Fault Detection and Isolation in Attitude Determination Systems

A simple buat powerful method to detect failures or changes in an attitude determination system employing a Kalman Filter is to monitor the filter residuals. By construction, the residuals should be a white sequence with zero mean and known variances. Departure of the residual statistics from the predicted values can be indicative of unknown parameter changes (gyro biases, star sensor alignment, etc.) and, therefore, used as an error flag to reconfigure the attitude system. Tbis paper presents an algorithm for detection and correction of failures by utilizing the filter residuals. An attitude determination simiulation is employed which uses gyros to propagate the inertial attitude while star sensor measurements provide updates to this attitude. The updates are processed by a decoupled Kalman Filter which is more efficient when fixed biases are part of the states and is also easily converted to a "fixed bias" filter when bias states are not to be eistimated. The nominal system filter would be a 3-state attitude only filter since the biases would not have to be estimated after an initial calibration. Only in the event of a "soft" failure (where a bias changes abruptly) or a periodic recalibration maneuver would the filter states be increased to estimate the biases. Under this "soft" failure condition, an autonomous correction can be obtained without ground support.