In light of the problems caused by complex equipment structure, working environment noise, and big data, a deep learning fusion model is proposed to achieve efficient and accurate fault diagnosis. First, the denoising autoencoder is used to process the random noise of the original signal and learn the low-level features. Then, the deep belief network is used to learn the deep features based on the learned low-level features. Finally, the fused depth features are fed into the PSO-SVM to train the intelligent diagnosis model. The proposed method is applied to the fault diagnosis of rolling bearings, The results show that the method proposed is more efficient and robust than the existing methods.