Skip to content

Loading...

Computer Vision 5 ay Computer Vision Engineer & Project Lead

Visual Quality Control System

Real-time defect detection system based on computer vision for an automotive supplier production line. Quality control within milliseconds with Edge AI.

%99.7
Defect Detection Rate
8ms
Inference Time
2.3M₺
Cost Savings
%96
Defect Reduction

Challenge

Keeping up with the production line speed (3 parts per second), varying lighting conditions, and detecting micron-level defects were the main challenges.

Solution

Reduced inference time to 8ms with TensorRT model optimization. Designed a controlled LED lighting system. Trained a model robust to different lighting conditions using data augmentation.

Highlights

1

Custom defect detection model training with YOLOv8

2

Edge AI deployment on NVIDIA Jetson

3

Real-time MES integration

4

Multi-camera synchronization system

Technology Stack

Python
YOLOv8
OpenCV
NVIDIA Jetson
TensorRT
MQTT
Docker

About the Project

Integrated into an automotive supplier's production line, this system scans every manufactured part with high-resolution cameras to detect defects within milliseconds.

Technical Details

  • YOLOv8-based object detection model
  • Edge AI deployment (NVIDIA Jetson AGX Orin)
  • Multi-camera synchronization
  • Real-time alarm and reporting system
  • MES (Manufacturing Execution System) integration

    • Impact

    The system reduced defective product rate by 96% and saved 2.3 million TRY annually in quality costs.