Real-time Anomaly Detection (RAD)

The Satellite Control Network (SCN) tracks and communicates with military and civilian satellites. Equipment failures result in several costs: the monetary cost of repair/replacement, the time and labor of technicians, and the operational cost of decreased functionality of damaged or degraded equipment (ranging from a slight dysfunctional antenna to a shutdown of multiple antennas). Our proposed artificial intelligence (AI) software would reduce all these costs. Our software aims to predict faults before they happen (including predicting Remaining Useful Life (RUL)), diagnose faults quicker than may be possible with human inspection, and recommend actions to schedulers if appropriate. To accomplish these goals, we propose Real-time Anomaly Detection (RAD). RAD builds on projects that we have already applied to other domains (DoD & non-DoD): Anomaly Detection via Topological feature Map (AD-TEAM); Management of consumables Adaptive Execution, SynchronizaTion, Replanning/rescheduling, Optimization system (MAESTRO); and Aurora, our intelligent scheduling and resource optimization architecture (see Related Work). RAD utilizes a combination of Self-Organizing Maps (SOMs), Case-Based Reasoning (CBR), Model-Based Reasoning (MBR), and Aurora’s near-linear scheduling and resource allocation algorithm. It will interface with the existing SCN system so that its predictions, detections, diagnoses, recommendations, and estimated RULs are presented in real-time to the operators/schedulers.