Fault Detection using Radial Basis Function Network and Polygonal Line

Authors:	Bhushan Bharat, The University of Sydney, Australia Romagnoli Jose A., The University of Sydney, Australia Wang Dawei, The University of Sydney, Australia
Topic:	6.4 Safeprocess
Session:	Fault Detection and Identification
Keywords:	Fault Detection, Fault identification, Radial base function networks, Genetic algorithms

Abstract

This paper proposes a novel approach for the reduction of the dimensionality of non-linear data based on radial basis function (RBF) network and polygonal line (PL). A method is suggested to find out the optimum number of nodes in the hidden layer which is mostly heuristic in case of other proposed methods. A hybrid optimization technique based on genetic algorithm (GA) and Broyden, Fletcher, Goldfarb, and Shanno (BFGS) Quasi-Newton algorithm is used for faster and effective training of the network. Kernel density estimation is used for finding the confidence limits. The method is applied for detecting the fault in a simulated continuous stirred-tank reactor (CSTR). The result shows that the proposed method is excellent for process monitoring in non-linear systems.