基於單維度卷積神經網路之遞迴終端滑動模式控制應用於雙軸運動平台精密定位

Abstract

本論文目標為開發一種具有高強健性、高精度之智慧型遞迴終端滑動模式控制器，用於音圈馬達雙軸運動平台定位控制，論文中首先介紹音圈馬達雙軸運動平台架構和運作原理，並對運動平台進行系統鑑別，得出系統模型參數並建立其動態模型。接著，本論文先以傳統滑動模式控制設計一雙軸運動平台控制系統，再以非線性滑動面之遞迴終端滑動模式控制解決傳統滑動模式控制無法在有限時間收斂之問題。為了降低控制力的高頻震盪，本論文引入雙曲正切函數取代符號函數的不連續性，改善控制過程中的抖顫現象。由於切換控制力需要得知不確定項的上限才可確保系統收歛，實務上並不容易取得，故引入單維度卷積神經網路估測器對系統不確定項進行估測與補償，提出智慧型遞迴終端滑動模式控制器，消除系統參數變化和外部干擾等不確定性影響，提高系統的強健性，最終採用Lyapunov函數證明系統的穩定性並得到網路權重之更新律。本研究以數位訊號處理器實現上述控制法則，並設計出兩種控制軌跡和雜訊，組合成四種控制情境進行分析對比，實驗結果得知與傳統滑動模式控制相比遞迴終端滑動模式控制最高改善了28.5%；智慧型遞迴終端滑動模式控制最高改善了39.3%，最終實驗證實所設計之控制系統具備優異的控制精密度的同時依然保有較佳的強健性。This paper aims to develop an Intelligent Recursive Terminal Sliding Mode Controller (IRTSMC) characterized by high robustness and precision for the positioning control of a Voice Coil Motor (VCM) two-axis motion platform. The paper begins by introducing the architecture and outlining the operational principles of the VCM dual-axis motion platform, followed by system identification to extract system model parameters and establish its dynamic model. Subsequently, a control system for the two-axis motion platform is designed using traditional Sliding Mode Control (SMC) in this paper. To address the convergence issue within a finite time associated with traditional SMC, we introduce Recursive Terminal Sliding Mode Control (RTSMC) with a nonlinear sliding surface. To reduce high-frequency oscillations in the control force, a hyperbolic tangent function is employed to replace the discontinuity of the sign function, improving the control process's jitter phenomenon. Since the switching control force requires knowledge of the upper limit of uncertainties for system convergence, which may not be easily obtained in practice, an One-Dimensional Convolutional Neural Network (1D-CNN) estimator is introduced to estimate and compensate for system uncertainties. This leads to the development of the IRTSMC, aiming to eliminate the impact of uncertainties such as system parameter variations and external disturbances, thereby enhancing system robustness. The stability of the system is ultimately proven using Lyapunov functions, accompanied by the update law for network weights.The implementation of the proposed control strategies is realized using a digital signal processor in this study. Two control trajectories and corresponding noise profiles are designed, creating four control scenarios for thorough analysis and comparison. Experimental results reveal that compared to traditional SMC, RTSMC demonstrates an improvement of up to 28.5%, while IRTSMC shows an even greater improvement of up to 39.3%. The final experiment validates that the designed control system exhibits outstanding control precision while maintaining superior robustness.