INSTRUMENT FAULT DETECTION AND IDENTIFICATION BASED ON ANALYTICAL REDUNDANCY

Abstract For the detection, localization, and estimation of ‘small’ sensor failures like biases or drifts a fault supervision method is presented that is based on the analytical redundancy between different measurements, given in form of a mathematical model of the process to be controlled. A general problem of such a computer-based instrument fault detection scheme is the separation of faults caused by the system itself and by faults of the sensors. For example, changes in process parameters have to distinguished from sensor failures to maintain normal system operation. The proposed failure detection scheme utilises state- augmented and hypothesis-conditioned Kalman-filters to track time-variant process parameters and to calculate a-posteriori probabilities for each hypothesis of a sensor fault. Least-squares algorithms are implemented to estimate the fault and to classify it as a bias, drift, or scale-factor deviation. Figures of merit are given for the detectability and separability of different sensor faults and for the observability of the parameter changes. Because very few practical applications were made in the field of instrument fault detection and reported results often rely only on simulated data, we have used real measurement data from an industrial steam generator. Some results from off-line processing of these data are given in this paper.