Glossary · Machine vision
Defect detection
Also known as: Surface inspection, Flaw detection
German: Fehlererkennung
In manufacturing and machine vision, defect detection is the automated identification of deviations such as scratches, cracks, dents, contamination or missing features on products, using rule-based image processing or machine learning models trained on examples.
- Machine vision
- AI
In one sentence
Defect detection finds scratches, cracks, contamination or missing features on products with rule-based vision or machine learning.
Example
A deep learning model detects porosity in die-cast housings that rule-based tools could not separate from the normal casting texture.
How it applies
- Engineering: Rule-based methods work when defects have clear contrast; machine learning helps with variable textures. Both depend on lighting that makes defects visible.
- Validation: Performance is measured on representative sets of good and defective parts. For machine learning, the Training data and test data must cover the product variants and defect types that occur in production.
- Documentation: The documentation team should document defect classes with images, the evaluation data and results, the model or rule version, and the procedure for retraining or adjusting limits. See Evaluation (evals) for how model quality is assessed.
Defect detection vs. anomaly detection
Defect detection looks for known defect types. Anomaly detection flags anything that deviates from learned normal parts and can find unknown defects, but it cannot say which defect it found.