Drag-reduction mechanisms in autonomous underwater vehicle formations: Physical analysis and physics-guided surrogate optimization

Abstract

The hydrodynamic interaction between closely spaced autonomous underwater vehicles (AUVs) in formation produces strongly nonlinear drag responses. In the present configuration space, two dominant pressure-related mechanisms are identified: stern pressure recovery at the leader and bow low-pressure immersion at the follower. To map these mechanisms over a wide formation parameter space, a physics-guided surrogate framework is constructed from 666 Reynolds-averaged Navier–Stokes simulations of dual-AUV configurations. Total resistance is decomposed into pressure and viscous contributions and modeled separately. Guided by the effective-boundedness principle of Learning and Inference assisted by feature-space engineering, the feature space incorporates bounded geometric descriptors, a regularized turbulent-wake power-law decay, a Gaussian lateral wake profile, and pressure-interaction proxies. The resulting surrogate achieves pressure-drag R2≈0.974–0.986 and enables formation optimization over the prescribed feasible design space. At V=5.14 m/s (approximately 10 kn), the system-optimal arrangement yields a 5.9% total-drag reduction relative to two isolated AUVs, with computational fluid dynamics verification errors below 1%. The individual optima are associated with geometrically incompatible mechanism-favorable regions: the leader optimum is located on the centerline (b/D=0), whereas the follower optimum occurs at b/D=1.5, where D denotes the maximum hull diameter. This lateral separation of 1.5D places the two individual minima in different regions of the static formation space, so they are not attained simultaneously by a single feasible configuration. A nondominated leader–follower Pareto front quantifies the resulting tradeoff and locates the system optimum as a compromise between the two competing drag objectives.