A MACHINE LEARNING-BASED PREDICTION MODEL FOR THE DISTRIBUTION AND INTERFACE STABILITY OF BANDED CARBIDES IN BEARING STEEL

The presence and stability of banded carbides significantly influence the mechanical performance and service life of bearing steels, particularly in high-demand applications. In GCr15 bearing steel, the formation and distribution of chromium-rich [Formula: see text] carbides during heat treatment can lead to anisotropic properties and premature failure if not properly controlled. However, traditional methods for predicting carbide behavior lack precision and adaptability across varied heat treatment conditions. Addressing this limitation, this research aims to develop a machine learning (ML)-based predictive model to estimate the spatial distribution and interface stability of banded carbides in GCr15 bearing steel subjected to controlled heat treatments, thereby improving material design and performance forecasting. GCr15 bearing steel samples underwent austenitization and quenching treatments under varied thermal parameters. These features were used to train Dynamic Gannet Optimized Intelligent Natural Gradient Boosting (DGO-INGB) to predict carbide banding and interface instability. Post-processing evaluation through Rockwell hardness (RWH) tests validated the predicted mechanical behavior. The proposed ML model achieved a prediction of austenitization with quenching for MAE (0.061, 0.049), RMSE (0.095, 0.087), and [Formula: see text] (0.86, 0.89) for interface stability under unseen conditions. The predicted results closely correlated with experimental hardness values, confirming the model’s reliability. This integrated ML-based approach enables efficient prediction of microstructural features in bearing steel, offering a novel tool for microstructural control.