Bearing remaining useful life prediction using self-adaptive graph convolutional networks with self-attention mechanism

Yupeng Wei,Dazhong Wu,Janis Terpenny
DOI: https://doi.org/10.1016/j.ymssp.2022.110010
IF: 8.4
2023-04-01
Mechanical Systems and Signal Processing
Abstract:Bearings are commonly used to reduce friction between moving parts. Bearings may fail due to lubrication failure, contamination, corrosion, and fatigue. To prevent bearing failures, it is important to predict the remaining useful life (RUL) of bearings. While many data-driven methods have been introduced, very few studies have considered the correlation of features at different time points, such a correlation could be used to identify and aggregate features at different time points for improving the robustness of predictive models. Moreover, many existing data-driven methods leverage neural networks with recurrent characteristics such as recurrent neural network (RNN) and long short term memory (LSTM). These methods are ineffective in processing long sequences and require longer training time due to the recurrent characteristics. To address these issues, a Siamese LSTM network is firstly introduced to classify degradation stages before predicting the RUL of bearings. Then we introduce a self-adaptive graph convolutional network (SAGCN) along with a self-attention mechanism in order to consider the correlation of features at different time points without using recurrent characteristics. Experimental results have demonstrated that the proposed method can accurately predict the RUL with a minimum average root mean squared error of 0.119, and outperforms existing data-driven methods, such as graph convolutional network, convolutional LSTM, convolutional neural network, and generative adversarial network.
engineering, mechanical
What problem does this paper attempt to address?