abtinmbm/andon

How abtinmbm/andon works

Andon is an anomaly-detection product that learns normal machine sounds and signals deviations through an andon light. Its principal workflow is audio/data ingestion → log-mel features → model training and anomaly scoring → control limits and light-state simulation. The repository also includes benchmark evaluation, int8 conversion/export, a dashboard, and a Pico-oriented C firmware path. README statements support the documented end-to-end capabilities; firmware internals are only partly sampled, so uncertain internal connections are omitted or marked unverified.

Main entry points: features.py Feature extraction/cache · train_idcnn.py CNN training · train.py Baseline training · score.py Anomaly scoring

Components 24

Data and features

DCASE dataset andon/data/dataset.py
WAV audio I/O andon/data/audio.py
Feature extraction/cache andon/data/features.py
Log-mel frontend andon/features/logmel.py

Detection models

Classic models andon/models/classic.py
Autoencoder andon/models/autoencoder.py
Machine-ID CNN andon/models/idcnn.py
Baseline training andon/train.py
CNN training andon/train_idcnn.py

Scoring and evaluation

Experiment configuration andon/experiment.py
Anomaly scoring andon/score.py
Benchmark evaluation andon/evaluate.py
Light state logic andon/light.py
Stream simulator andon/simulate.py
Int8 conversion andon/quantize.py

Embedded device

C model export andon/export_c.py
Firmware controller firmware/src/main.c
Microphone stream firmware/src/mic_stream.c
C log-mel frontend firmware/src/logmel.c
C inference model firmware/src/model.c

Interfaces

Dashboard dashboard/app.py

Other components

Machine audio
DCASE recordings
Fan hardware

Connections 31

DCASE recordings → DCASE dataset: provides clips
DCASE dataset → Feature extraction/cache: selects clips andon/data/features.py
Feature extraction/cache → WAV audio I/O: reads WAV andon/data/features.py
Feature extraction/cache → Log-mel frontend: extracts features andon/data/features.py
Feature extraction/cache → CNN training: supplies features andon/train_idcnn.py
Feature extraction/cache → Baseline training: supplies features andon/train.py
Baseline training → Autoencoder: trains andon/train.py
Baseline training → Classic models: trains
CNN training → Machine-ID CNN: trains andon/train_idcnn.py
CNN training → DCASE dataset: loads normal clips andon/train_idcnn.py
Baseline training → Experiment configuration: uses run config andon/train.py
CNN training → Experiment configuration: uses run config andon/train_idcnn.py
Anomaly scoring → CNN training: uses scoring helpers andon/score.py
Anomaly scoring → Benchmark evaluation: evaluates scores andon/score.py
Stream simulator → Anomaly scoring: scores windows andon/simulate.py
Stream simulator → Light state logic: drives states andon/simulate.py
Benchmark evaluation → DCASE dataset: scores test clips andon/evaluate.py
Benchmark evaluation → Experiment configuration: logs results andon/evaluate.py
Machine-ID CNN → Anomaly scoring: produces ID scores
Int8 conversion → Machine-ID CNN: converts model andon/quantize.py
Int8 conversion → Benchmark evaluation: checks scores andon/quantize.py
C model export → Int8 conversion: exports int8 model andon/export_c.py
C model export → C inference model: writes C headers
Machine audio → Microphone stream: streams sound
Firmware controller → Microphone stream: uses stream firmware/src/main.c
Microphone stream → C log-mel frontend: feeds frames
C log-mel frontend → C inference model: provides features
Firmware controller → Fan hardware: controls fan firmware/src/main.c
Dashboard → Stream simulator: uses demo flow dashboard/app.py
Dashboard → Experiment configuration: uses configuration dashboard/app.py
Dashboard → Light state logic: uses light logic dashboard/app.py

Drawn by GitDiagram on November 12, 2026 from the repository's file tree, README and source files. Read as Markdown ·