Automotive QC Vision

Computer vision quality control for automotive components

A computer vision system that inspects every component at 120 fps, diverts defects without stopping the line and gives OEMs a full audit trail.

Case auto qc — Quibo Technologies

The challenge

An automotive components manufacturer with strict dimensional and surface tolerances relied on 34 inspectors doing end-of-line visual checks over 10-hour shifts. Manual inspection caught only 71% of defects, and accuracy fell 40% in the last three hours of each shift. Three major OEM clients had issued formal quality notices, putting Rs. 22 Cr in annual contracts at risk with a six-month cure period.

Our solution

We trained a deep learning model on 140,000 labelled component images covering surface scratches, dimensional deviations, porosity and thread integrity failures, and deployed it on a multi-angle camera rig running at 120 frames per second on the live line. An automated reject lane diverts flagged parts without stopping production.

A defect analytics dashboard classifies every defect by type, shift, machine and operator so engineers can fix root causes, and an automated report for each batch, with timestamped logs, defect images and rejection rates, gives OEM clients the traceability they need to close quality notices.

What we delivered

  • Deep learning defect detection at 120 fps
  • Real-time reject diversion without line stoppage
  • Defect analytics and root cause dashboard
  • Automated OEM reporting and audit trail
“Our OEM clients now ask us to share our QC methodology. We went from a liability to a benchmark in the supply chain.”
Quality HeadManufacturing · Automotive client

Key results

99.1%Defect detection rate, vs 71% manual
120 fpsInspection speed on the live line
-94%Defects reaching OEM clients
Rs. 22 CrOEM contracts retained
  • 99.1% detection accuracy across all component types, against 71% under manual inspection.
  • All three OEM quality notices closed within 60 days of deployment.
  • Zero line stoppages for inspection: automated reject diversion kept full throughput.

Results measured with the client. The client’s name is withheld at their request.

Services used

Technology stack

PythonPyTorchOpenCVNVIDIA TritonC++ReactPostgreSQL
Ready to build?

Your project could be next.

Share what you want to build and we will reply within one business day with questions and a suggested next step.