Abstract: Series arc faults in aircraft AC systems pose a critical threat to flight safety because the current amplitude does not exceed protection thresholds, signal distortion is subtle, the detection process is highly susceptible to environmental disturbances and under coupled vibration-humidity conditions, the difficulty of fault detection increases significantly. To address the challenge of robust identification in 115 V/400 Hz systems under multiple operating conditions, this study constructs a series arc fault experimental platform capable of simulating combined vibration and humid-thermal environments, covering representative resistive and resistive-inductive load scenarios. Time-domain statistical analysis, frequency-domain feature analysis, and db4 wavelet multi-scale decomposition are employed, combined with correlation-thresholding, F-value ranking, permutation importance validation, and PCA dimensionality reduction, to obtain a low-redundancy, multi-domain feature set. A grid-optimized RBF-SVM classification model is then applied for fault recognition. Results show that vibration increases the spectral peak of a 10 A series arc fault under resistive-inductive load from 1 767.92 to 9 120.03. Humidity exhibits a dual “suppression-sharpening” effect on transient arc behavior: although wavelet energy decreases by 47.7%, pulse factor and kurtosis increase, indicating enhanced intermittency of discharges. Vibration demonstrates a “promotion-dispersion” effect, raising reignition frequency but dispersing pulse intensity. When coupled, nonlinear competition occurs, leading to an extreme suppression of arc transients, with wavelet energy reduced to 284.69. The proposed multi-domain fusion model achieves an F1-macro score of 0.99, improving by 62.3% compared with time-domain features alone.

Key words: aviation power supply, series arc fault, feature extraction, wavelet analysis, integration of multi-domain features