AI quality inspection uses cameras and deep-learning models to detect visible defects and classify each unit as OK or not OK at line speed, keeping an image record of every decision.
Steps
- Defect catalogue: which defects are critical, which cosmetic; example images for each.
- Camera and lighting: the smallest relevant defect determines resolution and optics; lighting is chosen for the surface.
- Data and labelling: good and defective examples are labelled; for rare defects, anomaly detection trained on good parts is used.
- Validation: the model is tested on images it has not seen; escapes and false rejects are reported separately.
- Integration: reject signals to the PLC, operator alerts and records in MES/ERP.
Metrics that matter
| Metric | Meaning | Why it matters |
|---|---|---|
| Escape rate | Defective units passed as OK | Customer complaints and returns |
| False reject rate | Good units rejected | Unnecessary scrap and re-inspection |
| Cycle time | Analysis time per unit | Keeping up with the line |
AI vs rule-based machine vision
Rule-based systems excel at fixed measurements and code reading. Deep learning handles natural variation — textures, castings, textiles — better. Many robust systems combine both. Solution page: visual quality inspection.
Not sure what is feasible on your site? Send a few camera frames and your use case.
Evaluate your camerasFrequently asked questions
Can existing CCTV be used for quality inspection?
Rarely. Inspection usually needs dedicated cameras positioned at the inspection point with controlled lighting.