模糊可微分小腦模型晶片系統之設計

Abstract

本研究將設計一個整合模糊控制器與可微分小腦模型控制器的晶片。 模糊邏輯控制器為仿效人類模糊判斷，採用模糊知識庫來描述受控系統的 控制邏輯，比一般傳統控制方法擁有更好的強健性與適應性，但其模糊知 識庫需採嘗試錯誤法來建立，且有穩態誤差，並無法保證達到精確的控 制，經由可微分小腦模型控制器的加入，可以改善模糊邏輯控制器的缺 點，縮短以嘗試錯誤法建立模糊知識庫的時間，進而提昇控制系統的效 能，降低系統的追蹤誤差，並且有效地提昇控制精確度。本研究係以FPGA 來設計本晶片，內含FLC 與DCMAC。本研究在FLC 方面以Mamdani 模 糊推論法做為推論機制，高度法為解模糊策略，DCMAC 方面，以高斯查 表法、非雜湊式映射位址運算，採FLC 與DCMAC 平行運算法的方式實 現。最後將本研究應用於超音波線性壓電陶瓷馬達的定位控制上，證實本 晶片系統，並驗證其具良好的控制效能The object of this thesis is building the FDCMAC chip system with integrating Fuzzy Logical Controller (FLC) with Differentiable Cerebellar Model Articulation Controller (DCMAC). FLC is imitating fuzzy judgment of human. It uses the fuzzy knowledge base to descript controlling logic of controlled system. Compared with general controller, FLC is more robust and suitable. But FLC has some steady state error, and the time of building fuzzy knowledge base is very long by try and error. It couldn’t be accurate in control. With integrating DCMAC and FLC into FDCMAC, it can improve the disadvantage of FLC, shortening the time of building fuzzy knowledge base, enhancing the performance of FLC, reducing error of tracing system, and making accuracy rising. In this study, it uses the FPGA to implement FDCMAC chip system. FLC, DCMAC and the main control is designed on the FPGA. In designing FLC, we adopt the Mamdani method to be the method of fuzzy inference. In designing DCMAC, we adopt lookup-table of Gauss function. And we design parallel FLC and DCMAC computing capability on FPGA by Verilog HDL. Finally, the FDCMAC chip system will experiment on controlling linear piezoelectric ceramic motor (LPCM) to prove that it has good performance of control.