Real-time detection for water pollutant based on triboelectric nanogenerators and machine learning

Water cleanliness and safety are fundamental to sustaining human activities and maintaining ecological stability. In this study, a self-powered water-quality sensing system is developed based on contact electrification and the distinct charge-transfer behaviors of different pollutants at the liquid-solid interface. When water samples containing heavy metal ions, microplastics, or rust flow through a conductive sponge, contact friction between the pollutants and the flexible porous structure generates differentiated triboelectric signals, which are continuously collected using an electrometer and a data acquisition card. By further integrating a Light Gradient Boosting Machine (Light GBM) model, a mapping relationship between signal features and pollutant types and concentrations is established for water-quality prediction. Experimental results demonstrate that the system can effectively identify heavy metal ions (Zn²⁺, Ba²⁺, and Al³⁺), polypropylene (PP) microplastics, and rust (Fe₂O₃), achieving an average classification accuracy of 86.67%. Validation experiments using municipal water samples from Kunming supplemented with quantified rust further confirm the reliability of the system. Under varying temperature (4.36-42.75 °C), pH (3-11), and turbidity conditions, the system maintains stable and accurate pollutant recognition, with detection accuracy reaching up to 100%. This study integrates liquid-solid triboelectric sensing with machine learning, providing a promising strategy for intelligent water-quality monitoring.