Definition
The average ratio of AI-detectable failure modes to total required failure modes across all inspected zones, measuring whether the computer vision model covers the full scope of damage types mandated by inspection standards. A score of 1.00 means the model can detect every required failure mode in every zone; a score below 0.70 means the model is a supplement to human inspection, not a replacement.
The Measurement Gap
Standard CV validation reports tell organizations whether the model performs well on detected defects. They do not tell organizations whether the model is looking for the right defects. A CV model may achieve 94% accuracy on surface corrosion detection while being entirely unconfigured for the fatigue crack signatures, disbond indicators, or lightning strike damage patterns also required by the inspection standard for that zone. Legacy quality metrics (detection accuracy, false positive rates, system uptime) do not ask this question.
CVFMCS surfaces the coverage gap directly. It compares the set of failure modes the CV model is trained and validated to detect against the set of failure modes inspection standards require to be detected, zone by zone. This is not a model performance question: it is a model scope question. An organization with a high-accuracy but low-coverage CV model is systematically under-inspecting its aircraft without knowing it. Regulatory approval of CV inspection assumes scope coverage; CVFMCS provides the evidence that scope coverage exists or quantifies the gap where it does not.
Formula
Plain language: For each inspected zone, divide the number of failure modes the CV model can detect by the number of failure modes the applicable inspection standard requires to be detected. Average these ratios across all zones to produce the score.
Notation:
CVFMCS = (1/Z) × Σ (M_detectable_z / M_required_z)
Variables:
| Variable | Definition | Source |
|---|---|---|
| Z | Total number of CV-covered inspectable zones | Zone mapping document |
| M_detectable_z | Failure modes in zone z the CV model is validated to detect | CV model validation report |
| M_required_z | Failure modes in zone z required by inspection standard | AMM/SRM/inspection task cards |
Calculation notes: "Validated to detect" requires documented evidence. Training data confirmation, detection threshold specification, and validation test results. A failure mode where the model was trained but not formally validated counts as not detectable for scoring purposes. The denominator M_required_z must be derived from the current revision of the applicable inspection document, not from historical task cards.
Proposed formula, pending validation pilots.