Fountain-inspired triboelectric nanogenerator as rotary energy harvester and self-powered intelligent sensor

Abstract

Toward the era of artificial intelligence (AI)-enabled smart rehabilitation and healthcare, wearable electronic devices that can accurately capture human motion and physiological signals are receiving more and more attention. However, existing devices still have limitations regarding energy supply, sensitivity, structural flexibility, fabrication cost, and system integration. Here, we proposed a wearable fountain-inspired triboelectric nanogenerator (FI-TENG) assisted by machine learning. The continuous sliding fountain-inspired structure can realize the effective amplification of the triboelectric layer displacement and positive pressure, and improve the shortcomings of the traditional TENG structure, such as poor electrical output performance, narrow sensing range and difficult to effectively sense the negative angle. By optimizing the design of the triangular displacement amplification angle, the tightened gap width, and the thickness of the sliding polyethylene terephthalate (PET) film, the performance of the optimal solution was improved by 70 % compared to the worst solution. The inconsistency between human body motion and TENG displacement direction was solved by introducing a slider-crank mechanism, which smoothly transformed the joint rotational motion into a linear motion of the slider. Due to its unique structural design, FI-TENG could efficiently harvest and accurately sense the positive and negative rotational motions of the human body's rotational joints, rehabilitation beds, and six-axis robots. As an energy application, when FI-TENG was installed in the wrist joint as a test environment, its maximum output power density could reach 64.65 mW/m² (rotation angle, frequency, and load resistance of 60°, 1 Hz, and 80MΩ). Based on the random forest (RF) machine learning method and intelligent microcontroller, an edge-AI intelligent system for human wrist rotation direction recognition was established. Finally, combined with the MobileNetV3-Small lightweight neural network, intelligent recognition based on two-dimensional (2D) images with higher accuracy (average accuracy of 97.56 %) was realized. The proposed FI-TENG shows potential application value in the fields of telemedicine monitoring, rehabilitation assistance devices and humanoid robots.