As a new achievement in the field of pattern recognition and machine learning, deep learning has broad prospects in the field of equipment fault diagnosis and health management. In this paper, a new method of fault diagnosis is proposed based on the characteristics of equipment fault big data and the advantages of deep learning theory. According to the principle of the denoising auto-encoder, the unsupervised feature learning of the training network is achieved, and the structuring of the whole neural network is completed. According to the type of fault, the output layer is determined. Using the BP algorithm, the supervised fine-tuning of the whole network is carried out, and thus the accuracy of fault classification is enhanced. By means of the above methods, the module-level fault diagnosis of a communications station is completed through experiments.