Research On PID Motion Control Method Of Reconfigurable Manipulator Based On Netural Networks

With the continuous development of industrial productivity, the lives of human begin to enter the stage of intelligence, and robots have come into the social industry and people's lives,reconfigurable robot has attracted people's strong attention among various kinds of robots.Reconfigurable robot owns the dual traits of both mechanical system and control system, and it can complete more than a definitely single task,which helps to reduce the application cost of the robot and make its efficiency greatly improved. This thesis mainly includes the research of reconfigurable robot's kinematics and dynamics analysis, and the study of neural network PID control algorithm as well as experiments for reconfigurable robot.Firstly, the kinematics model of six degree of freedoms (DOFs) reconfigurable robot is established and its relative analysis is done in the thesis, then simulation experiments are performed to verify its correctness. The general forward kinematics formula of reconfigurable robot's different modules is derived, which is applicable to four kinds of robot's modules.Based on the configurations of robot, the position and attitude of robot can be obtained by using MATLAB and MATLAB robotic Toolbox respectively which verifies its value. An inverse kinematics solution method is proposed on the basis of configuration planes, which can calculate the joint angles of reconfigurable robot's different joint modules in the case of known operating points.Secondly, the dynamics model of six degree of freedoms (DOFs) reconfigurable robot is established. According to the concept of configuration planes and its division method, the traditional dynamics is divided into the dynamics analysis between two configuration planes and dynamics analysis within a configuration plane. Then the novel dynamics model is constructed by adopting Newton-Euler equations based on configuration planes, the joints'velocities and accelerations can be obtained by forward iteration, both forces and torques can also be gotten by inverse iteration. Through comparing the dynamics simulation results of configuration plane-based Newton-Euler and traditional Newton-Euler method, the former one can reduce the calculation amount and save time.Thirdly, the dynamic model of three DOFs reconfigurable manipulator is identified by RBF neural network. Both of the result after identification and the input of single neuron PID would be input of overall manipulator' control system. Then according to ideal trajectory given, trajectory tracking control of reconfigurable manipulator can be completed. RBF network identification adopts adaptive iterative control algorithm to adjust the output weights of the network. Single neuron PID control model adopts supervised Hebb learning rule to adjust weights of single neuron. Because the RBF neural network can achieve approximation of arbitrary precision to any function and its own advantages, the local approximation is very good especially for nonlinear control system. The single neuron PID controller overcomes the shortcomings of the traditional PID controller which is time-consuming and the control precision is not ideal caused by the constant parameters.