Remaining useful life prediction of rolling bearings based on dual-branch global-local cross fusion network

Abstract The prediction of the remaining useful life (RUL) of industrial systems is of paramount importance in ensuring the stability and safety of such systems. Addressing the issues of incomplete signal feature extraction and neglect of inter-feature influence in traditional bearing life prediction methods, this paper proposes a dual-branch global-local cross-fusion network for bearing life prediction. Firstly, the raw vibration signal is subjected to preprocessing in order to construct health indicators and divide samples, and large kernel convolution is utilized for the purpose of extracting signal features. Then, two branches are employed to obtain global and local features respectively. Subsequently, a cross-attention mechanism is employed to fuse both local and global features, thus improving the accuracy and versatility of the model. Finally, the fully connected layer and regression function are utilized for bearing life prediction. The validity of the proposed method is established through the utilization of two benchmark datasets (i.e. IEEE PHM 2012 challenge datasets and XJTU-SY datasets). The experimental results clearly demonstrate that the method proposed in this paper exhibits significant superiority in the field of RUL prediction compared to other models.