Motors, fans, pumps and compressors always give warnings before they fail — subtle changes in vibration that nobody catches until it's too late.
Pulse is a palm-sized embedded device that learns your machine's normal vibration signature in ~30 seconds and alerts you in real-time when something changes. No cloud. No internet. No subscription. Everything runs on a STM32L412RB microcontroller with 40KB RAM.
Built with Zephyr RTOS and STMicroelectronics NanoEdge AI. Training data persists to onboard flash with CRC32 validation — so Pulse remembers what it learned even after a power cycle.
Currently tested on a Xiaomi air purifier fan. Looking for people with real industrial equipment — motors, pumps, CNC machines — to test this further.
https://github.com/Ayushkothari96/pulse
Details
[View on GitHub](https://github.com/Ayushkothari96/pulse)
Got the project to a point where I'm happy sharing it publicly. Tested anomaly detection on a Xiaomi air purifier fan — attaching the STM32 + LIS2DH12 directly to the casing and introducing a physical disturbance (tapping/scratching) gets detected within one inference cycle (~600ms).
Flash persistence is working reliably — training data survives power cycles with CRC32 validation on the header.
Next goal: find someone with an industrial motor or pump to test on real machinery.