Chinese Space Science and Technology ›› 2021, Vol. 41 ›› Issue (1): 113-119.doi: 10.16708/j.cnki.1000-758X.2021.0014

Previous Articles     Next Articles

Design of controller for gravity unloading system based on neural network

SUN Yibin,  WANG Limei   

  1. School of Electrical Engineering, Shenyang University of Technology, Shenyang110870, China
  • Online:2021-02-25 Published:2021-02-02

Abstract: The hanging system is one of the important methods to simulate the gravity unloading experiment of a space manipulator on the ground. To overcome the shortcomings of slow response and poor robustness of traditional PID control modes, an intelligent control method based on radial basis function (RBF) neural network was presented. This method has strong non-linear fitting ability and simple learning rules.It can map any complex nonlinear relationship, and is convenient for computer implementation. Taking advantage of this feature, a controller with higher precision of gravity unloading than PID control was designed. The control model identified by the controller using orthogonal least squares method, updates the weight RBF neural network by using a gradient descent method under the mathematical model of the servo motor with load. Finally, the simulation image of Matlab was obtained by writing s-function. Compared with PID control method, the simulation results have fast response, strong robustness and higher accuracy of gravity unloading (98%).

Key words: hanging, gravity unloading, gradient descent method, least square method, RBF neural network