Materials Science
• Magnetism
• Data Science

Background:

• M-type hexaferrites are crucial rare-earth-free permanent magnets.
• Coercivity (H_c) in these magnets arises from complex interactions between processing, microstructure, and intrinsic magnetic properties.

Purpose of the Study:

• To develop a physics-informed machine learning framework to predict coercivity (H_c) in M-type hexaferrites.
• To identify optimal processing conditions, microstructural features, and magnetic parameters for maximizing H_c.

Main Methods:

• Trained multiple machine learning models, with extreme gradient boosting (XGBoost) showing the highest accuracy.
• Utilized uniform manifold approximation and projection (UMAP) for visualization and analysis of a five-dimensional input space.
• Constructed a prediction function and heatmap to derive actionable design rules for H_c optimization.

Main Results:

• Identified optimal conditions for high H_c: single-domain grain size (500-700 nm), moderate processing temperatures and durations, maximized magnetocrystalline anisotropy constant (K₁), and reduced saturation magnetization (M_s).
• Demonstrated that simultaneous control over processing, microstructure, and intrinsic magnetic properties is essential for enhancing H_c.
• Validated ML-derived trends against established micromagnetic principles.

Conclusions:

• The developed machine learning framework successfully captures physically meaningful processing-structure-property correlations.
• This approach provides a powerful tool for the rational design and optimization of rare-earth-free permanent magnets.
• Simultaneous optimization of multiple parameters is key to achieving high coercivity in M-type hexaferrites.