Anomaly Detection on the Factory Floor

December 29, 2025

Testing anomaly detection in a factory setting.

Components of this build include: Raspberry Pi 4B (4GB on-board memory), Continuous Servos (2x), Positional Servo, Raspberry Pi 4B (4GB on-board memory), Adafruit Crickit HAT, USB 3.0 Boot, a USB camera, and a whole lot of cardboard, paper, and velcro.

The actual anomaly detection and conveyor belt control happens on my desktop computer, which is running the “Factory Control Panel” app displayed in the video. Data is sent to / from my desktop via a communications layer between it and the Raspberry Pi.

One of the interesting tradeoffs that I discovered existed in the relationship between conveyor belt speed, reliability of the system to detect anomalies, and overall system degradation. When I brought up the speed of the system, as showcased in the video, I noticed that anomalies were detected a little too late, thereby causing a drop in reliability.

Furthermore, with the conveyor belt moving so aggressively, wear and tear on the physical components of the system were amplified. Repairs were straightforward, but almost every repair meant bringing down the belt.

A hefty price. Speed of the belt should be balanced with reliability of the anomaly detection and uptime of the belt.

Additional investigation may warrant: a longer belt, more cameras, more positional servos to help clear the belt, and / or further improving anomaly detection speed and accuracy.

You can view more info on this project and more here.