Automated Learning and Monitoring of Limit Functions

In practice, automated monitoring of spacecraft relies heavily on limit-sensing and simulation. Unfortunately , limit-sensing tends to be too imprecise (missing alarms) and simulation tends to be too precise (false alarms). To help overcome those disadvantages, we present an anytime algorithm called envelope learning and monitoring via error relaxation (ELMER). It can incrementally generate successively tighter hi/low limit functions envelopes, essentially moving from the wide static limits typical of limit-sensing toward the precise predictions of simulations, while avoiding unacceptable false alarm rates. We summarize the techniques and motivations underlying ELMER and illustrate its performance on telemetry data from the NASA spacecraft TOPEX.