Based on a limited experimental dataset, a machine learning-based model has been developed for the prediction of properties and design of composition-process of NiAl-based nanoprecipitates strengthened non-oriented silicon steel. Validation experiments were conducted to evaluate the model accuracy and understand the coordination between magnetic and mechanical properties. Among different machine learning algorithms, the random forest algorithm exhibited superior accuracy with the determination coefficient over 90%. Based on the objective function set in the alloy development framework, three new alloy composition and processing routes were rapidly designed. The explored non-oriented silicon steels exhibited recrystallization microstructure with average grain size of 22 – 31.8 μm and medium γ -fiber texture. After aging, a high density of dispersed B2-type NiAl-based precipitates with average size of ∼5 nm and low lattice misfit of less than 0.4% was formed. High yield strengths of 826 and 854 MPa were achieved, with elongations of 9.8% and 4% for the two candidate steels, respectively, showing high reliability in strength prediction. The high strength increment of ∼250 MPa was attributed to the NiAl-based nanoprecipitates, which likely resided within the critical size range for cutting-to-bypass transition. High electrical resistivity and combined effects of grain size and texture contributed to high magnetic induction B 50 of 1.63 – 1.64 T and low iron loss P 1.0/400 of ∼24 W/kg. Paramagnetic NiAl-based nanoprecipitates with low lattice misfit had a negligible effect on magnetic properties. This approach offers a feasible route for prediction and integrated design of high-strength non-oriented silicon steels with tunable composition–process combinations.
Yang et al. (2026) studied this question.