To achieve the remaining life prediction of bearings in ship mechanical equipment, a multi-task gated networks prediction model based on the Bidirectional Gated Recurrent Unit (BiGRU), Variational Autoencoder (VAE), and Multi-gate Mixture-of-Experts (MMoE) is proposed.

First, the time-domain features of the bearing signals are calculated to characterize the basic degradation trends in the monitoring data. Then, a multi-task gated networks prediction model composed of bearing Health State (HS) assessment and Remaining Useful Life (RUL) prediction subtasks is established. In the subtasks, BiGRU and VAE are used to extract the degradation information from the trend signals of the time-domain features, and then MMoE is utilized to adaptively separate the distinctive features of the subtasks. Finally, the effectiveness is verified on the XJTU-SY bearing dataset.

The results show that, compared with classic time-series data prediction models such as Long Short Term Memory (LSTM), the multi-task gated networks prediction model has higher prediction accuracy, with the error metrics Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) improved by 62.5% and 67.81% respectively.

The proposed method can achieve the prediction of the remaining life of bearings and has certain reference value for the health management and intelligent operations and maintenance (O&M) of ship mechanical equipment.