Rolling bearing fault diagnosis method based on GJO–VMD, multiscale fuzzy entropy, and GSABO–BP neural network

Accurate fault diagnosis of rolling bearings is hindered by the weak nature of early fault signals and the limited availability of labeled data, especially under small-sample conditions. To overcome these challenges, this paper proposes a novel method combining golden jackal optimization (GJO) with improved variational mode decomposition (VMD), enhanced feature extraction, and optimized classification. First, GJO is used to optimally determine the key decomposition parameters of VMD, thereby improving the accuracy of vibration signal decomposition. A comprehensive discrimination factor algorithm then selects fault-sensitive intrinsic mode functions, and the signal is reconstructed to enhance fault characteristics. Multiscale fuzzy entropy is calculated from the reconstructed signals at multiple scales to build distinct state feature vectors. These vectors are fed into a back-propagation neural network optimized via the golden sine subtraction-average-based optimizer for precise fault classification. The method’s effectiveness is verified through simulation and experimental data. Compared with conventional approaches, it shows superior performance in extracting weak fault features and maintaining high diagnostic accuracy under small-sample scenarios. This integrated framework presents a robust solution for rolling bearing fault diagnosis.