A method based on symplectic geometric modal decomposition (SGMD), sensitive parameters and kernel fuzzy C-means clustering (KFCMC) was proposed for fault diagnosis of inner and outer rings of rolling bearings. The actual measured multi-mode fault vibration signal of hydraulic pump was studied based on SGMD. Based on the proposed similarity analysis method, the modal component with rich information of operation characteristics was reconstructed and used as the data source. The time-domain and frequency-domain parameters were extracted based on the data source, and the kurtosis, margin index, peak index and other sensitive parameters were selected as the feature vectors by using the popular learning method. The KFCMC was used to diagnose different faults of inner and outer rings of rolling bearings. Through the simulation and measurement of vibration signals of inner and outer ring faults of rolling bearings, it was validated that the method can effectively diagnose different faults of rolling bearings.