Bidirectional self-Attention Gated Recurrent Unit for Health Index Prediction of Rolling Bearings

In rotating mechanical equipment, rolling bearings are commonly used mechanical components, the failure of which can lead to serious accidents and significant economic losses. Therefore, it is necessary to construct health index to reflect the health status of rolling bearings, and the prediction of health index is an indispensable part of predictive maintenance. In order to accurately predict the health index of rolling bearings, this paper proposes a new gated recurrent unit neural network that can generate an initial hidden state containing important information, namely, a bidirectional gated self-attention unit. Firstly, the time-domain features of the rolling bearings life-cycle vibration data are extracted, and then fused as health index using deep denoising autoencoder. Secondly, a bidirectional self-attention gated recurrent unit is used to predict future health index sequences based on the existing health index vectors. Experimental results show that the proposed bidirectional gated self-attention unit can effectively predict the health index of rolling bearings, and has higher prediction accuracy and convergence speed than traditional prediction methods.