Anomaly detection for satellite power subsystem with associated rules based on Kernel Principal Component Analysis

a r t i c l e i n f o The paper presents an implementable method of anomaly detection for satellite power system. Specifically, a data-driven anomaly detection method for sensor data integrated Kernel Principal Component Analysis (KPCA) and association rule mining is demonstrated. Establishing associated rules among sensor monitoring data sets, this approach analyses the structure of measure space via its Eigen matrix with KPCA, and identifies the anomaly. Especially, different anomalies from satellite system and sensors can be distinguished with the changes of association rules. The effectiveness of the method is proved on sensor data from Feng-Yun satellite power subsystem. The power supply subsystem provides sustainable and reliable energy for satellite in order to ensure the normal operation. The performance of power supply subsystem will directly affect the other subsystems and dominate their performance of the satellite. Hence, it is significant to detect anomalous states of the power subsystem to ensure the system health [1,2]. Moreover, anomaly detection is the basic function of prognostics and health management (PHM) which has been applied widely in space engineering [3–5]. In general, the in-orbit anomaly detection of satellite requires detailed analysis of the large-scale data by monitoring sensors. However, the sensor is one of the elements or units with high failure risk, as the sensor anomaly on spacecraft may lead to state estimation error and false alarm. Thus, the anomaly detection method for satellite power subsystem must have the capability of anomaly identification. For a complex system (e.g. satellite power subsystem), a single sensor is incapable of collecting enough information for accurate condition monitoring. Multiple sensors are needed in order to complete this task [6,7]. The relationship between the multiple sensor monitoring data is often complex and nondeterministic. While detecting anomaly with monitoring sensor data, large amount of sensor data as multiple time series is redundant and correlative. Thus, the system and sensors can be described more comprehensively. It should be noticed that, the relationship between the multiple sensor monitoring data can no longer keep constant when sensor anomaly occurred. Therefore, it is quite critical to isolate and locate the sensor anomaly and system anomaly via the associated rules between sensor data. In this work, we propose a novel anomaly detection method for satellite power subsystem based on Kernel Principal Component Analysis (KPCA) and association rule mining. After establishing associated rules between multiple monitoring data, the structure of measure space …