以數位訊號處理器實現之智慧型音圈馬達定位控制系統

Abstract

本論文目標為針對非線性時變之音圈馬達，設計一具強健性與高精度之控制系統，本論文首先提出基於比例積分微分型細菌覓食模糊類神經網路控制系統，由於傳統的類神經網路控制系統，網路參數初值設計會導致控制系統陷入區域最佳解，所以本篇論文以最佳化演算法改良型細菌覓食演算法在馬達運動前先進行歸屬函數最佳化，避免系統陷入區域最佳解。 為了簡化控制系統計算複雜度，進一步提出具動態參數估測能力之補償型模糊類神經網路，此控制系統利用動態粒子群演算法於控制過程中即時最佳化Jacobia項，可有效提高系統控制指標性能。在此架構中，主控制器為補償型模糊類神經網路，另使用Elman類神經網路即時估測音圈馬達動子位置。 為提高系統之強健性，本論文提出智慧型分數階滑動模式系統，此系統以補償型類神經網路對不確定項估測，可解決傳統分數階滑動模式控制中切換控制之抖動現象，另外亦設計一平滑補償器，可補償估測誤差與確保系統之漸進穩定。 本論文以數位訊號處理器實現上述控制法則，並設計兩種追隨軌跡與兩種測試情況。實驗結果顯示所提出之控制系統確實能有效控制音圈馬達之動子位置。This paper aimed to design robust and precise control systems for the position control of a non-linear and time-varying voice coil motor (VCM). First, a proportional-integral-derivative based fuzzy neural network with elitist bacterial foraging optimization (PIDFNN-EBFO) control strategy was proposed in which the initial parameters of the network were optimized to avoid falling into local optimal solutions. Subsequently, a compensatory fuzzy neural network with dynamic particle swarm optimization (CFNN-DPSO) control approach was further developed to simplify the computational burden of the PIDFNN-EBFO control system. In CFNN-DPSO, the CFNN and DPSO were used to derive the control effort and estimate the Jacobian term, respectively. Besides, an Elman neural network (ENN) was used to serve as a virtual VCM system for the online estimation of the mover position. In order to improve the robustness of the VCM control system, an intelligent fractional order sliding mode control (IFOSMC) system was proposed in this study. In the IFOSMC, a CFNN observer was designed to observe a lumped uncertainty and replace the hitting control directly. Besides, a switching compensator was employed to compensate the observation error considering the system stability and control smoothness. All the real-time control systems were implemented via the digital signal processor (DSP). Moreover, two reference trajectories and two test conditions were provided to evaluate the control performances in this study. From the experimental results, the VCM can be controlled to track the reference trajectories efficiently and accurately via the proposed control systems.