Definition
The average number of scan cycles or calendar days between a defect's estimated first detectability and its actual first detection by the automated inspection system.
The Measurement Gap
The machine vision system misses 3% of defects per scan, which appears acceptable. But the same defect passes through the portal 40 times between overhauls. The probability of catching it on the first pass, when it is still minor, is low. Defects progress from minor to critical between detection events. Legacy metrics measure detection rate per scan, not per defect lifecycle. Bad-order ratio counts the eventual catch, not the delay.
Decision Supported:
Is the vision/detection system catching defects when they are still minor, or is it consistently missing them on early passes and only flagging them once they have progressed to a more severe state?
Formula
Plain language: For each defect detected by the automated system, estimate when it first became detectable (based on defect progression models, severity at detection, and scan frequency). Count scan cycles between estimated first detectability and actual detection. Average across all defects.
Notation:
DDL = SUM(D_detected_i - D_estimatedfirst_i) / N_defects
Variables:
| Variable | Description | Unit |
|---|---|---|
| D_detected_i | Date/scan cycle of actual detection for defect i | Date or count |
| D_estimatedfirst_i | Estimated date/scan cycle when defect first became detectable (back-calculated from severity at detection and progression rate) | Date or count |
| severity_at_detection | Measured severity when detected (wear depth, crack length, deformation) | Varies by defect type |
| progression_rate | Engineering-estimated progression rate for the defect type | Units/day |
| N_defects | Total defects detected in evaluation period | Count |
Sign Convention: Lower is better. 0 = defects caught at first opportunity.
Proposed formula, pending validation pilots.