Deep learning-based dual-reference triboelectric sensor for direct surface potential prediction

Quantitatively determining a material’s tendency to gain or lose electrons is crucial for triboelectric devices but remains challenging. Here, we introduce a dual-reference triboelectric sensor integrated with deep learning to rapidly estimate surface potential. An unknown material is contacted with two reference surfaces of opposite triboelectric polarity, producing paired electrical signals that act as internal calibration. A deep neural network maps these dual signals to the material’s effective surface potential, capturing interaction patterns that conventional analytical models cannot resolve. The system reliably quantifies surface-potential differences across diverse materials, achieving prediction errors below 8% and clearly distinguishing materials across the triboelectric series. The dual-reference design enhances robustness by compensating for environmental and measurement variations, yielding ~85% improved accuracy over single-reference methods. Overall, our results show that combining nanogenerator-based sensing with data-driven analysis enables accurate, quantitative interpretation of triboelectric responses and significantly broadens the functional capabilities of triboelectric sensors. Measuring a material’s electron-gain or loss tendency is essential for triboelectric devices but remains challenging. Here, the authors present a dual-reference triboelectric sensor combined with deep learning that rapidly estimates surface potential, achieving less than 8% prediction error and about 85% greater accuracy than single-reference methods by compensating for environmental and measurements variations.