Deep Learning-Enhanced High-Precision Wind Field Concurrent Triboelectric Sensing.

Abstract

Reliable and real-time wind field sensing is critical for environmental monitoring and distributed meteorological forecasting. However, conventional solutions often suffer from structural complexity and poor adaptability to harsh environments. In this work, a magneto-vortex triboelectric sensing system (MVTS) is developed by coupling a triboelectric nanogenerator (TENG) with a vortex-induced vibration (VIV) structure and magnetically reinforced elastic elements. The system converts wind-induced oscillations into electrical signals and supports full 360° wind direction decoding through a dual-channel frequency difference mechanism. Material-level optimization using FEP, nylon, and rabbit-fur electrostatic compensation enhances environmental resilience and long-term signal stability. A deep learning model, Regression Transformer (ReT), is constructed to extract temporal and frequency domain features from multichannel TENG signals, enabling high-accurate prediction of wind speed and direction. Controlled indoor experiments confirm a maximum wind speed error of 0.69 m s^-1, with prediction errors consistently below 5% and a wind direction error within 1°. Additional validations under -28 °C low-temperature conditions and wind-sand environments demonstrate the system's robust operation and strong environmental adaptability. This work provides a resilient, intelligent, and fully integrated solution for autonomous wind field monitoring in data-scarce, infrastructure-limited, and extreme outdoor scenarios.