Arc Fault Detection

Table of Contents

• Introduction
• Supported Combinations
• Dataset and Model Details
  • Dataset
  • Model Architecture
  • Input Features
  • Output
• Feature Extraction Configuration
• Steps to Run the Example
• See Also
• Sample Output

Introduction

An arc fault is an unintended electrical discharge that occurs when electrical current flows through the air between conductors or from a conductor to ground. Arc faults generate intense heat that can ignite surrounding materials and cause electrical fires. Unlike short circuits that typically trigger circuit breakers immediately, arc faults may draw current below the trip threshold while still producing dangerous heat. In DC systems like solar installations, arc faults are particularly concerning as they can sustain more easily than in AC systems.

This project demonstrates implementation of an AI-based arc fault detection system on AM26x microcontrollers. It showcases how to deploy machine learning models for real-time electrical arc fault classification in embedded systems, helping prevent electrical hazards through early detection.

Supported Combinations

Parameter	Value
CPU + OS	r5fss0-0 nortos
Toolchain	ti-arm-clang
Board	am261x-lp
Example folder	examples/ai/arc_fault/

Dataset and Model Details

Dataset

TI has created a specialized arc fault dataset containing DC current measurements under various voltage and current configurations.

Parameter	Value
Sensor	Current Meter
Channels	1 (DC Current)
Samples per File	250,000
Total Files	24 files (12 files in each of 2 classes)
Voltage Configurations	312 V, 318 V, 607 V
Current Configurations	3 A, 8 A, 8.5 A

Output Classes:

• Class 0: Normal Operation
• Class 1: Arc Fault

Model Architecture

This lightweight classification model CLS_1k_NPU contains approximately 1,000 parameters and follows a streamlined architecture consisting of four convolutional layers (each enhanced with BatchNorm and ReLU activation functions) followed by a single linear layer.

Input Features

The model takes 4D input (N,C,H,W):

• N (1): batch size which is restricted to 1
• C (1): channels which is 1 for single DC current
• H (256): samples of timeseries DC current which is 256 in this example
• W (1): width of samples is restricted to 1 for timeseries applications

Output

This model produces a 1D output representing the two possible classes - normal operation or arc fault condition - providing a straightforward binary classification decision.

Feature Extraction Configuration

The feature extraction pipeline is configured for frequency analysis of DC current signals:

• FE_WIN: Applies windowing to reduce spectral leakage
• FE_FFT: Performs Fast Fourier Transform (1024-point)
• FE_NORMALIZE: Scales FFT output by frame size
• FE_BIN: Groups frequencies into 256 bins
• FE_LOG: Applies logarithmic scaling
• FE_CONCAT: Combines outputs from multiple frames

The pipeline takes 1024-sample frames, computes FFT, and bins to 256 features. Arc faults exhibit distinctive frequency signatures that differ from normal operation.

Steps to Run the Example

• When using CCS projects to build, import the CCS project for the required combination and build it using the CCS project menu (see Using SDK with CCS Projects).
• When using makefiles to build, note the required combination and build using make command (see Using SDK with Makefiles)
• Launch a CCS debug session and run the executable, see CCS Launch, Load and Run
• The application will classify current readings as normal or arc fault

See Also

AI Examples

Sample Output

Arc Fault Detection Example Started ...
Feature extraction mismatches 0
All tests have passed!!
Golden vectors matched: 2 not matched: 0