Anti-skid braking system is the key component of civil aircraft, it is necessary to investigate the different fault modes and conduct fault diagnosis research on the system to ensure the safety of aircraft take-off and landing. In this paper, we introduce a novel fault diagnosis method that integrates one-dimensional convolutional layer-wise variation sparse autoencoder (1D-CVSAE) with multi-head self-attention mechanism to solve the difficulties encountered in extracting features from high-dimensional civil aircraft dataset and insufficient utilization of feature information. Firstly, we establish a fault simulation model for the anti-skid braking system to obtain relevant fault data. Then, we use one-dimensional convolutional autoencoder (1D-CAE) to extract effective features from the data, and combine it with the softmax classifier to build a semi-supervised network structure. Following that, we introduce the layer-wise variation sparse strategy to realize adaptive model structure optimization, and we also introduce the multi-head self-attention mechanism to further optimize the feature extraction process. Finally, the simulation results verify the superiority of the proposed method in the fault diagnosis of anti-skid braking system.