Abstract:  Rotating machinery vibration signals had nonlinear and non-stationary characteristics, and the early weak fault signals were easily disturbed by noise and difficult to be extracted fault features and identified fault types in fault diagnosis. Aiming at these problems, a method for fault diagnosis of rotating machinery based on variational modal decomposition (VMD), singular value decomposition (SVD) and support vector machine (SVM) was proposed. First, the original vibration signal was decomposed by variational modal decomposition to obtain several component signals. Then, the signal was reconstructed for each component signal, and the singular value feature vector of the reconstructed signal was extracted using singular value decomposition (SVD). Finally, these extracted feature vectors were inputted to the support vector machine (SVM) for pattern training and the test set was used to complete the fault diagnosis. The effectiveness of the proposed method was verified by the measured experimental data of the double-span rotor fault simulation experimental platform. The experimental results show that the singular values of the intrinsic modal function (IMF) component matrix obtained based on the VMD-SVD method show good stability, and show good separability in the three-dimensional characteristic scatter plot. Under variable working conditions and different speeds, the proposed method has a higher recognition accuracy rate than other combined methods, and the average classification recognition rate reaches 95.96% and 95.95% respectively, which can effectively identify the type of bearing fault.

Key words:  variational modal decomposition (VMD); singular value decomposition (SVD); support vector machine (SVM); fault diagnosis