When the existing research results of bearing fault diagnosis are applied to actual production and life, there are some problems such as complicated fault data samples and numerous parameters adjusted by traditional intelligent optimization algorithm. Aiming at the problems of low accuracy and complicated parameter adjustment when using bearing fault data directly to establish diagnosis model, a rolling bearing fault diagnosis method based on singular value entropy and Northern Goshawk algorithm combined with convolutional neural network optimization support vector machine was proposed. The vibration signal and acoustic emission signal of bearing fault are selected as data set, and the mode decomposition is carried out in different ways. By using kurtosis, Euclidean distance and mean square deviation, the effective eigenmode function component recombination signal is selected, and the singular value entropy is extracted as the eigenvalue. The Northern Goshawk algorithm combined with convolutional neural network to optimize the algorithm model of support vector machine was used for fault diagnosis. The experimental results show that the proposed method is more than 97% accurate in the vibration signal data set classification of bearing faults, and 99% accurate in the acoustic emission signal data set classification.