Weld seams are the lifeline of body-in-white strength and safety, yet their defects are varied and low-contrast—precisely the blind spot of manual inspection. A body-shop plant used DaoAI deep-learning vision to lock down its seam criteria.
The hundreds of weld seams on a body-in-white directly bear on structural strength and crash safety. This plant previously relied mainly on manual visual checks of seam appearance, but cracks, lack of fusion, undercut and porosity are complex in form and low in contrast, and criteria varied between inspectors, leaving both misses and false calls. The shift to high-strength and galvanized steel made seam surfaces more reflective and manual judgment harder, undermining quality consistency.
Teaching the Model to Read Seams
DaoAI AI-AOI surface inspection is built on deep learning. Trained on large volumes of seam samples, the model learns the feature signatures of typical defects—cracks, lack of fusion, undercut and porosity. The system images and infers each seam segment by segment, automatically reporting defect class, location and severity, converting experience-based subjective judgment into a unified, reproducible objective standard, and recording results as traceable quality data.
- Multi-defect coverage: cracks, lack of fusion, undercut and porosity recognized uniformly
- High accuracy: deep-learning detection of low-contrast defects clearly outperforms manual eyes
- Unified standard: judgments no longer vary by person, consistent across shifts
- Traceable data: each seam result is archived to support process and equipment root-cause analysis
Turn the experience in an inspector's head into a unified standard that is reproducible and traceable inside the model.
After go-live, defect detection became more stable, the risk of missed defects fell, and judgment consistency improved markedly. The defect data also fed back into continuous optimization of welding parameters.