Abstract

In this report, we present a simple yet powerful approach for combining dynamic system information with historical information and beliefs about the failure behavior of the grid system modules for locating faulty components.. In this approach, the failure of each component within the grid is considered to be an event with an associated probability distribution. Bayesian Belief Networks are used to update these probability distributions as new information is collected about the behavior of the system. The updated probabilities reflect the most likely cause of a failure. We present a thorough literature review of Fault Diagnostics Techniques and Bayesian Belief Networks (BBN). We consider the architecture of the GriPhyn project and specify the changes that need to be made to the architecture to incorporate the necessary modules for the BBN analysis. Further, we demonstrate the application of BBN’s for the fault diagnosis of grid systems using two examples. The first example is solved manually to show the underlying techniques and the second using the Hugin Expert software package. For the second example, we simulate a data transfer operation on the grid and present some initial experimental results.

Date

2002

Authors

Leila Meshkat, William Allcock, Ewa Deelman, Carl Kesselman

Publisher

Technical Report GriPhyN-2002-8, GriPhyN Project