This paper uses the general model Arc-L to simulate the fault arc of 220V/50Hz AC arc. Firstly, characteristic analysis is conducted on the simulated fault arc current and voltage signals, and a diagnostic method for fault arc current was determined. In terms of the architecture of the diagnostic model, this paper combines CNN with Transformer to enable the model to quickly capture local features while effectively capturing long-distance dependencies and integrating global information. The model applies a position encoding method specifically proposed for time series data, which can better capture the input order of time series data and effectively improve the robustness and generalization of the model. The accuracy index F1 scores of the CNN-Transformer model proposed in the paper are 97.42% and 88.46% respectively, on high and low resolution data in the Integer arc detection public dataset.