Novel Methodology for Partitioning Complex Systems for Fault Diagnosis Purposes
Abstract
The criterion used for partitioning is d-separation: a parallel between causal independency property and vertex separation in digraphs. If X, Y and Z represent the vertex subsets of two neighbouring partition regions and respectively the border between them, the d-separation criterion is used to decide if “knowing Z renders Y irrelevant to X”. It follows that diagnosis may be performed locally, inside each region, without communicating, via partition borders, with other regions. If borders are affected by faults, communication is needed. The described partitioning provides minimal borders between regions. It follows that communication process has minimal computational complexity.
