As offshore wind farms move into deeper waters, jacket support structures increasingly rely on welded tubular joints whose fatigue life is governed by local stress concentrations. This study introduces, for the first time, a fully automated surrogate-assisted workflow for weld-geometry optimization of tubular joints, uniquely incorporating realistic manufacturing constraints derived from physical mock-ups. The proposed “design-simulate-learn-optimize” loop is fully automated, generating Bézier spline weld seams in Rhino 7 with the Grasshopper plugin, transferring geometry via a Python interface to ANSYS Mechanical (release 2023 R2), and executing standardized model setup, meshing, solving, and post-processing without manual intervention. Effective notch factors are assessed using the Neuber/Radaj reference notch approach (recommended by IIW) to account for micro-support effects, while radial basis function optimization (RBFOpt/RBFMOpt) surrogates efficiently guide the search. A sensitivity assessment confirms that the axially optimized weld shape remains largely transferable to combined load cases (e.g., axial force, in-plane bending, and out-of-plane bending), with bending mainly affecting peak stress localization and magnitude. These optimized weld shapes can reduce notch factors by up to 56% and significantly decrease the required weld material. The proposed optimization framework represents a significant advancement beyond existing methodologies, as it combines software integration, application to welded tubular joints, and practical manufacturing constraints. Additionally, it is adaptable with potential extension to other welded tubular joints and hollow-section components in offshore and civil infrastructure, enabling seamless integration of automated, data-driven weld-shape optimization into parametric CAD/FEA-based digital design workflows.
Brömer et al. (2026) studied this question.