Find visible damage before downstream operations.
Inspecting near the process that creates the surface can support earlier review, containment and process investigation before more value is added.
Use Case
AI surface defect detection for manufactured parts
Give quality and manufacturing teams a consistent way to identify configured scratches, cracks and surface discontinuities under controlled inspection conditions.
Direct answer
AI surface defect detection learns the visible appearance of acceptable parts and configured discontinuities, such as scratches, cracks or marks, under controlled imaging conditions. United Vision helps plant teams capture examples, train the inspection, deploy the model near the line and update it as parts, finishes or defect definitions change.
i
Production qualification matters. This use case must be evaluated with representative evidence, the intended equipment, defined acceptance criteria and the required plant response.
Inspection sequence
Expose and locate scratches under controlled raking light.
The film shows why illumination is part of the inspection, not decoration. The gear is presented in a repeatable position, directional light reveals subtle discontinuities, and the configured model identifies the regions that require a plant-defined response.
- Gear indexes beneath fixed cameras
- Controlled light reveals surface discontinuities
- United Vision identifies each scratch location
Operational value
Why plants evaluate AI surface defect detection
Surface decisions often involve natural material variation, changing finishes and defect shapes that are difficult to express as a fixed rule. A trained visual model can support a more adaptable inspection when the signal is made consistently visible.
Learn the plant definition of an unacceptable surface.
Representative examples help separate relevant scratches, cracks or marks from acceptable texture, machining patterns, reflections and cosmetic variation.
Show where the condition appears.
Classification and localization can present the affected region for review, correction and a more useful production record when the workflow is configured to retain it.
Plant-owned lifecycle
How United Vision makes surface inspection plant-owned
United Vision connects this use case to the same plant-controlled workflow used to implement, update and extend visual inspection across the factory.
Capture
Collect acceptable and suspect surfaces with the production camera, optics, light and part presentation.
Define
Let plant experts label defect types, severity boundaries and acceptable natural variation.
Train
Train the visual model and challenge it with withheld parts, finishes and borderline conditions.
Deploy
Run the approved model on a supported edge device and connect the result to the chosen response.
Maintain
Add new defect appearances and revalidate when material, tooling, finish or lighting changes.
Explore the complete plant-controlled inspection workflow, deployment-readiness questions and customer data and IP position.
Qualification
What determines whether surface defect detection will work?
A defect must be optically visible and distinguishable from acceptable texture across the full production range.
Design light around the surface physics.
Raking, diffuse, coaxial or other controlled illumination should be evaluated against reflectivity, curvature, finish and the orientation of the smallest relevant discontinuity.
Keep the inspection view repeatable.
Qualify part position, rotation, focus, motion, field of view and occlusion so the model receives comparable evidence at the decision point.
Separate appearance from severity.
Agree which types, sizes, locations and severities require action, and preserve difficult borderline examples for validation and review.
Buyer questions
Questions buyers ask about AI surface defect detection
Direct answers for quality, operations, automation and manufacturing engineering teams evaluating this use case.
What surface defects can AI visual inspection detect?
AI visual inspection can be trained for visible, consistently imaged conditions such as configured scratches, cracks, dents, stains, pits or discontinuities. Feasibility depends on whether the defect can be separated from acceptable texture and production variation.
How is AI surface inspection different from rule-based machine vision?
Rule-based vision is effective when a stable visual condition can be described with deterministic features. A trained model can help with more variable appearances, but it still requires controlled imaging, representative examples, acceptance criteria and production validation.
Can reflective or textured parts be inspected?
They can be evaluated, but reflections and texture often become the main imaging challenge. Lighting geometry, polarization, optics, part presentation and the range of acceptable finishes must be qualified before model performance can be judged.
How many defect images are required?
There is no defensible universal number. The useful dataset must represent the defect types, severities, locations, part variants, finishes, lighting conditions and normal variation the production inspection will encounter.
What happens when the finish or part geometry changes?
The plant can capture the new condition, review labels, retrain the model and validate an updated version before redeployment. Changes that alter the image outside the qualified range should trigger that review.
Where does United Vision run production inference?
Production inference always runs on a supported factory-owned edge device close to the line. The camera-to-verdict path does not depend on a round trip to a vendor cloud. Edge hardware, latency, interfaces and failure behavior are qualified for this inspection.
Does this use case require United Vision cameras or a United Vision-owned cloud?
No. United Vision is hardware- and cloud-agnostic within supported configurations. The plant can bring qualified cameras and edge devices, and United Vision cloud services run in a supported cloud account chosen and owned by the customer.
Does adding this inspection require another United Vision software license?
Not within the licensed factory. One factory license supports unlimited United Vision use cases. The inspection still requires suitability assessment, imaging, integration and production validation, and any hardware or services are scoped separately.
Related guidance
Plan the use case as part of a scalable inspection system.
Use the supporting guidance to evaluate readiness, select the right technical approach and keep the inspection current after deployment.
AI visual inspection versus traditional machine vision
Understand when deterministic rules are enough and when learned visual variation matters.
Read moreHow to keep visual inspection models current
Build production change, error review and controlled redeployment into the operating model.
Read moreAI inspection for metals and machining
Explore surface, dimensional and thermal inspection questions for metal components and processes.
Read moreEvaluate surface defect detection against your actual parts.
Bring acceptable and suspect samples, defect definitions, surface finishes, current lighting and camera details, line speed and the required operating response.