NASA Logo, National Aeronautics and Space Administration

+NASA Home

+Ames Home

Automatic Anomaly Detection Knowledge Base Updates Demonstrated Using Live Orion EFT-1 Data Feed
Intelligent Systems Division Banner

Automatic Anomaly Detection Knowledge Base Updates Demonstrated Using Live Orion EFT-1 Data Feed

A filtering scheme to construct and update the Orion battery subsystem of the Inductive Monitoring System (IMS) Knowledge Bases (KBs) was demonstrated during the Orion Exploration Flight Test 1 (EFT-1) mission using a live telemetry data feed. An initial battery system KB was constructed using Orion simulation and test data. This KB was used to monitor the first 15 minutes of Orion EFT-1 in-flight data and the results were automatically filtered to produce a baseline KB based solely on telemetry data (disregarding previous simulation and test data). Throughout the course of the mission subsequent 15-minute periods were monitored with the most recently created KB with the filtered results incorporated to produce the next KB. IMS Outlier Detection Via Estimating Clusters (ODVEC) monitoring results for each hour were graphically displayed for comparison.

Preliminary assessment showed the filter and update procedures were effective in producing viable KBs for mission monitoring. Monitoring deviation scores generally decreased throughout the mission, indicating successful model refinement, but small data variations were still clearly reflected in the ODVEC results, showing that sensitivity was maintained in the automatically updated models. Further analysis of the data filtering and model updates will be performed to confirm initial observations, but these first results show promise for the approach.

BACKGROUND: The Spacecraft Autonomous Fault Evaluation and Management (SAFEM) task is developing capabilities to support next-generation mission control operations and enable autonomous spacecraft operations. An important step toward this goal is to reduce the effort required to develop anomaly detection models. This will allow the data-driven model update process to occur onboard spacecraft, enabling incorporation of all spacecraft data — including non-telemetered data, and facilitation of crew updates of the KBs describing nominal operations as the vehicle flies toward an exploration target. The same technology can facilitate KB updates in Mission Control by flight controllers — at the touch of a button — using telemetered data. SAFEM is extending the Inductive Monitoring System (IMS) software (known as the Anomaly Monitoring Inductive Software System, i.e., AMISS, at JSC and currently used by Mission Control Center flight controllers to monitor the International Space Station), particularly IMS’ ODVEC algorithm.

As a step toward automated IMS model updates, or “online” learning, data filtering functions were developed to detect and remove suspected non-nominal system data samples based on ODVEC results. These filter functions can be applied to real-time data to clean the data stream for subsequent incorporation into the IMS KB for that system. The filtering scheme was used to construct and update Orion battery subsystem KBs during the EFT-1 mission using a live telemetry data feed.

TEAM: David Iverson and Lilly Spirkovska

NASA PROGRAM FUNDING: Autonomous Systems (AS) project, Game Changing Development (GCD) program, Space Technologies Mission Directorate (STMD). Real-time data was provided through the ACAWS demonstration task of the Autonomous Mission Operations (AMO) project (Advanced Exploration Systems program, Human Exploration and Operations Mission Directorate)

Contact: David Iverson

First Gov logo
NASA Logo - nasa.gov