Remaining useful life prediction method based on gated dilation causal convolution

Rolling bearings are a crucial element of rotating mechanical equipment. Therefore, predicting the remaining useful life (RUL) of this equipment is vital. Monitoring data for rolling bearing operation usually consists of a long-life cycle sequence covering the entire life cycle of the bearing. This process requires deep learning models to model global and local modeling capabilities. However, traditional convolutional neural networks are not very good at modeling global features. We proposed a novel RUL prediction framework based on gating inflation causal convolution to address the above shortcomings. This framework includes a multi-feature squeeze excitation unit that adaptively proofreads feature responses from local and global perspectives. A sinusoidal position coding is also designed to allow the network to obtain information between the distant time steps. The framework also features a gated dilated causal convolution (GDCC) network to reduce the vanishing gradient probability, retain the nonlinear ability of the network, and dynamically realize information changes. Finally, the effectiveness of this method is verified through comparative experimental analysis, ablation study experiments, and visual analysis.