Abstract:

Aiming at the problems caused by the data increase in the early fault monitoring of modern rolling bearings，a fault diagnosis model for deep bidirectional long and short memory neural network（DBiLSTM）based on variational mode decomposition（VMD）and TEO energy window was put forward. Firstly，the instantaneous energy characteristics of bearing vibration signals were extracted by the VMDTEO window function that optimized by the improved fruit fly algorithm. Meanwhile，the characteristic matrix with time characteristic was constructed. Secondly，DBiLSTM network model was trained by using the training set to determine the model parameters. Finally，the trained model was applied in the test set to generate fault recognition results. The test used Case Western Reserve University bearing fault data set，and the results show that this method can effectively identify vibration signals of rolling bearings with various fault types and different damage levels when dealing with large amounts of data problems.

Key words: variational modal decomposition, TEO energy window function, DBiLSTM, bearing failure diagnosis