Machine Learning‐Based Prediction of Microstructure and Mechanical Properties of Ce‐Microalloyed 490CL Wheel Steel Flash Welded Joints

Flash butt welding is the predominant joining technique for automotive wheel manufacturing, yet welded joints often exhibit degraded mechanical properties due to grain coarsening and deleterious inclusion formation. This study integrates Ce microalloying (10–150 ppm) with machine learning (ML) to predict and optimize the microstructure and mechanical properties of 490CL wheel steel flash welded joints. Welding experiments were conducted following a central composite design (current: 280–360 A, upset pressure: 2.5–4.5 MPa), and microstructures were characterized via optical microscopy (OM), scanning electron microscopy–energy‐dispersive X‐ray spectroscopy (SEM–EDS), and electron backscatter diffraction (EBSD). A stacking Ensemble model combining RF, XGBoost, and BP‐ANN was developed using 12 input features with 10‐fold cross‐validation. Ce addition modified elongated MnS inclusions to spherical Ce 2 O 2 S (shape factor: 0.42→0.85) and refined weld zone grain size from 36 to 24.5 μm. At optimal 100 ppm Ce, yield strength increased 5% (475→498 MPa) and elongation improved from 18.5% to 21.0%. The Ensemble model achieved superior prediction accuracy. SHapley Additive exPlanations (SHAP) analysis identified Ce content and welding current as dominant factors. The optimal process window (Ce 80–120 ppm, current 305–340 A) was established for reliable quality prediction in automotive wheel manufacturing.