Teensy Vibration Classifier (Embedded ML)
Built a real-time, vibration-based object classification system that detects an impact, extracts features, and classifies the object using a neural network running entirely on a Teensy 4.1 microcontroller.
The project demonstrates a full embedded ML pipeline—from physical sensing to on-device inference.
How it works (Nutshell)
A steel plate mounted on vibration-isolating feet rings after impact. An IMU captures the response, the firmware detects the event, extracts compact features, and performs real-time inference on the device.
Trained on eraser and screw impacts, the system achieved >99% accuracy under controlled conditions. The project demonstrates an end-to-end embedded ML pipeline: sensing, triggering, feature extraction, and on-device classification.
Skills: Embedded systems • Signal processing • Neural networks • Real-time firmware
To see a comprehensive report on this project, look here.
The full repository can be found here: https://github.com/FriChaFri/Arduino-Vibration-Object-Classifier