氣壓人造肌肉致動器應用於單自由度機械臂設計及控制之研究

Abstract

氣壓人造肌肉致動器(以下簡稱氣壓肌肉)擁有傳統氣壓致動器之優點，藉由控制氣壓肌肉內部壓力可改變其長度及出力為其特色。其材料及結構特性使其能夠表現多數致動器所沒有的順應性，因而相當適用於外骨骼及復健設備等與人體有直接接觸之器材。然而，氣壓系統由於氣體的可壓縮性以及其他非線性特性，使其較難以準確的數學模型描述進而進行控制。近年來，許多研究提出各式不同形式的經驗數學模型企圖改進對氣壓肌肉的行為描述。 本論文旨在研究氣壓人造肌肉致動器於氣壓伺服控制系統上之應用，其中包含兩個不同設計之氣壓人造肌肉伺服系統。分別為水平單軸雙氣壓肌肉系統，以及單自由度氣壓肌肉對致動機械臂。前者使用非線性自調式類神經PID控制器。後者使用之控制器採用基於模型之串聯結構控制器分別進行力矩及角度之控制，並且將作為本實驗室往後相關研究之基礎。 本文首先對機械臂之機構部分以及氣壓系統部分進行建模，氣壓肌肉部分採用M. Eichhorn，C. Ament，及 T. T. Nguyen提出之經驗模型，為進行模型之參數鑑別，本論文建構一測試平台以量測氣壓肌肉的壓力、收縮率、以及出力之關係，並且以最佳化方法估算模型之最佳參數。由於該模型之特性，使其易於根據所需之出力及當下之人造肌肉收縮率計算出所需壓力，而成為一扭矩前饋控制器。為補償模型之誤差，本論文亦加入PID回饋控制器以增進扭矩的控制效果。以該力矩控制器為基礎，本論文將之延伸應用於阻抗控制，以達到重力補償之效果，使該系統對於與之互動者而言等效為純慣性系統。 更進一步，本論文使用反饋線性化方法與LQR控制器，並結合扭矩控制，以控制機械臂之角度。由於該系統採用之角度編碼器的精度不足以進行數值微分，使得角速度訊號無法直接量測，為克服該問題，本論文設計LTR觀測器以估算系統之狀態，並成功達成角度之追蹤控制。本實驗室以人體輔具應用為未來之研究方向，在此應用上，系統的作動順暢與否，相較於系統的控制精確度而言更為重要，因此本論文對於機械臂之控制目標為達成順暢的追蹤結果，同時將誤差保持於可接受之範圍內。

Pneumatic artificial muscles (PAM) inherit all the advantages provide by the conventional pneumatic actuators. Due to the structure and the material of PAMs, they possess a special property of compliance. Thus, they are suitable for the devices that have directly interaction with human such as exoskeletons and rehabilitation equipment. However, the compressibility of air and other nonlinearities result in the difficulty to describe pneumatic system accurately with mathematical model and control them. This thesis focuses on the application of PAMs on the pneumatic servo control systems, including two different designs of PAM systems. One is a single-axial dual-PAM system, and the other is a 1-DOF forearm robotic system driven by a PAM-pair. The single-axial dual-PAM system is controlled by a modified model-free self-tuning PID controller based on neural network, while the 1-DOF forearm robotic system adopts a model-based cascaded controller to control the torque and the rotational angle of the system, which also will become the basis of the related future research in our laboratory. This thesis firstly models the mechanism and the pneumatic components of the 1-DOF robotic system. The modeling approach adopted is proposed by M. Eichhorn , C. Ament, and T. T. Nguyen. To identify the parameter values of the model, a test bed is constructed to measure the relationship among the pressure, the contraction ratio, and the force of the PAMs, and an optimization approach is used to estimate the values of the parameters. Due to the structure of the model, it is easy to calculate the pressure needed to produce a certain desired force under different contraction ratio. With the help of the model, we design a feedforward torque controller. Also, a feedback PID controller is added to compensate the modeling error. Based on the torque controller, impedance control is achieved to compensate the gravity, making the system equivalent to a pure inertia system. Moreover, the combination of feedback linearization and LQR control is integrated with the torque control to control the rotational angle of the forearm robotic system. An LTR observer is designed to estimate the angular velocity and overcome the insufficient resolution of the encoder used in the system. With the overall control system, angle tracking control is successfully achieved. The future research can be made for rehabilitation devices. Under this condition, smoothness is more important than accurate control performance. Thus, the objective to control the forearm robotic system is to track the desired path smoothly while maintaining the error in an acceptable range.