Anytime Diagnosis Using Model-Based Methods for Satellite Diagnostics

Many satellite anomalies manifest themselves slowly over time and go undetected until they reach critical and possibly unrecoverable status. Because modern satellite systems are relatively reliable, the ground controller must perform the almost impossible task of attending carefully over long periods of time to telemetry readouts which almost always indicate nominal operation, while maintaining a constant readiness for the onset of failure. Humans are notoriously poor at that type of task. On the other hand, intelligent systems have been successfully used to monitor emerging trends in data streams and have been used successfully for tireless monitoring of manufacturing processes, changes in structure configurations, and deterministic failure patterns. The problem then becomes to develop an "intelligent" real-time system for fault detection, diagnosis, and recovery/resolution of anomalies in satellite telemetry streams. This Model Based Reasoning Diagnostic Engine (MBRDE) is aimed at the combination of adaptive and knowledge-based intelligent methodologies that provides a satellite diagnostic assistant for incorporation into future satellite ground stations. We present a framework for a diagnostic methodology that combines characteristics of model-based reasoning with those of anytime algorithms. We illustrate a bottomup modeling method that constructs a hierarchical device model, and a top-down traversal method that constructs a tree of potential component diagnoses for a given anomaly in the device’s observed outputs. These methods combine to form an innovative framework for providing diagnostic assistance based on model-based principles given any amount of processing time. This work was performed under a Small Business Innovation Research contract No. F30602-94-C-0089, Air Force Research Laboratory, Space Vehicles Directorate, Kirtland AFB, NM.