压电陶瓷执行器的智能预测控制方法研究

CASIA OpenIR > 毕业生 > 博士学位论文

	压电陶瓷执行器的智能预测控制方法研究
	刘伟川
	2016-06-02
学位类型	工学博士
中文摘要	压电陶瓷执行器在精密加工、高精度定位等场合广泛应用。然而，压电陶瓷执行器本身具有的迟滞非线性特性严重影响执行器的定位精度。同时，迟滞特性又具有频率相关性，使得压电陶瓷执行器的输出受其输入电压信号频率的影响。针对这些问题，本文以模型预测控制技术为基础，提出了压电陶瓷执行器的神经网络建模法和模糊模型建模法、以及基于这些模型的智能预测控制方法。在上述方法的基础上，实现了“粘滞-滑动”压电微动平台的精密定位控制。具体来说，本文的创新之处包括以下四个方面： • 无迟滞逆补偿的双神经网络非线性预测控制：针对逆补偿控制依赖于迟滞逆模型精度的问题，提出了一种基于双神经网络串联模型的非线性预测控制方法。首先，利用神经网络良好的函数逼近特性分别对迟滞特性以及频率相关性进行建模，并构成描述压电陶瓷执行器动态特性的串联型压电陶瓷执行器模型。第二，在上述双神经网络串联模型基础上对压电陶瓷执行器的位移进行预测，并基于此设计了非线性神经网络预测控制器，实现了对压电陶瓷执行器的高精度定位控制。相比于文献中最为常用的基于迟滞逆补偿的控制方法，本文提出的方法无需求解迟滞特性的逆模型，从而有效地避免了迟滞逆模型精度对定位控制的影响。实验结果表明所提出的双神经网络模型比经典Duhem模型具有更高的建模精度。在控制精度方面，所提出的非线性神经网络预测控制器具有比PID控制器更好的控制精度。 • 无迟滞逆补偿的动态线性化神经网络预测控制：针对非线性预测控制需在线求解优化问题和物理约束较难处理等问题，提出了一种基于动态线性化神经网络的预测控制方法。首先，利用单个神经网络对频率相关的迟滞特性进行建模，并在每一个采样时刻对神经网络模型进行了线性化处理。在此线性模型的基础上，设计了解析的预测控制律，有利于实现压电陶瓷执行器的高频跟踪控制。此外，还考虑了压电陶瓷执行器的实际物理约束。实验结果表明，单个神经网络模型与双神经网络模型的建模精度基本一致，所提出的控制器在定位精度上具有更好的性能。同时，解析预测控制器与文献中已有的方法进行了对比，取得了更高的跟踪控制精度。 • 基于自适应模糊模型的模糊预测控制：针对解析预测控制器对离线模型 ii 压电陶瓷执行器的智能预测控制方法研究训练精度依赖性强的问题，提出了基于自适应T-S模糊模型的模糊预测控制器。每个模糊规则下的线性子模型通过离线辨识得到模型初值。同时，引入了模型估计器在线调整模型参数，从而避免初始离线T-S模型精度对控制精度的影响。之后在该自适应模型基础上设计了模糊预测控制器。实验结果表明，在初始模型精度不高的情况下，自适应模糊模型预测控制器仍能够有效地根据跟踪误差调整模型参数，并取得较高的控制精度。此外，所提出的方法在与文献中其他方法的对比实验表明：本方法能够取得更好的控制性能。 • \粘滞-滑动"压电微动平台的定位控制：针对迟滞特性对“粘滞-滑动”压电微动平台定位控制的影响，提出了考虑迟滞特性的平台定位控制方法。首先，研究了基于自抗扰控制的无模型“ 粘滞- 滑动”微动平台定位控制器，该控制器无需对“粘滞-滑动”微动平台本身进行建模，可直接实现对该平台的高精度定位控制。其次，研究了基于神经网络模型的“ 粘滞- 滑动”微动平台预测控制器，实现了对“ 粘滞-滑动”微动平台的闭环定位控制。实验结果表明上述所提出的两种方法均能实现对“粘滞-滑动”微动平台的精密定位控制。最后，针对粘滞时执行器与末端器之间的相对滑动问题，提出一种基于正-逆神经网络目标位移估计的定位控制方法，该方法利用正-逆神经网络对压电陶瓷执行器的目标位移进行估计，之后控制执行器运动到目标位移从而实现对末端器的定位控制。实验结果表明该方法能够有效地处理执行器与末端器之间的相对滑动问题。
英文摘要	Piezoelectric Actuators (PEAs) are widely applied for high-precision operations and positioning. However, the inherent hysteresis nonlinearity of PEAs seriously deteriorates the accuracy of the PEAs’ motion. Meanwhile, the hysteresis nonlinearity has the rate-dependent property, which means the output of PEAs is affected by the frequency of its input voltage. For these issues, this thesis proposes the neural-network modeling and T-S fuzzy modeling method for modeling of PEAs. And based on these models, this thesis proposes the neural-network based predictive control methods and T-S fuzzy predictive control methods to realized the tracking control of PEAs. Besides, based on the methods above, this thesis studies the modeling and control of Piezoelectric-Actuated Stick-Slip Micro-motion Device (PASSMD). Specifically, the innovations of this thesis can be summarized as follows: • The inversion-free dual-neural-networks model based nonlinear predictive control: The inversion-based method is highly depended on the precision of the inverse model of hysteresis. To deal with this problem, this thesis proposes the dual-neural-networks model based nonlinear predictive control method of PEAs. First, two neural-networks are used to approximate the hysteresis nonlinearity and the rate-dependent property, respectively. Then the two neural-networks are constructed as the cascaded model of PEAs. Second, based on the cascaded model, the nonlinear model predictive control method is studied to realize the high-precision control of PEAs. For the proposed method, the inversion of the hysteresis nonlinearity is no longer needed. Therefore, the proposed method can avoid the affection of the inversion to the control accuracy. The experiments results suggest that the proposed neural-network model outperforms the Duhem-based model of PEAs. For the control performance, the proposed nonlinear predictive controller has better positioning accuracy than the PID controller. • The inversion-free dynamical linearized neural-network based iv 压电陶瓷执行器的智能预测控制方法研究 explicit predictive control: For the nonlinear predictive control, it needs to solve an on-line optimization problem and this is a computational burden of real-time