雙軸機械手臂之適應性神經網路滑動模式控制

Abstract

本研究之目的是針對機械手臂之循軌控制提出適應性神經網路滑動模式控制方法。於系統模型部份已知的情況下，運用極點配置法來設計標稱控制器指定機械手臂之理想動態，並透過滑動模式干擾估測器及適應性神經網路補償器將系統的不確定性及外部干擾予以補償，以實現指定的理想動態。 系統控制架構中之滑動模式干擾估測器用於提昇整體控制架構之初始性能，並對於未知的干擾給予快速有效的補償，以提升系統的強健性能。相對於適應性神經網路控制器透過自訂的適應性法則，將未知干擾建模於神經網路規則庫；當建模完成便可依據系統之狀態，查得目前的系統干擾，以達到即時的干擾補償，可進一步改善滑動模式干擾估測器補償的相位落後問題。 本文實驗平台方面，採用美國德州儀器公司(Texas Instruments Incorporated, TI)所生產之TMS320C6713 DSP搭配具FPGA之自製擴充子板為控制器核心。在FPGA方面，以硬體描述語言(VHDL)撰寫Encoder, ADC與DAC等週邊界面程式；在控制法則實現上，利用TI所提供的Code Composer Studio (CCS)發展環境，以C/C++撰寫控制器程式並下載到DSP上執行。藉由本實驗室自製的雙軸機器手臂實驗平台進行追圓軌跡控制，結果顯示能有效提升循軌的表現及降低循軌誤差。

A scheme of adaptive neural network sliding-mode control is proposed in this paper to deal with highly nonlinear dynamics of robotic manipulators for trajectory tracking. By using a simplified model, a nominal controller is obtained by pole placement design to specify ideal closed-loop dynamics. Then, an adaptive neural network compensator augmented with a sliding-mode disturbance observer (SDOB) compensator for system uncertainties and external disturbances. The SDOB ensures well transient performance and compensates well for unknown perturbation. In addition, the adaptive neural network compensator is used to model an unknown perturbation according to the proposed adaptive law. When the perturbation has been well modeled, the control system can efficiently compensate for the perturbation, avoiding the phase-lag problem associated with the SDOB. The experimental system consists of a two-link robotic manipulator and a DSP/FPGA system, that is the control kernel. We employ the C language and VHSIC hardware description language (VHDL) as tools for developing a servo control system. The experimental results of tracking a circular trajectory show that the proposed scheme improves the tracking performance and decreases the tracking error.