Development of False Alarm Mitigation Techniques

An essential premise motivating the Joint Strike Fighter Prognostics and Health Management (JSF-PHM) approach is the capability of correctly detecting faults in components or subsystems. Whether a fault is declared according to time series recordings by monitored on-board sensors, or by sophisticated reasoning algorithms that recognize slow changes in behavior as a component degrades over time, fault detection must be performed correctly. Otherwise, failed components may not be replaced as necessary, or one may be unnecessarily serviced due to a false alarm. Both types error interfere with the attainment of AL objectives of minimizing aircraft support and logistics costs. JSF-PHM demands novel algorithmic methods to eliminate false alarms. This is a challenging task, since classification algorithms intended to detect anomalies (unrecognized faults) learn data probability distributions in known fault condition space. Extrapolation outside the training data is problematic; however this is exactly the region where anomalies are expected to occur. Faced with an anomalous condition, proper discrimination between "fault" and "no-fault" will determine the overall false alarm rate of the PHM system. This proposal describes a novel means to achieve a minimal false alarm rate at a given, fixed level of statistical confidence.