A MODEL-BASED METHOD FOR RANKING MEASURED CHANNELS

A quantitative method for ranking measurements is presented. The technique can be used to identify faulty measurements or redundant measurement channels. The method assumes the knowledge of a model from a previous modelling or identification stage. The implementation is in the time domain, using a suboptimal fault-magnifying observer structure to validate the internal connectivity of the measured responses. Both simulated and experimental cases are presented, showing good sensitivity to the detection of various measurements faults, e.g. bad placement of sensors (on nodal lines) or low signal to noise ratios. Standard tests such as observability and controllability measures fail to detect such phenomena in realistic situations. © 1992.