The study focuses on analyzing anomalies in telemetry data from the EDEN ISS bio-regenerative life support system (BLSS) prototype during its third operational year in 2020. The authors employ two unsupervised anomaly detection methods, MDI and DAMP, to identify anomalous subsequences in the univariate and multivariate time series data from the EDEN ISS subsystems.
To categorize the detected anomalies, the authors extract four sets of features from the anomalous subsequences and apply K-Means clustering and Hierarchical Agglomerative Clustering (HAC). The quality of the clustering results is evaluated using the Silhouette Score and a novel Synchronized Anomaly Agreement Index (SAAI) that assesses the temporal alignment of anomalies assigned to the same cluster.
The results show that the MDI and DAMP methods produce complementary anomaly detection results, with DAMP identifying the majority of anomalies. K-Means clustering outperforms HAC in generating more balanced cluster sizes, which is desirable for isolating diverse anomaly types.
The analysis identifies several anomaly type candidates, including peaks, anomalous day/night patterns, drops, and delayed events. Some of these anomaly types exhibit recurring behavior, warranting further investigation into their underlying causes. The authors also discuss the challenges in interpreting multivariate anomalies due to the diverse sensor readings in each subsystem.
The insights gained from this study are crucial for refining the risk mitigation system for future BLSS iterations, as it helps identify systematic issues that could impact food production and operational efficiency in long-duration space missions.
Ke Bahasa Lain
dari konten sumber
arxiv.org
Wawasan Utama Disaring Dari
by Ferdinand Re... pada arxiv.org 09-27-2024
https://arxiv.org/pdf/2406.09825.pdfPertanyaan yang Lebih Dalam