摘 要 本篇論文利用變結構控制、模糊滑動模式控制及模糊類神經網路等控制理論設計來抑制軸向移動繩系統的振動問題,在驅動的方法上,分別藉由調整繩子的軸向張力以及移動邊界的質量阻尼機構系統。在單純的模糊控制器設計過程中,針對非線性系統控制規則資料庫的建立有一定的困難度,另外對於系統的穩定性及強健性並不是必然的結果。所以我們結合了滑動模式控制理論以及模糊推論來建立起系統控制的規則資料庫,期望對於受控系統的穩定性及強健性有一定的幫助。此外,為了提高系統的優良性能,我們希望藉由基因演算法來搜尋模糊滑動模式控制器中最佳的控制參數。除了上述的模糊滑動模式控制器結合了基因演算法的應用,對於擁有線上學習功能的模糊類神經網路並兼具有滑動模式特性的控制器也在本論文中被發展出來並成功運用在軸向繩系統的振動控制上。在切換控制器中利用李雅波諾夫的方法我們可以解決系統穩定性的問題。另外針對我們所設計出來線上自我學習的演算方法,模糊類神經網路控制器中可調變的參數將朝著振動位移量收斂的方向作自我的修正並最終找到一個最佳參數。針對各種設計出來的控制器使用於系統的減振效果將在數值模擬的部分呈現出來。在模擬結果方面我們可以很清楚的看到結合了模糊滑動模式控制與基因演算法的新型控制器相較於其他的有較佳的效果。但是在可行性方面,模糊類神經網路控制器在設計流程中較為簡單,且其線上即時學習調變參數的特性優於其他離線搜尋最佳參數的方法。
Abstract The variable structure control (VSC) via Lyapunov direct method, fuzzy sliding mode control (FSMC), fuzzy neural network control (FNNC) methods, in the present study, are applied to suppress vibration of the axially moving string system, which is driven by adjusting the axial tension and boundary Mass-Damper-System (MDS), respectively. Due to the difficulty of utilizing fuzzy logic control (FLC) to obtain the linguistic control rules, the stability and robustness are not guaranteed. The sliding-mode control is then incorporated in the scheme of FLC to generate the control rule bases in order to meet the requirement of stability and robustness. In additions, for the purpose of improving system performance, the Genetic Algorithm (GA) is applied to search for the optimal parameters of the fuzzy sliding mode controller. Beside the aforementioned method of fuzzy sliding mode control plus GA, the on-line learning algorithm Fuzzy Neural Network (FNN) accompanied with sliding mode characterization is also developed and applied to achieve control goal. By using the Lyapunov direct method in the alternative control law, we can resolve the stability of the fuzzy control method, and the on-line intelligent control schemes, the parameters of FNN are adjusted in the direction that minimizes sliding mode or amplitude variable . Simulations are conducted to verify the effectiveness of control designs. The results show the performance of FSMC with GA in general is favorable to others, but the practicability of FNN is more simple and easy than FSMC with GA due to the on-line learning nature of FNN and off-line turning of control law for FSMC with GA.