A methodology for knowledge acquisition and reasoning in failure analysis of systems

The problem of modeling knowledge about the fault behavior of a system and utilizing this model for reasoning about and diagnosing failures is addressed. A solution that merges graph and fault-tree-based failure analysis with rule-oriented reasoning is presented. Failure analysis is divided into two phases, a failure source location phase and a failure cause identification phase. Each phase consists of a failure model and a process that operates on it. The failure models for the first and second phases are based on lesel-structured fault propagation digraphs and augmented fault trees, respectively. The augmented fault tree (AFT) is a conceptual structure that encodes probabilistic, temporal, and heuristic information in addition to the causal aspects of failures modeled by conventional fault trees. The two models are combined to form a novel hierarchical failure knowledge representation scheme. Upper levels of this hierarchy are made up of the fault propagation digraphs. Each level represents a view of the system under a particular granularity, and the granularity increases with levels. This feature permits control over the resolution of fault diagnosis. The lowest level consists of a set of cause-consequence knowledge bases containing production rules. These production rules are derived from augmented fault trees and represent the cause-effect relations among failure events that lead to the corresponding subsystem's failure. A knowledge acquisition procedure to generate these failure models and failure analysis processes that operate on them are described. The methodology proposed is inherently parallel as the processes may operate on different levels independently