Inverse Modeling of Nonlinear Artificial Muscle Using Polynomial Parameterization and Particle Swarm Optimization
Inverse Modeling of Nonlinear Artificial Muscle Using Polynomial Parameterization and Particle Swarm Optimization
复制标题
DOI：
10.1155/2020/8189157
复制
发表时间：
2020-12
影响因子：
--
通讯作者：
M. A. Mat Dzahir;Shin-ichiroh Yamamoto
中科院分区：
文献类型：
--
作者：
M. A. Mat Dzahir;Shin-ichiroh Yamamoto
气动人工肌肉（PAM）具有良好的功率重量比和自然顺应性，使其在医疗工程中具有广泛的应用前景。然而，它在控制机器人操作器时具有不期望的滞后效应。这种行为是准静态的，并随激励频率和外力的变化而变化。除此之外，它还继承了具有非局部记忆效应的摩擦预滑动行为。需要补偿这些非线性以实现控制系统的最佳性能。尽管PAM的逆建模具有有限的应用，但是对于需要逆问题的解决方案的某些控制系统实现是重要的。本文提出了以活化压力形式对PAM进行反演的方法。该启动压力模型是根据静态压力导出的，并从压力/长度滞后中提取滞后分量。静压模型的推导遵循基于唯象的三阶多项式模型。它能够表征PAM在低压和高压下的非线性区域。提取的迟滞模型的推导遵循了动摩擦的机理。在此基础上，提出了由静压模型的回归系数和提取的迟滞模型的动摩擦系数来确定启动压力模型的方法。这些系数的回归常数，其特征在于从滞后数据集，通过使用模型参数识别和粒子群优化（PSO）方法。模型仿真结果表明，在不同的激励频率和外力下，均方根误差（RMSE）值小于10%的误差评估。这种PAM的逆建模实现了一种简单的方法，但它应该是有用的控制设计应用，如康复机器人，生物医学系统和人形机器人。
The properties of pneumatic artificial muscle (PAM) with excellent power-to-weight ratio and natural compliance made it useful for healthcare engineering applications. However, it has undesirable hysteresis effect in controlling a robotic manipulator. This behavior is quasistatic and quasirate dependent which changed with excitation frequency and external force. Apart from this, it also inherits frictional presliding behavior with nonlocal memory effect. These nonlinearities need to be compensated to achieve optimal performance of the control system. Even though an inverse modeling of PAM has limited application, it is important on certain control system implementation that requires the solution to the inverse problem. In this paper, the inverse modeling of PAM in the form of activation pressure was proposed. This activation pressure model was derived according to static pressure and extracted hysteresis components from pressure/length hysteresis. The derivation of the static pressure model follows the phenomenological-based model of third-order polynomial. It is capable of characterizing the nonlinear region of PAM at low and high pressure. The derivation of extracted hysteresis model follows the mechanism of dynamic friction. In this principle, the activation pressure model was dependent on regression coefficient of the static pressure model and dynamic friction coefficients of the extracted hysteresis model. The regression constants of these coefficients were characterized from the hysteresis dataset by using model parameter identification and the particle swarm optimization (PSO) method. The result from model simulation shows the root mean square error (RMSE) value of less than 10% error was evaluated at various excitation frequencies and external forces. This inverse modeling of PAM implemented a simple approach, but it should be useful in control design applications such as rehabilitation robotics, biomedical system, and humanoid robots.