A Digital Twin and Probabilistic Optimization Framework for Resistance Spot-Welded Joints in Automotive Applications

Abstract

This study formulates a digital twin–driven probabilistic model to forecast and optimize the tensile behavior of resistance spot-welded (RSW) galvanized steel auto-body joints. Experimental experimentation was performed on multi-spot cross-tension specimens by a Taguchi L27 design with the considerations of welding current (8959–9554 A), sheet thickness (0.19–0.30 mm), electrode force (350–497 N), weld spot number (2–8), and radial distance (12–16 mm). Tensile strength varied from 4.08 Kgf to 150.91 Kgf. A linear regression model accomplished ±5% prediction accuracy in this range, with thickness and number of welds as the most significant parameters, and radial distance as having a significantly negative effect on strength. To offset the weakness of regression, finite element analysis (FEA) was systematically incorporated, allowing for a mechanistic understanding of stress concentrations due to off-center weld placement and to confirm the observed empirical trends. The FEA also showed predictive validity, with calculations at an unseen parameter set (9554 A, 0.27 mm, 350 N, 8 spots) approximating tensile strength to 145.2 Kgf versus the tested 150.9 Kgf (3.7% error). Parameter uncertainty was quantified using Monte Carlo simulations with Latin Hypercube Sampling (170 runs), and Global Sensitivity Analysis indicated sheet thickness and weld radius as overall factors on strength variability. Dynamic tensile testing (1.32–2.00 m/min) validated the model's insensitivity to service-relevant strain rates. Together, the synergy from the combination of regression, FEA, and probabilistic simulations provides a predictive digital twin approach that is both statistically sound and mechanistically interpretable. The approach improves weld performance assessment reliability and provides scalable applicability towards lightweight and safe automotive structures.