Fault severity assessment of rolling bearings method based on improved VMD and LSTM

In order to solve the problem of selection of appropriate wavelet basis function and clearly show the physical meaning of Empirical Mode Decomposition (EMD), an improved Variational Mode Decomposition (VMD) method with Long Short-Term Memory (LSTM) neural network is proposed. With the Cuckoo Search (CS) algorithm, the central frequency updating rules of VMD are optimized. And the low efficiency and local optimum problem is avoided. Meanwhile the decomposition layer number is found by the instantaneous frequency theory. For improving the prediction accuracy in traditional regression prediction methods, a LSTM neural network is designed for regression prediction of time sequence characteristics. The proposed method is implemented on actual bearings data which is derived from the bearing laboratory of Case West Reserve University in the United States and the University of Cincinnati Bearing Data Center. The experimental results showed that the improved VMD method was more robust and more accurate than the other traditional methods. And it has some practical value for real application and guiding significance for theory.