Artificial Neural Network Based Control Systems 2003-01-0359

This paper reports on an Artificial Neural Network (ANN) based approach to the design of controllers. The system is developed from a truly parallel hardware implementation of a Self-Organizing Map (SOM) type network resulting in a controller suitable for embedded ‘real-time’ non-linear applications. Both open and closed loop control are described with the aid of their application to Active Valve Train (AVT) control. The utilization of such ANN based systems has been found to reduce development time and lead to increased system flexibility, thereby increasing the application opportunities. These advantages come about due to the nature of SOM based systems which enable controllers to be developed without requiring an analytical solution to be derived. Instead, these SOM based systems learn by example and create a map of a non-linear transfer function.