CASTeC - CONTEXT AWARE SPACECRAFT TELEMETRY CHECKING
Satellite Health monitoring by means of Machine Learning
CASTeC is a software tool intended to ease the labour-intensive task of spacecraft telemetry checking, by automating the telemetry signals trend analysis and the detection of anomalous behaviours and novelties.
It provides a predictive and proactive monitoring based on data mining and innovative machine learning techniques, so allows to relieve the Flight Control Engineers from manually setting alarm and warning thresholds over the thousands of parameters of housekeeping telemetry shaping satellite’s health status.
CASTeC learns from the nominal system behaviour, derived either from models and simulations, or from real telemetry data labelled by operators during routine operation.
The telemetry checking is performed by evaluating a large number of statistical features over distributed time intervals, identifying the significant ones and comparing them with reference values.
When a feature deviates from these references values, autonomously defined, the tool highlights the novelty or the trend anomaly in the parameters, raising alarms and warnings based on smarter criteria than usual simple signal range check, and, in addition, with thresholds that are autonomously tailored by the application.
CASTeC output information consists in the S/C nominal behavior characterization and the check output, which is a priority score assigned to all telemetries indicating the degree of novelty of the parameters in the checked period. This allows users to focus on most relevant events and novelties first.
A further great advantage of CASTeC is that it works “context aware”, which means it is able to distinguish several different nominal behaviors, related to satellite position in the orbit track (e.g. Earth/ orbiting body distance, sunlight or umbra…) and subsystems’ status. Knowledge of the operational conditions increase the effectiveness of AI algorithms, allowing to significantly enhance telemetry checking.
Its graphical interface provides a toolset intended to a quick and friendly browsing and drill-down of the (huge) telemetry time series; provides navigable views like heatmaps (also animated as video sequences) and 3D satellite models highlighting sub-systems according their detected health status.
- Autonomous mining of monitoring key features
- Autonomous thresholds learning
- Autonomous anomalies detection
- Contexts management
- Synoptic views of the health status of the spacecraft
- Housekeeping drill down and cross-comparison
- 3D models and videos for advanced investigation
- Collaboration tools for analyses sharing and discussion
- Report generation