Autonomous PCB Defect Inspection Agent

Closed-loop PCB defect detection system using OpenCV 5 & AWS Serverless. Uses an agentic perception-decision loop for real-time quality control.

Comment

💡 Inspiration

Traditional Automated Optical Inspection (AOI) systems in electronics manufacturing are rigid and static they detect anomalies but cannot adapt or autonomously trigger workflow actions. We wanted to bridge computer vision with agentic automation to create a true closed-loop PCB defect inspection agent that doesn't just see defects, but acts on them in real time.

⚙️ What it does

Our system provides end-to-end, dynamic PCB quality control powered by OpenCV 5 and AWS Serverless infrastructure:

  • Perception Layer (OpenCV 5):- Pre-processes high-resolution PCB image feeds using adaptive thresholding, denoising, and contour extraction to detect defects like solder bridges, missing pads, and trace cracks.
  • Agentic Decision Loop:- Dynamically evaluates defect severity and confidence metrics to trigger context-aware execution actions such as halting virtual lines or routing alerts.
  • AWS Cloud Integration:- Automatically archives high-res defect visual crops in AWS S3, logs structured time-series telemetry in AWS DynamoDB, and dispatches high-priority event notifications via AWS SNS & Lambda.

🛠️ How we built it

  • Computer Vision Core: Built using Python 3 and OpenCV 5 for lightning-fast image processing and visual contour isolation.
  • Agentic State Engine: Developed a Python-based decision controller that maps perception outputs into programmatic tool calls.
  • Cloud Architecture: Integrated AWS SDK (Boto3) to seamlessly handle asynchronous telemetry logging, image storage, and alert routing.

🚧 Challenges we ran into

Simulating real-time closed-loop hardware feedback without expensive industrial camera hardware was a challenge. We resolved this by engineering a robust synthetic PCB image feed generator and mock tool-invocation pipeline.

🎓 What we learned

  • Harnessing new features in OpenCV 5 for accelerated feature extraction.
  • Designing event-driven agentic loops that minimize cloud API latency during real-time processing.

🚀 What's next for Autonomous PCB Defect Inspection Agent

  • Deploying light-weight model inference on edge devices (NVIDIA Jetson / Raspberry Pi).
  • Adding automated camera pan/ tilt/ zoom adjustment triggers directly into the agent decision loop.

Built With

Share this project:

Updates

Submission history