Tensor-Based Anomaly Detection for Satellite Telemetry Data

Satellites is the bird’s-eyes that enable us to view massive areas of earth at the same time, satellites can gather more data, more quickly, than tools on the ground. Satellites also can view into space better than telescopes at earth’s surface. Development of such artificial satellites which composed of many subsystem, requires a lot of time and money that any deadly failure is unacceptable, the satellite operating in a remote environment, so it is practically very hard or impossible to repair it once a severe failure occurs, so detecting the anomalies of the subsystems measurement values (Telemetry data) is the first step in satellite failure protection and early warning. Traditional spectral-based methods like PCA is traditional for detecting anomalies in a variety of domains and problems. However, if the collected data contains tensor (multiway) structure, for example space-time-measurements values, such as the satellites subsystems measurement, some significant anomalies may stay hidden with these traditional methods. Tensor-based anomaly detection (TAD) applied in a variety set of disciplines over the recent years, although it is not recognized yet as an official category of anomaly detection techniques. This work target to highlight the candidate of tensor-based technique as a new approach for identification and detection of abnormalities and dud in the satellite telemetry data.