Mechanical equipment fault detection is of great significance in industrial applications. The traditional fault diagnosis method based on vibration signal processing and analysis relies on rich professional knowledge and artificial experience, and it is difficult to achieve accurate feature extraction and fault diagnosis. In this paper, the deep learning method can be used to automatically learn the characteristics of deep features from the data. A qualitative and quantitative diagnosis method for rolling bearing faults based on improved convolution deep belief network is proposed. First, in order to provide better shallow inputs, the original vibration signal is converted to the frequency domain signal by the fast Fourier transform. Secondly, in the process of model training, the Adam optimizer is introduced to speed up the model training and improve the convergence speed of the model. Finally, in order to make full use of the characterization capabilities of each layer, the model structure is optimized to come up with a multilayer feature and fusion learning structure is proposed to improve the generalization ability of the model. The experimental results show that the proposed improved model has better diagnostic accuracy than the traditional stack autoencoder (SAE), artificial neural network (ANN), deep belief network (DBN) and standard convolution deep belief network (CDBN). It has better diagnostic accuracy and effectively realizes qualitative and quantitative diagnosis of bearing faults.