Model-based fault detection using RBF networks and Extended Kalman Filter

A model-based fault detection method is developed using two Radial Basis Function (RBF) Neural Networks. Two RBF neural networks are used as process output models and process variables at normal conditions are used for training the networks. One RBF network estimates the process outputs with a positive error and the other one estimates the process outputs with a negative error for all training data. Extended Kalman Filter (EKF) algorithm is used to train neural network parameters. Outputs and variables of the penicillin fermentation simulator are used as practical data for testing the performance of the algorithm.