Overcoming Critical Challenges Hindering Resistance Spot Welding of Dissimilar Advanced High Strength Steel Joints: LME Mitigation and Weld Class Prediction

Abstract

Reducing carbon dioxide emissions from the transportation sector has driven demand for lighter vehicles. Advanced high-strength steels (AHSS) enable the use of thinner gauges without compromising crashworthiness due to ability to absorb high fracture energy. Because these materials are exposed while in-service AHSS are typically zinc-coated for corrosion protection. However, excellent mechanical strength is insufficient for these materials to be used for automotive application, they must also be capable of being welded into the automotive structure. Resistance spot welding remains the dominant joining method in automotive body-in-white production, yet two challenges affect weld quality in dissimilar stack-ups: liquid metal embrittlement (LME) cracking and the absence of reliable offline quality prediction. Most existing studies have focused on similar stack-ups, leaving dissimilar joints inadequately addressed. This research examines both challenges using dissimilar configurations representative of industrial practice. The first part investigates LME mitigation in two-sheet joints of zinc-coated 3G-980 AHSS and interstitial-free steel. A short high-current pre-pulse (16 kA, 20 ms) reduced the crack index from 0.56 to 0.14, a 75% reduction. Cross-sectional analysis revealed that the pre-pulse shifted the nugget toward the IF sheet, increasing the distance between the susceptible 3G-980 surface and the fusion boundary. This geometric shift reduced the overlap between liquid zinc and tensile stresses, suppressing crack formation. Contrary to welding made in similar material joint configurations where high-current pre-pulses intensified cracking, the same approach effectively mitigates LME in dissimilar configurations. The second part develops a machine learning framework for weld quality classification in three-sheet dissimilar AHSS stack-ups. Each weld was classified as acceptable (Ok), No weld, or Expulsion based on online assessments. Random Forest and XGBoost classifiers were trained on a 137-sample dataset, with XGBoost achieving 89.3% accuracy and superior performance near class boundaries. The trained models enabled identification of no weld regions and provided a basis for adaptive parameter selection. Both LME severity and weld class are critical indicators of joint integrity yet have rarely been addressed together for dissimilar coated AHSS. This thesis provides an experimentally grounded vii framework linking welding parameters to quality outcomes, offering practical pathways for process optimization in automotive resistance spot welding.