由於醫學進步與科技日新月異，以及面對老年化社會勞力成本的增加，近年來醫療器材與與輔助義肢的需求越來越大。一種新的肢體輔具則是利用肌電訊號的高時間解析度，作為機械輔具的控制訊號。然而，機械輔具的角度控制與穩定性，成為此種肌電訊號為基礎之輔具設備的成功關鍵。 本研究主要目的是擷取表面肌電訊號來預測出受測者在步行時手肘關節、髖關節和膝關節角度的變化，讓肌電訊號將來可能控制機器人或者是應用在醫療輔助肢體器材上。我們使用彎曲曲率檢知器來量測關節角度變化，它擁有的線性關係來得到相對電壓，人體的肌電訊號和肢體的角度變化之關係則是使用倒傳遞類神經網路理論來預測結果，而角度預測的結果和實際量測的結果做比較；手肘關節均方根誤差約1^o~2^o，髖關節均方根誤差約7^o~8^o和膝關節角度均方根誤差約5^o~7^o。 Due to the coming of aging society and high labor cost, modern society has increased demands on medical instruments and assistive prostheses. One novel prosthesis which has drawn great attention is the use of electromyography (EMG) as control signal to control the movements of a prosthesis. Nevertheless, one key element to the success of these intelligent prostheses is the reliability and stability of articular angle during movements. This study aims to develop a method for predicting the changes of articular angles during movements. The predicted articular angle can be used to facilitate the rotation stability of motor control in an artificial prosthesis. In this study, the articular angles were measured by a thin bending flex sensor. The measured articular angles were used as ground truth for performance evaluation. By recording the EMG signals as inputs and measured articular angles as outputs to an artificial neural network (ANN), the ANN predicts the angle position at next time point based on past information. The root Mean Square Error of predicted hand, hip ,and knee joint angles is 1^o~2^o , 7^o~8^o , and 5^o~7^o , respectively.
