Prediction and mechanism explain of austenite-grain growth during reheating of alloy steel using XAI

Abstract

Austenite-grain growth is an important factor in heat treatments, such as annealing and normalizing, for controlling the microstructures and overall properties of alloy steels. Thus, several researchers have proposed empirical equations for predicting austenite-grain growth in the reheating process. However, it is still important to improve the accuracy of the prediction model and analyze the model mechanisms and variable importance. Machine-learning models are key to enhancing prediction accuracy without the need for additional experiments. Therefore, machine-learning models are applied to predict austenite-grain growth with greater accuracy. The explainable artificial intelligence (XAI) is adopted to discuss the variable importance and mechanisms of the machine-learning model. 458 useable data points are collected from the literature, and then analyzed and eliminated outliers using a boxplot. The hyperparameters are adjusted using five-fold cross-validation and a grid search. Random forest regression (RFR) is selected based on its accuracy. The RFR is compared with an empirical equation to confirm the enhancement of the model accuracy. The variable importance and mechanisms of the machine-learning model are then discussed using the SHAP analysis.

Keywords

• Alloy steel
• Heat treatment
• Austenite
• Grain growth
• Explainable artificial intelligence
• Machine learning