Computer vision inspection is the automated checking of parts or products for defects using cameras and image analysis, increasingly with deep learning models. It replaces or supports manual visual inspection on a line.
An inspection system captures images at a fixed station or on a moving line, then analyzes them to decide pass or fail and, often, the defect type. Two broad approaches exist. Rule-based: measure edges, compare with a golden template or CAD, apply tolerances. Learning-based: train a model on labeled images of good and bad parts, or on good parts only for anomaly detection.
A well-known example is automated optical inspection (AOI) of printed circuit boards, which checks for missing components, solder bridges, skew, and wrong parts. Deep learning is now common for surface defects on textured or variable materials such as castings, fabric, wood, or food, where rules break down.
Limits: models need representative defect images, which are rare by definition. False rejects cost product and false accepts cost customers, so thresholds are a business decision. Changes in lighting, supplier, or product variant can degrade accuracy. A human review step for borderline cases is common.
Questions about Computer vision inspection
What is the difference between rule-based and learning-based inspection?
Rule-based inspection measures edges, compares the part with a golden template or CAD and applies tolerances. Learning-based inspection trains a model on labeled images of good and bad parts, or on good parts only when it is set up as anomaly detection.
What do you need before deep learning inspection works?
Representative images of defects, which are rare by definition. You also need to decide the thresholds as a business matter, since false rejects cost product and false accepts cost customers. Borderline cases are commonly sent to a person for review.
Computer vision inspection in real plants
- La Próxima Estación and KEYENCE: La Próxima Estación runs two KEYENCE IV4 AI vision sensors on its line in Irapuato to detect foreign matter and off-color vegetables, linked to a rejection system it built itself.
- Toast and Instrumental: Toast installed 6 Instrumental stations across 2 hardware programs to inspect each assembly stage of its restaurant point-of-sale hardware at its factory partners.
- NVIDIA and Instrumental: NVIDIA uses Instrumental's visual AI across its manufacturing network to catch assembly errors on GB200 and GB300 NVL72 server trays before functional test.
- BorgWarner and Cognex: BorgWarner's Changnyeong plant inspects EV power train assemblies with Cognex 3D vision and checks Park-Lock assembly with In-Sight 2800 edge learning systems.
- Audi and Siemens: Audi and Siemens brought the AI-based Weld Splatter Detection check into series production in the Neckarsulm body shop; it detects possible weld splatter on vehicle underbodies.
Vendors linked to Computer vision inspection
Eigen Innovations (3) · Instrumental (3) · Cognex (2) · MVTec Software (2) · Elementary (1) · FANUC (1) · KEYENCE (1) · OMRON (1) · Siemens (1) · Teledyne DALSA · UnitX
Related terms
Machine vision
Machine vision is the use of cameras, lighting, and image-processing software to inspect, measure, identify, or guide automatically in industrial settings.
Anomaly detection
Anomaly detection is the identification of observations or patterns that deviate significantly from normal behavior in data.
Synthetic data
Synthetic data is artificially generated data that mimics some statistical properties of real data.
Statistical process control (SPC)
Statistical process control (SPC) is the use of statistical methods, mainly control charts, to monitor a production process and detect when its variation goes beyond what its normal, stable behavior would produce.
Root cause analysis
Root cause analysis (RCA) is a structured method for finding the underlying cause of a failure or defect, so the corrective action prevents recurrence instead of only treating the symptom.