The main goal is anomaly propagation in satellite telemetry. There are several approaches to solve the problem, but the system reflects the case based on tracer implementation. The idea of applying this approach is inspired by the article. The implementation of the tracer is based on the project
Unfortunately, the real raw telemetry data cannot be displayed here due to
security concerns. However, the data.tar file contains already
preprocessed info to implement the tracing pipeline and, additionally,
./data/raw folder contains two samples real data represented already.
In this case, the necessary information about the satellite
system are angles, estimated and measured wheel velocities.
The components form a state vector,
and in this vector we can find anomalies by component.
The key steps of the preprocessing are:
-
get LOWESS curve to align raw telemetry
-
get absolute values for quantile barrier application and binarization.
-
create representation of state vector in the tracing problem form.
Such a baseline is used by the telemetry of the on-board ARO system, which provides a solution to the same problem. According to the ROCAUC metrics, the ARO system quality is 54%, while the tracer quality is 82%.
