Skip to content

ac_arc_fault

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

ac_arc_fault

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
AFCI_LabellingScript.py
AFCI_LabellingScript.py
 
 
config_MSPM0.yaml
config_MSPM0.yaml
 
 
readme.md
readme.md
 
 
readme_labelling.md
readme_labelling.md
 
 

readme.md

AC Arc Fault Detection

Laavanaya Dhawan, Nathan Nohr, Vinamra Shrivastava, Akshat Aggarwal

Overview

The AC Arc Fault Detection application is an Edge AI solution that runs on the MSPM0G5187 microcontroller with integrated Neural Processing Unit (NPU). It detects series arc faults in residential and commercial electrical systems, which are a leading cause of electrical fires. This reference design combines the TIDA-010971 analog front end with machine learning inference to achieve high detection accuracy while maintaining immunity to masking loads per UL 1699 guidelines. This application enables engineers to develop Arc Fault Circuit Interrupter (AFCI) products for compliance with National Electrical Code (NEC) requirements while leveraging the power efficiency and performance of Edge AI on microcontrollers.

Problem and Solution

  • Arc faults cause over 30,000 home fires annually, resulting in hundreds of deaths and over $1 billion in property damage
  • Traditional circuit breakers fail to detect arc faults because they operate at different frequencies
  • Edge AI enables complex pattern recognition, multi-feature analysis, and adaptive detection

Key Performance Targets

  • Less than 10ms response time
  • Greater than 95% detection accuracy
  • Compliance with UL 1699 requirements

System Components

  1. Hardware:

    • MSPM0G5187 microcontroller with integrated NPU Link
    • TIDA-010971 Analog Front End with PCB Rogowski coil Link

  2. Software:

    • Code Composer Studio 12.x or later
    • MSPM0 SDK 2.08.00 or later
    • TI Edge AI Studio

Dataset Labelling

For preparing and labelling your AC arc fault dataset, use the AFCI_LabellingScript.py script. Refer to readme_labelling.md for detailed instructions on dataset format, labelling modes, and usage.

Dataset Operations

Data collection requires multiple scenarios:

  • Normal Operation: Inductive loads, LED lighting, power supplies, mixed loads
  • Arc Fault: Series and parallel faults
  • Masking Loads: Vacuum cleaners, drills, dimmers

Our example dataset(ac_arc_fault_log300.zip) that is used in the config file includes:

  • 100+ captures per load type
  • Appliances: Dimmer, SMPS, Drill, Compressor, Resistive, Vaccum

Feature Extraction Pipeline

  1. ADC Sampling: 512 samples at 107 kSps (~4.76ms per frame)
  2. Real FFT: 512-point FFT using ARM CMSIS-DSP
  3. Complex Magnitude Calculation
  4. DC Removal
  5. Binning: Average 8 adjacent FFT bins → 32 features
  6. Normalization to INT8 range
  7. Frame Concatenation: Stack 8 frames (256 total features)

Model Architecture Options(GUI only Options)

Four pre-configured model architectures:

Model Parameters Flash(MSPM0G5187) Inference Time Accuracy Notes
ArcFault_model_200_t ~200 3.6 KB - 99.60% Simplest, smallest & fastest
ArcFault_model_300_t ~300 3.9 KB - 99.60% -
ArcFault_model_700_t ~800 4.5 KB - 99.42% Sweet spot between speed & memory
ArcFault_model_1400_t ~1600 5.6 KB 0.7 ms 99.88% Recommended - Most accurate

Model Architecture Options(Available on Tensorlab CLI Tools)

Eleven pre-configured CNN/ResNet models available (~100 to ~20K parameters), most with NPU support.

See tinyml-modelzoo/README.md for full details.

Performance Metrics for MSPM0G5187 with NPU

  • End-to-end latency: <150ms (including 8-frame voting)
  • Detection accuracy: >99%
  • False positive rate: 0.01%
  • Precision: 99.97%
  • Recall: 99.72%
  • F1-Score: 99.84%

Training and Deployment Process

NOTE: Running the config yaml takes care of everything including feature extraction, training, quantization and compilation.

  1. Training:
  • Use TI Edge AI Studio (GUI) or tinyml-tensorlab (CLI)
  • Batch size: 50, Learning rate: 0.04, Optimizer: SGD
  • Enable Quantize-Aware Training for INT8 accuracy
  1. Quantization:
  • INT8 quantization required-Enabled in the config file by default aswell as in Edge AI Studio
  • ~4x reduction in model size
  1. Compilation:
  • TI Neural Network Compiler converts trained model
  • Generates model.a, interface headers, and configuration

How to Run

After completing the repository setup, run the following command from the tinyml-modelzoo directory:

Windows:

.\run_tinyml_modelzoo.bat examples\ac_arc_fault\config_MSPM0.yaml

Linux:

./run_tinyml_modelzoo.sh examples/ac_arc_fault/config_MSPM0.yaml

References