A Novel Methodology for Series Arc Fault Detection by Temporal Domain Visualization and Convolutional Neural Network.

Area of Science:

• Electrical Engineering
• Machine Learning
• Fire Safety Engineering

Background:

• AC arc faults are a primary cause of residential electrical fires, posing significant safety risks due to high temperatures.
• Existing arc fault detection methods face challenges in efficiency and accuracy, driving the need for advanced solutions.
• Machine learning is increasingly explored for reliable arc fault diagnosis.

Purpose of the Study:

• To propose and validate a novel Temporal Domain Visualization Convolutional Neural Network (TDV-CNN) for accurate AC arc fault detection.
• To assess the performance of the TDV-CNN methodology across various electrical loads and fault conditions.
• To demonstrate the potential of TDV-CNN for enhancing electrical safety in residential settings.

Main Methods:

• Collected current data from series arc faults using a current transformer and high-speed data acquisition system.
• Filtered the acquired current signals and converted them into time-sequential grayscale images.
• Developed and applied a TDV-CNN model to classify arc fault signals from five different electrical loads.

Main Results:

• The TDV-CNN methodology achieved classification accuracy of 98.7% or higher for identifying the working states of five distinct electrical loads across ten categories.
• Experimental validation demonstrated the method's effectiveness with varying signal characteristics.
• The proposed technique proved reliable for series arc detection.

Conclusions:

• The TDV-CNN methodology offers a highly accurate and reliable approach for detecting series AC arc faults.
• This machine learning-based technique has significant potential for improving electrical fire safety.
• The developed methodology can be adapted for fault diagnosis in other related fields.