A Real-Time Digital Control Environment Based on MATLAB, xPC-Target Along With a Magnetic Levitation Device for Neural Network Control Law Implementation and Verification

Abstract

The objective of this article is to present the development and hardware implementation of a robust control system for a custom made magnetic levitation device. This hardware in the loop (HIL) device was developed for the purpose of real time control experimentation and education. The digital controller requires a C++ complier and MATLAB, Simulink, Real Time Workshop, and xPC Target by MathWorks. The controller itself consists of a classical lead/lag compensator augmented by static (apriori trained) and dynamically trained neural network (NN) controllers. The need for a dynamically trained NN is supported by the effects of nonlinearities and time changing unmodeled dynamics as compared to an apriori trained static NN. The digital controller may be quickly reconfigured to use either NN allowing for quick comparison. The steady state error of each NN control technique will be compared and the effects of unmodeled changes to the HIL will be discussed in support of the superior performance of the dynamic NN.