AI Visual Inspection

HULK ROBOT starts AI visual inspection projects with the inspection object and defect samples. For optical lens appearance inspection, imaging and recognition algorithms connect to sorting and data recording. The inspection scope depends on sample assessment and project configuration.

Can one system detect every defect?

A generic accuracy figure cannot replace a project-specific assessment. Define defect categories, positions, sizes and acceptance boundaries, then verify that imaging consistently reveals the differences. Materials, transparency and surface conditions require separate tests.

How should samples be prepared?

Provide normal parts, each defect class and borderline samples with documented manual judgments. State specifications, inspection surfaces and production cycle time. Keep acceptance samples separate from tuning samples; a small set of familiar parts is not sufficient evidence.

How are cycle time and detection performance confirmed?

Measure missed defects, false rejects, repeatability and actual cycle time including handling within the agreed sample scope. The 2,400 lenses/hour shown refers to an existing optical lens inspection configuration, not a uniform specification for every material or defect.

What do the gallery images represent?

The railway-rail surface defect and dimensional inspection illustration belongs to that separate project. Its inspection object differs from lenses, so it must not be used as evidence of lens-inspection performance. Each image describes only its corresponding system or equipment.

Sample testing workflow

1. Define defect classes, surfaces, positions, sizes and severity. Agree on borderline cases, and specify dimensional and appearance requirements separately.
2. Number samples and record specifications, manual decisions and defect positions. Include normal batch, orientation and surface variation. Record imaging-condition changes for transparent or reflective parts.
3. Verify imaging before recognition. Test lighting, views and positioning to ensure differences remain visible. Record cameras, lighting, workpiece position and the environment so results can be reproduced.
4. Reserve independent acceptance samples. Count missed defects and false rejects by class, including sample counts and proportions. Define reinspection for inconsistent, borderline or indeterminate results. One accuracy figure cannot represent untested classes.
5. Check the complete production cycle: feeding, positioning, inspection, result output and sorting. Test abnormal parts and equipment interfaces. Evaluate inspection quality and effective output together, not algorithm runtime alone. Agree on stored data and fields for the project.

Information for an enquiry

Provide specifications, materials, inspection surfaces, the defect list, numbered samples, manual decisions and borderline examples. Include cycle time, site lighting, positioning, handling, missed-defect and false-reject limits, sorting/reinspection requirements and existing interfaces.

Sample testing and acceptance guide · Discuss an inspection project

Optical lens inspection equipment shipment

This video shows packing and shipment. It is not a demonstration of inspection operation or recognition results.

Watch the equipment shipment — Douyin, Chinese content