01
Scratches and stains
Detect scratches, spots, contamination and color differences on metal, plastic, glass and packages.
Back to products
Defect Inspection
Surface defect inspection solution
Use optical imaging, rule-based algorithms and deep learning inspection to detect scratches, stains, gaps, deformation, foreign objects and burrs.
Object
Defects
Scratches, stains, gaps, foreign objects, deformation and burrs.
Algorithm
Rules/AI
Choose rule-based or deep learning methods by defect stability.
Acceptance
Samples
Normal, abnormal and borderline samples must be covered.
Scenarios
For complex defects and costly manual reinspection
Defect inspection depends heavily on samples and acceptance boundaries, especially for borderline defects.
02
Gaps and deformation
Inspect edge gaps, dents, warping, deformation, missing material and assembly anomalies.
03
Foreign objects and mix-ups
Detect foreign objects, mixed materials, wrong parts and missing components in production or packaging.
System
Make defects visible first, then make judgement accurate
Different defects are sensitive to lighting, angle, background and algorithm route, so imaging must be designed around defect features.
01
Optical imaging route
Use ring, bar, back, coaxial, low-angle or line-scan lighting to emphasize defect features.
02
Rules and deep learning
Use explainable rules for stable defects and deep learning for irregular texture or complex defects.
03
Review and sample management
Store defect images, classes, locations and review results for parameter tuning and model iteration.
Process
Let samples and acceptance criteria drive the algorithm route
Defect inspection needs real production variation, not only one or two sample images.
01
Defect definition
Clarify classes, size, location, acceptable range, false reject / miss requirements and review flow.
02
Optical experiments
Test lighting, angle, background and exposure to make defects appear reliably.
03
Algorithm validation
Compare rule-based, deep learning or hybrid methods on real samples.
04
On-site iteration
Collect false rejects, misses and borderline samples after launch to tune rules or models.
Deliverables
Keep defect boundaries and review mechanisms explicit
New samples may appear after launch, so delivery should include review, sample supplement and iteration mechanisms.
01
Defect classes and acceptance rules
Defect classes, size thresholds, region rules and acceptable boundaries.
02
Sample library and review records
Normal, abnormal, false reject, missed and borderline samples for later review.
03
Iteration and maintenance strategy
Parameter tuning, model updates, version records and issue closure process.
FAQ
Defect inspection requires expectation management
Defect shape, sample coverage and acceptance boundaries directly affect stability.
Can all defects be recognized reliably at once?
Not always. It depends on stable imaging, sufficient samples and clear acceptance boundaries.
When is deep learning suitable?
When defects are complex and hard to describe with rules, and enough samples are available.
What if new defects appear after launch?
Keep review and sample supplement flows, then expand coverage through parameter tuning or model iteration.
Sample Validation
Defect projects should start with sample validation
Send normal parts, defective parts, borderline samples and judgement standards so we can evaluate imaging and algorithms.