Application of Neural Networks for Diagnosing and Predicting the Condition of an Industrial Furnace

(Draft Paper) This paper discusses an industrial application of a neural network based automatic scheme for fault detection and diagnosis. Faults in a lime kiln are isolated and detected, using two multilayer feedforward networks. One is used to model the industrial process according to its nonlinear structure. The fault detection method is based on the output prediction error between the real simulated response and the neural network model response. A second neural network is fed with the residual vector from the previous phase for which each speci c output corresponds to each speci c fault. The good performance obtained demonstrates the e ectivness of the strategy employed.