The Challenge
A Tier 2 automotive stamping supplier in Pune producing pressed metal components for dashboard brackets and body panels was experiencing a defect escape rate of 1,800 ppm — causing monthly customer returns, rework costs, and pressure on IATF 16949 audit scores. Eight quality inspectors running 6 shifts were producing inconsistent results driven by fatigue. Root cause analysis took 3–4 days because there was no defect image history.
Why DeepVision Was Selected
The plant evaluated three automated options. DeepVision was chosen because it trained on existing defect samples without a photographic studio, integrated with their Siemens S7 PLC without line modification, and Indus Vision committed to a working pilot within 14 days.
Deployment
Three GigE Vision stations were installed at press output, after deburring, and before packing. DeepVision trained on 2,400 labelled images covering 7 defect classes. Parallel validation ran for 18 days — achieving 98.7% detection accuracy and 0.6% false positive rate.
Results After 6 Months
| Metric | Before | After | Change |
|---|---|---|---|
| Customer returns (ppm) | 1,800 | 220 | −88% |
| Monthly rework cost | ₹14.2L | ₹6.1L | −57% |
| QA headcount | 8 FTEs | 3 FTEs | −62% |
| Total quality costs/month | ₹22.8L | ₹9.8L | −57% |
| Inspection throughput | 320 parts/hr | 680 parts/hr | +113% |
“DeepVision caught defect types we did not even know we were producing. Within two months our customer returns dropped so significantly that one OEM upgraded our supplier rating. ROI was clear by month three.” — Quality Head, Tier 2 Automotive Supplier, Pune