A Predictive Approach to Failure Estimation and Identification for Space Systems Operations
Ivano Verzola, Anne-Emmanuelle Lagny, Janos Biswas · SpaceOps 2014 Conference · 2014
In this work an intelligent agent to predict future failures occurrences is presented. The agent is intended to support the Columbus Flight Control Team real-time operations. The first step is the selection of the object of the study: the occurrences of Single Event Upset caused by radiation on Columbus Orbital Laboratory Mass Memory Units. An exploratory analysis is performed to determine the metrics used by the agent to foresee the future failures. The exploratory analysis results constitute an interesting set of measurements consistent with the literature on the matter. To compute probabilities, a probabilistic model is created to support the algorithm of the agent. The implementation strategy includes elements of incremental machine learning techniques, scalability and real-time interfaces. The results obtained from an analysis of about two years of data are evaluated, showing that the intelligent agent can effectively support decision on preventive maintenance operations to reduce the occurrences of failures.