Manufacturing · Case study
Edge AI visual inspection for manufacturing defect detection
A real-time visual inspection system for an industrial parts manufacturer. A computer-vision model trained on the factory's own images runs on a Raspberry Pi with a camera at the line, flags defective parts as they pass, alerts supervisors and keeps an inspection log with images. No cloud is involved.
The challenge
Checking industrial parts for defects by eye is slow and inconsistent. The factory wanted anomalies caught in real time as parts moved through production, without streaming images to the cloud.
Our approach
We trained an object-detection model on the factory's own image data and deployed it on a Raspberry Pi with a Pi Camera. Inference runs on the device, supervisors get an alert the moment a defect is found, and every flagged part is logged with its image.
What we built
- Object-detection model for industrial parts, trained on factory-specific datasets
- On-device inference on Raspberry Pi with a Pi Camera
- Real-time defect alerts for line supervisors
- Inspection-log interface with an image of every defect
Highlights
- Edge-only processing with no cloud dependency, so images stay on site
- Every flagged part is stored with its image for later review
Technology
- TensorFlow
- Python
- OpenCV
- Raspberry Pi
- Flask
Manual inspection depends on who is on shift. We replaced the first look with an object-detection model trained on images of the factory’s own parts, so it learns what normal and defective look like for that product rather than a generic dataset.
The model runs entirely on a Raspberry Pi with a Pi Camera at the inspection point. When it spots an anomaly it alerts a supervisor and saves the image to an inspection log, which gives the quality team a searchable record of every defect without sending a single frame off site.
Client details are kept confidential. On a call we can walk you through this project in more depth, under NDA.