Vision
Deep-learning defect detection
Vision inspection approach that uses trained neural network models to flag defects that are hard to describe with traditional rule-based tools.
Traditional vision tools rely on programmed rules like edge contrast or blob area, which struggle with defects that vary in shape, such as scratches, dents, or inconsistent flash on molded parts. Deep-learning tools instead train a model on labeled example images of good and bad parts, then classify new images based on learned patterns.
Cognex, Keyence, and several dedicated software vendors now offer deep-learning modules that integrate with their existing camera hardware, so the trained model runs alongside conventional vision tools in the same inspection station.
Model accuracy depends on the size and quality of the training image set. A change in part design or lighting conditions can require retraining, so this approach carries more ongoing maintenance than a fixed rule-based tool.
Related integrations
Common questions
- What does Deep-learning defect detection do?
- Vision inspection approach that uses trained neural network models to flag defects that are hard to describe with traditional rule-based tools.
- What does Deep-learning defect detection work with?
- It works with Cognex In-Sight, Keyence vision, Basler cameras, MES integration.
- What should I watch for with Deep-learning defect detection?
- Needs a substantial labeled training image set to start. Model retraining required after part or lighting changes. Explaining a rejection to operators is harder than with rule-based tools.