Modeling and control for pneumatic manipulator based on dynamic neural network

Abstract

This paper studies the modeling and motion control of manipulators driven by single-rod pneumatic actuators. The dynamics model of pneumatic manipulator is analyzed profoundly at first. Then aim at the highly nonlinear, strong coupling, time-various of pneumatic manipulator dynamics, a new internal model controller for pneumatic robot servo system is presented, which has a three-layer feedforward neural network as controller (NNC) and a diagonal recurrent neural network (DRNN) as model predictor (NNM). The idea of the proposed control strategy is to make the system has self-adaptability and strong robustness for parameters variations, model error and various outer disturbances by updating weights of NNC real-time based on on-line model predicting of NNM. Dynamic learning algorithms of both NNM and NNC networks are discussed in this paper. Computer simulation results indicate that the system has strong robustness and significantly improves the control performances of pneumatic manipulator.