Split Detection System

Overview

Built a scalable, distributed vision system leveraging Python, PyTorch, and high-resolution GigE camera streams. Containerized each component with Docker Compose and orchestrated an asynchronous ZeroMQ pipeline to achieve real-time split classification across multiple production lines. Developed modular services to reliably detect metal splits with 98%+ accuracy, and learned to architect robust distributed systems under real-time constraints.

Key Features

• Real-time split classification at 100+ frames per second.
• Asynchronous messaging with fault tolerance.
• Custom HMI & End-of-Line (EOL) applications built with Python/FastAPI backend and Dart/Flutter frontend for intuitive operator control.
• Modular microservices architecture in Docker containers.
• Automated data logging, analytics integration, and live status dashboards.

Tech Stack

UI & Deployment

The custom HMI offers real-time visualization of split detections, system health metrics, and historical reports. EOL stations provide pass/fail decisions with seamless integration into factory automation PLCs. Deployed on-premises with minimal dependencies and remote upgrade capabilities.

Source Code

Unavailable