An AI-driven triaxial self-powered sensing system for attitude safety monitoring of aerial robots

This study reports a triaxial self-powered sensing system for attitude safety monitoring (TSPS-AM) in aerial robots. The system uses a three-axis, four-channel architecture that couples triboelectric nanogenerator (TENG) and electromagnetic generator (EMG) mechanisms, enabling its sensing units to detect triaxial acceleration, pitch, and roll without an external power supply. Validation confirms that TSPS-AM exhibits excellent linear fitting relationships within the horizontal acceleration range of 1–15 m/s2 and vertical acceleration range of 1–10 m/s2, and can monitor pitch and roll angles. Integrated with a lightweight CNN + GRU + Attention deep learning model, the system enables high-resolution multi-parameter identification, achieving resolutions of 0.1 Hz, 0.1 mm, and 2° for vibration frequency, amplitude, and excitation direction, and 0.1 Hz, 1°, and 2° for tilt frequency, tilt angle, and excitation direction. The identification accuracy of all parameters ranges from 93.34% to 99.33%. When used as an energy-harvesting unit, it delivered a peak output power of 0.0711 mW, corresponding to a power density of 273 mW m−3. After 60,000 cycles, the coefficient of variation of the output voltage for each channel remained below 5%, indicating good output stability. Finally, the trained model is deployed at the computer edge node with a Python-based terminal interface, and experimental tests confirm it effectively identifies and visualizes 9 typical aerial robot flight motions with 99.96% accuracy. In summary, TSPS-AM integrates self-powered sensing, data transmission, identification, and monitoring, laying a foundation for future attitude monitoring and collision-risk assessment in aerial robots.