串聯式機械手臂背隙誤差與關節順應性之鑑別與補償

Abstract

機械手臂在自動化產業需求逐年上升，然而受限於絕對精度不足，主要應用始終侷限於搬運、塗佈或鑽孔等精度要求相對較低的工序。關於機械手臂之誤差來源，可以分為來自零件尺寸、安裝公差的幾何誤差，以及各軸關節與桿件在不同運行條件所產生的非幾何誤差，包括背隙、順應性（彈性）、摩擦、溫升等現象。過去文獻對於順應性鑑別多需仰賴高成本的力矩感測器，或與各軸馬達連線溝通取得手臂關節受力情形資訊，否則僅能得到順應性係數結合手臂重量與重心資訊的混合參數。而本研究聚焦於機械手臂關節的背隙及順應性現象，整合相關文獻提出一種僅需利用位置量測資訊，能夠同時鑑別幾何、背隙與順應性誤差參數的建模架構，並能透過所鑑別之參數對各軸不同運行方向、手臂末端負載條件進行誤差預測及補償。 本研究提出之鑑別架構以Modified Denavit-Hartenberg（MDH）運動學模型為基礎，利用機械手臂各軸來回轉動位置差異鑑別背隙現象，並由兩種以上的設計載重造成之位置差異鑑別關節順應性係數，再以結合參數篩選的Levenberg-Marquardt（LM）演算法同時鑑別幾何參數誤差以及手臂連桿之重量與重心資訊。由數值模擬結果顯示所提出方法在隨機誤差干擾下，表現較傳統文獻方法更加穩定，而應用在實驗平台HIWIN RT605-710機械手臂路徑補償，能夠將目標路徑之最大誤差改善61%、平均誤差改善84%；且當載重條件改變時，也能將目標路徑之最大誤差改善51%、平均誤差改善75%。

The demand of robot manipulator keeps increasing in the field of automation industry recently. However, due to lack of position accuracy, the usage of industrial robot has usually been limited to the process that requires less accuracy such as pick and place, spray painting, and drilling. The error sources of industrial robot are classified into two categories, the geometric error that comes from size tolerance or assembling error, and the non-geometric error composed of other errors caused by different operating conditions at joints or links, including backlash, hysteresis, compliance (or elastic), and thermal deformation. Some of traditional techniques require more information to estimate the torque applied to joints, which can be obtained by executing identification process with communication to joint motor or setting up F\T sensor with higher cost. The other traditional techniques could only identify parameters that represent the fusion of compliance parameter and link mass and center of gravity. This study presents an approach to the identification of geometric error, joint backlash, and compliance error simultaneously with positional measurements only, and the identified parameters are used to predict and compensate for the position error of the robot with different moving direction or attached payload. The proposed approach is based on modified Denavit-Hartenberg (MDH) model, firstly the backlash angles are calculated with the position difference of opposite rotating direction of each joint, and the compliance coefficients are identified by applying two kinds or more designed payloads, then the geometric error and the parameters consisting of link mass and center of gravity are formulated by a Levenberg-Marquardt algorithm combined with parameter selecting method. Compared to recent literatures, the proposed method shows more stability under random disturbance during numerical simulations. The proposed approach is applied to a 6 degree-of-freedom industrial serial robot HIWIN RT605-710, the experiment demonstrates that the maximum and mean position error can be improved by 61% and 84% with same payload condition, and the maximum and mean position error can be improved by 51% and 75% after changing payload condition.