Application of interval models to the detection of faults in industrial processes

A major difficulty when analytical redundancy is applied to fault detection is taking into account the uncertancies assosiated with the system and measurements. In this paper, this uncertainty is considered via the use of intervals for the parameters of the model and the measurements is analyzed. The coherence between the model and measurements is analyzed if they are inconsistent, then ther is fault. The reference behavior for fault detection is obtained by simulation of the interval model. This probelm of simulation is reformulated as a range computation problem, which is a hard problem but can be softened using error-bounded estiamations. To carry out the interval range computation. Modal Interval Analysis is used. Results are improved by the use of several sliding time windows. The major advantage of this technique is the absense of false alarms if the uncertaincies associated with the system and the measurements are taken into account in a guaranteed way. This method is being used to detect faults in academic examples and real processes like the ones used with the European project CHEM.