Controller-dynamic-linearization Based Data-driven Iterative Learning Control And Applications

This dissertation focuses on the controller-dynamic-linearization based data-driven iterative learning control for a class of unknown nonlinear non-affine discretetime repetitive systems.The systematic design of the learning controller structures and the uniform automatic tuning method of the learning controller parameters are established.Some problems of the learning control system design and analysis are discussed.The simulations and experiments are conducted for verification.The main works and key contributions are summarized as follows.1.Four controller-dynamic-linearization based data-driven iterative learning control methods are investigated for unknown single-input single-output nonlinear non-affine discrete-time repetitive systems based on the nonlinear complexity of the controlled plants.Based on the unknown nonlinear ideal learning controller,four learning control laws are formulated through the compact form,partial form and full form dynamic linearization methods in the iteration domain,and the full form dynamic linearization method both in the iteration and time domains,respectively.By virtue of the equivalent data model for the controlled plant in the iteration domain,the automatic tuning of the learning control gains in the iteration domain is achieved through the Newton-type optimization method using only the measured input-output data of the controlled plant.Then the corresponding four data-driven iterative learning control methods are formulated integrating with the estimations of the pseudo partial derivative and pseudo partial gradients.The pointwise convergent properties of the proposed control methods in the iteration domain are established in the sense of 2-norm.The simulation examples and compared analyses are conducted to verify the effectiveness of the proposed control methods.2.For the aforementioned unknown single-input single-output nonlinear repetitive systems,the radial basis function neural network(RBFNN)is introduced to approximate the desired learning control gain,which enhances the nonlinear tracking ability of the controlled plant for the proposed data-driven iterative learning control method.Based on the learning control law and equivalent data model of the controlled plant obtained by the full form dynamic linearization method in the iteration domain,the estimations of the weight matrix of the RBFNN and the unknown information related to the controlled plant are achieved utilizing the steepest descent method.Then the improved data-driven iterative learning control method based on the RBFNN is formulated.It is shown that the uniform ultimate boundedness of the proposed control method in the iteration domain is guaranteed through rigorous theoretical analysis in the sense of 2-norm.The simulation example and the simulation on the high-speed train model are implemented to demonstrate the effectiveness and applicability of the proposed control method.3.Two controller-dynamic-linearization based data-driven iterative learning control approaches are investigated for unknown multi-input multi-output nonlinear non-affine discrete-time repetitive systems.For an unknown nonlinear multi-input multi-output ideal learning controller,two learning control laws are constructed using the compact form and partial form dynamic linearization methods in the iteration domain,respectively.Their existences are proved.Based on the equivalent data model of the controlled plant,the automatic tuning of the pseudo Jacobian matrix and the pseudo partitioned Jacobian matrix for the two constructed learning control laws is designed using the steepest descent method,which avoids the matrix inversion problem of using the Newton-type optimization method.Then the corresponding two data-driven iterative learning control schemes are formulated integrating with the estimations of the pseudo Jacobian matrix of the controlled plant.The pointwise convergent properties of the two proposed control methods are established in the sense of a generic norm.The effectiveness and application are verified by a simulation example and an experiment on the Gantry linear motor system.4.Since the aforementioned data-driven control methods are only designed for a single agent,they are not capable of managing the effect of the relationship between the dynamics of each agent and the communication topology for a multi-agent system on the global control objective.Then the controller-dynamic-linearization based distributed iterative learning control methods are investigated,focusing on the consensus tracking problem of unknown nonlinear non-affine discrete-time heterogeneous repetitive leader-follower multi-agent systems.Based on the graph theory,three distributed learning controllers are formulated by virtue of the unknown nonlinear ideal distributed learning controller,that is,the distributed iterative learning control laws using the compact form,partial form and full form dynamic linearization methods in the iteration domain respectively,and their existences are analyzed.Combining with the equivalent data model of each agent,the learning control gains are automatically tuned over the distributed control objective through the Newton-type optimization method using only the local measurement information among neighboring agents,and three data-driven distributed iterative learning control schemes are formulated.Theoretical analyses demonstrate that the tracking errors of all the agents are convergent in the iteration domain in the cases of iteration-independent and iteration-varying communication topologies using the three proposed distributed control approaches.The simulation examples and the simulation on the multiple permanent magnet linear motors are utilized to demonstrate the effectiveness and applicability of the proposed distributed control approaches.