The intelligent diagnosis method has been successfully applied in the field of mechanical equipment bearing fault recognition. Aiming at the problems of low diagnosis accuracy and insufficient manual feature extraction of existing methods， an intelligent recognition method based on convolutional deep belief network and improved kernel-based extreme learning machine （CDBN-IKELM） is proposed. Firstly， the convolutional deep belief network （CDBN） is used to extract the fault features from the original signal， and then isometric feature mapping is adopted to reduce the dimension of the extracted multi-dimensional features for removing redundant feature information. Finally， improved kernel-based extreme learning machine （IKELM） is utilized to classify rolling bearing fault under variable conditions， and particle swarm optimization （PSO） is used for optimizing the important parameters of the model. Through the identification of various bearing faults under different working conditions， it is verified that the method can effectively diagnose the bearing faults under varying conditions with high accuracy， and the effect is better than existing intelligent fault diagnosis methods.