摘 要
本研究應用於非線性磁浮系統，針對平衡點建立線性化的數學模型，且分析系統操作點的穩定性，並根據系統動態設計控制器。為了使鐵球穩定懸浮於每個氣隙高度，本文運用徑向基函數類神經網路(radial basis function neural network，RBFNN)結合比例積分微分(PID)控制器，建構出一種線上學習且非線性增益的智慧型控制器，將受控體與徑向基函數網路之輸出所產生的誤差透過類神經網路的自我學習能力調整出最佳化的控制參數。對於磁浮系統的控制加入離散化的低通濾波器，濾除感測器雜訊以改善系統的動態響應，形成具有數位濾波且非線性增益的回授控制架構。除了懸浮於每個高度的定位控制，本文嘗試以1Hz與10Hz正弦波的位置命令進行追蹤控制，控制效果皆能達到理想的系統響應，實驗結果顯示本文所提出的控制器優於傳統PID控制器，大幅提昇系統的性能。
關鍵詞：磁浮系統、比例積分微分(PID)控制器、RBF類神經網路、數位低通濾波器
Abstract
In this study, a nonlinear gain controller, based on radial basis function neural network (RBFNN), is presented to manipulate a linearized magnetic levitation system (MLS). The controller was developed in accordance with the system dynamic and equilibrium point. To control the stable levitation of the steel ball in the air gap, the proposed controller exploited RBFNN structure with online learning to identify PID gains. Through the RBFNN self-learning ability, the controller had optimal tuning parameters. Due to linearized system and uncertainties of disturbance by the position feedback sensor device, the digital low pass filter was utilized in the system and nonlinear gain feedback control was designed. To validate the proposed MLS controller, many experiments were performed to evaluate system performance and compared with the conventional PID controller. Experimental results demonstrate the proposed controller, positioning and tracking different frequencies of sine-waves with amplitudes of 1 and 10 Hz, delivers high performances.
Keywords: Magnetic Levitation System, PID Controller, RBFNN, Digital Low Pass Filter