A Novel Multi-Fidelity Optimization Method for the Design of Large Displacement Unmanned Underwater Vehicles

Large displacement unmanned underwater vehicle (LDUUV) serves as a critical underwater platform with enhanced endurance and advanced autonomous navigation capabilities. In LDUUV shape optimization design, two prominent challenges arise: high computational costs and the black-box nature of the optimization problem, where the relationship between design variables and objective functions is obscure and nonlinear. To address these challenges, a systematic approach is implemented by constructing a parametric model of the LDUUV shape, establishing high- and low-fidelity simulation models through grid-independence validation, and developing a multi-fidelity optimization algorithm tailored for computationally intensive black-box problems. The algorithm utilizes a radial-basis-function-based multi-fidelity surrogate model to bridge multi-fidelity datasets while integrating an adaptive search range mechanism. This mechanism dynamically adjusts the search boundaries and concurrently defines localized search domains, enabling efficient identification of optimal solutions through balanced global exploration and local refinement. Application of this methodology achieves a significant drag coefficient reduction of 17.7% relative to the LDUUV baseline configuration.