Physics-enhanced machine learning for predicting strength of high-carbon chromium steel during thermomechanical processing and spheroidizing annealing

• Physics-enhanced machine learning framework developed for strength prediction of high-carbon chromium steel. • Deep learning-based microstructural segmentation using U-net to extract carbide morphology from SEM images. • Microstructural feature-driven prediction improves accuracy over process-parameter-only models. • Physics-informed loss function enhances model generalization and consistency with strengthening mechanisms. • SHAP-based interpretability analysis reveals key microstructural descriptors affecting strength. High-carbon chromium steels are essential for bearing manufacturing due to their exceptional hardness, wear resistance, and contact fatigue strength. Understanding the microstructural evolution during thermomechanical processing and spheroidizing annealing is critical for optimizing mechanical properties. However, conventional models struggle to capture the complex interactions between process parameters, microstructure, and strength. This study presents a physics-enhanced machine learning framework to predict yield strength and ultimate tensile strength. A U-net deep learning algorithm extracts key microstructural features—spheroidal, short rod-like, and lamellar carbides—from SEM images, enhancing predictive accuracy. To mitigate dimensionality issues, pearson correlation and random forest rankings are applied for feature selection. A physical loss function is integrated into the neural network, ensuring alignment with metallurgical principles. The proposed ANN-Phys model outperforms SVR, GBR, and ANN, achieving higher R 2 and lower RMSE . SHAP analysis identifies the volume fraction of lamellar carbides, spheroidal carbide count, and carbide size as dominant factors influencing strength. This data-driven approach bridges process, microstructure, and property relationships, offering a robust tool for optimizing high-carbon chromium steel processing.