A NEURAL NETWORK MODEL-BASED CONTROL METHOD FOR A CLASS OF DISCRETE-TIME NONLINEAR SYSTEMS

Jerry Sweafford Jr. and Farbod Fahimi

Keywords

Datadriven system identification, discrete sliding mode control, neural networks, nonlinear modelbased control, robotic manipulators,trajectory tracking

Abstract

State-space model-based controllers are generally preferred over neural network model-based controllers, in part because many neural network model-based controllers lack proof of stability. However, the physics-based modelling needed for the derivation of state-space models, and the identification and verification of the model parameters are challenging tasks that are easier when neural network models are used. A model-based state-space controller with neural network modelling would combine the existence of proof of performance with a less difficult modelling task. We present a simple method of system identification and control based on a discrete, nonlinear model of the plant, in which a neural network state predictor is used to estimate the terms of the state-space model for a discrete sliding mode controller. The proposed approach was demonstrated and validated by effective trajectory tracking in a simulated two-degree-of-freedom (DOF) robotic arm and the actual robotic arm on which the simulation was based.

Important Links:



Go Back