The global call for a carbon-neutral society accelerates the growth of green energy technologies, increasing the demand for high-performance Nd-Fe-B-based magnets which play a critical role as a torque source for electric motors in e-mobility solutions and direct-drive wind turbines. However, Dy, Tb, and Nd, which are used in these magnets to deliver the performance at elevated operating temperatures of 100-200°C have limited natural rare earth resources.

This is renewed interest in searching for alternative magnets. For this purpose, SmFe12-based compounds are considered a potential candidate due to their high Fe content compared to other 4f-3d compounds and their superior intrinsic magnetic properties [1]. However, their practical applications are hindered due to the challenges of obtaining the optimum microstructure where hard magnetic grains will be isolated with the intergranular phase. This is achieved in high-performance Nd-Fe-B magnets through phase equilibrium between the hard magnetic Nd2Fe14B phase and low melting point phases. Hence, it is essential for magnet development to explore the equilibrium between the hard magnetic 1:12 phase and phases with low melting points in Sm-Fe-M phase diagrams, where M represents phase stabilizer elements such as V or Ti [2].

Therefore, in this work, we build data-driven methods by re-examining the equilibrium Sm-Fe-V phase diagram. This method will be adaptable to unknown Sm-Fe-M phase diagrams. For this purpose, first, we digitized the existing data from the literature group to existing phases and predicted the missing parts by using random forest, k-nearest neighbor (KNN), and Neural Network - Multi-Layer Perceptron (NN-MLP). The most uncertain regions in the phase diagram are determined from the probability distribution P(p│x) of the phase region labeled by p at each point x [3]. New samples are then synthesized experimentally from the most uncertain region.

Sm-Fe-V ingots were prepared by induction melting and annealed under Ar at 1100°C for 20 h. Subsequently, the microstructure, crystallographic, and magnetic properties characterization are performed. The obtained results were incorporated into the dataset. After the initial iteration, it was revealed that the 1:12 and liquid phase equilibrium can exist in the V-lean region with a composition of Sm12Fe77V11 (Fig. 1). This iterative cycle minimizes the necessity of numerous experiments and accelerates the development of an optimal microstructure.