As the volume of telemetry data generated by satellites and other complex systems continues to grow, there is a pressing need for more efficient and accurate anomaly detection methods. Current techniques often rely on human analysis and preset criteria, presenting several challenges including the necessity for expert interpretation and continual updates to match the dynamic mission environment. This paper critically examines the use of deep anomaly detection (DAD) methods in addressing these challenges, evaluating their efficacy on real-world spacecraft telemetry data. It exposes limitations in current DAD research, highlighting the tendency for performance results to be overestimated and suggesting that simpler methods can sometimes outperform more complex DAD algorithms. By comparing established metrics for anomaly detection with newly proposed ones, this paper aims to improve the evaluation of DAD algorithms. It underscores the importance of using less accuracy-inflating metrics and offers a comprehensive comparison of DAD methods on popular benchmark datasets and real-life satellite telemetry data. Among the DAD methods examined, the LSTM algorithm demonstrates considerable promise. However, the paper also reveals the potential limitations of this approach, particularly in complex systems that lack a single, clear predictive failure channel. The paper concludes with a series of recommendations for future research, including the adoption of best practices, the need for high-quality, pre-split datasets, and the investigation of other prediction error methods. Through these insights, this paper contributes to the improved understanding and application of DAD methods, ultimately enhancing the reliability and effectiveness of anomaly detection in real-world scenarios.

@en

Spacecraft systems collect health-related data continuously, which can give an indication of the systems’ health status. While they rarely occur, the repercussions of such system anomalies, faults, or failures can be severe, safety-critical and costly. Therefore, the data are used to anticipate any kind of anomalous behaviour. Typically this is performed by the use of simple thresholds or statistical techniques. Over the past few years, however, data-driven anomaly detection methods have been further developed and improved. They can help to automate the process of anomaly detection. However, it usually is time intensive and requires expertise to identify and implement suitable anomaly detection methods for specific systems, which is often not feasible for application at scale, for instance, when considering a satellite consisting of numerous systems and many more subsystems. To address this limitation, a generic diagnostic framework is proposed that identifies optimal anomaly detection techniques and data pre-processing and thresholding methods. The framework is applied to two publicly available spacecraft datasets and a real-life satellite dataset provided by the European Space Agency. The results show that the framework is robust and adaptive to different system data, providing a quick way to assess anomaly detection for the underlying system. It was found that including thresholding techniques significantly influences the quality of resulting anomaly detection models. With this, the framework can provide both a way forward in developing data-driven anomaly detection methods for spacecraft systems and guidance relative to the direction of anomaly detection method selection and implementation for specific use cases.@en