Stewart主动隔振平台的神经网络控制
文献类型:学位论文
作者 | 马嘉 |
学位类别 | 工学博士 |
答辩日期 | 2009-05-31 |
授予单位 | 中国科学院研究生院 |
授予地点 | 中国科学院自动化研究所 |
导师 | 谭民 |
关键词 | Stewart并联机构 主动隔振 近似动态规划 神经网络控制 扰动观测器 Stewart parallel mechanism active vibration isolation approximate dynamic programming neural network control disturbance observer |
其他题名 | Neural Network Control of a Stewart Mechanism-Based Active Vibration Isolation Platform |
学位专业 | 控制理论与控制工程 |
中文摘要 | 随着科学技术的发展,精密仪器仪表、精密加工设备、光学有效载荷等振敏设备对力学环境提出了更高的要求。为了减小多维低频振动造成的负面影响,具有六自由度运动能力的Stewart主动隔振平台近年来引起了广泛的关注。Stewart主动隔振平台是一个存在不确定性、强耦合性、高度非线性的多输入多输出系统,其控制问题非常复杂。本文从对单杆主动隔振平台的研究入手,进而推广到对Stewart主动隔振平台的研究,采用神经网络控制作为主要控制方法,分别针对单一方向和六个自由度方向的低频振动主动控制问题展开了研究。 首先,阐述了Stewart主动隔振平台的研究意义,对它的研究现状进行了综述,并对主要的振动主动控制方法进行了总结。 第二,将单杆主动隔振平台的振动主动控制问题转化为一个优化控制问题来处理。针对该隔振平台的线性模型,提出了一种单网络自适应评价振动主动控制器,并设计了离线和在线两种训练算法。进一步考虑该隔振平台中的非线性因素,针对其非线性模型,提出了一种双启发式动态规划振动主动控制器。仿真研究验证了这两种振动主动控制器对单一方向振动抑制的有效性。此外,采用音圈电机作为作动器,搭建了单杆主动隔振平台的原型样机并进行了系统性能测试实验。 第三,考虑振动对下平台运动的影响,利用Kane方程建立了Stewart主动隔振平台在工作空间中的动力学模型。在动力学模型的推导过程中,考虑了各支杆运动产生的附加惯性力和惯性力矩对上平台动态特性的影响,进而推导得到针对隔振应用的较完整的系统动力学模型。 第四,针对Stewart主动隔振平台的振动控制问题,提出了一种基于径向基函数(RBF)神经网络的稳定自适应控制器。基于e1-修正算法设计了RBF神经网络的权值矩阵、高斯基函数中心和宽度的在线自适应调节律。并且应用Lyapunov稳定性理论证明了在扰动力和神经网络逼近误差有界的条件下,整个闭环控制系统为一致最终有界稳定。基于Stewart主动隔振平台的动力学模型进行了仿真研究,结果表明该控制器能够有效抑制六个自由度方向的低频振动。 第五,基于考虑音圈电机动态特性和模型不确定性的Stewart主动隔振平台的动力学模型,提出了一种基于RBF神经网络的扰动观测器,并进一步结合反推(Backstepping)设计方法,提出了一种六自由度振动主动控制器。在模型存在不确定性的情况下,整个闭环控制系统的一致最终有界稳定性可通过Lyapunov理论得到保证。并通过仿真验证了该振动主动控制器的有效性。 最后,对本文所取得的研究成果进行了总结,并指出了需要进一步开展的工作。 |
英文摘要 | With the development of science and technology, vibration-sensitive equipments such as precision instruments, precision manufacturing devices, optical payloads, etc., have put forward higher requirements for mechanical environment. In order to alleviate negative impacts caused by low frequency vibrations in multi-dimensions, a Stewart mechanism-based active vibration isolation platform(SMAVIP), which has a six-degree-of-freedom movement capability, has attracted extensive attention. A SMAVIP is a strong coupling, high nonlinear, multi-input multi-output system with uncertainties. Its control problem is very complex. This paper begins with the research on a single-strut active vibration isolation platform(SSAVIP), then extends to the research on a SMAVIP. Mainly using neural network control methods, active control problems of low frequency vibrations in a single direction and in directions of six degrees of freedom are studied respectively in this paper. Firstly, the research significance of a SMAVIP is introduced. And its previous research work is overviewed. Moreover, main active vibration control methods are summarized. Secondly, the active vibration control problem of the SSAVIP is changed into solving an optimal control problem. For the linear model of the SSAVIP, a single network adaptive critic(SNAC) based active vibration controller is proposed. And two training algorithms, i.e., the off-line and online algorithms are designed to adapt the controller respectively. Furthermore considering the nonlinear characteristics of the SSAVIP, a dual heuristic programming(DHP) based active vibration controller is proposed for its nonlinear model. Simulation results have demonstrated that the two developed controllers can attenuate low frequency vibrations in a single direction effectively. Moreover, a prototype of the SSAVIP is constructed and some system performance tests are carried out. Thirdly, considering vibration effects on the lower platform, a dynamic model of the SMAVIP in the work space is developed based on Kane's equation. During the derivation of the dynamic model, the additional inertial forces and inertial torques caused by the movement of each strut are taken into account and the almost complete dynamic model is obtained for vibration isolation applications. Fourthly, focusing on the active vibration control problem of the SMAVIP, a stable adaptive controller based on a radial basis function neural network(RBFNN) is proposed. An online adap... |
语种 | 中文 |
其他标识符 | 200518014628014 |
源URL | [http://ir.ia.ac.cn/handle/173211/6201] ![]() |
专题 | 毕业生_博士学位论文 |
推荐引用方式 GB/T 7714 | 马嘉. Stewart主动隔振平台的神经网络控制[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2009. |
入库方式: OAI收割
来源:自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。