control of PEAs. Meanwhile, the nonlinear predictive control is hardly to handle the physical constraints of PEAs. To deal with these problems, This thesis proposes the single neural-network modeling method of PEAs. And this method implements the dynamical linearization of the neural-network model in each sampling interval. Based on this dynamical linearized model, the predictive controller with explicit form is studied, which can address the high-frequency tracking control of PEAs. Furthermore, the constrained predictive controller is also proposed to deal with the physical constraints of PEAs. The experiments results suggest that the dynamical linearized model has highly modeling accuracy, which is same as the dual-neural-networks model of PEAs. Notably, the proposed controller has better control performance than the nonlinear predictive controller. Meanwhile, the comparison experiment results suggest that the proposed controller can achieve higher control precision than the existing methods in the literature. • The adaptive T-S fuzzy model based predictive control: For the explicit predictive controller, the dynamical linearized model needs to be obtained in an on-line way. And the control accuracy is highly depended on the offline obtained neural-network model. To deal with these problems, This thesis proposes an adaptive T-S fuzzy model based predictive controller. The linear submodel of each fuzzy rule can be directly identified in an off-line way. Meanwhile, the model estimator is introduced to on-line adjust the sub-models. Therefore, the accuracy of the off-line obtained T-S fuzzy model can be avoid to affect the control performance. Then the fuzzy predictive controller is designed to achieve the high precision control of PEAs. The proposed method can adjust the model to realize good control performance when the initial off-line obtained model does not have high modeling precision. If the initial model has relatively good accuracy, the proposed method can further improve the control performance. Besides, the proposed method has better control performance than the existing methods in the literature. • The modeling and control methods of Piezoelectric-Actuated ABSTRACT v Stick-Slip Micro-motion Device: To deal with the affection of hysteresis to the control of Piezoelectric-Actuated Stick-Slip Micro-motion Device, this thesis studies the positioning control methods, which are considered the hysteresis property. First, the Active Disturbance Rejection Control (ADRC) based model-free controller is proposed, and it can be directly designed without considering the model of PASSMD. Second, the neural-network based predictive controller for PASSMD is studied. This controller can realize the closed-loop positioning control of PASSMD. The proposed controllers are both verified on the prototype of PASSMD. The experiment results suggest that the proposed controllers are both effective to implement the positioning control of PASSMD. Finally, to deal with the relative slip problem between the PEA and the end-effector, this thesis proposes a positioning control method based on the positive-contrary neural-network model. The dynamics between the displacement of PEA and the motion of endeffector is modeled by the positive neural network model first. Then the reference of the PEA is on-line estimated by the contrary neural network. Therefore, the positioning control of end-effector can be realized by actuating the PEA to the reference position. The experiment results suggest that the proposed method can deal with the relative slip problem effectively.
关键词	压电陶瓷执行器微动平台迟滞非线性神经网络模糊系统预测控制粘滞-滑动
文献类型	学位论文
条目标识符	http://ir.ia.ac.cn/handle/173211/11667
专题	毕业生_博士学位论文
推荐引用方式 GB/T 7714	刘伟川. 压电陶瓷执行器的智能预测控制方法研究[D]. 北京. 中国科学院大学,2016.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
Thesisfinal.pdf（17578KB）	学位论文		限制开放	CC BY-NC-SA