【作者】 李敏；

【导师】 朱煜；

【作者基本信息】 清华大学 ， 机械工程， 2017， 博士

【摘要】 超精密工件台是光刻机核心部件之一,其轨迹跟踪性能是保证光刻机产率和分辨率的关键,因此开展超精密工件台的运动控制方法研究具有重要的理论意义和工程应用价值。针对超精密工件台运动控制的三大技术难题:小建立时间与变轨迹运动需求难以兼顾、扫描段轨迹跟踪误差的移动平均偏差(Moving Average,MA)与移动标准偏差(Moving Standard Deviation,MSD)相互矛盾、动力学特性及外部扰动随位置变化致使性能鲁棒性难以实现,本文分别对前馈控制、变增益反馈控制及离散滑模控制的数据驱动参数整定方法展开了系统深入的研究。针对复杂且非最小相位的动力学特性导致前馈控制器难以兼顾小建立时间与变轨迹运动需求的难题,提出了一种数据驱动零相位误差跟踪前馈控制方法。综合模型信息构建了一种全新的前馈控制器参数化结构,结合辅助变量辨识法,以参考轨迹与跟踪误差的相关性作为参数整定的优化目标,给出了一种噪声环境下数据驱动参数无偏整定方法。通过岭估计法解决了工件台非最小相位特性引起的Hessian矩阵近奇异问题,实现了参数整定的快速稳定收敛。针对线性反馈控制的固有局限性,为了同时改善工件台的扫描段MA与MSD,提出了一种基于误差频率特性的变增益反馈控制方法,根据不同轨迹阶段跟踪误差的频率特性设计附加变增益环节。对于该非线性系统,采用Lyapunov稳定理论给出了低保守性的绝对输入-状态(鲁棒)稳定判据。针对变增益反馈控制器的参数优化难题,为了实现扫描段MA与MSD的最优化,提出了一种变增益反馈控制数据驱动加速迭代参数整定方法。创新性地提出以MA与MSD的加权2-范数作为目标函数,针对该带约束的非凸优化问题,采用列文伯格-马夸尔特法作为参数整定律以保证收敛稳定性,并给出了非线性系统下简便且精确估计目标函数梯度及Hessian矩阵的方法。为进一步加快收敛速度,提出了一种多参数加速迭代算法。最后,针对动力学特性及外部扰动随位置变化致使性能鲁棒性难以实现的难题,提出了一种数据驱动变增益离散滑模控制方法。通过揭示离散滑模控制本质上由相互独立的线性反馈控制、前馈控制及非线性切换控制组成的重要属性,提出了设计离散滑模控制的新方法,即依次独立设计上述三项控制项。利用本文前述研究结论与成果,综合变增益与数据驱动方法,同时消除了离散滑模控制对状态信息及精确模型信息的依赖性,并克服了鲁棒性与抖振抑制之间的矛盾。更多还原

【Abstract】 The ultra-precision motion stage is an important mechatronic unit of industrial lithographic scanners for manufacturing integrated circuits,and its excellent tracking performance is the key to ensure the throughput and resolution.Consequently,the motion control study for the ultra-precision motion stage possesses important theoretical significance and engineering application value.However,the motion control of the ultra-precision motion stage is confronted with the following three technical challenges: it is hard to take into account both small settling time and insensitivity to reference variations;moving average(MA)and moving standard deviation(MSD)of the tracking error during the exposure phase are contradictory to each other;and it is quite difficult to achieve the performance robustness due to the position-dependent dynamics and disturbances.To solve these problems,this dissertation focuses on the systematic and deep research on the data-driven parameter tuning methods of the feedforward control,variable-gain feedback control,and discrete sliding mode control(DSMC).For the ultra-precision motion stage,the complex and non-minimum phase dynamics lead to the difficulty in simultaneously achieving small settling time and insensitivity to reference variations.To realize small settling time regardless of reference variations,a data-driven zero phase error tracking feedforward control method is synthesized.A new parametric structure containing the model information is first proposed for the feedforward controller.On the basis of instrumental-variable identification method,the correlation between the reference and the tracking error is selected as the optimization criterion,and a data-driven parameter tuning method is developed to achieve the unbiased parameter optimization in noisy environment.Furthermore,the ridge estimate method is employed to guarantee a fast convergent iteration in the case of ill-conditioned Hessian matrix,which results from the non-minimum phase dynamics of the ultra-precision motion stage.To overcome the inherent limitations of linear feedback control and simultaneously improve MA and MSD during the exposure phase,a frequency-dependent variable-gain feedback control method is proposed.The add-on variable-gain elements are designed according to the error frequencies of different reference trajectory phases.For the corresponding nonlinear closed-loop system,the input-state(robust)stability criterion with less conservatism is established via Lyapunov stability theory.Aiming at the parameter optimization problem of the above nonlinear variable-gain feedback controller,a data-driven parameter tuning method with multi-parameter accelerated iterative algorithm is synthesized to achieve the simultaneous optimization of MA and MSD during the exposure phase.A weighted 2-norm regarding MA and MSD is significantly selected as the objective function,and the corresponding parameter optimization problem is constrained and nonconvex.Consequently,the Levenberg-Marquardt algorithm is employed as the parameter tuning law to ensure the convergence stability,and a simple method is proposed that provides the precise estimation of the gradient and Hessian matrix of the objective function in the nonlinear control system.Moreover,a multi-parameter accelerated iterative algorithm is proposed to further accelerate the convergence rate.For the ultra-precision motion stage,the position-dependent dynamics and disturbances make it difficult to achieve the performance robustness.To suppress the unmolded position-dependent dynamics and disturbances,a novel data-driven variable-gain DSMC is proposed.The essence that DSMC consists of linear feedback control term,feedforward control term and nonlinear switching control term,is revealed.As a result,a new design idea is put forward for DSMC where the above three control terms are separately designed in sequence.On the basis of the foregoing research,concepts of “variable-gain” and “data-driven” are newly introduced into DSMC.Therefore,the dependence on the full-state information and the accuracy model are simultaneously eliminated,and the trade-off between the performance robustness and the chattering alleviation is well balanced.更多还原