Learning Control Algorithms: Convergence Performance Improvement and Application
|School||Zhejiang University of Technology|
|Course||Control Theory and Control Engineering|
|Keywords||iterative learning control initial condition problem repetitive control sliding-mode control linear motors|
The convergence performance of learning control algorithms is of great importance. The initial states error degrades the convergence performance of the learning control systems. It’s worthwhile to discuss how to relax its initial condition without great loss in system perfomace. When unrepeatable uncertainties exist, learning variable structure control strategy would be a good choice for the assurement of the system’s convergence. More attention should be paid to the chattering phenomenon in variable structure control system. For discrete-time repetitive sliding-mode systems, the characteristics of the sliding-mode dynamic deserve dedicated analysis. Taking account of the issues mentioned above, this thesis focuses on the following aspects:Iterative learning control with finite time deadzone modification is presented for a class of high-order nonlinear time-varying systems in the presence of initial condition errors. With the introduction of finite-time dead-zone, the developed controllers are able to confine the specified error function to zero over a pre-specified time-interval. A Gramian-based initial rectifying action is used so as to realize the the complete tracking over another pre-specified time-interval. Parametric time-varying uncertainties, norm-bounded uncertainties, as well as non-parametric uncertainties are under consideration separately, and iterative learning controller, robust learning controller and neural network iterative learning controller are designed to deal with corresponding uncertainties. It is shown by the theoretical analysis and numerical simulation that, system states can follow the desired ones totally since a pre-specified time. All the signals in the closed-loop are proved to be bounded.For a class of nonlinear systems, the terminal attractor is adopted in the control design to deal with initial state error. The iterative learning algorithm is designed to deal with parametric time-varying uncertainties, while a periodic one is given to cope with time-invariant counterpart. A neural network iterative learning controller is specially designed to handle non-paremetric uncertainties. It is shown in the theoretical analysis that system states completely follow the desired trajectories after a pre-specified time, and all the signals in the closed-loop are proved to be bounded. Numerical simulation is conducted to illustrate the validity of the proposed method. A new learning variable structure control strategy is proposed to improve the convergence performance for nonlinear uncertain continuous-time systems. The learning part helps to improve tracking accuracy, while the variable structure part enforces the robustness. Replacing the sign fuction with a continuous function, the designed control can eliminate chattering. Replacing the sign fuction with a sat function also works. The stability of the close-loop system is proved and numerical simulation is given.A repetitive sliding-mode control strategy is proposed to address the problem of periodic trajectory tracking and periodic disturbance rejection. By the virtue of the designed repetitive controller, the s-trajectory is confined to a slimmer quasi-sliding mode bandwidth within a predefined duration, and the bandwidth is proportional to the variability of the disturbance. The disturbance estimation error is forced to a specified residual region since another predefined duration. The same method is applied to the design of traditional sliding-mode controller for slow-varying disturbance rejection. Theoretical analysis and numerical simulation show that the designed controller is chattering-free and effective to achieve the desired control performance, while the stability of the closed-loop is guaranteed.A repetitive sliding-mode control strategy is proposed. With the introduction of disturbance compensation part, the switching function dynamics is desired. The notion of attractive layer and quasi-sliding mode are specified. Theoretical analysis shows that, attractive layer bounds and quasi-sliding mode bandwidths can be different. Numeric simulation based on the mathematical model of the linear motor control system is given to verify the theoretic result. And experiments are conducted on the linear motor control system.