Study On The Intelligent Control Of Flexible Joint Manipulator Based On Dynamic Pattern Recognition

In recent years,the combination of pattern recognition and control theory has attracted much attention.With the aid of neural network,the pattern recognition theory has been applied widely in real life.Meanwhile,flexible joint manipulator with some significant features,such as light mass,small volume and low energy consumption,has a great application prospect in flexible manufacturing and aerospace fields.With the improvement of the intelligent level of the manipulator system,the pattern recognition theory is also gradually applied to the practical manipulator control.Therefore,it is greatly significant to study the flexible joint manipulator control problem based on pattern recognition,which may make flexible joint manipulator more effective for the high-technology industry.In this thesis,pattern-based intelligent control is developed for flexible joint manipulators by using deterministic learning theory.Firstly,the novel pattern recognition method is proposed to identify and recognize the closed-loop system dynamics and reference trajectory,respectively.In order to acquire,store and reuse experience knowledge of flexible joint manipulator dynamics from different tracking tasks,the main works of this thesis are concluded as follows:Firstly,the multi-link flexible joint manipulator model is described by Lagrange dynamic equation,and then the adaptive neural tracking control problem is considered for the multi-link flexible joint manipulator.Based on the structural features of the considered manipulator model,the dynamic surface control method is used to design the adaptive neural controller,which reduces the calculation burden of the proposed method.With the help of the deterministic learning theory,the weights of neural networks are stored as constant values in the steady-state control process.As a result,a set of neural learning controller group is constructed by using the stored experience knowledge for different reference trajectories.Secondly,in view of the pattern recognition problem of different reference trajectories,this thesis gives a similarity definition of dynamical patterns and designs the corresponding estimators of dynamical patterns.Subsequently,the dynamical pattern library is constructed based on dynamical features of different reference trajectories.By comparing the norm of the residual error between the dynamical estimator and the test reference trajectory,the pattern prerecognition and recognition strategy are respectively proposed to achieve the real-time monitoring and rapid recognition of the test reference trajectory.According to the recognition results,the switching strategy of the controller is presented by using the neural learning controller group.The proposed switching control strategy ensures the continuity of control signals and improves the transient-state performance of the control system during the pattern switching instant.Simulation results of a two-link flexible joint manipulator are given to verify the effectiveness of the proposed pattern-based control scheme.Finally,this thesis focuses on the prescribed tracking performance of the manipulator control system.By introducing the transformed function,the original constrained control problem is converted into a non-restricted one,and a novel neural learning control scheme is developed based on the deterministic learning theory.In order to verify the effectiveness of the proposed scheme,simulation results of a two-link flexible joint manipulator are given in this thesis.Moreover,this thesis discusses the feasibility on whether the proposed control scheme with prescribed performance can be extended to pattern-based control mentioned above.