Enhancing resistance spot welding predictive accuracy through automated calibration and surrogate modeling
Resistance Spot Welding (RSW) performance is a critical enabler for high stack ratio (HSR) and multi-sheet (3T/4T) joints in next-generation battery electric vehicle (BEV) structures, where weld robustness directly influences crash performance and certification outcomes. The Welding Performance Prediction (WPP) framework advances weld parameter selection from empirically driven practices to a physics-informed, data-centric methodology.
This work presents an integrated simulation–machine learning framework to improve RSW predictive accuracy through three core components:
(1) automated calibration of the electrical contact resistance curve,
(2) development of a bias-corrected multi-fidelity response surface model (RSM),
(3) structured integration of virtual and physical design of experiments (Sim–Exp).
Automated curve fitting is applied to calibrate the temperature-dependent electrical resistance within finite element simulations, reducing prediction error in nugget diameter and replacing manual tuning with a systematic optimization workflow. Following calibration, computationally efficient low-fidelity simulations are executed with improved confidence and incorporated into a multi-fidelity RSM. High-fidelity physical test data, though resource-intensive, provide physics-consistent benchmarks. The multi-fidelity strategy leverages these datasets to correct systematic bias from low-fidelity simulations, achieving high predictive accuracy while significantly reducing experimental requirements. Sensitivity analysis and feature screening are employed to reduce dimensionality and enhance model robustness.
Virtual Weld Quality Assessment (VWQA) enables efficient exploration of welding schedules, while targeted physical testing provides metallurgical and mechanical validation. The resulting Sim–Exp framework expands design space coverage, lowers testing burden, and strengthens predictive reliability, supporting scalable digital twin deployment and early weld robustness assessment in advanced BEV development programs.
