Research On Control Of Multi-link Chain Manipulators Based On Deep Learning

To implement complex and sophisticated operating duty,the manipulators require the flexible mechanism design combined with the corresponding control methods.However,the introduction of flexibility brings great challenge to the precise position control of the manipulators.With the rapid development of computer technology,as an excellent big data processing technology,artificial intelligence has prominent research significance and application value.Deep learning as a crucial technique for realizing artificial intelligence,become one of the important approaches for deep integration of mechanical research and artificial intelligence.This dissertation has established a hybrid control method of flexible manipulators based on deep learning neural network,using the end trajectory of the manipulators to predict the driving force torque of the joints in feedforward control,and the driving force torque is used as the input of Proportional Integral Derivative(PID)feedback control.Chapter 1 of this dissertation introduces the development background and research status of flexible manipulators and deep learning.A rigid-flexible coupling dynamics model of the spatial multi-link chain flexible manipulators system composed of n flexible links and n flexible rotating joints is established in Chapter 2.Chapter 3 proposes an effective hybrid control method based on Temporal Convolutional Network in series form for flexible manipulators terminal trajectory control.In this method,the end trajectories of samples obtained through a large number of dynamics simulations are used as the initial input values of the network,and the network parameters are optimized according to the error function between the driving torque and the expected driving torque.In order to improve the control accuracy,the joint driving torque predicted by the network is taken as feedforward input,combined with kinetic equation and PID control,to form a hybrid control method suitable for the flexible manipulators,and thereby the end trajectory of the flexible manipulators is controlled.In Chapter 4,some control simulations are carried out for the trajectory tracking problems of a planar two-link flexible manipulators.Four different driving forces of joints,constants,linear,harmonic and random,were used to generate simulation samples.Simulation results show that the accuracy that the proposed hybrid control method predicting the end trajectory of the manipulator can reach 90%.