Integrating machine learning and multi-objective optimization to investigate the magnetic and mechanical properties of FeSiCr soft magnetic composite processed by selective laser melting

Soft magnetic composite (SMC) products have been widely used in electromagnetic applications due to their unique properties, including extremely low eddy current loss and relatively low total core loss at medium and high frequencies. However, there were few studies that investigated the contradictory of mechanical and magnetic properties to process FeSiCr SMC powders. To ensure that the parts fabricated by selective laser melting (SLM) satisfy the application requirements and to assist manufacturing engineers in choosing optimal process parameters, the relationship of four key process parameters (oxygen concentration, laser power, scanning speed, and hatch distance) and three fabricated part properties (permeability, core loss, and ultimate strength) was established by the machine learning models. Three experiments based on the L9 orthogonal array design were conducted to optimize the material properties and to collect data for modeling. Five machine learning algorithms were investigated, and the better ones were adopted to estimate the part properties with the input SLM parameters, which were outperformed the inference based on the regression method. A manufacturing parameter suggestion system was developed with the selected machine learning models and a multi-objective optimal algorithm, NSGA-II, to assist engineers in determining the optimal process parameters regarding the conflictive mechanical and magnetic property requirements.