Stacking fault energy prediction for austenitic steels: thermodynamic modeling vs. machine learning

Area of Science:

• Materials Science
• Computational Materials Science
• Metallurgy

Background:

• Stacking fault energy (SFE) is a critical microstructure attribute governing deformation mechanisms and mechanical properties in austenitic steels.
• Accurate and straightforward computational tools for modeling SFE are currently lacking, hindering material optimization.
• Understanding SFE is vital for designing advanced high-performance steels.

Purpose of the Study:

• To develop and compare computational tools for predicting SFE in over 300 austenitic steels.
• To evaluate the reliability of thermodynamic and machine learning models for SFE prediction.
• To identify key alloying elements influencing SFE and their complex effects.

Main Methods:

• Application of thermodynamic modeling approaches.
• Implementation of ensembled machine learning algorithms.
• Statistical analysis of experimental data for over 300 austenitic steels.

Main Results:

• Ensembled machine learning algorithms demonstrated superior prediction accuracy for SFE compared to thermodynamic and empirical models.
• Thermodynamic model reliability can be enhanced by improving low-temperature CALPHAD databases and interfacial energy predictions.
• Nickel (Ni) and Iron (Fe) show a moderate monotonic influence on SFE, while other elements exhibit composition-dependent effects.

Conclusions:

• Machine learning offers a more reliable approach for predicting SFE in austenitic steels.
• Further development of thermodynamic databases is essential for accurate SFE calculations.
• Alloying element effects on SFE are complex and require careful consideration of overall composition.