Research On The Neural Network Adaptive Control Strategy Of Parallel Robot

The parallel robot has many excellent characteristics, like its simple structure, high stiffness, high precision, and fast dynamic response. It is especially adapted to the occasion which required high precision, heavy load capability and limited space. But the control method of parallel robot was a principal problem, because of its serious nonlinearity, coupling and uncertain mathematical model. Almost all the traditional control methods failed to get a satisfying performance on the parallel robot. Therefore, the focus of the paper is to find a position closed-loop control strategy of the parallel robot which is suitable for medical and rehabilitation, to ensure the parallel robot works accurately and efficiently with inaccurate measurement and modeling, load changes and external disturbances.Firstly, this paper completed the kinematics analysis and trajectory planning of parallel robot with Matlab by studying the theoretical of6-DOF parallel mechanism. The working space of parallel robot is analyzed, and the simulation shows that the parallel robot can meet the position and posture requirements of the rehabilitation training. In addition, the trajectory planning of the parallel robot was simulated by CP trajectory planning algorithm with the given trajectory parameters of the platform for a fixed point.Secondly, this paper focused on the parameters self-tuning of PID based on BP neural networks and PSO algorithm, and made improvement of them. Based on these researches, a new neural network adaptive control algorithm modified by PSO was proposed. It consisted of the traditional PID, BP neural network and the PSO global optimization algorithm which was used to optimize the initial weights of BP neural network. The optimized BP neural network was then used to adjust PID parameters on-line. Variation operation was introduced to the optimization process and the comprehensive influence on PSO and BP introduced by the choice of the activation function gain and the number of hidden layers was considered. The algorithm can reasonably determine the initial weights of BP neural network and can more effectively improve the problem that neural network easily goes into the local minimum value and has slow convergence speed, simulation results show that the proposed method has greatly improved in accuracy and real-time.Finally, model of the6-DOF stewart parallel robot was built by Matlab/SimMechanics. Then the simulation was made through with the traditional PID control strategy. This paper has designed the neural network adaptive controller modified by PSO, and also verified the correctness and the effectiveness of the controller on the simulation platform. The simulation results showed that the neural network adaptive controller modified by PSO can not only adjust the PID control parameters online, but also assure superior efficiency and stable performance, which is better than PID controller.