Optimal Design of the Austenitic Stainless-Steel Composition Based on Machine Learning and Genetic Algorithm

Area of Science:

• Materials Science
• Computational Materials Science
• Metallurgy

Background:

• The materials genome paradigm offers a new approach to accelerate materials research and development.
• Austenitic stainless steel is a critical material, but its development can be time-consuming and costly.

Purpose of the Study:

• To optimize the chemical composition of austenitic stainless steel.
• To reduce production costs and speed up the development of new steel grades.
• To achieve these goals without sacrificing mechanical properties.

Main Methods:

• Collected experimental data for austenitic stainless steel.
• Developed machine learning models, specifically gradient boosting regression (GBR), for predicting mechanical properties.
• Utilized Bayesian optimization to tune GBR hyperparameters.
• Employed the NSGA-III algorithm for multi-objective optimization of chemical composition.

Main Results:

• The GBR model achieved high prediction accuracy for yield strength (R²=0.88), ultimate tensile strength (R²=0.99), elongation (R²=0.84), and reduction in area (R²=0.88).
• Feature importance and SHAP values provided insights into the factors influencing mechanical properties.
• The NSGA-III algorithm identified optimal non-dominated solutions for chemical composition.

Conclusions:

• Machine learning and genetic algorithms can effectively optimize austenitic stainless steel composition.
• This approach accelerates R&D cycles and reduces costs while maintaining material performance.
• The study demonstrates the power of computational methods in advancing materials science.