Real-Time Series Arc Fault Detection and Appliances Classification in AC Networks Based on Competing Convolutional Kernels

This paper presents a method to detect and classify series arc faults affecting domestic AC electrical circuits by the analysis of electric current time series data, based on the HYDRA (HYbrid Dictionary-Rocket Architecture) algorithm, a fast dictionary method for time series classification employing competing convolutional kernels. The key novel contributions are twofold: <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$i)$</tex-math></inline-formula> competing convolutional kernels are suitable to effectively extract features representing an effective set of arc fault detection indicators, and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$ii)$</tex-math></inline-formula> the classification performed in this way is feasible to be executed in real time. The proposed method is validated using a public database, where data from 13 different types of loads is collected according to the IEC 62606 standard. To reduce inference time and optimize the algorithm for embedded control units, a feature reduction strategy is employed. The effectiveness of the proposed method is demonstrated through experimental tests conducted under both arcing and non-arcing conditions and across different load types. Moreover, its accuracy is also tested in case of transients caused by operational changes in common electrical appliances. Achieved results show a detection accuracy of approximately 99%, with appliance classification performance around 98%, with inference times ranging from 2.8 ms to 172.0 ms while executing the algorithm on an ARM Cortex-based board.