Analysis of Series Arc Detection Using PCA

The risk of series arc faults presents a growing safety concern, as they are inherently undetectable by conventional overcurrent circuit breakers. This paper proposes a real-time series-arc detection technique based on principal component analysis(PCA). From current signals sampled at 100[kHz], we extract nine time-domain features once per 60[Hz] cycle (mean, variance, skewness, kurtosis, maximum, minimum, interquartile range, RMS, and peak-to-peak). These features are z-score standardized using parameters derived from a baseline of normal operational data and then projected onto the top three principal components. We define the Q-statistic (PCA residual variance) as the anomaly score and declare a series arc when it exceeds a predefined threshold for three consecutive cycles. Experiments show that incorporating RMS current variation enables reliable discrimination between series arcs and inrush currents demonstrating robust performance suitable for practical deployment.