Adaptive Limit Checking for Spacecraft Telemetry Data Using Kernel Principal Component Analysis

In space development, early anomaly detections of the space systems, especially detecting symptom of anomaly, and early measures against them are very important. In this study, we introduce a model acquisition method for spacecraft telemetry data using multivariate statistical tools and auto anomaly detection with the acquired model. We expand conventional limit checking to more adaptive one by considering relationship of telemetry data series using Kernel PCA. This study is aimed for reducing the cost of the modeling of spacecraft systems and improving the anomaly detecting accuracy. We applied our study to a real satellite telemetry data provided from Japan Aerospace Exploration Agency (JAXA), and detected not only anomalies but symptom of anomalies that conventional limit checking can not detect.