• Seven geometric parameters predict spot weld durability via notch and electrode stresses. • Quadratic regression with two-way interactions advances stress prediction accuracy. • Simulation boundary conditions critically influence key parameter significance. • Nominal stress calculations reduce sensitivity to sheet thickness variation. • Geometric variability in models improves the reliability of fatigue life predictions. Fatigue failures in resistance spot welds represent a critical challenge in automotive design, yet current evaluation methods typically consider only nominal geometries and neglect manufacturing‑induced variability. This study investigates the influence of geometric variations on fatigue-life prediction by combining linear-elastic notch stress analysis using a simplified notch and regression-based surrogate modeling. The approach targets high-cycle fatigue dominated by elastic stress concentrations requiring precise simulated peak stress evaluation. A D‑optimal design of experiments was conducted with seven geometric design variables, including sheet thickness, notch radius, electrode indentation, nugget diameter, electrode diameter, and the inner and outer gaps. In total, 80 simulations were performed to assess notch stress and electrode indentation stress as response variables. Quadratic regression models achieved coefficients of determination R 2 > 0.99 for both responses. The models reveal that parameter significance changes with boundary conditions, such as constant force or stress, that affect the critical area. The 95 % confidence band for notch stress widens from about 3 % in the central design space to approximately 6 % for thinner sheets and increased electrode indentation. Incorporating realistic parameter deviations substantially improves fatigue-life prediction reliability and enables quantitative assessment of uncertainty. The proposed framework establishes a variability-aware methodology that integrates simulation and regression analysis into spot weld design, advancing beyond current nominal-value-based practices. While the present work focuses on stress prediction and variability, the simulated notch stresses can subsequently be combined with experimental S–N data by reproducing the test geometry and loading in the simulation, thereby enabling direct fatigue-life assessment within the same conceptual framework.
Baer et al. (2026) studied this question.