Temporal dependence Mahalanobis distance for anomaly detection in multivariate spacecraft telemetry series

Area of Science:

• Spacecraft engineering
• Data science
• Time series analysis

Background:

• Spacecraft telemetry data are crucial for monitoring health and performance.
• High-dimensional, dependent, and pseudo-periodic telemetry data challenge traditional anomaly detection.
• Existing Mahalanobis distance (MD) methods lack temporal evolution capture, leading to detection inaccuracies.

Purpose of the Study:

• To develop an advanced anomaly detection method for multivariate spacecraft telemetry data.
• To address the limitations of fixed-threshold MD methods in capturing temporal dynamics.
• To enhance the detection of contextual and collective anomalies in real-time spacecraft operations.

Main Methods:

• Implementation of temporal dependence Mahalanobis distance (TDMD) using multi-factor prediction.
• Construction of dynamic upper and lower bounds for MD based on time series correlation.
• Online testing and validation using simulated and real spacecraft telemetry data.

Main Results:

• TDMD effectively detects contextual and collective anomalies in multivariate telemetry series.
• The proposed method demonstrates improved accuracy compared to traditional fixed-threshold approaches.
• Experiments confirm the effectiveness and applicability of TDMD on diverse telemetry datasets.

Conclusions:

• TDMD offers a robust solution for anomaly detection in complex spacecraft telemetry.
• The method enhances spacecraft health monitoring by accurately identifying subtle and evolving anomalies.
• TDMD is a valuable tool for ensuring the reliability and safety of space missions.