Prediction of Hole Expansion Ratio in Advanced High-Strength Steels Using Physics-Informed Machine Learning.

Area of Science:

• Materials Science
• Computational Materials Science
• Mechanical Engineering

Background:

• The hole expansion ratio (HER) is crucial for assessing the formability of advanced high-strength steels (AHSS) in automotive manufacturing.
• Experimental determination of HER is resource-intensive, hindering rapid material development.
• Data scarcity poses a significant challenge for developing accurate predictive models for HER.

Purpose of the Study:

• To develop a machine learning (ML) framework for predicting the HER of AHSS.
• To address data scarcity by employing physics-informed synthetic data generation.
• To provide a computationally efficient tool for AHSS design and optimization.

Main Methods:

• Generated a synthetic dataset of 300 AHSS conditions using validated empirical relationships.
• Incorporated chemical composition, microstructure fractions, and mechanical properties into the dataset.
• Trained and evaluated multiple ML algorithms, optimizing a Gradient Boosting model.
• Performed feature importance analysis to validate the physical meaningfulness of the model.

Main Results:

• The optimized Gradient Boosting model achieved high predictive accuracy (R² = 0.80, RMSE = 5.81%, MAE = 4.93%) on an independent test set.
• Feature importance analysis identified ultimate tensile strength, bainite, martensite fractions, and strain hardening exponent as key predictors.
• The model's predictive errors were comparable to experimental variability, validating the synthetic data approach.

Conclusions:

• Machine learning models trained on physics-informed synthetic data can accurately predict HER values for AHSS.
• This approach offers a practical and accelerated method for AHSS design and optimization in the automotive industry.
• The physically meaningful feature rankings confirm the validity of using synthetic data for materials modeling.