Research On Feedforward Control Techniques For High Precision Motion Control Systems Based On Iterative Learning

The ultra precision motion system is the key part of the lithography machine,as a result its control technology becomes one of the core technologies.In order to meet the key technical specifications of the lithography machine,such as critical dimension,overlay and productivity,the motion system must satisfy the high dynamic and ultra precision motion requirements,simultaneously.At present,the2-DOF(degree of freedom)control structure combining feedback and feedforward is usually adopted in ultra precision motion control.While the feedback control bandwidth is limited by the mechanical structural modes,the feedforward control can improve the response speed and motion accuracy of the motion control system by compensating the reference trajectory and external disturbances in time.However,for the controlled plant such as the motion system of the lithography machine which has a complex model and suffers from complex disturbances,the problem that traditional feedforward control technologies are too model dependent and cannot compensate the unknown disturbance,has become the main technical bottleneck of the realization of the high dynamic and ultra precision motion.The purpose of this thesis is to improve the feedforward compensation ability of the motion control system of the lithography machine,so as to cope with the actual working condition that the motion system of the lithography machine has a complex model and suffers from complex disturbances.For this reason,this thesis studies the feedforward control technology based on iterative learning,taking the ultra precision motion stages of the lithography machine as the research object.By proposing an effective method to improve the accuracy of the discrete frequency domain model of the finite-length system,the analysis,design and calculation accuracy of iterative learning control(ILC)method in discrete frequency domain are ensured.By proposing a data-driven robust optimization ILC method for uncertain motion control systems,the design conservativeness of the ILC method caused by the model uncertainty is reduced.By proposing the iterative learning based feedforward control method for non-rigorously repetitive motion control systems,the tolerant capacity to nonrepetitive interferences and the flexibility to nonrepetitive reference trajectories of the ILC method are improved.The main contents of this thesis are stated detailedly as follows:Firstly,aiming at the problem that ILC cannot be accurately analyzed,designed and calculated in discrete frequency domain due to the truncation error of the discrete frequency domain model of the finite-length system,this thesis proposes an effective method to improve the accuracy of the discrete frequency domain model of the finite-length system.To this end,the accuracy of the discrete frequency domain model of the finite-length system is analyzed qualitatively and quantitatively,and then a method to reduce the truncation error is proposed,which improves the accuracy of the discrete frequency domain model of the finite-length system.On this basis,the discrete frequency domain model,z-domain model and the lifted domain model of the iterative learning motion control(ILMC)system are compared and analyzed.In addition,with the help of the discrete frequency domain model,the performance of the ILMC system in discrete frequency domain is comprehensively analyzed,including the robust monotonic convergence,the steady-state performance,the convergence speed and the learnable bandwidth.The analysis results are applicable to all ILMC systems using the standard form ILC method.Secondly,aiming at the problem that the traditional robust ILC methods only focuse on the robustness of the ILMC system and ignores the steady-state performance and the convergence speed,this thesis studies the robust optimization ILC method under the condition of certain model uncertainties.Firstly,according to the discrete frequency domain performance analysis results of the ILMC system,this thesis presents the robust optimization design principles for the ILC method within the maximum learnable bandwidth.According to design results,this thesis introduces two typical ILC methods which can achieve robust optimal performance.In practice,model uncertainty is inevitable.The robust optimization ILC method can reduce the conservativeness of the ILMC system performance under the same model uncertainty as much as possible.Considering improving the accuracy of the model is the fundamental way to reduce the design conservativeness of the robust ILC method,this thesis directly uses the discrete frequency response as the system model and studies the measurement method of the discrete frequency response of finite-length system.The main goal is to reduce the impact of the disturbance,noise and truncation error on the model accuracy.Thirdly,aiming at the problem that non-repetitive interferences and non-repetitive trajectories degrade the performance of the ILMC system,the ILC method and the iterative feedforward tuning(IFFT)method based on the optimal estimation theory are studied in this thesis.An ILC method based on the kalman filtering theory is first proposed to improve the convergence performance of the ILC system subject to random interferences.However,stochastic interferences do not conform to the actual situation,thus the thesis proposes an ILC method based on set-membership estimation theory considering bounded non-repetitive interferences.With the help of the set-membership estimation theory,the thesis reanalyzes the kalman filtering ILC method,revealing its unknown convergence performance under the condition of bounded non-repetitive interferences.Considering the practical application,the above research results are extended to the uncertain system.In addition,in order to enhance the flexibility of the ILMC system to the varying reference trajectories,and improve the estimation accuracy of feedforward control parameters,especially reduce the variances of the parameter estimates,this thesis studies the IFFT method based on the kalman filtering theory.Fourthly,the main methods proposed are studied and analyzed experimentally on the motion stages of the lithography machine.Firstly,the effectiveness of the accuracy improvement method of the discrete frequency domain model is verified via the experimental studies on the discrete frequency response measurement method of the finite-length system and the discrete frequency domain robust inverse-model feedforward control method.Secondly,the longitudinal comparitive experiments of the robust inverse model ILC method,the ILC method based on the kalman filtering theory and the ILC method based on set-membership estimation theory are carried out,and the transverse comparitive experiments are carried out with the existing robust inverse-model ILC methods and the traditional feedforward control method.In addition,the kalman-filtering instrumental-variable IFFT method proposed in this thesis is compared with the existing instrumental-variable IFFT method and the ILC method.All comparative experimental results verify the effectiveness and superiorities of the proposed methods.The main experimental results show that in the case that the reticle stage tracks the reference trajectory with the displacement of 0.1m,the maximum speed of 2.4m/s and the maximum acceleration of 95m/s~2,compared to the traditional feedforward control method,the iterative learning based feedforward control method reduces MA from 3nm to 2nm,MSD from 5.5nm to 4.7nm(when settling time is set as 8ms),and reduces the actual settling time by 6ms(when the allowable error bound is set as MA=3nm).The performance indexes of MA and MSD are improved by not less than 33%and 14%,respectively.