本文研發電子手系統,包括形狀記憶合金(Shape Memory Alloy,簡稱SMA)、電子手硬體本身、肌電訊號(Electromyographic,簡稱EMG)、類神經演算法及希爾伯特黃轉換四大部分。第一部分介紹形狀記憶合金,包括材料結構及記憶特性原理;第二部分包含電子手設計與架構,包括關節間的連動性及SMA在電子手間驅動方式;第三部份則是人體手臂前肢的肌肉構造介紹與電極貼片的黏貼處以及訊號擷取的處理過程,其中包含訊號的放大、濾波、整流等步驟,達到取得乾淨的肌電訊號的目的;第四部分主要為訊號擷取後之相關特徵值運算,並以類神經網路架構原理及希爾伯特黃轉換原理,來比較經過希爾伯特黃轉換後及原始EMG訊號之辨識率。
This thesis main topic is to build a system of SMA-driven cyberhand, which includes the SMA design, mechanical design, EMG signal acquisition, and artificial neural network and Hilbert Hung Transform. First of all, we introduce the principle and characteristic of SMA. We also introduce how does it work and what type of SMA we choose. Second, we introduce mechanism design and the mobility of every finger. The working principle of SMA inside the cyberhand will also be shown. Third, we introduce the muscle tissue of front hand; where we stick surface electrodes; and the EMG signal processing methods, that involves signal amplifier, filter, and rectification for obtaining pure EMG signals. Fourth part is to apply neural network and Hilbert Huang Transform to discriminate different hand gestures, when the pure EMG signals are being inputted to the controller. We compare the recognition rates between using HHT and without using HHT for EMG signal conditioning.