A data-driven miniature water-flow vector sensor via dynamic pressure gradient inversion

Accurate flow velocity and direction measurement is essential for AUV/UUV hydrodynamic awareness, supporting sideslip estimation, path planning, and flow field perception. To overcome existing limitations of large volume, high power consumption, and poor dynamic adaptability, this study proposes a method based on dynamic pressure gradient inversion. Using Bernoulli’s theorem, a circular sensor array (42 mm radius) with MS5837-30BA pressure sensors was constructed. A lightweight CNN-LSTM hybrid network extracts spatiotemporal pressure features for joint velocity–direction measurement. The 3.28 MB model runs on an STM32F429VIT6 microcontroller with RS-485 output. Fluid simulation shows velocity error of 0.05 m/s (0.1–1.0 m/s) and direction error of 4.7° (0°–359°). Flume experiments yield velocity error ≤ ± 0.1 m/s (0.2–0.8 m/s) and direction error ≤ ± 15 ° (0°–359°). With miniaturization, ∼ 0.5 W power consumption, and strong anti-interference, the system offers an effective solution for marine flow field measurement.