CONTINUOUS SURFACE INSPECTION: VALIDATE DEFECTS AND COVERAGE
Yenra | Updated September 11, 2026
Article: https://yenra.com/surface-inspection-system/

Purpose: connect a continuous-surface inspection specification to validation and material disposition.
Use with agreed product acceptance rules, qualified reference labels, actual camera documentation and an approved test plan. Keep evaluation samples separate from tuning/training material.

DEFECT CATALOG AND COVERAGE
Material/grade/finish range / width mm / speed profile and units:
Defect class | size/severity/orientation/location | acceptance boundary | reference sample:
Acceptable texture and borderline cases / label reviewers / disagreement resolution:
Test variation: speed, flutter, edge position, working distance, contamination, duration:
Sampling unit (segment, defect, area or other) / independence / selection method:

IMAGING CONFIGURATION
Camera/lens/lighting identities / software/model/threshold versions:
Working distance / focus / exposure us / gain / pixel format / interface:
Cross-web field mm / pixels / travel spacing mm per line:
Encoder scaling / trigger / supported resulting line rate / processing capacity:
Fictional geometry: 1,000 mm / 4,096 pixels = about 0.244 mm/pixel.
At 2 m/s = 2,000 mm/s and 0.25 mm/line, required rate = 8,000 lines/s.
At 50 us exposure, travel = 2,000 * 0.000050 = 0.10 mm.
These are planning calculations, not a smallest-detectable-defect claim. Validate contrast, blur, focus, noise, orientation and decision behavior.

ERROR RECORD (repeat by class and relevant production condition)
Actual defective: flagged TP ___ ; missed FN ___ ; total TP+FN ___
Actual acceptable: flagged FP ___ ; accepted TN ___ ; total FP+TN ___
Detection = TP/(TP+FN); miss rate = FN/(TP+FN).
False-reject rate = FP/(FP+TN); correct flags = TP/(TP+FP).
If a denominator is zero, report undefined, not zero percent.
Fictional held-out set: TP 92, FN 8, FP 18, TN 882.
Detection 92%; miss 8%; false reject 2%; correct flags 92/110 = about 83.6%.
These are inspection segments under an agreed label rule. Class prevalence affects the proportion of flags that are correct. Record uncertainty and sampling limits.

LOCATION, FAILURES AND DISPOSITION
Roll/coil ID / coordinate origin / cross-web and travel positions / image and timestamp:
Splice and restart behavior / marker or reject-station offset verification:
Lost lines/images/triggers/encoder/processing tests and coverage-gap handling:
Affected-material hold or disposition / quality owner / acceptance evidence:
Reviewer / unresolved cases / release decision / date:
Revalidate affected conditions after optical, speed, material, software or threshold changes.

PRIMARY REFERENCES
Edmund Optics' illumination-geometry explanation
https://www.edmundoptics.com/knowledge-center/video/eo-imaging-lab/eo-imaging-lab-the-w-of-illumination-geometry/
Basler's line-scan use cases
https://docs.baslerweb.com/line-scan-gige-use-cases
resulting line-rate documentation
https://docs.baslerweb.com/resulting-acquisition-line-rate