Abstract: Aiming at the problem that the prediction of remaining useful life (RUL) of rolling bearings lacks reliable indicators to describe their health status due to the nonlinearity and non-stationarity of vibration signals, a prediction method combining graph convolutional network (GCN) and long short-term memory (LSTM) networks is proposed in this paper. Firstly, the bearing vibration signal is decomposed by empirical mode decomposition (EMD) to obtain the intrinsic mode functions (IMF), followed by normalization of the IMFs, and computation of adjacency and characteristic matrix; Secondly, the adjacency and feature matrix are used as the inputs of GCN to capture the data relationship and mine the deep features. Then, these deep features, along with the IMFs, are input into the LSTM network to realize time series relationship modeling and construct health indicators (HI). Finally, after using the moving average filter to eliminate the oscillation, the HI was polynomial fitted to calculate the threshold moment and determine the bearing RUL. Taking the IEEE PHM 2012 data challenge data set and XJTU-SY rolling bearing accelerated test data set as objects, the proposed method is verified. The experimental results show that when using HI constructed by GCN-LSTM model for rolling bearing RUL prediction, the prediction results closely approximate the real value, which has certain advantages in practical application.

Key words: rolling bearing, graph convolution network, long short-term memory network, remaining useful life prediction