A differentiable optimization framework for automated design of offshore jacket structures under varied scenarios

Conventional design methods for offshore jacket platforms rely heavily on engineers’ experience, resulting in conservative designs. This study employs a beam element-based topology optimization approach for jacket platforms, integrating multidisciplinary knowledge and simultaneously optimizing node coordinates and cross-sectional sizes. The objective is to minimize the weighted sum of volume and compliance. Given that marine loads are design-dependent, a differentiable formulation of Morison’s equation is developed. Pile-soil interaction is incorporated through a hybrid sensitivity scheme. Displacement constraints, diameter-to-thickness ratio constraints, and symmetry are considered. A Gumbel-Softmax-based strategy is employed to enable differentiable optimization of discrete standard sections for manufacturability. A member removal strategy is proposed to enable diversified design. The proposed method accommodates various multi-leg spatial frame configurations and can complete the design of a 4-leg jacket within 3-5 minutes. In addition, its effectiveness has been validated on a real engineering case.