大行程精密定位平台之即時模型估測與控制切換

Abstract

本論文發展一套模型估測技術，結合多重控制器切換機制，在響應預測器中使用最接近系統之模型，來預測未來響應及切換控制器，並將其應用於大行程精密定位平台。 隨著科技快速發展，許多高科技產業與產品走向精密化、微小化的趨勢，例如：半導體產業製程、微機電系統製程、雙光子製程等等。壓電材料由於具有高精密度與響應快等優點，已被廣泛的應用在精密定位系統，然而受限於壓電材料的特性，平台行程受限，因此本論文整合精密壓電平台及大行程步進馬達平台，其中壓電平台採用壓電材料進行精密定位，步進馬達則可以增加平台整體行程。為了提升整體控制性能，我們在壓電平台與步進馬達平台控制架構中分別加入了模型估測器，結合控制器切換機制，最後將其整合為大行程且高精密之整合平台。 壓電平台的部分，模型估測器通過當下系統的輸入與輸出，即時從十組模型估測出系統，接著結合控制器切換機制，將其放入響應預測器中，預測系統未來響應，決定當下最佳控制器。我們進行了十次系統識別的實驗，得到了十組不同的系統模型，透過間隙值分析，找出其標稱系統，利用迴路成型的概念設計標稱系統的快、中、慢強韌控制器，藉由粒子群演算法將強韌控制器降階後，得到了標稱系統的快、中、慢閉迴路轉移函數。接著我們推導讓十組模型有相似響應的控制器，在模型估測器判斷出最接近之系統模型後，響應預測器會替換成該模型之控制器，透過使用較精準的系統模型，改善壓電平台響應。 馬達平台的部分，模型估測器通過當下系統的輸入與輸出，從十組馬達模型找出最接近系統的模型，運用此模型作為前饋控制器參數，使前饋控制器之參數更準確，並且加入增益調變控制，調配馬達的速度，改善馬達平台響應。 最後我們整合兩種平台，使用雙迴圈控制架構，透過整合平台輸入與輸出，決定壓電平台響應預測器之模型，修正整合平台的誤差，最後透過模擬以及實驗，展現加入模型估測器後，可以改善整合平台定位精度。

In this paper, a model estimation mechanism combined with multiple switching controllers is developed. By estimating the system model in real time and replacing it with the closest model in the response predictor, the current optimal controller is determined by the predictor to predict the future response, and it is applied to the long-stroke precision positioning stage. With the rapid development of science and technology, many high-tech industries and products are moving toward precision and miniaturization, such as: semiconductor industry process, MEMS process, two-photon process and so on. Piezoelectric materials have been widely used in precision positioning systems due to their advantages of high precision and fast response. However, due to the limitation of material travel, it cannot reach large-scale manufacturing. Therefore, this paper integrates precision piezoelectric stage and long-stroke stepper motor stage, in which piezoelectric stage adopts piezoelectric materials for precision positioning, and stepper motor can increase the overall travel of the stage. In order to improve the overall control performance, a model estimator is added to the control architecture of piezoelectric stage and stepper motor stage respectively. It is integrated into a long-stroke and high-precision combined stage. In the part of piezoelectric stage, the model estimator online estimates the system from ten sets of models through the input and output of the current system, and then integrates the controller switching mechanism to put it into the response predictor, predict the future response of the model, and determine the best controller. By analyzing the gap value, we find out the nominal system and design the fast, medium and slow robust controller of the nominal system. After reducing the order of the robust controller by particle swarm optimization algorithm, we get the fast, medium and slow closed loop transfer function of the nominal system. We then make ten sets of controllers whose models have similar responses, and after the model estimator determines the closest system model, the response predictor replaces the controller of that model, improving the piezoelectric stage response by using a more accurate system model. In the part of the motor stage, the model estimator finds the model closest to the system from ten groups of motor models through the input and output of the current system, and uses the parameters of this model as the feedforward controller to make the parameters of the feedforward controller more accurate, and adds the gain scheduling control to adjust the motor speed and improve the response of the motor stage. Finally, we combined the two stages, use the double-loop control architecture, through the input and output of the combined stage, determine the model of the piezoelectric stage response predictor, correct the error of the combined stage, and finally through simulation and experiment to show that the combined stage response is improved by adding the model estimator.