Purpose <p>Setting parameters for variational mode extraction (VME) has been a challenge in fault feature extraction. Previous studies have often optimized these parameters using fitness functions like entropy value or kurtosis to evaluate the VME decomposition results. However, these methods overlook the impact on subsequent fault classification. This paper aims to propose a more effective method for fault feature extraction in rotating machinery.</p> Methods <p>A fault feature extraction method based on VME and the synthetic detection index (SDI) is introduced. The VME parameters are optimized using the particle swarm optimization (PSO) algorithm, with the maximum value of SDI as the objective function. The optimized VME is then used to extract features from the original signals and build a fault feature set for fault classification.</p> Results <p>The proposed method was tested on two datasets, and the numerical results demonstrate its superior performance in fault classification compared to other methods.</p> Conclusion <p>The results show that the proposed method, which optimizes VME parameters using PSO, enhances fault feature extraction and classification accuracy in rotating machinery, outperforming other comparative techniques.</p>