Research On Neural Network Trajectory Tracking Control Of Space Robot And Micro-Gravity Simulation Method

Space Robots are playing a more and more important role in the human exploration of space, they could repair and recovery of the satellites, carry out construction and maintenance of the space station,execute danger task instead of astronaut and so on. Hence, study of space robots has important significance.The free-floating space robots are as the control objects; the paper studies in four conditions. Namely , uncertainty model, parameters saltation, non-speed information feedback and joints dead zone condition. Three-dimensional ground microgravity simulation methods of space robots based on gas suspend have been studied.This paper introduces the current domestic and international overview of the development of space robots, and summarizes the research of space robot trajectory control, then reviews the space ground microgravity simulation method and points out the advantages and disadvantages of each method. For attitude control and not control free-floating space robots, kinematics and dynamics model are established, then control problems of trajectory tracking are studied base on dynamics model.First, as the actual space robot system model always contains the unmodeled unknown uncertainties, this paper derives the system uncertainty model, which affects the system stability by designing the ideal controller. Then the uncertainty model is estimated by the neural networks with a strong ability of learning, linear parametric technology is used to design the on-line real time adjust learning law including weights value and network parameters. For the neural network accumulation model error, the robust controller is used to adaptive compensation. Moreover, considering the engineering application, PD controller is used to parameters adjustment, the system is proved to be uniformly ultimately bounded based on Lyapunov theory.Further, for the parameter saltation problems caused by space robot extreme condition, this paper designs a robust controller based on neural network. And during the control finds out: parameters saltation caused uncertainty model sudden changes, the neural network with the ability of learning couldn't approximate the changed uncertain model in short time, this caused the neural network failure, and the system stability was destroyed. For this situation, this paper proposes a adaptive H∞robust control strategy based on neural sliding model integrated, this integrated controller is constitude of classification strategy of the neural network approximation region, sliding mode variable structure control free from system parameter variations and disturbances, rapid response and good dynamic characteristics. In the early control stage, once it happens parameters saltation conditions, this integrated controller quickly starts sliding mode controller that has redundant characters to compensate the bland region out of the neural network approximation region, compensation error is suppressed as external disturbance by the robust controller. All the controllers together accelerate the convergence velocity, let the system L2 gain less than the given index. It ensures the system stability in parameters saltation condition.In addition, the space robot joint motors are usually run on the low-speed conditions. It makes the quality of the signal poor, and additional weight of tachometer also affects the efficiency of space robots. Hence, research of the non-velocity information feed back space robot is more important. This paper puts forward a output feedback control strategy based on neural network, neural network serves as the dual roles of observer and controller, namely, while the neural network compensates the uncertain part of controller and the unknown nonlinear part of observer,it estimates the joint angle velocity with the error measure. The proposed control method doesn't need the dynamic model and inertia inverse matrix, it only need the joint angle information. Adaptive algorithms of network weights and network parameters can achieve real-time online adjustment. It proves the whole closed loop system stability based on the Lyapunov theory.Subsequently, for the space robot actuators dead zone problem, this paper proposes adjustment compensation strategy. The system uncertain model adopts neural network to approximate, the approximation error as the robust item is eliminated. For the actuators dead zone, two neural networks separately are used for dead zone estimation and adaptive compensation. The mathematical relationship among the dead zone output, dead zone compensation and controller is deduced. The neural network compensator and dead zone estimator adaptive learning law are designed. The system is proved to be uniformly ultimately bounded based on Lyapunov theory.Finally, for the problem of the proper space microgravity simulation means based on floating, this paper proposes passive and active two novel three-dimensional simulation of micro-gravity flotation. Passive system mainly uses gravity compensation and the torque balance to achieve compensation, and active system uses a series of mechanical transmission and motor drive system to achieve compensation. The two simulation methods both could solve the complex three-dimensional microgravity simulation problems.