Abstract: Objectives Aiming at the problem that it is difficult to realize accurate fault diagnosis of ship propulsion shaft system under the constraint of edge resources, a new cloud-edge-end collaborative fault diagnosis method of ship propulsion shaft system is proposed. Methods Firstly, the monitoring signal data of the ship propulsion shafting is obtained at the device end. Then, wavelet transformation is performed on this data at the edge end to generate a time-frequency diagram, thereby constructing a spatially structured input. Secondly, a wide residual module is built on the cloud to efficiently capture the local texture and structural information features in the input image, and at the same time, a global attention mechanism module is utilized to obtain the dependency relationships between global data, enabling the iterative update of the model and the distribution of the model. Finally, fault diagnosis is implemented at the edge end, and the effectiveness of the proposed method is verified based on the simulation test bench of the ship propulsion shafting. Results The results show that the proposed method can achieve cloud model training and model distribution, as well as edge fault diagnosis, with an accuracy rate of 98.52%, which is at least 4.2% higher than other methods. Conclusions This method can provide a reference for the fault diagnosis of the ship propulsion shafting under the cloud-edge-end collaborative framework.