A Lightweight Edge-Enabled Airspeed Estimation System for Fixed-Wing UAVs Using Distributed Flexible Pressure Sensors

Abstract

Reliable airspeed information is fundamental to flight control and safety for fixed-wing uncrewed aerial vehicles (UAVs), yet conventional protruding air-data probes can impose installation and integration constraints on small platforms. This article presents a distributed wing-surface pressure-sensing and edge airspeed-estimation system comprising eight commercial flexible pressure sensors, a custom STM32 acquisition board, an i.MX6ULL platform, and a lightweight temporal convolutional network. The estimator uses only eight-channel digitized time series acquired from the pressure sensors; Pitot airspeed is used solely for supervised training and evaluation. Three outdoor flights yielded 15 380 samples after alignment at 10 Hz. Over 4879 target timestamps from an independent test flight, the fixed TinyTCN instance used for embedded deployment achieved a mean absolute error (MAE) of 0.664 m/s, a root-mean-square error (RMSE) of 0.963 m/s, and a coefficient of determination of 0.963 against the raw Pitot reference. Replay-based execution on the i.MX6ULL produced a mean absolute difference of 0.00169 m/s relative to the computer reference pipeline and a mean software-pipeline computation latency of 9.94 ms; the computation time of every software window was below 100 ms. These results support the effectiveness of the integrated sensing and estimation system under the tested flight conditions and show that the causal preprocessing and inference pipeline fits the computational budget of the target embedded platform.