Model-Based Reasoning in the Detection of Satellite Anomalies
Abstract
Automatic fault detection and recovery would be a mandatory requirement for a satellite where some degree of autonomy is required. This thesis reviews some AI techniques used for the detection of satellite anomalies, and concludes that the model-based reasoning paradigm is best suited for automated on-board fault detection because it can cope with situations not necessarily programmed into the knowledge base. Using the Scheme language and its SCOOPS object oriented extension, development of software is described that models the pitch control channel in the attitude and velocity control subsystem of a typical geo-stationary communications satellite. This model is used by the model-based reasoning algorithm to diagnose faults in the real system. The algorithm used, is based on Scarl's 'Full Consistency Algorithm', which is suitable for systems that have many sensors, but has limitations when applied to systems that are dependent on time or have feedback loops. These limitations were overcome by using a model that did not include time dependent objects and by 'breaking the loop'. It was found, for this problem domain, that the reasoner's model did not have to be identical to the real system to be able to successfully detect the cause of an anomaly.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1990
- Accession Number
- ADA230535
Entities
People
- Ralph W. Dries
Organizations
- Air Force Institute of Technology