What types of defects can AI detect in photos?

Surface defects (scratches, cracks, color deviations, inclusions), dimensional deviations (position and size measurement of features), assembly errors (missing component, incorrect orientation, wrong part substitution), and printing/laser errors (unreadable code, incorrect label).

Can I submit photos manually instead of using a line camera?

Yes — the system accepts photos uploaded via mobile or tablet. Suitable for custom production or incoming inspection. For serial production above 200 pieces/day we recommend an industrial camera on the line.

How accurate is the AI model?

After training on 300+ samples (both good and defective pieces) it achieves 95–99% accuracy. False-positive rates (good pieces rejected as defective) are configurable based on your quality tolerance.

AI Visual Defect Detection: quality under control on every piece

Manual visual quality inspection has a systemic problem: a fatigued worker misses 15–25% of defects after 4 hours. Your customer finds the defect instead of you — and you pay with a warranty claim, a return and damaged reputation.

Types of defects AI detects

Surface defects

Scratches, cracks, gouges
Discoloration and color deviations (color tolerance configurable in dE units)
Surface roughness out of tolerance
Point contamination, bubbles, inclusions

Dimensional deviations

Hole in wrong position (position measurement with ±0.1 mm accuracy)
Dimension out of tolerance
Warping, bending, deformation

Assembly errors

Missing component (screw, gasket, label)
Incorrect part orientation
Wrong part — variant substitution (e.g. M5 vs. M6 screw)

Identification and traceability

Readability of QR code, barcode, laser-engraved serial number
Correctness of label and printed text

How the system works

Photo capture or upload

A photo can come from an industrial camera on the production line, from a mobile application (for incoming inspection or custom production), or from a flatbed scanner (for flat parts).

AI inference in milliseconds

Part: HŠ-2245-B | Batch: J-226 | 03.05.2026 14:32:11

INSPECTION RESULT: ❌ NOK

Detected defects:
  1. Scratch — location: left side, size: 12×0.3 mm
     Severity: CRITICAL (exceeds 8 mm tolerance)
  2. Color deviation — dE = 4.2 (tolerance: 3.0)
     Severity: WARNING

Recommended action: SET ASIDE for manual review

Statistics and trends

The system logs every inspected piece. Result: “2.8% of pieces in batch J-226 had a scratch on the left side — it occurred between 14:00 and 15:30.”

ROI for a manufacturing company

Scenario	Without AI	With AI
Manual inspection (8h/day)	2 FTE × €1,200/month = €2,400	0.2 FTE oversight
Defects missed	15–25%	<3%
Warranty claims/year (1,000 pcs/day)	30–50 cases	3–5 cases

Investment in AI visual inspection: €8,000–25,000. Payback period: 3–8 months.

AI Visual Defect Detection — quality control without manual inspection