// the find
KankariAvinash/Predictive_Maintenance_for_Industrial_IoT_Using_EdgeIntelligence
A student/research project demonstrating on-device anomaly detection for industrial machinery using an ESP32-S3-EYE board, audio classification via Edge Impulse, and temperature sensing. It classifies machine states into four condition categories (normal, abnormal speed, abnormal temp, both) without any cloud dependency. The target audience is embedded ML beginners and IIoT hobbyists, not production engineers.
The ESP32-S3-EYE choice is smart — the integrated digital mic eliminates external I2S wiring, and 8MB PSRAM gives enough headroom for the Edge Impulse TFLite runtime. The four-state decision matrix (speed + temperature combined) is more useful than a binary classifier alone. The full Edge Impulse SDK is vendored in-tree, so the build is self-contained and doesn't require a cloud account at compile time. Audio dataset is included, which means you can actually reproduce the training run.
This is clearly a university project — 0 stars, ~10 audio samples per class, and the dataset folder contains files named 'New Recording 21.wav', which tells you exactly how the data was collected. The model is trained on one specific machine under two speed conditions; it will almost certainly misclassify on different equipment without retraining. No documentation on model architecture, accuracy metrics, or false-positive rate — the README has a diagram but no numbers. The temperature threshold appears to be hardcoded rather than configurable per-machine, which makes the yellow/red state logic brittle for any real deployment.