The folks at Modzy got in touch to tell us about their defect detection platform that uses Raspberry Pi Zero W and a Raspberry Pi Camera Module to pick up mistakes on factory production lines. Their message specifically stated their love for Raspberry Pi, so I was sold.
Modzy deploys machine learning models in the cloud and at the edge. They built the demo above to show their manufacturing customers how easy and affordable it is to detect defects using machine learning in a factory.
How it works
Hardware:
- Raspberry Pi Zero W
- Raspberry Pi Camera Module 3
- NVIDIA Jetson Nano running a computer vision model
- A “conveyer belt” consisting of a cake decorating turntable with a motor attached
A Raspberry Pi Camera Module acts as the eyes of the production line monitor, feeding real-time images from the production line. A computer vision model analyses the images to detect broken teeth, scratches, and dents in 3D-printed spur gears as they roll through on the homemade “conveyer belt”. When the model detects a defect, Modzy’s system updates and timestamps a log, recording what was wrong with the spur gear and exactly when it rolled past the detection point on the production line.
Low-cost efficiency
This defect detection platform cost less than $150 to build.
Modzy is particularly proud that this model achieved three out of three on speed, security, and cost:
- It’s fast thanks to low-latency GPU-powered inference (meaning it can spot defects as quickly as the production line rolls)
- It’s secure, since it can operate completely offline
- It’s cost effective, since no cloud computing is needed and the hardware is affordable
More machine learning from Modzy
The Modzy folks are big into Raspberry Pi for machine learning and have built two other apps:
- Their Air Quality Index Prediction detects current air quality with Raspberry Pi 3B+, and uses that data to generate a prediction for the next hour.
- Their Hugging Face NLP Server deploys and runs a hugging face model on Raspberry Pi with Docker.
