Title: 平行機械運動控制與CMAC學習控制器設計
Motion Control of a Parallel Machine and CMAC Learning Controller Design
Authors: 羅文期
Wen-Chi Luo
Kai-Tai Song
Keywords: 平行機械;學習控制;逆向運動學;動力學;遺傳基因演算法;Parallel machine;Learning control;Inverse kinematics;Dynamics;Genetic algorithm
Issue Date: 2000
Abstract: 本論文主旨在於研究三軸平行機械以及其控制器設計。為了了解此三軸平行機械之特性,我們分析此三軸平行機械之逆向運動學及動力學,並且以電腦模擬驗證其正確性。針對平行機械之高性能控制需求,我們提出基於學習的控制器架構—小腦模型控制器(Cerebellar Model Articulation Controller, CMAC)。由於缺少平行機械實體直接實驗亦沒有順向運動學可供模擬驗證,故以二軸機械臂為控制對象,利用電腦模擬驗證CMAC之性能。本論文以追蹤誤差、適應性以及雜訊影響為測試項目,結果顯示CMAC有良好的性能,適合應用於平行機械上。在CMAC控制器設計方面,我們提出隨著學習次數降低之學習速率以加快收斂速率。並以遺傳基因演算法(Genetic algorithm, GA)來替代人工搜尋最佳的學習速率,模擬結果顯示皆可達到預期的功效。
This thesis presents analysis and design of the motion control of a parallel machine. The analysis and formulation of inverse kinematics and dynamics of the parallel machine have been completed. Computer simulation results confirm the effectiveness of the theoretical formulation. For the controller design, we proposed a learning control architecture – Cerebellar Model Articulation Controller (CMAC). In lack of the parallel machine for experiment and forward kinematics for simulation, we use a two-link robot manipulator instead to verify the performance of the proposed design. Computer simulation results show that the tracking error performance, adaptability and the ability of anti-noise of CMAC are fairly good. Therefore CMAC is a good candidate for the controller of the parallel machine. In addition, we use an exponentially reducing learning rate to accelerate the speed of convergence. Finally, a GA-based approach is proposed to search the best learning rate. Simulation results show it has a great potential to search the best CMAC parameters.
Appears in Collections:Thesis