Being able to quickly identify anomalous behaviour allows to reduce downtimes and keep our spacecrafts healthy. The current approach built in our Mission Control System is to use out-of-limits checks. However, many behaviours are anomalous even if they are within limits. Novelty detection is a new monitoring paradigm that allows to detect unusual behaviours in telemetry parameters. Unusual behaviour is usually a signature of anomaly in their way to develop. The novelty detection monitoring approach requires very little engineering knowledge as it learns from given examples of nominal behaviour. It also features very seldom false alarms and is complementary to the out-of-limit paradigm.
Keywords: monitoring, diagnostics, performance analysis
The Novelty Detector has been validated with Venus Express, XMM and Cryosat2 anomaly cases. The Novelty Detector managed to detect the anomaly way before the out of limits checks did. In some cases, these anomalies were not even detected by the out of limits. Flight Control Engineers check the Novelty Detection results daily as part of their monitoring tasks.
The monitoring technique based on Novelty Detection uses outlier detection techniques based on density. The assumption is that a new behaviour is often the signature of an anomaly in the way to develop. The ESA Patent Group has decided to protect this novelty detection based monitoring technique by filing an international patent application.
Novelty Detection has been developed in ESOC by the ESA Advanced Mission Concepts and Technologies Office.
The most widely extended approach for automatically detecting anomalous behaviour in Space Operations is the use of Out-Of-Limits (OOL). The OOL approach consists of defining an upper and lower threshold so that when a measurement goes above the upper limit or below the lower one, an alarm is triggered for engineers. Then engineers will inspect the parameter that is out of limits and determine if it is an anomaly or not and decide which action to take.
While Out-Of-Limits are useful and they successfully trigger alarms when parameter readings go out the defined thresholds, they suffer from the following limitations:
- Some behaviours are anomalous even if they are within the defined limits.
- OOL are not defined for every parameter. Engineers only define OOL for a subset of parameter for which they want to receive alarms if they exceed the limits. Therefore, OOL is not systematic in the sense that it does not cover every parameter.
- Quite often engineers receive OOL alarms that are completely expected. A typical example is the OOL defined for the Automatic Control Gain (AGC) during a pass. At Acquisition of Signal (AOS) and Loss of Signal (LOS) the AGC goes outside limits. However, it is expected to happen and in every pass these two OOL alarms will be raised.
- It requires effort to adapt OOL to useful values as the mission goes through different phases or simply ages.
The Novelty Detector project has been developed to cope with the current OOL limitations. Its main goal is to automatically detect anomalies and report them to engineers for further investigation. The Novelty Detector characteristics are the following:
- Systematic: it scans every parameter in search for novel behaviour.
- Realistic: it takes into account that parameters can behave nominally in several different ways.
- Robust: it does not make any assumption on what kind of behaviour or how many different behaviours a parameter will have.
- Versatile: it works with any kind of parameter.
- No prior knowledge required: the only inputs required are a time period that makes sense (e.g. orbital period) and examples of nominal periods.
The aim with the Novelty Detector was to automatically identify anomalies. However, we did not reach that far yet. At this stage we can only detect new behaviour. In this sense we can state: “What is happening today in parameter P never happened before”. In most of the cases this signals an anomaly, but could also mean a new nominal behaviour.
The Novelty Detection is a MUST Client in the sense that it uses time series data provided by the MUST repository. However, it can be easily adapted to work with any kind of time series data.
- Martinez-Heras, J.A., Donati, A., Kirsch, M. G. F., Schmidt, F., New Telemetry Monitoring Paradigm with Novelty Detection; In SpaceOps 2012 Conference, Stockholm, Sweden, June 11 – 15, 2012