Composition design and optimization of Fe-C-Mn-Al steel based on machine learning

Area of Science:

• Materials Science
• Computational Materials Science
• Metallurgy

Background:

• Fe-C-Mn-Al steels are a promising class of advanced high-strength steels.
• Optimizing their composition and heat treatment is crucial for achieving desired mechanical properties.
• Traditional methods for material discovery are time-consuming and costly.

Purpose of the Study:

• To explore the composition space of Fe-C-Mn-Al steel using machine learning.
• To identify material compositions and heat treatment parameters yielding high ultimate tensile strength (UTS) and total elongation (TE).
• To accelerate the development of high-strength Fe-C-Mn-Al steel.

Main Methods:

• A literature-derived dataset of 580 Fe-C-Mn-Al steel samples was compiled.
• Eight machine learning models were trained to predict UTS and TE.
• Random Forest Regression (RFR) was selected for its superior predictive performance.

Main Results:

• The RFR model achieved high accuracy in predicting UTS (average absolute error ~90 MPa) and TE (average absolute error ~7.9%).
• Model validation with external data confirmed its strong predictive capabilities.
• The RFR model identified 50 optimal Fe-C-Mn-Al steel compositions and heat treatments for high UTS.

Conclusions:

• Machine learning, specifically RFR, is effective for predicting mechanical properties of Fe-C-Mn-Al steels.
• This approach significantly expedites the discovery of advanced high-strength steel compositions.
• The identified material combinations offer valuable directions for future steel development.