Objective To address the challenges of extracting fault features and achieving accurate fault diagnosis for rolling bearings a method based on the optimized sparrow search algorithm variational mode decomposition VMD was proposed to efficiently extract rolling bearing fault features. Methods Firstly the sparrow search algorithm is improved by integrating the Osprey and Cauchy mutation resulting in the Osprey-Cauchy-Sparrow Search Algorithm OCSSA . Secondly the OCSSA was used to optimize the parameters of VMD for decomposing bearing signals. Finally the feature vectors were input into a convolutional neural network-bidirectional long short-term memory network CNN-BiLSTM for identifying rolling bearing fault types. Results Experimental results showed that the fault diagnosis accuracy of the diagnostic model based on OCSSA-VMD feature extraction was 99. 333%. Compared with the Sparrow Search AlgorithmVMD Grey Wolf Optimizer-VMD Particle Swarm Optimization-VMD and traditional VMD feature extraction methods the fault diagnosis accuracy of the proposed method was improved by 3. 666% 5% 6. 667% and 9% respectively. Conclusion This method effectively captures fault features and significantly enhances the fault diagnosis accuracy.