Deciphering the composition-microstructure correlation in low-density FeMnAlC steels with machine learning

[Display omitted] FeMnAlC steels have been regarded as prospective candidates for the advanced high-strength steel with reduced density. However, the correlation between the complex steel composition and the resulting microstructure remains unclear. This study presents a robust and efficient approach for establishing the correlation between the composition and microstructure of low-density FeMnAlC steel using a machine learning algorithm. By employing classification and regression models, we successfully predicted four significant characteristics: the existence of the κ phase, a dual-phase region (γ + κ) at the aging temperature, a solitary γ phase region, and the volume fraction of the κ phase. Notably, the elements Al and C, with a particular emphasis on Al, exert significant influence on the phase structure. For example, when the Mn element content is 30 wt%, an estimated minimum Al content of around 4 wt% is required to guarantee the existence of the κ phase. Furthermore, the anticipated content range of Al and C required for the existence of the (γ + κ) duplex region at the aging temperature spans from 4.5 to 9.3 wt% and 0.4 to 1.1 wt%, respectively. Ultimately, the Fe-30Mn-8Al-1C steel was chosen for experimental validation, and the predicted characteristics exhibited excellent agreement with the experimental results. Our research offers a roadmap for the development of low-density steel, and the approach we devised holds promise for application in other metallic materials as well.