Definition
The rate and magnitude of disagreement among multiple AI models monitoring the same equipment, measured as the proportion of assessment windows in which models issue materially conflicting condition assessments or recommendations.
The Measurement Gap
There is no metric that tells you how often independent AI models monitoring the same equipment disagree with each other, which means conflicting machine assessments are resolved ad hoc by whoever notices, without any record of which model was right.
Formula
Plain language: For each asset monitored by two or more models, divide time into assessment windows. In each window, compare the models' condition assessments on a common severity scale. Count windows where any pair of models differs by more than one severity band, or where recommendations conflict (one says run, another says intervene). EMFI is conflicting windows over total multi-model windows.
Notation:
EMFI = W_conflict / W_multi × 100
Variables:
- W_conflict = assessment windows with a material inter-model conflict: severity delta > 1 band, or contradictory recommendations (count, DERIVED)
- W_multi = total assessment windows where 2+ models produced an assessment for the same asset (count, DERIVED from model output logs)
- Severity mapping = each model's native output normalized to a common 4-band scale (normal / watch / plan intervention / intervene now) (mapping, MANUAL: defined once per model)
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
When the vibration model, the thermography model, and the process-data model all watch the same compressor, are they telling one story or three, and who is accountable for reconciling them before an operator has to choose?