A Radial-Basis-Function-Network-Based Controller Derived from a Neural-Network-Based Controller and its Application to Controlling Mechanical Systems

Abstract

In this paper we describe a method of designing a radial-basis-function-network-based(RBFN)controller. The RBFN controller is derived from a neural-network-based(NN)controller using differential operator focusing on the sigmoid function derivative of the NN controller. To overcome the Jacobian problem, the RBFN controller uses a learning algorithm and a neural identifier which uses an adaptive algorithm to estimate the plant Jacobian. A conventional feedback controller is incorporated into the RBFN controller to ensure both robustness and stability at the beginning of the learning process. Simulation results for mathematical plants demonstrate the applicability of the RBFN controller for controlling nonlinear systems and experimental results for 1-degree-of-freedom robots demonstrate its usefulness for controlling practical systems. Application to controlling an ODD positioner and a tunneling machine demonstrates the effectiveness of the RBFN controller for controlling mechanical systems.