Diagnostic alarm sequence maturation in timed failure propagation graphs

Diagnostic model development presents a significant engineering challenge to ensure subsequent diagnostic processes using such models will yield accurate results. One approach to developing system-level diagnostic models that has been receiving attention is the Timed Failure Propagation Graph (TFPG), developed at Vanderbilt University. Unfortunately, developing TFPG models is also difficult and error-prone. In this paper, we extend previous work in using historical maintenance and diagnostic information to identify potential errors in the TFPG-based diagnostic models and recommend ways of maturing these models. This is done by extending the maturation process to incorporate historical alarm sequences and to model these sequences using a probabilistic transition matrix (similar to a Markov chain). The resulting sequence model is compared to the causal relationships identified in the original TFPG to discover discrepancies between the two. Potential sequence modeling errors with recommendations are given back to an engineer or analyst. We report on the maturation process and algorithms and also provide preliminary experimental results.