Vision-Based Quality Inspection System
Real-time defect detection on an automotive production line — cut manual inspection effort by ~40%.
Manual inspection of automotive parts was slow, inconsistent between shifts, and rejection data was never recorded anywhere it could be analysed.
Built a full detection pipeline — camera setup, dataset labelling from line footage, model training, and ONNX-based edge inference — then wired results into a rejection log and dashboard.
Reduced manual inspection effort by around 40% and gave the quality team a defect history they can analyse by part, shift and defect type.
What I built
- Camera and lighting setup for the inspection station
- Defect detection model trained on real line footage
- ONNX edge inference for real-time decisions
- Reject logging into SQL with dashboard integration
Domain knowledge used
Quality consistency, tolerances, and what a real defect looks like on a real part — plus the discipline of keeping the system running across shifts.
More projects
PPE Detection System
Computer vision + deep learning watching the shop floor for missing helmets and vests, in real time.
Production Monitoring Dashboard
Inspection outputs, PLC/SCADA event logs and OEE metrics in one centralised real-time reporting platform.
Multi-Camera RTSP Streaming Infrastructure
RTSP/FFmpeg pipeline with automated watchdogs and failover — above 99% availability across shifts.