Multiple filter methods for detection of hazardous states in an industrial plant

Model‐based detection systems are presented and classified according to their complexity. The main emphasis is put on multiple Kalman filters. A two‐stage approach including Bayes formula and inhomogeneous Markovian processes is suggested to achieve a fault discrimination. Methods are presented to use the information of nondisjunctive models in this context. To show the proper working of the detection algorithms simulation studies and on‐line laboratory experiments with runaway reactions are considered first. A report on an implementation at a large‐scale industrial process unit concludes the discussion.