A fault diagnosis technique for rolling bearings is proposed, combining fine-grained multi-scale symbolic entropy (FGMSE) with whale optimization algorithm-multi-class support vector machine (WOA-MSVM). The method involves decomposing vibration signals using FGMSE to extract multi-dimensional fault features, followed by optimizing the WOA-MSVM model parameters to enhance diagnostic accuracy. The technique is validated using a bearing fault dataset from Jiangnan University, achieving an accuracy of 99.33%. The FGMSE method reduces signal sampling point loss across different scales, providing a more comprehensive representation of signal information. Symbolic entropy effectively quantifies the complexity of signal patterns, and when combined with FGMSE, it enhances the accuracy and stability of fault diagnosis. The WOA-MSVM model optimizes parameters such as penalty factors and kernel functions, improving diagnostic performance. The method outperforms traditional models like KNN, DT, and RF in accuracy and efficiency. The results demonstrate the effectiveness of the proposed technique in fault diagnosis of rolling bearings under unbalanced sample conditions. The study highlights the potential of FGMSE-WOA-MSVM for engineering applications in rotating machinery.A fault diagnosis technique for rolling bearings is proposed, combining fine-grained multi-scale symbolic entropy (FGMSE) with whale optimization algorithm-multi-class support vector machine (WOA-MSVM). The method involves decomposing vibration signals using FGMSE to extract multi-dimensional fault features, followed by optimizing the WOA-MSVM model parameters to enhance diagnostic accuracy. The technique is validated using a bearing fault dataset from Jiangnan University, achieving an accuracy of 99.33%. The FGMSE method reduces signal sampling point loss across different scales, providing a more comprehensive representation of signal information. Symbolic entropy effectively quantifies the complexity of signal patterns, and when combined with FGMSE, it enhances the accuracy and stability of fault diagnosis. The WOA-MSVM model optimizes parameters such as penalty factors and kernel functions, improving diagnostic performance. The method outperforms traditional models like KNN, DT, and RF in accuracy and efficiency. The results demonstrate the effectiveness of the proposed technique in fault diagnosis of rolling bearings under unbalanced sample conditions. The study highlights the potential of FGMSE-WOA-MSVM for engineering applications in rotating machinery.