HP Labs Technical Reports

TRANSCEND: A System for Robust Monitoring and Diagnosis of Complex Engineering Systems

Manders, Eric; Mosterman, Pieter; Biswas, Gautam; Barford, Lee

HPL-1999-37

Keyword(s): diagnosis; signal processing; abrupt faults; qualitative reasoning; change detection; fault detection; fault identification

Abstract: This paper describes TRANSCEND, our system for monitoring and diagnosis of abrupt faults in complex dynamic systems. The key to this work has been our ability to model the transient behavior in response to faults in a qualitative framework, which enables us to overcome significant complexity and convergence issues that arise in numerical processing, especially for non linear systems. In our qualitative framework, the predicted future behavior of hypothesized faults is captured in the form of signatures, and analyzed by a progressive monitoring scheme. However, generating qualitative features from real signals, the signal to symbol transformation problem, is a challenging task. This paper discusses model-driven methods for generating symbolic feature descriptions of magnitude and slope changes in noisy, continuous data. To test our integrated framework for monitoring, prediction, and diagnosis, we have developed an automobile engine testbed. Sensors installed on the engine are connected to a PC workstation through a real-time data acquisition system. Experiments have been successfully conducted on faults introduced into the cooling system of the engine.

27 Pages

Back to Index