Abstract: Because of its compact structure and good friction performance, coated spherical plain bearing has a wide application prospect in the field of aerospace equipment. The effective prediction of its remaining useful life (RUL) can provide a certain theoretical basis for equipment maintenance. Therefore, a prediction method of RUL based on Long Short-term memory neural network (LSTM), variational mode decomposition (VMD) and ensemble empirical mode decomposition (EEMD) is proposed. First of all, VMD and EEMD are used to extract the features of bearing friction torque signal. Features were selected according to Pearson correlation coefficient between features and bearing swing times and 3 groups of feature sequences with high correlation coefficients were selected. The selected features are relatively normalized as the model input to reduce the influence of friction torque amplitude changes under different working conditions. Finally, the hyperparameter optimization interval is selected to perform Bayesian optimization on LSTM, so as to obtain the Bayesian optimization-LSTM model and this model is constructed to predict the RUL of coated spherical bearing. The results show that proposed the model integrates multiple signal features that can characterize the degradation information of coated spherical bearings, and has high prediction accuracy of RUL for bearings under different working conditions, and also shows its good generalization performance.

Key words: coated spherical plain bearing, variational mode decomposition, ensemble empirical mode decomposition, long short-term memory neural network, Bayes optimization