AFCI offline kit - Part 4: TinyML model zoos & AFCI reference code (ST + TI)
Offline development kit for an arc-fault detection / electrical-fire prevention system (current sensor -> STM32F407 ADC+DMA -> FFT/STFT features -> quantized DNN -> trip). Target offline PC: Windows 10/11 x64, CPU-only Python training.
Contents: TinyML model zoos & AFCI reference implementations (ST + TI)
| Folder / file | What | Why |
|---|---|---|
01_ST_model_zoo/stm32ai-modelzoo-services_*.zip |
ST STM32 model zoo services (GitHub STMicroelectronics/stm32ai-modelzoo-services, no submodules). Contains a complete arc_fault_detection/ use case: Hydra/YAML training, evaluation, int8 quantization, prediction, benchmarking scripts (TensorFlow), docs, and the AFD test-bench dataset arc_fault_detection/datasets/afd_test_bench.zip (Arc/Normal CSV train/val/test). |
The most direct starting point for an STM32 arc-fault model. Needs its own Python env -> Part 5. |
01_ST_model_zoo/stm32ai-modelzoo_arc_fault_detection_and_doc_*.zip |
Sparse copy of STMicroelectronics/stm32ai-modelzoo: only arc_fault_detection/ (+ doc/). Includes the real (not LFS-pointer) pretrained INT8 TFLite models st_conv (time/freq, 1 & 4 channels, 512 samples) and st_dense (freq, 1 & 8 channels) with their READMEs/metrics. |
Baselines to deploy on the F407 with ST Edge AI Core / X-CUBE-AI or TFLite Micro, and to compare against your own models. |
02_TI_edge_AI/tinyml-tensorlab_*.zip |
TI Tiny ML Tensorlab (modelmaker, model zoo, training tools; BSD-3). Includes DC arc-fault and electrical-fault classification/anomaly-detection examples. | Second reference pipeline (PyTorch-based) for time-series fault models. |
02_TI_edge_AI/tinyml-modelzoo_*.zip |
TI tinyml-modelzoo (model definitions/configs). | Model architectures for time-series classification. |
02_TI_edge_AI/mspm0-sdk-2.11.00.07_edgeAI_subset.zip |
Subset of TI MSPM0 SDK 2.11: all LP_MSPM0G5187/edgeAI examples incl. ac_arc_fault_data_capture, ac_arc_fault_detection_ai_npu, ac_arc_fault_detection_live_preview, electrical_fault_ai_npu, nilm_appliance_usage_classification_npu, waveform_classifier_ai; the edge-AI/NPU library; MSPM0 Edge AI user guide. |
Firmware reference for TI TIDA-010971 (edge-AI AC arc detection): sampling, feature extraction, inference and decision logic you can port to the STM32F407. Reference only; the rest of the SDK is not included. |
Licenses: ST model zoo = ST SLA0044 / per-folder LICENSE.md (models/dataset have their own LICENSE.md); TI tensorlab = BSD-3-Clause; MSPM0 SDK = TI license (see license_mspm0_sdk_2_11_00_07.txt). Git history removed.
Rebuilding split files
Files over 100 MB were split into *.part001, *.part002, ... pieces. After downloading the whole repo:
- Windows:
powershell -ExecutionPolicy Bypass -File .\join_parts.ps1 - Linux/macOS:
bash join_parts.sh
Both scripts rebuild the files and check them against SHA256SUMS.txt.
Files (5 files, 702.9 MB total)
| File | Size | Split into |
|---|---|---|
01_ST_model_zoo/stm32ai-modelzoo-services_2026-10-07.zip |
273.7 MB | 3 |
01_ST_model_zoo/stm32ai-modelzoo_arc_fault_detection_and_doc_2026-04-21.zip |
7.2 MB | - |
02_TI_edge_AI/mspm0-sdk-2.11.00.07_edgeAI_subset.zip |
11.3 MB | - |
02_TI_edge_AI/tinyml-modelzoo_2026-10-07.zip |
48.2 MB | - |
02_TI_edge_AI/tinyml-tensorlab_2026-10-07.zip |
362.6 MB | 4 |
Each file keeps the license of its original publisher (see the notes above).
Master index
This is one of 19 repos (AFCI_part1 … AFCI_part19). Overview of all parts: AFCI_part6 / 00_START_HERE_MASTER_INDEX.md · Installation: AFCI_part6 / 00_INSTALLATION_GUIDE.md