Rotary Bearing Fault Diagnosis Based on Improved VMD Algorithm and ELM

The performance of the traditional variational modal decomposition (VMD) method is greatly affected by the number of modalities decomposed by artificial settings when the signal is modally decomposed, and the traditional learning method has a slow training speed. It is easy to trap in the local minimum and is sensitive to the selection of the learning rate. In view of the above problems, this paper proposes a method for fault diagnosis of rotating bearings based on an improved VMD algorithm combined with extreme learning machine (ELM). First, the vibration signal is decomposed by using VMD according to the number of different modes, and the information entropy of each mode obtained after each decomposition. At the same time, the minimum value of each information entropy is selected, and Compared with the minimum values of the information entropy under different modes, the minimum information entropy is selected, the modal number corresponding to the minimum information entropy is selected as the best modal number, and the information of each intrinsic mode function (IMF) corresponding to the optimal modal number is finally selected. Information entropy is sent to ELM as a feature for modal recognition. Experimental results show that this method can classify more than 93% of faults in rotating bearings.