Diagnosis of continuous valued systems in transient operating regions

The complexity of present day embedded systems (continuous processes contro lled by digital processors), and the increased demands on their reliability motivate the need for monitoring and fault isolation capabilities in the e mbedded processors. This paper develops monitoring, prediction, and fault i solation methods for abrupt faults in complex dynamic systems. The transien t behavior in response to those faults is analyzed in a qualitative framewo rk using parsimonious topological system models. Predicted transient effect s of hypothesized faults are captured in the form of signatures that specif y future faulty behavior as higher order time-derivatives. The dynamic effe cts of faults are analyzed by a progressive monitoring scheme till transien t analysis mechanisms have to be suspended in factor of steady state analys is. This methodology has been successfully applied to monitoring of the sec ondary sodium cooling loop of a fast breeder reactor.