Validation of autonomous fault diagnostic software

A model based diagnostic engine called Marple, developed at TRW, relies on nominal models to perform fault detection and isolation. Nominal model-building and validation and verification play a critical role in Marple's ability to correctly isolate faults and to report nominal system behavior. Recently developed Marple models of a portion of the Attitude Control Subsystem of the Total Ozone Mapping Spectrometer-Earth Probe satellite have been validated with both simulated and flight data. Sensitivity studies performed with simulated data show that the models are particularly sensitive to sensor noise. Initial comparison of Marple models to fault-free measured data revealed the need for additional model parameter tuning to eliminate diagnoses of faults when none actually occur in the data. This paper develops a methodology for building Marple models and describes experiences with using simulated and measured data to validate models.