- Client
- Industrial manufacturer (under NDA)
- Platform
- Rockchip NPU single-board computer
- Goal
- Safety alerts and anomaly detection on-site
- Constraint
- No server, no cloud dependency
The problem
The client wanted to detect unsafe situations and abnormal machine behaviour on their production floor. Their IT policy did not allow video or sensor data to leave the site, and they did not want to run and maintain a server for it. Everything had to happen on one small device next to the line.
How I solved it
Vision on the NPU
Camera frames go through hardware-accelerated capture and resize into a detection model on the Rockchip NPU, which flags safety events such as people entering restricted zones.
Anomaly detection on sensor data
Machine signals feed a lightweight model that learns normal behaviour and flags anything unusual, instead of relying on fixed thresholds.
Local rules and alerts
A rule engine combines both signals and triggers alarms, relay outputs and on-screen alerts locally.
Local storage and dashboard
Events, short clips and trends are stored on the device and shown on a local web dashboard.
Built for 24/7
Watchdogs, read-only root file system, log rotation and automatic recovery after power loss.
What the client got
Results
- Safety alerts in under a second, entirely on the device.
- Anomaly detection that adapts to each machine.
- No server, no cloud and no data leaving the factory.
- A system the plant team can run without specialist IT support.
Embedded systems consultant at TechProgeny, Bengaluru. 7+ years and 29+ shipped products across firmware, embedded Linux, edge AI, apps and IoT servers.