- 學位論文
直流無刷馬達之線上模型參考模糊控制器設計
On-line Model Reference Fuzzy Controller Design for a DC Brushless Motor
摘要
本論文之主要目的在於設計模型參考模糊控制器（簡稱MRFC）進行直流無刷馬達之線上控制。 此線上MRFC的設計概念取材自模型參考學習型模糊控制器（簡稱FMRLC）的架構。目前常見的FMRLC含有參考模型、模糊控制器（簡稱FC）及模糊反模型（簡稱FIM）三個部份；其中FIM的輸出用來修改FC的規則庫；FC的規則庫與FIM相同，僅正、負號相反；且FC的輸入通常選用閉路系統之響應誤差及其變化率。本文所提之線上MRFC以FIM之輸出取代誤差變化率當作FC的輸入，且以此訊號來線上調整控制量以取代複雜的規則庫調整機構。 一般而言，FC之規則庫、歸屬函數（membership function）和尺規因素（scaling factors），對閉路系統的響應、收斂性以及穩定性都有決定性的影響。雖然這些控制器參數均可藉由離線搜尋技巧，例如適應性基因演算法（簡稱AGA），取得最佳值，由於模擬時所用之馬達系統模型誤差等因素造成實作響應不佳，因此僅能當作控制器之初值，再加以手調或以線上型控制器自動調整之。本文之線上MRFC採用正三角形歸屬函數，僅搜尋尺規因素，並使其逼近最佳值的附近即可，因而可節省搜尋時間且顯示線上調整之有效性。 經由模擬證實，在離線MRFC架構中，即使因為尺規因素的變動或規則庫未適當設計，造成系統響應不佳，而以線上MRFC來加以修正後，確實能得到令人滿意的系統響應。
並列摘要
The main purpose of this thesis is to study the design of a model reference fuzzy controller (MRFC) for on-line control of a DC brushless motor. The idea of designing the on-line MRFC comes from the structure of a fuzzy model reference learning controller (FMRLC). In general, a FMRLC contains three parts: a reference model, a fuzzy controller (FC) and a fuzzy inverse model (FIM). The output of the FIM is used to adjust the rule base of the FC. The only difference between the FIM and the FC is the reversed sign of the rule base. And the inputs of the FC are the error and error rate of the closed-loop system. The on-line MRFC proposed in this thesis uses the output of the FIM instead of error rate for one of the inputs of the FC. The on-line control force is adjusted by the same signal instead of the complex rule base tuning mechanism. In general, the rule base, the membership functions and the scaling factors of FC have decisive effects for response, convergence and stability of the closed-loop system. Although the optimal controller can be obtained by the off-line searching technique (e.g. adaptive genetic algorithm (AGA)); however, the simulation result causes poor response in the experimental system. Hence the optimal values from simulation serve only as the initial design of the controller, and the advanced adjustment should be made by hands or by the on-line controller. In this thesis, the on-line MRFC uses the triangular membership functions, and only rough searching for the scaling factors is needed. Therefore, the searching time is greatly reduced and the proposed controller is shown to be effective. The simulation results show that poor responses for the off-line MRFC caused by unsuitable designing of the scaling factors or the rule base can be effectively improved and become satisfactory by the proposed on-line MRFC.