SmCo₅ magnets are the material choice in aerospace engineering and satellite communication due to their unique properties, such as resistance to extreme temperature (-300°C to 550°C), rapid temperature changes, and being less reactive and corrosive. By substituting B in SmCo₄B, the anisotropy field can be enhanced from 40 T to 90 T (estimated) at 300 K, leading to ultrahigh coercivity while also reducing production costs [1-2]. However, this substitution significantly reduces the magnetization (µ₀M_s). Nonetheless, their low remanence (µ₀M_r = 0.3 T) remains the primary bottleneck limiting their applications. Optimizing processing conditions and chemical composition, such as through strategic elemental substitution, can help us maximize coercivity H_c and M_s in their performance. Integrating experimental results with machine learning (ML) algorithms will accelerate the discovery of the optimal balance between these properties.

A dataset consisting of 60 entries related to SmCo₄B-based compounds was compiled from the literature and combined with our experimental results. Features used for machine learning models included chemical composition, wheel speed, annealing temperature and time, remanence ratio, coercivity. To identify the optimal regression models, we trained various machine learning algorithms, including linear regression, partial least squares (PLS) regression, adaptive boosting (AdaBoost), support vector machine (SVM), K-nearest neighbors (kNN), and decision trees. For a given composition, we observed that AdaBoost and kNN provided superior predictive performance. The validation has been done on a separate test set (K-Fold) as well as experimentally.

This approach enables realistic H_c predictions for permanent magnets. High degree of correlation between the actual and predicted values is an indication that our models have good predictive ability. We are focusing on reverse permanent magnets design which can be characterized by exceptionally high H_c and M_s. Considering the growing database, it should be possible to improve the current limitations of the developed model and predict new permanent magnet materials.

Acknowledgment: The research leading to these results has received funding from the ERC under the European Union's Seventh Framework Programme (FP/2007-2013) / ERC Grant Agreement n. 57400056 and by the DFG, German Research Foundation Project ID No 405553726, CRC/TRR 270.

References

[1] H. Ido et al., J. Appl. Phys., 1993, 73, 6269.

[2] X. Jiang et al., J. Alloys Compd., 2014, 617, 479-484.