Bearing Fault Diagnosis Based on SMA-VMD and CNN-LSTM

Abstract

To address the challenges of manual parameter setting dependency and low accuracy in bearing fault diagnosis using Variational Mode Decomposition (VMD), a method integrating Slime Mould Algorithm (SMA), VMD, and Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) network is proposed. Firstly, the SMA is employed to optimize the critical parameters of VMD using minimum information entropy as the fitness function, which resolves the issues related to manual parameter settings in VMD. Subsequently, the optimal parameter combination is used to process the bearing signals and extract the relevant Intrinsic Mode Functions (IMF). Finally, these IMFs are input into the CNN-LSTM network for fault diagnosis. The results demonstrate that this method significantly enhances the accuracy of motor bearing fault diagnosis, further improving the fault diagnosis accuracy.