Reflective metal surfaces are a recognized vision challenge: the same scratch appears and vanishes with angle. Add scarce defect samples and conventional methods stall. DaoAI's good-sample-only route sidesteps the sample problem, handing uncertainty to the model.
A consumer-electronics plant produces metal mid-frames and housings finished by anodizing, brushing or polishing, giving strongly reflective surfaces. Appearance defects include scratches, dents, pitting, color variation, oxidation spots and knocks, requiring grade sorting (A ships, B downgrades, defects rejected). Strong reflection makes defects appear and disappear across lighting angles, leaving imaging highly unstable.
The difficulty compounds on samples: high-value defects such as fine scratches occur rarely and are hard to collect in volume for training; meanwhile models are many and turn over fast, so each new housing needs redeployment. The plant had to solve reflective imaging while keeping stable grading under scarce defect samples and frequent changeover.
DaoAI solution: go live trained on good samples only, with zero-code changeover for micron-level grading
DaoAI adopts a good-sample-only strategy: build a normal-appearance baseline from qualified housings, and flag any deviation as a defect—no need to pre-collect scarce defect samples. Paired with a multi-angle optical setup that suppresses reflection, the system identifies and grades scratches, pitting and color variation at micron-level precision.
- Goes live trained on good samples only, bypassing the scarce-defect-sample bottleneck of reflective metal
- Micron-level precision on scratches, dents, pitting, color variation and oxidation spots
- Zero-code changeover lets a new housing switch grading baselines on site
- Auto-grades into A/B/reject, cutting defect escapes 94% versus manual inspection
Reflection wears out human eyes chasing the light; the model cares about one thing—does it look like a good sample.
After go-live, reflective housings are graded automatically and stably, with A/B/reject calls no longer hostage to inspector fatigue or viewing angle. The most critical fine scratches and pitting are reliably captured, defect escapes fell 94% versus the prior manual process, downgrades and customer complaints dropped together, and new models go live with zero-code changeover.
FAQ
What are the difficulties in the appearance grading of reflective metal casings?
There are many difficulties in the appearance grading of reflective metal casings. The same scratch may appear or disappear at different angles, resulting in unstable imaging. High - value defect samples are scarce, hard to collect for training, and frequent model changes require redeployment.
How does DaoAI solve the problem of appearance grading for reflective metal casings?
DaoAI adopts a strategy of training only with good products. It establishes a normal appearance benchmark with qualified casings and judges deviations as defects. Combined with a multi-angle optical solution to suppress reflection, it achieves micron-level accuracy in recognition, with zero-code operation for model changes and automatic grading.
What are the effects after implementing the DaoAI solution?
After the implementation of the DaoAI solution, the stable automatic grading of reflective casings is achieved without relying on inspectors. Minor scratches are stably detected, the defect escape rate drops by 94%, downgraded products and customer complaints are reduced, and new models can be quickly launched with zero-code operation.