LIFE PREDICTION OF ROLLING BEARING BASED ON MULTI-RESOLUTION SINGULAR VALUE DECOMPOSITION AND ECNN-LSTM

Abstract

In order to solve the problem of the lack of life characteristic characterization ability in the prediction of the remaining life of the rolling bearing,a method of the residual life prediction of the rolling bearing based on multi-resolution singular value decomposition and ECNN-LSTM is proposed. Firstly,the time-frequency information of vibration signals in onedimensional multi-space scale is extracted by using the multi-resolution singular value decomposition method,and the health stage is divided according to the standard deviation at the initial time. Secondly,a high-efficiency channel attention mechanism module was added to the two-layer one-dimensional convolutional neural network structure,and the convolution kernel was adaptively adjusted for multi-channel interaction without dimension reduction,so as to fully extract bearing degradation characteristics and establish effective life degradation indicators. Finally,MSE loss function is used to achieve the residual life prediction on LSTM.The feasibility and effectiveness of the proposed method are verified by Cincinnati whole life data.

Keywords

• Rolling bearing;Remaining life prediction;Efficient channel attention mechanism;Multi-resolution singular value decomposition;Long and short term memory network（LSTM）