Title: 以增強性學習為基礎的模糊邏輯用於近場光碟機的控制
Fuzzy Control for Near-Field Optical Disk Drives Based on Reinforcement Learning
Authors: 張萬坤
Wan-Kun Chang
Tzong-Shi Liu
Keywords: 模糊控制;增強性學習;近場光碟機;Fuzzy Control;Reinforcement Learning;Near-Field
Issue Date: 2000
Abstract: 近場光碟機應用近場光學理論之飛行讀寫頭設計,突破了以往光碟機技術瓶頸,可以大幅度提高光碟片儲存容量和密度,然而光碟機讀寫性能要求相對提高。為了達成讀寫頭的控制,本研究結合模糊控制特性與增強式學習的特點,發展以增強式學習為基礎之模糊控制。模糊控制對於數學模型不確定性之系統、非線性現象與實現上都有優異的表現。增強式學習法是一種間接教導式學習法,適用於粗略的回授訊號學習,然而光學讀取頭於聚焦、尋軌、循軌時,通常所能量測到訊號型態正是如此。設計此模糊控制器不需事先推導讀取頭之數學動態模型,而是根據讀取頭的實際輸入輸出資料建立模糊模型,然後以增強式學習方法調整控制器的會員函數。本研究以雷射都卜勒振動儀量測壓電材料位移特性,設計適用於近場光碟機之兩階段光學讀取頭,並以系統模擬來驗証所提出的控制方法有效可行。
The flying head design in near-field optical disk drives applies the near-field optics theory to overcome the limit of conventional optical disk drive techniques, thereby substantially increasing data storage capacity. To that end, enhancing control performance to improve tracking speed and accuracy is required. This study aims to develop fuzzy control based reinforcement learning, which incorporates characteristics of reinforcement learning control into fuzzy control. Fuzzy control has excellent characteristics of dealing with model uncertainty, nonlinearity and easy implementation. Reinforcement learning is specially suitable for rough feedback signals; however, the measured feedback signals for an optical head is also rough in focusing, seeking and following. To carry out the present controller, it is not necessary to obtain beforehand the mathematical model of the pickup head. This work constructs fuzzy rules based model based on input-output data of the pickup head and tune fuzzy membership functions by reinforcement learning. During voltage excitation, a bimorph PZT simultaneously undergoes two axes deformation, which can be used for both tracking control and flying height control. To validate the proposed bimorph PZT method, this work conducts measurements.
Appears in Collections:Thesis