Failure Mode Classification Rule

There is no failure classification system. When a technician closes a work order, they write a free-text description: 'fixed the thing,' 'replaced motor,' 'adjusted.' The same bearing failure gets described as 'bearing,' 'brg replacement,' 'noise from spindle,' or 'SKF 6205 swap' by different technicians. Failure mode analysis is impossible because there's no consistent language for describing what went wrong.

AI cannot detect failure patterns because the same failure type is described differently every time. Pattern recognition across thousands of free-text descriptions yields noise, not signal.

A basic failure code list exists — 10-15 broad categories like 'bearing failure,' 'motor burnout,' 'leak,' 'sensor fault.' Technicians select a code from a dropdown when closing work orders. But the list is too coarse — 'bearing failure' covers everything from a contaminated grease nipple to a catastrophic spindle seizure. Root cause codes don't exist. 'Other' accounts for 25% of selections. The classification tells you something failed but not why or where specifically.

AI can produce Pareto charts of failure categories but cannot perform meaningful root cause analysis or failure mode effects analysis because the classification is too coarse to distinguish failure mechanisms.

A structured failure taxonomy exists with hierarchical codes: asset class (pump, motor, conveyor) → component (bearing, seal, impeller) → failure mode (wear, contamination, overload) → root cause (inadequate lubrication, foreign object, thermal stress). Technicians navigate a guided selection tree. The taxonomy is documented and consistent. But it's a standalone reference — not linked to equipment BOMs, maintenance procedures, or historical failure data for validation.

AI can perform hierarchical failure analysis — trending failure modes by component type and root cause. Cannot validate classification accuracy because the taxonomy isn't linked to equipment data that could confirm or contradict the technician's selection.

The failure classification system links to equipment context. When a technician reports a failure on Pump 7, the system presents only the failure modes valid for that pump type's components. Root cause selections link to the maintenance procedures that address them. Historical failure records for similar equipment validate that the selected classification is consistent with past patterns. The classification becomes context-aware rather than a generic code list.

AI can perform context-validated failure classification — confirming that the technician's selection is consistent with the equipment type and historical patterns. Can identify misclassifications and suggest corrections.

The failure classification is a formal FMEA ontology with machine-readable relationships. Every failure mode maps to specific degradation mechanisms, physical root causes, detectable precursor signatures, and effective preventive actions. An AI agent can ask: 'for bearing contamination failures on high-speed spindles, what are the detectable vibration signatures that precede the failure by 2-4 weeks, and what maintenance actions have the highest prevention success rate?' and compute the answer from the formal ontology.

AI can perform comprehensive reliability engineering from the formal failure ontology — automated root cause analysis, failure prediction from precursor signatures, and optimized maintenance strategy selection.

The failure classification taxonomy is self-evolving. When a failure occurs that doesn't match existing categories, the system creates a provisional new failure mode, links it to the equipment and conditions, and seeks similar patterns across the fleet. When enough similar incidents accumulate, the system proposes a formal taxonomy addition. When failure modes become obsolete (equipment retired), the taxonomy prunes itself. The classification system grows and adapts with the maintenance operation's learning.

Fully autonomous failure classification management. AI maintains a self-evolving taxonomy that discovers new failure modes, validates existing classifications, and optimizes the classification structure from operational data.

Ceiling of the CMC framework for this dimension.