To address the issue of low diagnostic accuracy caused by the severe imbalance between normal and fault samples in marine diesel engine fault diagnosis, this paper proposes a fault diagnosis method that integrates Adaptive Semi-Unsupervised Weighted Oversampling (A-SUWO) with Probabilistic Neural Network (PNN). In this method, the A-SUWO algorithm is first employed to adaptively augment the minority-class fault samples. While generating new samples, it preserves the topological structure and distribution characteristics of the original data, effectively alleviating the imbalance between normal and fault categories and overcoming the drawback of traditional oversampling methods that tend to cause boundary overlapping among samples. Subsequently, the balanced samples are fed into a probabilistic neural network for training. Leveraging the advantages of PNN, such as high training efficiency, good classification stability, and suitability for real-time computation, a fault classification model for marine diesel engines is constructed to achieve high-precision identification of fault types. Using typical fault data for verification, the experimental results show that under severe imbalance conditions with a normal sample to fault sample ratio of about 1:10, the diagnostic accuracy of the proposed method has increased from 82% of the original imbalance model to over 98.5%, and the F1 score and G-mean have significantly improved synchronously. This indicates that the method can effectively identify the types of ship diesel engine faults, and has high diagnostic accuracy and good classification stability under sample imbalance conditions, providing an effective technical means for ship diesel engine fault diagnosis.

Marine diesel engines, A-SUWO, PNN, Fault diagnosis,Class imbalance