Electrical grid-independent machine learning-assisted wearable gait analysis device with triboelectric-electromagnetic hybrid energy harvester.

In this study, an Electrical grid-independent Machine learning-assisted Wearable device for Gait analysis (EMWG) with a ground reaction force sensor is presented. For gait analysis, a multi-layer perceptron is identified as the optimal model among various Artificial Intelligence (AI) algorithms in terms of analysis performance (accuracy of 88 % and model conversion rate of 84.22 %) and power consumption (operation for 1 h 47 min with a 110 mAh battery) when deployed on a microcontroller. A triboelectric-electromagnetic (TENG-EMG) hybrid energy harvester was developed to supply power to the AI-embedded microcontroller by harvesting biomechanical energy. The hybrid energy harvester generated 150.28 μW and 158.61 μW from TENG and EMG, respectively, with walking, and 224.2 μW and 594.75 μW with running. The microcontroller can operate for 33.33 min using the 110 mAh battery, which charges in 126 min via the hybrid energy harvester operating at 5 Hz. The power efficiency (harvested vs. consumed) is 26.45 %. As a proof-of-concept, drunken person determination was performed using EMWG-mounted smart shoes with an accuracy of 88 %. The proposed wearable gait analysis device with embedded, grid-independent machine learning may serve as a technical guide for developing next-generation motion analysis wearables.