Artificial intelligence -driven triboelectric nanogenerators for healthcare: Devices, algorithms and systems

Triboelectric nanogenerators (TENGs) can convert biomechanical and biochemical stimuli into electrical signals without an external power source, thus enabling wearable and implantable health monitoring systems. However, triboelectric signals are sparse, nonstationary, and highly sensitive to motion, humidity, contact changes, and device aging. Their clinical value therefore depends not only on device sensitivity, but also on algorithms that transform unstable self-powered waveforms into reliable physiological information. This article reviews the intelligent TENG health-monitoring system from the perspective of the collaborative design of devices and algorithms. We first survey self-powered sensors for sweat and body-fluid analysis, pulse and blood-pressure monitoring, respiration, ocular and facial-muscle motion, implantable cardiac monitoring, and self-powered pacing. We then trace the algorithmic evolution from hand-crafted-feature machine learning to convolutional, recurrent, attention-based, multimodal, and neuromorphic models, linking each model family to the signal bottleneck it addresses. We argue that in the next stage of TENG health monitoring will no longer mainly be defined by peak output or isolated recognition accuracy, but will more reflect aspects such as calibration stability, cross-subject generalization, motion artifact suppression, energy efficiency, and clinical interpretability. Emerging architectures like state-space models and reservoir computing may provide the computational basis for low-power, real-time health systems.