Subscribe to DSC Newsletter

ACM Talk in Palo Alto, CA 4/8/09: Using Supervised and Unsupervised Learning to Detect and Isolate Faults in Rocket Engines
Date: Wednesday, 08 April 2009, 6:30 PM
Location: SAP LABS, Building 1 (previously aka "Building D")
3410 Hillview Avenue, Palo Alto, CA
RSVP: Venky at [email protected] and
Greg at [email protected]

We have used two classes of algorithms to automatically detect and isolate faults in rocket engines. The first class of algorithms is known as supervised learning algorithms. Examples of supervised learning algorithms include decision trees and support vector machines. These algorithms require training data consisting of a large number of labeled examples of sensor data from both nominal operation and from failures. They learn a model that can distinguish among nominal data and data from each failure mode, and can thus perform both fault detection and fault isolation. In real rocket engine sensor data, there are not enough failures to allow supervised learning to be used, so we have only been able to use this class of algorithms with simulated data. The second class of algorithms is known as unsupervised anomaly detection algorithms. These algorithms are trained using only nominal data, learn a model of the nominal data, and signal a failure when future data fails to match the model. They are not able to identify the failure mode, but they can be trained using real data that does not include any failures. Examples of unsupervised anomaly detection algorithms include the Inductive Monitoring System (IMS), Orca, GritBot, and one-class support vector machines. We will present results of applying unsupervised anomaly detection to detecting faults in real data from the Space Shuttle Main Engine, and of applying supervised learning to detecting and isolating faults in simulated data from the J-2X rocket engine.

About the Speaker
Mark Schwabacher earned his Ph.D. in computer science in 1996 from Rutgers University . His thesis work applied artificial intelligence to engineering design. He has worked at NASA Ames Research Center for ten years, where he has worked on several systems health management activities. He served as the Software Lead of the NASA X-37 Integrated Vehicle Health Management Experiment, and is currently leading the development of the Ares I-X Ground Diagnostic Prototype in collaboration with NASA Kennedy Space Center, NASA Marshall Space Flight Center, and the Jet Propulsion Laboratory. He has also applied anomaly detection algorithms to Earth science and to aviation security.

Tags: anomaly, data, detection, engine, learning, machines, rocket, sensor, supervised, support, More…unsupervised, vector

Views: 61

On Data Science Central

© 2021   TechTarget, Inc.   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service