AI On‑line Visual Inspection Project Implementation for Automotive Chassis Stamping Workpieces

Case Cycle: September 2025‑August 2026

Enterprise Background: The enterprise is a tier‑2 auto‑chassis stamping supplier in South China, operating 9 automatic stamping lines, mainly producing chassis bracket and reinforcing plate components. Before transformation, full finished‑products relied on manual offline inspection. Frequent staff turnover and night‑shift fatigue caused occasional defect outflow, leading to customer returns and re‑inspection costs. The enterprise decided to implement on‑line AI visual retrofitting for two core high‑volume stamping lines.

• Finished‑product defect outflow rate: 5.12%
• Average daily manual QC staff for two lines: 8 persons
• Monthly customer return quantity: 216 pcs
• Monthly cost for re‑inspection & rework: USD 4,120
• Overtime‑work proportion of QC team: 32%
• Defect missing mainly: tiny surface indentation, micro‑crack at bending corner

1. Deploy multi‑angle high‑resolution industrial cameras and customized diffuse light‑source beside press discharge port, eliminate reflection interference on metal stamping surfaces;
2. Collect more than 12 000 real‑defect image samples from production site to train AI segmentation model, continuously add new‑found defect samples for iterative optimization;
3. Link visual system with conveyor‑line rejecting actuator; once defects are identified, trigger automatic rejection to isolate non‑conforming parts;
4. Realize data interface between vision platform and factory MES, record defect type, quantity and time‑stamp for every inspected workpiece;
5. Optimize SOP: re‑position manual inspectors from full‑piece screening to spot‑checking and model‑maintenance work, reduce repetitive visual labour.

This real‑production case proves that AI on‑line visual inspection can effectively solve long‑standing pain points of manual‑inspection fatigue and unstable judgement in stamping workshops. It significantly reduces defect outflow, cuts re‑work cost and lowers dependence on large‑scale QC labour. Meanwhile enterprises must be aware that visual system needs continuous sample iteration and on‑site debugging rather than one‑time installation. In next phase, the factory will push defect‑data linkage to press equipment, realizing early warning of mold wear according to rising crack‑defect frequency and moving toward predictive quality control.