Abstract: Objectives To avoid information loss during manual feature extraction and the unreliability of single-sensor diagnosis, a new method integrating multi-sensor deep learning models is proposed. Methods This method first regards the sensors at four different positions as the nodes of a graph structure. Then, a parallel four-channel one-dimensional convolutional neural network (1DCNN) is used to extract the deep features of the original data of each node respectively. Meanwhile, a dynamic adjacency matrix learning module is employed to learn the spatial correlations among the nodes, so as to obtain the adjacency matrix of each node. Subsequently, a graph convolutional neural network (GCN) is used to fuse the deep features of the four nodes. Finally, the fused features are dimensionally reduced and classified to obtain the diagnostic results. Results Experimental results show that the proposed multi-sensor fusion model achieves a fault diagnosis accuracy of over 98.82% on the marine screw pump dataset, and its performance is significantly better than that of other models.Conclusions After the deep features extracted by the parallel four-channel 1DCNN are fused by GCN, the diagnostic accuracy and fault tolerance are significantly improved.（