摘 要：在氣動人工肌肉的靜態(tài)建模中，為尋找拉力與氣壓和位移的函數(shù)關(guān)系，該文利用訓(xùn)練后的BP神經(jīng)網(wǎng)絡(luò)預(yù)測氣動人工肌肉輸出力。將準(zhǔn)靜態(tài)實驗獲得的氣壓、位移和對應(yīng)的輸出拉力代入BP神經(jīng)網(wǎng)絡(luò)進行訓(xùn)練，得到氣動人工肌肉的BP神經(jīng)網(wǎng)絡(luò)靜態(tài)模型。預(yù)測結(jié)果表明，預(yù)測拉力與試驗測得拉力相關(guān)系數(shù)達0.99以上，且通過BP神經(jīng)網(wǎng)絡(luò)預(yù)測拉力與實測拉力誤差率在較大收縮范圍內(nèi)維持在較低水平，從而證明根據(jù)BP神經(jīng)網(wǎng)絡(luò)預(yù)測拉力的靜態(tài)模型是可行的。
關(guān)鍵詞：氣動肌肉；驅(qū)動器；BP神經(jīng)網(wǎng)絡(luò)；輸出力；預(yù)測
文獻標(biāo)志碼：A 文章編號：1674-5124（2015）12-0115-04
Application of BP neural networks in force prediction of pneumatic muscle actuators
GU Baotong， LIU Kai， MA Tao
（College of Mechanical and Electrical Engineering，Nanjing University of Aeronautics and Astronautics，
Nanjing 210016，China）
Abstract: To find out the function relationship among force， pressure and displacement in static models of pneumatic muscle actuators， the trained BP neural network is used to predict the force of pneumatic muscle actuators in the paper. To be specific， these data obtained through quasi-static experiment are trained in the BP neural network to get a BP neural network-based static model for pneumatic muscle actuators. Prediction results show that the correlation coefficient between the predicted and experimental force is higher than 0.99 and the error rate is confined in a relative low level for a wide range of contraction. The static model is therefore proven feasible.
Keywords: pneumatic muscle； actuator； BP neural network； output force； prediction
