Deep Learning-Aided Synthetic Airspeed Estimation of UAVs for Analytical Redundancy With a Temporal Convolutional Network

A synthetic air data system (SADS) is an analytical redundancy technique that crucial for unmanned aerial vehicles (UAVs) and used as a backup during sensor failures. Unfortunately, the existing state-of-the-art approaches SADS require GPS signals or high-fidelity dynamic UAV models. To address this problem, novel airspeed estimation method leverages deep learning unscented Kalman filter (UKF) proposed. Our fusion-based only requires inertial measurement unit (IMU), elevator control input, airflow angles while GPS, lift/drag coefficients, complex aircraft models are not required. Additionally, we demonstrate our proposed temporal convolutional network (TCN) more efficient model than renowned models, such ResNet bidirectional long short-term memory (LSTM). learning-aided UKF was experimentally verified on long-duration real flight has promising performance compared with methods. In particular, it confirmed robustly estimates even under conditions where of conventional methods degraded.