Fault Diagnosis for Marine Two-Stroke Diesel Engine Based on CEEMDAN-Swin Transformer Algorithm

The state information of marine diesel engines is strongly time-varying under the interference of multiple internal and external excitations. Fault features are difficult to extract adaptively. Fault diagnosis accuracy is not high. For this problem, a fault diagnosis method is proposed combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), signal-to-image conversion and Swin Transformer network. Firstly, vibration signals are decomposed by CEEMDAN. Arrange the intrinsic mode functions from low to high by center frequency. The time–frequency matrix can be obtained. Secondly, the time–frequency matrix is converted into a 2-D color image by signal-to-image conversion and pseudo-color coding. The color, shape and texture features of the 2-D images fully reflect the operating conditions of the diesel engine. Finally, use 2-D images as input to Swin Transformer to fully extract feature information for fault diagnosis. The experiments show that the proposed method can effectively extract the fault feature information, and the average fault diagnosis accuracy can reach 98.3%. Under the interference of different noises, it is verified that the model has certain adaptability. As a basic exploratory study, the proposed method can provide a reference for the crew to judge the faults.