Development of Smart Wearable Devices with AI-Enhanced for Smart Healthcare Applications Truong Tien Vo

Abstract

본 논문은 인공지능(AI)과 사물인터넷(IoT)을 기반으로 한 스마트 웨어러블 기술을 활용하여 성능 향상 및 건강 모니터링을지원하는혁신적인접근방안을제안한다.연구는골프스윙생체역학분석,침치료중활력징후모니 터링,그리고홈피트니스환경에서의근육활성도평가에중점을둔다.제안된시스템들은고급센싱기술,딥러닝 알고리즘,그리고 IoT기능을통합하여실시간사용자맞춤형솔루션을제공한다. 제1장에서는골프퍼포먼스를향상시키기위한데이터기반셀프코칭도구로서스마트골프장갑(Smart Golf Glove, SGG) 시스템을 제안한다. 본 시스템은 딥 뉴럴 네트워크와 IoT 플랫폼을 통합한 하드웨어–소프트웨어 프 레임워크를 기반으로 하며, 스윙 단계 및 이벤트 구분을 위한 알고리즘과 비지도 학습 모델(Unsupervised Learning Model, ULM)을 통해 소량의 라벨링 데이터를 효과적으로 처리한다. SGG 시스템은 초보 골퍼의 비정상 동작을 92.4%의정확도로검출하여,언제어디서나가능한셀프코칭도구로서의유용성을입증하였다. 제2장에서는심박동도(Ballistocardiogram, BCG)분석을기반으로한스마트침대센서(Smart Bed Sensor, SBS) 시스템을 제시한다. 이 시스템은 매트리스 하부에 장착된 PVDF(Polyvinylidene Fluoride) 필름 센서와 GRU(Gated Recurrent Unit), MHSA(Multi-head Self-Attention)를기반으로하는다중작업학습모델을적용한다. SBS시스템은 활동인식 98.7%,자세분류 97.6%의정확도를달성하였으며,심박수및호흡수는각각분당 0.77회, 0.43회의낮은 오차로추정함으로써침치료중실시간임상판단을효과적으로지원한다. 제3장에서는홈피트니스사용자를위한팔착용형스마트의류(Arm-worn Smart Clothing, ASC)시스템을제 안한다. 본 시스템은 표면 근전도(sEMG) 센서와 관성 측정 장치(IMU)를 통합하고, 다중 모달 데이터 융합 및 AI 기반 햅틱 피드백을 통해 운동 중 근육 활성도 및 활동을 인식하며, 근 피로를 감지하고 진동 피드백을 제공하여 운동의 정확성과 안전성을 향상시킨다. 근 활성도 인식(MAR)과 활동 인식(HAR)에서 각각 98% 이상의 정확도를 기록하였다. 종합적으로,본논문에서제안한 AI기반웨어러블시스템들은재활,스포츠트레이닝,개인건강관리등다양 한분야에서실시간 ·정확한모니터링과분석이가능한고도화된솔루션을제공하며,스마트헬스케어및피트니스 기술의발전에기여할수있는가능성을보여준다.|This thesis presents innovative approaches for smart wearable technologies driven by the artificial intelligence (AI) and Internet of Things (IoT) and to improve performance and health monitoring. It focuses on golf swing biomechanics, vital signs monitoring in acupuncture, and muscle activation assessment in fitness training. The proposed systems integrate advanced sens- ing, deep learning, and IoT capabilities to deliver real-time and user-centric solutions. In the first chapter, the research introduces the Smart Golf Glove (SGG) system, developed to meet the growing need for accessible, data-driven tools to enhance golf performance. Com- bining a deep neural network with an IoT platform, the system analyzes swing biomechanics through a comprehensive hardware–software framework. It incorporates a novel algorithm for segmenting swing phases and events, along with an Unsupervised Learning Model (ULM) to ex- tract deep features and handle the challenge of limited labeled data. The SGG system achieves 92.4% accuracy in detecting abnormal motions among beginner players, highlighting its effec- tiveness as a valuable tool for beginner golfers in self-coaching anytime and anywhere. In Chapter 2, the Smart Bed Sensor (SBS) system is introduced, utilizing ballistocardiogram (BCG) analysis, addresses non-invasive vital signs monitoring needs for acupuncture. It employs polyvinylidene fluoride (PVDF) film sensors under a mattress, coupled with multi-task learning using Gated Recurrent Unit (GRU) and Multi-head Self-Attention (MHSA) mechanisms. The system surpasses FDA-approved devices by means of 98.7% accuracy in activity recognition, 97.6% in categorizing laying positions, and provides heart rate and respiration rate estimations with errors as low as 0.77 and 0.43 beats per minute, respectively, and thereby supporting real- time clinical decision-making. Chapter Three presents the Arm-worn Smart Clothing (ASC) system, inspired by the grow- ing demand for home fitness and the need for real-time muscle feedback. The system integrates surface electromyography (sEMG) and inertial measurement unit (IMU) sensors with multi- modal data fusion and AI-driven haptic feedback. It achieves over 98% accuracy in muscle ac- tivation recognition (MAR) and human activity recognition (HAR), detects muscle fatigue, and provides vibratory cues to guide form and effort, enhancing safety and efficiency of home-based workouts. Together, these AI-driven wearable technologies offer precise, real-time solutions for di- verse applications. With significant implications on rehabilitation, athletic performance, and personal health, they have great potential to transform smart sports training, medical treatment, and fitness.