標題: 自調式摩擦力誤差補償器設計
The Design of Friction Compensators by Applying the Self-Learning Algorithm
作者: 蔣明潔
Ming-Jye Jeang
徐保羅
Pau-Lo Hsu
電控工程研究所
關鍵字: 摩擦力;類神經網路;學習自動機方式;運動控制;friction;neural network;learning automata method;motion control
公開日期: 1998
摘要: 在絕大多數的工具機中都會受到摩擦力的影響,而降低工具機的加工精度。傳統的摩擦力誤差補償方式是改變馬達的扭矩來克服摩擦力的作用。但此種方式需要變更驅動器的設計,並不適合使用在現有的工業馬達伺服系統上。因此,我們提出了新的摩擦力誤差補償方式,希望能將由摩擦力所造成的追跡誤差 ( tracking error ) 在命令輸入端就可以消除掉,藉由觀察追跡誤差與速度的關係,我們發現由摩擦力所造成的追跡誤差與速度有關,而與位置無關。所以我們設計了一個以速度為輸入,以補償的位置命令為輸出的補償器,來補償由摩擦力所產生的誤差。 首先,為了能將位置迴授控制系統的相位落後消除,我們也使用了零相位誤差追跡控制器,同時提供系統較佳的增益響應。進一步,在我們所設計的補償器中,我們提出了兩種補償器設計方式,一種是使用學習自動機來調整 sing function 的補償方式;另一種是使用類神經網路來做為補償的方式。兩種方式都是以速度做為輸入,以獲得最少的追跡誤差為學習目標。 在我們隨後的模擬與實驗中,我們都驗證了所提出的補償器架構可以有效的將摩擦力造成的追跡誤差與輪廓誤差加以消除,第一種方式可將誤差降為原來約18%,第二種方式更可進一步可達5%。
The friction force dominants the contouring performance of most CNC machines. Traditionally, friction is compensated by a torque compensation scheme which is difficult to be implemented on industrial AC servo systems. According to the relation between the tracking error and the velocity, three newly developed friction compensation techniques are proposed in this paper. By applying compensation only for the reference position command, the tracking error produced by friction are thus successfully reduced. The zero phase error tracking controller (ZPETC) is primarily applied to degrade the influence of servo lag in this position loop. Then, three compensation methods are designed by applying, the learning automata, neural network and the linearization technique. These compensators use the velocity as the input and the trained compensation position signal is generated as output to minimize tracking errors. These compensators are successfully implemented on the DYNA CNC machine and show that the contouring errors are reduced to about 18% and 5% by applying the learning automata algorithm and by applying either the neural network compensator or linearization technique, respectively.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT870591116
http://hdl.handle.net/11536/65000
Appears in Collections:Thesis