Abstract
Techniques are proposed herein for detecting and validating recurring failure patterns across a fleet-scale corpus of causal event graphs. A pattern-discovery engine groups taxonomy-normalized node labels and categories, computes support as the number of distinct incident graphs containing a pattern, and classifies retained patterns into new taxonomy categories and known but unvalidated patterns. Engineers confirm or reject ranked patterns through a human-in-the-loop workflow. Confirmed patterns are recorded in a persistent validated-pattern registry that is structurally independent of the graph index and reapplied after index rebuilds. The registry acts as a partial graph coloring and active-learning oracle: future discovery operates on its complement, so validated patterns are suppressed and the unvalidated frontier is monotonically non-increasing. A complementary field-analysis (FA) layer counts recurring outcome-field values and supports cross-tier queries that correlate causal graph patterns with failure modes, closure codes, case status, and component part numbers.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Gupta, Nilay and W Chen, Wayne, "SELF-EXHAUSTING FLEET-LEVEL FAILURE PATTERN DETECTION VIA PERSISTENT GRAPH COLORING AND HUMAN-IN-THE-LOOP VALIDATION", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/11523