Edge-Inference Pipeline for a Smart-Factory Vibration Monitor
Overview
What this challenge is about.
Architect a pipeline that runs on an ESP32-S3 + STM32 combo (provided): (1) sample 3-axis accelerometer at 3.2 kHz, (2) compute windowed FFT features on-device every 1s, (3) run a small anomaly detector (isolation forest distilled to under 200KB or a tiny CNN via TFLite Micro), (4) escalate to MQTT cloud only when score crosses threshold or every 6h health-check. Train and evaluate against 6 weeks of recorded data covering 4 fault classes + healthy baseline. Measure: detection latency, false-positive rate at 1-week horizon, MQTT egress kB/day. Deliver: firmware repo, trained edge model, 6-page evaluation report, 5-page deployment architecture spec.
The Brief
What you'll do, and what you'll demonstrate.
Run vibration anomaly detection on-device on ESP32-S3 with under 1 percent false-positive weekly and under 1 MB/day MQTT egress per machine.
Earning criteria — what you'll demonstrate
- Design a multi-stage edge-inference pipeline on constrained MCUs
- Quantize/distill ML models to fit edge memory + latency budgets
- Evaluate cyber-physical systems with operationally relevant metrics
- Specify deployment architectures that respect network + power constraints
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Career mappings coming soon.