Anomaly Detection via Topological Feature Map

We propose a machine-learning technology that significantly expands NASA’s real-time and offline ISHM capabilities for future deep-space exploration efforts. Our proposed system, Anomaly Detection via Topological fEAture Map (AD-TEAM), will leverage a Self-Organizing Map (SOM)-based architecture to produce high-resolution clusters of nominal system behavior. What distinguishes AD-TEAM from more common clustering techniques (e.g., k-means) in the ISHM-space is that it maps high-dimensional input vectors to a 2D grid while preserving the topology of the original dataset. The result is a ‘semantic map’ that serves as a powerful visualization tool for uncovering latent relationships between features of the incoming points. Thus, beyond detecting known and unknown anomalies, AD-TEAM will also enable space crew to semantically characterize the clusters discovered. In doing so, personnel will better understand how faults propagate throughout a system, the transitional states of subsystem degradation over time, and the dominant features (and their relationships) of subsystem behavior. In addition to analyzing single subsystem datasets, we also propose to cross-correlate subsystems in order to capture the cascading effect of faults from one subsystem to another, as well as discover latent relationships between subsystems. Such analysis would significantly aid in the maintenance and overhauling activities of NASA’s deep-space missions.