根據小波網路之追蹤控制方法

Abstract

本論文旨在應用小波性質與發展小波網路的方法於追蹤控制。當重現的輸入或干擾作用於機器人、伺服機構及其他循跡的控制系統時，容易發生重現性誤差，先前文獻採用反覆控制或學習控制以消除系統的重現性誤差。本研究應用小波轉換，提出小波反覆控制器的方法，俾減少傳統反覆控制器所需的記憶容量。此外，本研究還發展小波網路控制器，消除重現性誤差。小波網路控制器採用梯度法線上學習網路權值，並以壓電彎片與碟機組成的系統，執行即時控制實驗驗證其效果，結果證實能有效減少重現性誤差。為了近場光碟機讀寫頭之飛行高度控制，最後並且對於壓電彎片、滑動片、以及碟機構成的系統，執行控制實驗，比較了比例積分控制器的結果，證實小波網路之追蹤控制方法，能使滑動片的飛行高度更穩定。

This study aims to exploit wavelet properties and develop a wavelet network approach to tracking control. Repetitive error commonly occurs in robotics, servomechanisms and other tracking scenarios, either in form of reference inputs or disturbances. Repetitive control and learning control have been adopted for such systems. In this study, wavelet repetitive controllers based on wavelet transform are proposed to reduce required memory size in conventional repetitive controllers. Further, wavelet network controllers are also developed for online learning using a gradient training rule and canceling repetitive errors. A PZT-disk system is implemented to demonstrate the present method. Finally, a PZT-disk-slider system is also used to validate wavelet network controllers in flying height control of pickup heads in near-field optical disk drives. Comparing with a proportional-integral controller, experimental results show that the proposed method enables the slider to more stably follow the disk surface deformation.