An Integrated Gearbox Fault Diagnosis Method Based on OCSSA‐Optimized VMD and CNN‐BiLSTM

ABSTRACT The integration of precise data feature extraction and deep learning represents a pivotal technical approach for effective fault diagnosis. This paper proposes an integrated fault diagnosis method based on Variational Mode Decomposition (VMD) optimized by an Osprey‐integrated Cauchy mutation Sparrow Search Algorithm (OCSSA), combined with a Convolutional Neural Network‐Bidirectional Long Short‐Term Memory (CNN‐BiLSTM) network. Firstly, a multi‐indicator weighted fusion fitness function—integrating kurtosis, envelope entropy, and the average correlation coefficients is designed. Subsequently, the OCSSA algorithm employs the fitness function to search the optimal VMD parameters (mode number K and penalty factor α ) to accurately isolate the optimal Intrinsic Mode Function(IMF) rich in fault information. Secondly, a nine‐dimensional feature vector is extracted from the optimal IMF to construct the feature dataset. Subsequently, a CNN‐BiLSTM network is constructed and trained using this low‐dimensional feature dataset to achieve precise fault classification. Finally, the proposed OCSSA‐optimized VMD‐CNN‐BiLSTM fault diagnosis method is validated using both a public dataset and a private measured dataset. The results demonstrate that the proposed methodology can accurately identify single and compound faults of gears and bearings, achieving an average accuracy of 98%. Compared with three typical intelligent fault diagnosis models, the proposed model exhibits superior diagnostic accuracy and robustness.