Optimized matching feature extraction method for bearing fault diagnosis

Feature extraction methods based on mode decomposition and dynamic metrics can characterize signal complexity in different frequency bands. However, existing methods usually ignore the complementarity among different metrics and the matching relationship between decomposed modes and metrics, which limits their generalization ability. To address these issues, this study proposes an optimized matching feature extraction method based on reconstructed complete ensemble empirical mode decomposition with adaptive noise (RCEEMDAN) and the optimized mode-metrics matching (OMMM) strategy. RCEEMDAN is developed to reduce redundant modes in CEEMDAN and obtain more accurate and refined modal components. OMMM is designed to determine the optimal matching relationship between decomposed modes and multi-source metrics. By introducing metrics generated from different evaluation mechanisms, OMMM enhances feature representation through metric complementarity. Simulation experiments demonstrate that RCEEMDAN achieves better decomposition performance than four comparison algorithms. Classification experiments on four types of chaotic signals further show that the proposed optimized matching feature extraction method provides superior classification performance compared with individual metrics. Real-world bearing signal experiments verify that the proposed method can effectively extract discriminative fault features. These results indicate that the integration of modes and metrics generated by different mechanisms improves the representation ability and generalization performance of feature extraction, making the proposed method effective for both simulated nonlinear signals and real bearing fault signals.