Remaining Useful Life Prediction of Rolling Bearings Based on Multi-Feature Fusion

As an important component of mechanical systems, the monitoring of the health state and remaining life prediction of bearings is of great significance in preventing failures and reducing downtime. In this paper, a bearing remaining life prediction method combining variational modal decomposition (VMD), symmetrized dot pattern (SDP) and Weighted Dynamic Localized Linear Embedding (WD-LLE) model is proposed. First, the bearing vibration signals are decomposed using VMD to obtain several intrinsic modal function (IMF) components; subsequently, the SDP method is used to extract valid features from each IMF component to ensure the validity of the features. Next, the time-domain, frequency-domain, and entropy features of the IMF components are extracted separately, and the trend factor is utilized for feature selection. In order to reduce the dimensionality and retain the key features, a Weighted Dynamic Local Linear Embedding (WD-LLE) algorithm is proposed to reduce the dimensionality of the features. Finally, the final trend data of the bearing is obtained by weighted fusion of the WD-LLE results of each IMF component, and the remaining life is predicted using the Bi-LSTM model. The experimental results show that the method has higher accuracy and robustness compared with the traditional prediction methods.