Diagnosis of Sugar Factory Processes Using GMDH Neural Networks

Abstract

This paper deals with GMDH (Group Method of Data Handling) neural networks and their application in Fault Detection and Isolation (FDI) systems. Such networks can be considered as feedforward networks with a ‘growing’ structure during the training process. To apply the GMDH networks in the FDI for real technological processes we propose some extentions of the GMDH theory to multi-output networks and dynamic modelling. The proposed networks and their extensions have been implemented in an industrial model-based diagnostic system for some units of the Lublin sugar factory in Poland. Finally, simulation results show the effectiveness of the proposed neural networks in FDI systems.