High-fidelity surrogate-based multidisciplinary multi-objective sequential optimization for AUV design

To address insufficient multidisciplinary coupling and the high computational cost of high-fidelity optimization in AUV path-tracking tasks, this study develops a CFD-informed surrogate-assisted multidisciplinary multi-objective sequential optimization framework. The framework integrates parametric geometry, CFD-derived hydrodynamic derivatives, motion dynamics, PID control, and ALOS-3D path guidance, with tracking accuracy and relative rudder actuation effort selected as the two objectives. To improve optimization efficiency, a Euclidean distance-based nearest-neighbor substitution strategy (ED-NNS) is introduced to transfer sparse CFD-derived hydrodynamic information from the geometric design space to the multidisciplinary design space. A hypervolume contribution-uncertainty method (HV-UNC) is further developed to identify and validate Pareto-sensitive solutions with high prediction uncertainty. The optimization process converges after six HV-UNC iterations, requiring 30 high-fidelity CFD validation samples. Compared with HV-based and uncertainty-based infill strategies, the proposed HV-UNC method reduces the CFD computational time by 33.3% and 45.5%, respectively. The final validation results show that the mean relative errors of both objectives are reduced to approximately 3%, indicating improved local surrogate accuracy near the Pareto front. For the representative optimized design CASE B, the trajectory-tracking RMSE decreases from 5.4831 m to 2.3471 m, while the relative rudder actuation effort, E total , decreases from 0.0224 to 0.0066, corresponding to reductions of 57.2% and 70.5%, respectively. These results demonstrate that the proposed framework can improve AUV path-tracking performance while reducing high-fidelity CFD cost and maintaining the reliability of the Pareto solution set.