부산대학교 도서관

Telemetry data received from satellites during in-orbit operation are typically analyzed by experts to identify faults that might lead to potential subsystem failures. A typical method is to apply limit-checking procedure to locate off-nominal features which do not fall within the expected normal range of values. Human-experts' analysis is usually limited. It focuses on analyzing selected group of features that would interpret a specific situation that might be taking place onboard the satellite. When the size of telemetry features and number of observations are in the order of hundreds or thousands, a full expert-based analysis is almost impossible to achieve. In this paper, expert-based analysis is leveraged by the automation of the telemetry analysis process based on a machine learning technique called the Logical Analysis of Data (LAD). LAD is a pattern recognition and classification approach that combines ideas and concepts from optimization, combinatorics and Boolean functions. One of the main advantages of LAD is its explanatory power, which offers a classification and an interpretation of the root causes of the events under study. Patterns generated via LAD are easily understood by experts as they are constructed from the features within the set of observations. Consequently, LAD is used in numerous practical applications. We are applying LAD in the process of fault diagnosis and prognosis in satellites by combining the domain expert's knowledge and the knowledge extracted by LAD. The procedure begins by performing a fault tree analysis (FTA) and a list of corrective actions, by domain human experts. This analysis is usually limited by the expert's knowledge to represent known faulty states of the system. However, operation data collected over a period of time can introduce meaningful hidden knowledge about faulty states that were not represented in the original FTA. We apply LAD to find this hidden knowledge in data collected through simulation, testing and in-flight operation. The results reflect significant interpretation power and an effective leverage of the obtained knowledge by the integrated telemetry analysis tool when compared to traditional limit-checking method.