Tempered Hardness Optimization of Martensitic Alloy Steels

A simple Gaussian process regressor (GPR) model is employed to predict steel hardness and toughness response for tempered martensitic steels. A dataset of over 2000 hardness values from over 250 distinct alloys was compiled, with the aim of incorporating a diverse set of quenched and tempered martensitic steels. The Izod impact toughness was included for over 450 of these alloy/temper conditions. The GPR exhibited an increase in accuracy for both the predicted hardness and Izod impact toughness over linear regression trained on the same dataset. Shapley additive explanations (SHAP) were used to assess the importance of the input features of tempering temperature, tempering time, and 15 elements. Tempering temperature and carbon content were the most important input features in all models. The relative importance of the other 14 alloying elements varied depending on the target property. The SHAP analysis highlighted the complex relationships between composition and mechanical properties that are able to be captured by machine learning approaches.

Automated summary

A simple Gaussian process regressor model was used to predict the hardness and toughness response of tempered martensitic steels, showing increased accuracy compared to linear regression. The analysis revealed that tempering temperature and carbon content were the most important input features affecting the model's performance. The SHAP analysis demonstrated the complex relationships between alloy composition and mechanical properties captured by machine learning approaches.