Research On Algorithm And Application Of Robot Manipulator Motion Control Based On RBF Neural Networks

Trajectory tracking control of robot manipulator is the important part in its motioncontrol. Control system of robot manipulator is a multi-variable, strong coupling, and highlynonlinear uncertain systems, trajectory tracking requires the robot manipulator to moveaccording to a desired trajectory that has been given. Model reference adaptive control(MARC) based on radial basis function neural networks (RBFNN) not only has strongcapability of dynamic approximation and adaptive by RBFNN, but also could improve thecontrol real-time and immunity interference immunity, thus it has been used in nonlinearcontrol widely. But the traditional learning algorithm of RBFNN, based on K-meansclustering, is very sensitive to algorithm initial value, and it also demands that the number ofall input samples and RBF should be given in advance. For this sensitivity problem of initialvalue, the thesis improves the learning algorithm of RBFNN based on K-means clustering,and uses a learning algorithm based on entropy clustering-radial basis function (EC-RBF) totrain the RBFNN. Through adopting this new method, neural Networks solution of robotmanipulator inverse kinematics is achieved. Using this method to train the two RBFNN thatthe neural Networks model reference adaptive control system of robot manipulator includes,the dynamic model identification and trajectory tracking control of unknown robotmanipulator are implemented. Comparing with the traditional K-means clustering algorithm,the simulation result indicates that this improvement algorithm is more effective and superior.The main contents are below:(1) To study the basic theory and structure of RBFNN, and study the1improvementlearning algorithm EC-RBF, which is based on entropy clustering of RBFNN. Using theentropy clustering, the initial value of traditional K-means clustering is optimized, and thenumbers of clustering are ascertained. Thus the structure of RBFNN is certain, and theresponse ability and generalization ability and of neural Networks are both enhanced.(2) To study the dynamic equations establishment of two different robot manipulator,and in order to achieve the neural Networks solution of robot manipulator inverse kinematics,use EC-RBF learning algorithm to train the RBFNN. The SCARA robot manipulator dynamicequation is analyzed and founded. The simulation result shows that the RBFNN usingEC-RBF learning algorithm has higher solution accuracy to robot manipulator inversekinematics and faster learning speed comparing with the traditional K-means clusteringalgorithm.(3) To study the application of RBFNN based on EC-RBF learning algorithm in robotmanipulator model identification. Radial basis function neural Networks identification(RBFNNI) structure is built and trained by EC-RBF learning algorithm. Using this RBFNNI, the dynamic model of unknown robot manipulator is identified off-line. The simulation resultshows that this RBFNNI has great identification ability, and it can approach the robotmanipulator dynamics model accurately.(4) To study the application of RBFNN based on EC-RBF learning algorithm in traje-ctory tracking control of robot manipulator. The system of neural Networks Model referenceadaptive control (NNMRAC) is designed. In this system, the identification result is used toreplace the unknown robot manipulator, and radial basis function neural Networks control(RBFNNC) is used to control the system. The two RBFNN both use EC-RBF learningalgorithm to learn and adjust on-line. The simulation result shows that the RBFNN based onEC-RBF learning algorithm has better tracking effect in trajectory tracking control of robotmanipulator comparing with the traditional K-means clustering algorithm.