A global attention based gated temporal convolutional network for machine remaining useful life prediction

As the core technique of the prognostic and health management field, data-driven remaining useful life (RUL) prediction generally requires abundant data to construct reliable mappings from monitoring data to machines' RUL labels. However, the diverse working conditions of machines can lead to their different degradation trajectories, which makes similar data indicate diverse RULs of different machines. When predicting RULs with monitoring data, the phenomenon causes a severe label confusion problem and limits the performance of datadriven RUL prediction methods. In this paper, a new gated-temporal-convolutional-network-based method is proposed for RUL prediction tasks of machines. To handle the label confusion problem, a novel global attention mechanism is proposed, which enables the proposed model to identify confused data by the difference in machines' global degradation tendencies. Besides, a new temporal convolutional network with a gating mechanism is proposed for better feature extraction performance. Moreover, a new nearest-neighbor-based data compensation strategy is designed to simplify data distributions. Both strategies also contribute to the solution of the problem. The proposed method is verified on an aircraft turbofan engine dataset and a bearing dataset. The experiment results show the effectiveness of the proposed method.