指導教授：林達德臺灣大學：生物產業機電工程學研究所蔡安智Tsai, An-ChihAn-ChihTsai2014-11-262018-07-102014-11-262018-07-102014http://ntur.lib.ntu.edu.tw//handle/246246/261794近幾十年來，外骨骼機器人已經成為一項熱門的研究主題，並已被廣泛運用在加強人類能力、輔助行動不便的人以及物理治療知應用上。這類的研究主要在於加強使用者本身的能力或彌補使用者所欠缺的能力，並讓其能以自身的生理訊號直覺地控制外骨骼機器人。因此，本研究提出了一個新穎的，使用短時距傅立葉轉換排序 (short-time Fourier transform ranking, STFT-ranking) 之特徵以代表多通道肌電訊號 (Electromyography, EMG)　中所蘊含的不同肌肉間的關係以代表其動作資訊；並且本研究也開發了一套以使用者肌肉訊號為主要控制訊號之外骨骼機器手臂系統以加強使用者的能力。此外，此系統利用作為主要訊號以辨識使用者動作與稱為順應控制 (Admittance Control) 之機器人控制方法配合使用以控制此外骨骼機器手臂，並提出順應控制中調整參數之方法做為控制用途。 此STFT-ranking 特徵可用來擷取使用者肌肉訊號中所代表的動作資訊，並透過主成分分析法以及支持向量機建立動作辨識的模型。藉由此動作辨識模型，外骨骼機器人系統將可以利用使用者本身的EMG訊號來辨識其動作，而使得使用者能夠直覺的指揮機器人動作。此外，本研究也提出了肌肉電訊號與關節力量之間轉換的計算方式並應用於所提出的順應控制方法中，以及在此控制方法中，如何調整其控制參數的方法，並且將省力效果以能量節省指標 (Energy saving index, ESI) 量化作為分析與比較。 針對在兩種不同的肌肉收縮 (等張與等長收縮) 情況下，本研究亦進行了使用STFT-ranking特徵以及EMG訊號傳統特徵 (包含時域與頻域特徵) 所得之動作辨識效能的比較與分析。在不同特徵的比較測試中，在訓練與驗證階段的EMG訊號皆來自於相同的肌肉收縮狀況下，STFT-ranking 特徵可提供準確率超過90% 的動作辨識準確率。平均來說，使用外骨骼機械手臂系統可降低使用者的能量損耗，其降低量約超過40 % (肩關節約43.7 %，肘關節約59.3%)。根據實驗結果，本研究所提出的方法與機器人系統，在未來將可應用在輔助人們 (如: 農民、勞工) 搬運重物以及減輕其工作負擔。In last few decades, exoskeleton robotics has become a popular research topic and used for enhancing human abilities, assisting disabled people, and physical therapy applications. This type of researches focuses on enhancing or recovering user’s abilities and enabling the user to intuitively control an exoskeleton robot using the user’s biomedical signals. In this study, a novel feature extraction method, using the short-time Fourier transform ranking (STFT-ranking) feature, was proposed to represent the motion pattern information in multichannel EMG signals. Moreover, an exoskeleton-type robot arm system is implemented to enhance human abilities using multi-channel electromyography (EMG) signals as the main control signals. Otherwise, a control strategy based on admittance control method and a selection method of how to define an appropriate range for the adjustable admittance control parameters are proposed. The STFT-ranking feature method was used to extract information from the user’s muscles; then in order to build a motion pattern recognition model, principal component analysis (PCA) and support vector machine (SVM) were applied. With the motion pattern recognition model, the exoskeleton robot system can recognize the user’s motion patterns then be controlled by the user intuitively. In addition, the estimated relationship between the EMG signals and the force produced was employed to control the robot arm through the admittance control method. Furthermore, a selection method of how to define an appropriate range for the adjustable admittance control parameters was proposed. The corresponding assistant performance quantized by an energy saving index (ESI) was compared and analyzed. Considering two different muscle contractions (dynamic and isometric contractions), the study also compared and analyzed the performances of applying STFT-ranking features and conventional EMG features including time-domain and frequency-domain features. Among the features tested, the STFT-ranking feature yielded an accuracy rate exceeding 90% when the EMG signals the same type of muscle contraction were used in the training and validation feature data sets . On average, the robot arm system saved user energy by over 40% (43.7% and 59.3% for shoulder and elbow joints, respectively). The proposed methods and exoskeleton robotic arm system can be applied in the future to assist people such as farmers and laborers in the case of handling heavy objects to reduce their workloads.誌 謝 v 摘 要 vii Abstract ix List of Figures xi List of Tables xiv Chapter 1 Introduction 1 Chapter 2 Literatures Review 5 2.1. Signal Processing for EMG Signals 5 2.2. EMG Signals for Muscle Skeleton Model 7 2.3. EMG Signals for Motion Pattern Recognition 8 2.4. EMG Signals during Different Muscle Contractions 12 2.5. Applications of Exoskeleton Robotics 14 2.5.1. BLEEX 15 2.5.2. Hybrid Assistive Limb (HAL) 16 2.5.3. EMG-Based Exoskeleton Robotics 17 Chapter 3 Materials and Methods 19 3.1. Data Collection 19 3.2. System Architecture 22 3.2.1. Exoskeleton Robotics Arm Subsystem 25 3.2.2. Signal Acquisition Subsystem and Computing Subsystem 28 3.3. EMG Signal Processing and EMG to Force Regression 34 3.3.1. The Signal Segmentation and Onset Detection Method 34 3.3.2. EMG to Force Regression Method 40 3.4. Motion Pattern Recognition with EMG Signals 41 3.4.1. Conventional EMG Features 41 3.4.2. The Proposed Feature – STFT-Ranking Feature 47 3.4.3. Principal Component Analysis (PCA) 54 3.4.4. Support Vector Machine (SVM) 55 3.5. The EMG-Based Control Method 60 3.5.1. Admittance Control Method 60 3.5.2. Energy Saving Index (ESI) 65 3.5.3. Strategy of 2 DoF Robot Arm Control 69 3.5.4. Safety Mechanism 71 Chapter 4 Results and Discussions 74 4.1. Experimental Protocol 74 4.1.1. Motion Pattern Recognition 74 4.1.2. Assistant Performance Evaluation 84 4.2. Results of Motion Pattern Recognition 86 4.2.1. Comparison of Using Different Features 86 4.2.2. Comparison of Using Different Muscle Contractions 99 4.3. Results of Assistant Performance Evaluation 117 Chapter 5 Conclusions and Suggestions 128 5.1. Conclusions 128 5.2. Suggestions for Future Work 131 References 133 Appendix 14113104737 bytesapplication/pdf論文公開時間：2014/08/22論文使用權限：同意有償授權(權利金給回饋學校)肌電訊號外骨骼機器人動作辨識順應控制短時距傅立葉轉換排序法thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/261794/1/ntu-103-D96631003-1.pdf