Machine Learning-Based Prediction of Stacking Fault Energy in High-Manganese Steels: A Comparative Study of Ensemble...

Area of Science:

• Materials Science
• Computational Materials Science
• Metallurgy

Background:

• High-manganese (high-Mn) austenitic steels offer exceptional strength-ductility, vital for automotive and structural applications.
• Controlling deformation mechanisms like transformation-induced plasticity (TRIP) and twinning-induced plasticity (TWIP) is key to optimizing these steels.
• Accurate prediction of stacking fault energy (SFE) is critical for tailoring these mechanisms.

Purpose of the Study:

• To systematically evaluate six supervised machine learning (ML) models for predicting SFE in high-Mn steels based on alloy composition.
• To identify the most influential compositional variables affecting SFE.
• To develop validated computational tools for targeted high-Mn steel alloy design.

Main Methods:

• A curated, outlier-cleaned experimental database of Fe-Mn-C-Si-Al-Cr-Ni-N compositions was used.
• Six ML models (MLR, RF, ETs, GB, SVR, stacking ensemble) were trained and evaluated.
• Nested 5-fold cross-validation with GridSearchCV and Z-score outlier removal (|Z| > 3) were employed.

Main Results:

• Extra Trees (ET) and Gradient Boosting (GB) models achieved high training R² values (0.988 and 0.990).
• The stacking ensemble model showed the best generalization on the test set (test R² = 0.603).
• Al, Fe, and Mn were identified as the most influential elements, with Al being the strongest linear predictor.

Conclusions:

• Stacking fault energy (SFE) in high-Mn steels is highly predictable from alloy composition alone using ML.
• The stacking ensemble model provides a validated computational tool for designing high-Mn steels with desired TRIP/TWIP characteristics.
• Composition-SFE design maps can guide the development of advanced high-Mn steels for specific applications.