Multi-model neural network sliding mode control for robotic manipulators

Abstract

A Multi-model neural network sliding mode controller (MNNSMC) is proposed for robotic manipulator in this paper. The proposed MNNSMC scheme combining the SMC (sliding mode control) and neural network technique. The multimodel ensures that when the working environments of robotic manipulator are changeful, we can choose the proper model to get a better control indicators. The controller applies the SMC to obtain high response and invariability to uncertainties and adopts neural network to estimate the switch gain in order to weaken the sliding mode chattering. The neural network is trained extensively with the state estimation error backpropagation learning algorithm. It consists of an input layer, hidden layer and output layer. Input layer of vector are errors and velocity errors and output layer of vector means to estimate the switch gain. In order to ensure the rationality of the switch, a new switching index is proposed which is a PID type with forgetting factor. The simulation results demonstrate the effectiveness and feasibility of the proposed control strategy.