Cosmoxyn
Back to Case Studies
Computer Vision

Real-Time Defect Detection Cuts Quality Reject Rate by 83% on the Production Line

PrecisionPartsIndustrial ManufacturingPune, IndiaDelivered in 9 weeks4 engineers

The Challenge

PrecisionParts manufactures metal components for the automotive sector. Their manual QC process reviewed only 12% of output due to throughput constraints, and a 4.2% defect escape rate was causing costly recalls and customer penalties. They needed 100% inspection coverage without slowing the line.

Our Solution

We designed and deployed an edge-based computer vision pipeline: high-resolution industrial cameras mounted at key inspection points feed frames into a custom YOLOv8 model trained on 14,000+ labelled defect images (scratches, cracks, dimensional mismatches, surface pitting). Defective parts are flagged and auto-diverted by a pneumatic reject gate in under 80ms. A live dashboard tracks defect trends by shift, machine, and defect type.

The Results

100%

Inspection coverage (was 12%)

83%

Reduction in defect escape rate

< 80ms

Detection latency per component

₹42L

Annual recall & rework cost avoided

We went from inspecting 12% of parts to 100% — and catching defects we didn't even know existed. The payback period was under 4 months.

VD

Vikram Desai

VP Operations, PrecisionParts

Tech Stack

YOLOv8PythonOpenCVNVIDIA JetsonFastAPIReactInfluxDB

Project Details

ClientPrecisionParts
IndustryIndustrial Manufacturing
LocationPune, India
Timeline9 weeks
Team4 engineers

Key Outcomes

  • 100%Inspection coverage (was 12%)
  • 83%Reduction in defect escape rate
  • < 80msDetection latency per component
  • ₹42LAnnual recall & rework cost avoided
Get a similar result

Free audit — no commitment