Study on the Spalling Behavior of Wear-Resistant Steels under Friction and Wear Conditions Using Image Segmentation

Abstract In this study, two representative wear-resistant steel specimens of high manganese steel (HM) and low alloy steel (LA) with different microstructures and mechanical properties were prepared through systematic heat treatment processes. A deep learning-based scanning electron microscopy (SEM) image segmentation framework based on the U-Net architecture was developed to investigate the surface morphology and mechanisms of wear behavior. The spalling regions on the sample surfaces were annotated and categorized into subsurface and surface spalling. Relying on a precisely labeled dataset and optimized through a hybrid loss function combining cross-entropy and weighted loss strategies, the proposed framework achieved high accuracy in pixel-level segmentation. The segmentation results directly quantified the spalling area ratio and revealed distinct wear and failure modes of samples. This study established correlations among spalling types, mechanical properties, and processing conditions, providing new insights into the mechanisms of the surface crack initiation. Overall, this work presents a scalable and objective data-driven approach that integrates artificial intelligence with metallographic characterization, offering a new pathway for analyzing failure mechanisms and predicting the service life of wear-resistant steels.