An algorithm has been developed for a computer-based electronic system that would aid human ground controllers in monitoring data from sensors in the electric-power system of the International Space Station. Among other things, the algorithm encodes knowledge of human experts; thereby harnessing the ability of computers to do diagnostic calculations that human experts would do but that take much longer when done manually. The algorithm also has potential for application to a variety of systems outside the aerospace industry.
The Space Station power system is characterized by two levels of redundancy: (1) two or more sensors measure the same value; and (2) synthetic (probable) measurements can be generated by taking account of physical laws that govern the monitored power system. Traditionally, ground controllers utilize the second level of redundancy in that they compare any suspicious sensor measurements against probable measurements, which, heretofore, they have calculated by hand.
Although NASA has employed autonomous electrical power systems for several years, computers have not been used to help ground controllers diagnose anomalies in sensor measurements. Control-center consoles have merely alerted controllers when sensor readings have exceeded set maximum or fallen below set minimum values.
The present algorithm systematically filters possibly conflicting and erroneous sensor data to determine the most probable values of critical measurements. The algorithm generates the most probable values of current and voltage for the sensors in question, draws conclusions from statistical information, and indicates, to ground controllers, one of three possible types of causes for any anomalies: sensor failures, soft faults, and sensor drift.
The algorithm, implemented in a C++ computer program, calculates the most probable values from all available measurements, real and calculated. If a value does not fall within the 3-standard-deviation (3σ ) range for all measurements, the measurement that differs most from the estimated value is eliminated from the calculation. This procedure is repeated until either the estimate falls within the 3σ range for the remaining measurements or only two disagreeing measurements are left, in which case the value is considered indeterminate. The most probable values of sensor measurements are calculated from the means and variances of redundant values, creating a weighted average based on the variance of each sensor.
This work was done by Wallace Kelly III of Rockwell Space Operations Company for Johnson Space Center. For further information, access the Technical Support Package (TSP) free on-line at www.nasatech.com/tsp under the Electronics & Computers category. MSC-22728