Artificial intelligence operated with machine learning was performed to optimize the amount of metalloid elements (Si, B, and P) subjected to be added to a Fe-based amorphous alloy for enhancement of soft magnetic properties. The effect of metalloid elements on magnetic properties was investigated through correlation analysis. Si and P were investigated as elements that affect saturation magnetization while B was investigated as an element that affect coercivity. The coefficient of determination R2 (coefficient of determination) obtained from regression analysis by learning with the Random Forest Algorithm (RFR) was 0.95 In particular, the R2 value measured after including phase information of the Fe-Si-B-P ribbon increased to 0.98. The optimal range of metalloid addition was predicted through correlation analysis method and machine learning.

The soft magnetic properties of Fe-based amorphous alloys can be controlled by their compositions through alloy design. Experimental data on these alloys show some discrepancy, however, with predicted values. For further improvement of the soft magnetic properties, machine learning processes such as random forest regression, k-nearest neighbors regression and support vector regression can be helpful to optimize the composition. In this study, the random forest regression method was used to find the optimum compositions of Fe-Si-B-C alloys. As a result, the lowest coercivity was observed in Fe80.5Si3.63B13.54C2.33 at.% and the highest saturation magnetization was obtained Fe81.83Si3.63B12.63C1.91 at.% with R2 values of 0.74 and 0.878, respectively.