← All resources
Technology comparison
Rule-based AOI versus deep-learning inspection
We already have AOI equipment. Is there a reason to move to AI inspection?
The difference shows up when the lighting shifts or a new defect appears. Rule-based means rewriting the algorithm; deep learning means training it. The case is strongest on lines where over-detection already sends work back to a human.
Terms
- AOI (Automated Optical Inspection)
- Equipment that photographs a part and judges its appearance
- Rule-based
- Judgement by human-defined conditions — thresholds, shape rules
- Over-detection
- Good parts judged defective. Adds re-inspection labour
- Under-detection
- Defective parts judged good. Leads to escapes
- Segmentation
- Splitting the defect region pixel by pixel to yield position and area
What actually differs
- Lighting and position change
- Rule-based is sensitive / deep learning is robust
- New defect type
- Rule-based needs an algorithm update / deep learning trains on ~5 images
- Complex textures (BGA, QFN)
- Rule-based struggles to classify / deep learning classifies
- Quantifying a defect
- Hard for rule-based / deep learning yields area and position per pixel
- Detection resolution
- Rule-based typically needs 5+ pixels / DSTEK resolves at 2–3
When not to change
- A single product, fixed lighting and an unchanged defect set for years — rule-based is sufficient.
- Processes that only measure dimensions, with no appearance judgement, likewise.
- Conversely, if product variants change often, defect types keep growing, or people are tied up re-checking over-detections, the case is clear.
Related products and solutions
Resources