Aiming at the problems that the traditional marine main engine fault diagnosis model is difficult to update with real-time data, and the marine main engine has many monitoring points but few fault samples, a fault diagnosis method which can handle unbalanced data and update the model online is proposed.

First, principal component analysis (PCA) is used to reduce and extract the features of the monitoring samples to reduce the complexity of the training model, and the SMOTETomek technique is used to construct fault samples to balance the training set. Next, to solve the problem that the diagnosis model is difficult to update in real time, the online sequential extreme learning machine with regularization (OSRELM) model which combines regularization method and can update online is introduced. Finally, the feasibility of the OSRELM model is verified by taking the main engine fuel system as an example, and the effectiveness of the overall model is verified by ablation experiments with unbalanced marine main engine data.

The results show that the proposed method can improve the diagnostic accuracy by 29.73% on the basis of the original model.

The proposed method has higher diagnostic accuracy, a smaller fluctuation range and better stability than other similar algorithms. In the case of unbalanced data, it still has a strong ability to identify fault samples, providing valuable references for research on marine main engine fault diagnosis.