In order to achieve fast and accurate identification of fault arc occurring in aircraft AC circuits, a combined fault detection algorithm of a one-dimensional convolutional neural network (1DCNN) based on multi-verse optimization (MVO), incorporating a similarity-based attention module (SimAM) with time-frequency domain as the feature input, was proposed to optimize model deployment efficiency and global search capability. Firstly, the experimental platform for aircraft fault arc under multiple working conditions was established according to the required standard, and the current data under fifteen different load conditions such as point contact, simulated vibration, and anti-interference was collected. Secondly, the time-domain signals were transformed into frequency-domain signals using fast Fourier transform for feature analysis. The frequency spectrum indicated that the 1 000 ~ 5 000 Hz components contained more feature information when the fault arc was generated. Therefore, the frequency-domain features within this range were fused with the original time-domain features into one-dimensional data, which were then normalized and used as model input. Finally, the MVO-SimAM-1DCNN detection model was built and trained. The optimal hyperparameters were obtained through MVO and then applied to the neural network, resulting in a test accuracy of 97.72%. And multiple performance parameters were introduced to evaluate the effectiveness of the model. The results show that this feature extraction strategy has certain effectiveness and this combined detection model is superior to the traditional model in terms of recognition accuracy and detection speed. Meanwhile, it has fewer network layers to improve deployment efficiency, providing ideas for the research and development of aviation fault arc detection devices.